2013 Book PreventiveMethodsForCoastalPro

Tarmo Soomere
Ewald Quak Editors
Preventive Methods
for Coastal
Protection
Towards the Use of Ocean Dynamics
for Pollution Control
Preventive Methods for Coastal Protection
Tarmo Soomere r Ewald Quak
Editors
Preventive Methods
for Coastal
Protection
Towards the Use of Ocean Dynamics

for Pollution Control
Editors
Tarmo Soomere Ewald Quak
Wave Engineering Laboratory Wave Engineering Laboratory
Institute of Cybernetics Institute of Cybernetics
at Tallinn University of Technology at Tallinn University of Technology
Tallinn, Estonia Tallinn, Estonia
and
Estonian Academy of Sciences
Tallinn, Estonia
ISBN 978-3-319-00439-6 ISBN 978-3-319-00440-2 (eBook)

DOI 10.1007/978-3-319-00440-2
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2013942501
Mathematics Subject Classification: 62P12, 62P30, 65C35, 65M06, 65M75, 76B15, 76F25, 86-08,
86A05, 86A10, 86A22, 91B76
© Springer International Publishing Switzerland 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of pub-
lication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

“Though this be madness,
yet there is method in’t”
William Shakespeare
Foreword
Shipping as means of transportation has been among the first and most direct ways
for humans to benefit from the marine ecosystem services. With the increasing vari-
ability of industrialized production on land and intensifying globalization of the
markets, the direct environmental pressure arising from shipping is ever increasing.
Through history, the ways of utilizing the offshore maritime resources have diver-
sified. Therefore, prevention of harmful impacts from shipping and other offshore
activities has a key position in today’s policies concerning the marine environment
and sustainability.
In spite of navigation made complicated by narrow straits, shallow waters,
labyrinths of skerries and islands and winter ice, the Baltic Sea is one of the busiest
shipping regions of the world. It is projected that the transportation of goods through
the Baltic Sea will double by 2017. Of the average of 2,000 ships at sea each day,
200 are tankers carrying oil or other substances harmful for the environment. Dur-
ing the first decade of the 21st century there were more than 50 shipping accidents
causing pollution in the Baltic. In addition around 200 illegal spills are registered
each year (HELCOM 2010).
This book summarizes results of BalticWay, one of sixteen projects funded by
BONUS (www.bonusportal.org), the joint Baltic Sea research and development pro-
gramme, in 2009–2011. Its aim was to examine how to make shipping as well as
offshore and coastal engineering safer by using new knowledge on semi-persistent
surface currents and to avoid transport of environmentally harmful spills to vulner-
able coastal areas. The aim was to produce science based advice to decision making
in order to reduce the environmental risk.
The BalticWay research is exemplary in linking science and policy. It is founded
on basic oceanographic research, using the modelling approach to make forecasts
of material flows in a highly dynamic hydrographic environment that spatially and
temporally varies across a multiple of scales. At the same time, the project responds
to many objectives and goals of today’s policies in consideration of environment
and sustainability issues as well as to related demands on research, development
and innovation.
vii
viii Foreword
In today’s reality, the emphasis of using research as the base for policy decisions
is a cross-cutting issue and frequently featured in environmental and sustainability
policy documents. The demand for strengthening the data and knowledge basis of
policy making is repeated in documents ranging from the global, UN led initiatives
(IOC/UNESCO 2011), to modern EU regulations (EC 2008b) and to the Baltic Sea
Action Plan (HELCOM 2007) of the Baltic Marine Environment Protection Com-
mission. The issue is also highlighted as a horizontal action in the overarching EU
Strategy for the Baltic Sea Region (EC 2008a). All in all, there is an urgent need for
targeted research fostering sustainable development.
From the wider perspective of research, development and innovation policy,
the BalticWay project demonstrates the added value and power of multinational
cooperation. Only recently have the national and EU level research governance
structures been adapted so that they can efficiently support multinational coop-
eration. The cradle of the BONUS programme that provided funding for all na-
tionalities involved in the BalticWay and other BONUS projects that were im-
plemented in 2009–2011, lies in the concept of the European Research Area
(ec.europa.eu/research/era/index_en.htm). It aims at a unified research area open to
the world based on the internal market, in which researchers, scientific knowledge
and technology move and circulate freely. The EU is fostering a better use of re-
search for the benefit of the society by various funding instruments. In order to lead
the efforts of marine sciences towards this very goal, the EU has issued its strategy
for the marine and maritime research (EC 2008a).
The above and many other policies are indicating how we humans should learn
more about nature in order to adjust our activities so that our common planet remains
healthy for future generations. Yet bringing it all into practice needs research. Also,
it is not enough any more to be an excellent researcher but there is also a need to
be conscious about societal needs and how the results from researchers’ dedicated
work can be utilized for the benefit of the citizens.
BalticWay puts into practice this new way of thinking concerning the role of sci-
ence in society and in an excellent way shows what the role and place of science
should be. It shows how science that seems very theoretical—fine-scale modelling
of semi-persistent surface currents—is capable of producing results that potentially
have very important practical applications. The results and the management sug-
gestions made by the project team are now under real-life proofing and feasibility
validation and this way most likely on their way to contributing to actual solutions
that are needed for the good of the future of the Baltic Sea.
Executive Director, BONUS Kaisa Kononen
Science for a better future of the Baltic Sea region
Helsinki, Finland
Foreword ix
References
HELCOM (2010) Maritime activities in the Baltic Sea—an integrated thematic assessment on mar-
itime activities and response to the sea pollution in the Baltic Sea region. Baltic Sea environ-
ment proceedings No 123
IOC/UNESCO, IMO, FAO, UNDP (2011) A blueprint for ocean and coastal sustainability. Paris,
IOC/UNESCO
European Commission (2008a, September 3) A European strategy for marine and maritime re-
search. A coherent European Research Area framework in support of a sustainable use of
oceans and seas. COM(2008) 534 final. Retrieved December 13, 2012, from eur-lex.europa.eu/
LexUriServ/LexUriServ.do?uri=COM:2008:0534:FIN:EN:PDF
European Commission (2008b) Directive 2008/56/EC of the European Parliament and of the Coun-
cil of 17 June 2008 establishing a framework for community action in the field of marine
environmental policy
HELCOM (2007) The Baltic Sea Action Plan. A new environmental strategy for the Baltic Sea
region. Helsinki, Baltic Marine Environment Protection Commission
Preface
The aim of this book is to present for non-specialist researchers and experts a com-
prehensive overview of the background, key ideas, basic methods, implementation
details and a selection of solutions offered by the novel technology for the opti-
mization of the location of dangerous offshore activities in terms of environmental
criteria that was developed in the framework of the BONUS cooperation.
The first part of the collection introduces the basic principles of ocean modelling
and depicts the long way from the generic principles to the practical modelling of
oil spills and propagation of other adverse impacts. The second part focuses on the
techniques for solving the inverse problem of quantification of offshore areas with
respect to their ability to serve as a source of environmental danger to vulnerable
regions.
The book first presents an overview of the essential features of the hydrography,
the functioning and the key properties of the dynamics of circulation and currents
of the major implementation area—the Baltic Sea and the Gulf of Finland. The ba-
sic principles of Eulerian modelling of 3D currents are introduced in terms of both
generic concepts and specific implementation for the circulation modelling. This is
complemented by an overview of the differences in circulation model implementa-
tions for other potential target areas (such as the Mediterranean or the Black Sea)
and of the limitations of circulation models when it comes to operational modelling
of adverse impacts, e.g., oil spills, and specific features of operational and/or oil
spill modelling.
The basic element of the technology—Lagrangian trajectories of adverse impacts
in the marine environment—is presented mostly from the viewpoint of their cal-
culation from Eulerian information about marine currents. The relevant modelling
efforts are complemented with results from the very first long-term experiments
with Lagrangian drifters in the surface and subsurface layers of the Baltic Sea and
the Gulf of Finland. Statistical analysis of properties of large pools of data about
Lagrangian transport is applied for approximately solving the inverse problem of
pollution propagation. This approach is a feasible way for a systematic quantifica-
tion of the offshore areas in terms of the potential of current-induced transport of
adverse impacts (released in these areas) to vulnerable domains. The maps charac-
xi
xii Preface
terizing this potential are used for the construction of optimal fairways and the best
locations of potentially dangerous activities. Finally, applications of the developed
technology for solving certain problems of maritime spatial planning are discussed
together with new possibilities for express methods for testing the vulnerability of
a particular sea area, the applicability of the entire approach and direct estimates of
the potential benefit from the new fairway designs.
The chapters are written in a tutorial style; they are mostly self-contained and un-
derstandable for non-specialist researchers and students. The goal was to highlight
all key steps, methods, models and data sets necessary to be combined in order to
produce the practically usable technology and/or decision support system for a par-
ticular sea domain. The book was designed not only as a description and a manual of
the particular technology but also as a roadmap highlighting complicated technical
issues of ocean modelling for practical purposes that are frequently hidden deep in
model manuals.
To ensure the scientific quality of the contributions, each chapter was carefully
reviewed by two to four independent experts, with at least one from outside the
Baltic Sea science community. Special thanks go to all authors and referees, without
whom the making of this book would not have been possible. Finally, financial and
mental support by the BONUS initiative during all phases of the preparation of this
book is warmly acknowledged.
Tallinn, Estonia Tarmo Soomere
March 2013 Ewald Quak
The BalticWay Project
The ever increasing impact of the marine industry and the increase in risks asso-
ciated with potential oil pollution from ship traffic or oil platforms, calls for novel
methods for mitigating beforehand the impact of risks on vulnerable areas. The
goal of the BONUS+ BalticWay initiative was to develop methods for preventive
reduction of offshore environmental risks caused by the maritime industry that are
transported by various metocean drivers to valuable sea areas. The idea was to char-
acterize systematically the damaging potential of the different open sea areas in
terms of the potential transport to vulnerable regions if faced by an oil spill or other
pollution. This way, by placing maritime activities in the safest offshore areas, the
consequences of potential accidents can be minimized before they occur.
Traditionally risks of maritime industry are associated with possible accidents
that may lead to loss of lives or property, or to environmental pollution. The man-
agement of environmental risks has been much focussed on small areas around the
installation or the ship in question. However, the technological progress has led to a
new paradigm in the treatment of such risks that are no more located in small areas.
The amounts of harmful substances potentially released to the sea have increased
to a level that are of acute danger to the ecosystem or to society, even in seemingly
remote and safe locations hundreds of kilometres from the accident site.
The approach employed in this initiative was based on a smart use of the ex-
istence of semi-persistent patterns of transport which affect considerably pollution
propagation as well as drift of various items such as vessels without propulsion,
rescue boats or lost containers. These patterns make the probability of transport of
dangerous substances or undesired items from different open sea areas to vulnerable
sections highly variable. For certain areas this probability is relatively small and re-
or directing activities to these areas would appear to be feasible as well with very
limited additional costs.
The central goal was thus to quantify the offshore areas in terms of the potential
of adverse impacts released in a particular area to be transported by metocean drivers
to certain vulnerable regions. The gain is expressed in terms of the probability of
pollution transport to the vulnerable areas and the time pollution takes to reach
xiii
xiv The BalticWay Project
these areas. As a first approximation, coastal areas were used as a generic model
for valuable regions.
The extreme complexity of water motions in marine environment calls for the use
of nontraditional mathematical methods to identify the persistence, properties, and
potential effect of favourable patterns of transport, and to establish generic criteria
for the existence of areas of reduced risk in different sea regions from the massive
pool of the results of numerical simulations. Moreover, the desired quantification
of offshore domains involves solving an inverse problem of pollution propagation.
Such problems are frequently ill-posed and no universal method exists for solving
them. An approximate solution was obtained by means of statistical analysis of
a large number of particular solutions (Lagrangian trajectories) of the associated
direct problem of propagation of water or pollution parcels.
The resulting quantification of the offshore areas was applied for optimization of
ship routes in terms of minimizing the risk of coastal pollution. For example, in the
Gulf of Finland the gain from the use of the optimum fairway is about 40.
The BalticWay team believes that information derived using the developed tech-
nology is of vital importance for institutions responsible for environmental protec-
tion and maritime spatial planning. It is directly usable in the decision-making pro-
cess in a crisis situation, e.g., about different search-and-rescue issues. It can also
be used as a tool supporting decisions about how far the fairway for ships carrying
dangerous cargo should be located from any vulnerable areas facing the open ocean.
In brief, by studying transport of pollution driven by surface currents, wind and
waves in the Baltic Sea, BalticWay has developed a new method to determine how
shipping as well as offshore and coastal engineering activities can be made envi-
ronmentally safer. This research identifies areas that are safer to use, distinguishing
them from those where marine activities are better avoided.
Last but not least, while addressing such complicated, frequently transbound-
ary environmental problems caused by industrial activities, it is also important that
BalticWay was not only driven by the needs of scientific research alone; rather it was
an initiative of the scientific community reflecting the cooperative research needs
towards sustainable development and effective stewardship of the fragile sea ar-
eas. The core objective was to establish key components of a reliable, robust and
low-cost technology for the environmental management of shipping, offshore, and
coastal engineering activities. In essence this means integrating marine ecosystem
management with other needs of society, and linking scientists, stakeholders and
decision-makers in the process of elaborating a scientific base for political and eco-
nomical decisions.
Wave Engineering Laboratory Tarmo Soomere
Institute of Cybernetics at Tallinn University of Technology
Tallinn, Estonia
Estonian Academy of Sciences, Tallinn, Estonia
The BalticWay Project xv
PROJECT PARTNERS
Estonia
• Institute of Cybernetics at Tallinn University of Technology (Coordinating part-
ner)
• Laser Diagnostic Instruments, Tallinn
Denmark
• Danish Meteorological Institute, Copenhagen
Finland
• Finnish Environment Institute, Helsinki
Germany
• Institute for Coastal Research, HZG Geesthacht
• Leibniz Institute of Marine Sciences at the University of Kiel
Sweden
• Department of Meteorology, University of Stockholm
• Swedish Meteorological and Hydrological Institute, Norrköping
Project coordinator: Professor Tarmo Soomere, email: soomere@cs.ioc.ee

xvi The BalticWay Project
Oleg Andrejev (1941–2012)
One of the grand old men in the modelling of mesoscale dynamics in semi-enclosed
seas, Oleg Alexandrovitch Andrejev (1941–2012) worked almost 20 years from
1993 onwards in Stockholm University in Sweden, in the Finnish Institute of Ma-
rine Research, Helsinki and finally in the Finnish Environment Institute (SYKE)
in Helsinki. His scientific work was devoted to 3D hydrodynamic modelling of the
Baltic Sea, especially the Gulf of Finland. Already in 1989 Oleg Andrejev published
together with Alexander Sokolov the 3D model OAAS. This model was extensively
used for several fundamental and case studies in the Baltic Sea and the Gulf of
Finland.
The most well-known results stemming from this model are those related to mean
water circulation, water age and upwelling studies. In particular, the discovery of the
highly ordered subsurface flow in the Gulf of Finland to the north of the gulf axis
served as a trigger for the attempts to design environmentally safer fairways in the
Gulf of Finland that led to the underlying idea of the entire BalticWay cooperation.
In 2009–2011 Oleg Andrejev worked as senior scientist in the BalticWay project
where he developed very high-resolution model version for the Gulf of Finland and
substantially contributed to the success of the entire BalticWay project.
Kai Myrberg
Contents
1 Towards Mitigation of Environmental Risks . . . . . . . . . . . . . . 1

Tarmo Soomere
Part I Modelling the Underlying Dynamics

2 Topography, Hydrography, Circulation and Modelling of the Baltic
Sea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Kai Myrberg and Andreas Lehmann
3 Introduction to Computational Fluid Dynamics and Ocean
Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Tomas Torsvik
4 Studying the Baltic Sea Circulation with Eulerian Tracers . . . . . . 101
H.E. Markus Meier and Anders Höglund
5 European Semi-enclosed Seas: Basic Physical Processes and Their
Numerical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Emil V. Stanev and Xi Lu
6 The Gulf of Finland, Its Hydrography and Circulation Dynamics . . 181
Kai Myrberg and Tarmo Soomere
Part II Lagrangian Dynamics and Inverse Problems

7 TRACMASS—A Lagrangian Trajectory Model . . . . . . . . . . . . 225
Kristofer Döös, Joakim Kjellsson, and Bror Jönsson
8 Evaluation and Tuning of Model Trajectories and Spreading Rates
in the Baltic Sea Using Surface Drifter Observations . . . . . . . . . 251
Joakim Kjellsson, Kristofer Döös, and Tarmo Soomere
9 Statistics of Lagrangian Transport Reveals Hidden Features of
Velocity Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
Tarmo Soomere
xvii
xviii Contents
10 Applications of the Inverse Problem of Pollution Propagation . . . . 319

Tarmo Soomere
11 Applications of an Oil Drift and Fate Model for Fairway Design . . . 367
Jens Murawski and Jacob Woge Nielsen
Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
Chapter 1
Towards Mitigation of Environmental Risks
Tarmo Soomere
Abstract This introductory chapter describes the background of a preventive tech-

nology for environmental management of remote risks associated with potentially
dangerous offshore activities. The notion of environmental risk caused by shipping
and offshore activities is introduced as the product of the probability of an accident
and the cost of its consequences. A new paradigm of remote environmental risks of
contemporary ship traffic is presented, with examples of the remote impact (such as
oil spill or wake waves) and a short outline of the challenges for ship routing based
on environmental criteria. Several pioneering solutions towards decreasing both the
local and the remote impact of ship traffic are discussed. The concept of the quan-
tification of offshore and remote areas in terms of their potential as starting points of
pollution that after some time hits vulnerable areas is introduced, and related chal-
lenges are discussed. While the impact of wind and waves on pollution transport is
relatively well understood, the largest challenge is to take into account the presence
of semi-persistent patterns of currents and quantify their role in pollution propa-
gation. An approximate solution to this inverse problem of pollution propagation
can be obtained using statistical analysis of a large pool of trajectories of selected
particles. There exist several meaningful ways of such a quantification of offshore
domains, for example, in terms of the probability of hitting vulnerable areas or of
the time it takes for the pollution to reach such areas. An engineering solution can be
developed using any of these quantifications. A prototype technology for the selec-
tion of optimum fairways combines four ingredients: a high-resolution circulation
model, a method for calculating Lagrangian transport, a measure of environmental
risk and a method for selecting an optimum sailing line.
T. Soomere (B)
Wave Engineering Laboratory, Institute of Cybernetics at Tallinn University of Technology,
Akadeemia tee 21, 12618 Tallinn, Estonia
e-mail: soomere@cs.ioc.ee
T. Soomere
Estonian Academy of Sciences, Kohtu 6, 10130 Tallinn, Estonia
T. Soomere, E. Quak (eds.), Preventive Methods for Coastal Protection, 1

DOI 10.1007/978-3-319-00440-2_1,
2 T. Soomere
1.1 Introduction
It has almost become a cliché that the marine environment is suffering from ever
increasing anthropogenic pressure. Regrettably, it is a fact that this pressure is much
more extensive and multi-faceted than we imagined. Historically, anthropogenic
pressure had been concentrated on the sea surface and in coastal regions, and mostly
limited to shipping and fishing industry. With the development of agriculture and
industry, it has rapidly spread over semi-sheltered and shelf seas, and today it has
reached the most remote parts of the World Ocean, even down to the bottom sed-
iments in its deepest chasms. While much of the impact is absorbed by the huge
water volume, the World Ocean is believed to severely react to the human activity,
for instance, through worldwide sea level rise and acidification of its water masses.
The sea surface plays a particular role in both the functioning of the sea and the
use of the marine environment. It is almost impossible to imagine contemporary
society without the services provided by this interface between air and water. As
one of many benefits, the sea surface offers the possibility for major transportation.
Shipping has been for a long time and evidently will also remain in the future an ef-
ficient and environmentally friendly mode of transportation. Today it is responsible
for over 80 % of the world trade by volume (Boissonnas et al. 2002). The ever-
increasing density of long-distance ship traffic, the modal shift of local rail and road
transport within the European Union to a waterborne one (Banister 2007), and the
growing size of ocean-going ships all suggest a further increase in ship transport.
This perspective contains also several perils. The largest direct danger to the
marine environment and ecosystems is the rapid growth of oil transportation in the
vicinity of valuable or vulnerable sea area. The concept of certain marine areas
being more valuable than others has obviously been used for millennia in the context
of fishing industry. Traditional fishing cultures around the world have engaged in
closure-based practices that have functioned to protect marine resources. Perhaps
the best-known example of this is the tabu or kapu concept established by Pacific
island cultures many centuries ago (Bowden-Kerby 2001).
The contemporary concept of (possibly other) marine areas being more vulner-
able or important, and the need for specific measures for their protection, proba-
bly originates from the mid-1800s. The pioneers were people from different parts
of the New World. The oldest marine protected area, Royal National Park, New
South Wales, Australia, was designated in 1879 (Anonymous 2002). Soon after-
wards North America (Breton National Wildlife Refuge, Louisiana, USA, 1904)
and Asia (Matang Forest Reserve, State of Perak, Malaysia, 1906) followed. In this
respect, Europe has the last place, with the Underwater Reserve of Monaco (1976)
(Anonymous 2002).
The listed examples and the majority of marine protected areas are actually
coastal or nearshore waters. The importance of protection of certain offshore areas is
less obvious. It has been legally recognized only two decades ago when the Interna-
tional Maritime Organization (IMO) adopted the Assembly Resolution1 720(17) on
1 http://www.imo.org/blast/blastDataHelper.asp?data_id=22581&filename=A720(17).pdf.
1 Towards Mitigation of Environmental Risks 3
November 6, 1991 about designating special zones in the sea and identifying partic-
ularly sensitive sea areas (PSSAs). This was an essential modification of the princi-
ple of the freedom of navigation legitimized and protected by the IMO towards mit-
igating risks to the marine environment. This resolution provided a new paradigm
to coastal states not only to protect specific areas within their territorial waters but
also to extend such protection to further offshore (Lefebvre-Chalain 2007). It has
marked the emergence of a new cooperation between maritime activities and the
protection of fragile marine ecosystems (Lefebvre-Chalain 2007). In a later resolu-
tion2 982(24), the IMO defines PSSAs as areas with “ecological, socio-economic, or
scientific” importance and, consequently, recognizes these areas as needing special
protection (Kachel 2008). With the increase in the understanding of the function-
ing of seas and oceans, the importance of different domains such as ecologically
sensitive locations (Burgherr 2007) or large marine ecosystems (Sherman and Duda
1999) has been highlighted. We denote all these locations as vulnerable areas.
More generally, the danger from potential pollution from the ship traffic is trans-
boundary as the propagation of pollution in the marine environment of course ig-
nores any legal division of the sea between different authorities. It is of great concern
for both semi-sheltered seas and along the oceanic coasts (Eide et al. 2007). Even
the most remote parts of the world cannot be considered as safe in this respect. Ship
traffic in the arctic regions—which contain large food resources but also are particu-
larly sensitive to environmental pollution—is increasing. Traces of pollution easily
reach also Antarctic waters. Thus, reliable design and improved safety are critical
to the acceptance of ship operations, especially in particularly vulnerable sea areas,
and have a strong environmental benefit for all.
The largest single oil spills released to the marine environment stem in fact from
offshore oil platforms. Spills from underwater pipelines also form a notable part of
the oil releases (Burgherr 2007). Underwater mining and dumping of material into
the sea may also lead to the deterioration of the local ecosystem. The sea bottom
at some places resembles a scrapyard and may contain dangerous concentrations of
chemicals used in the recent past (Verta et al. 2007). In many aspects the buffering
capacities of the World Ocean may be exhausted. Keeping it sustainable and rea-
sonably managing dangers associated with its exploitation is, therefore, one of the
major challenges of this century.
1.2 The Notion of Risk, Probability and Consequences

Contemporary scientific and engineering literature use the term ‘risk’ in many dif-
ferent meanings. To avoid ambiguity and possible confusion of risk with uncer-
tainty, in what follows we shall rely on the classical concept of risk as a quantita-
tive measure of a potential adverse event. The notion of risk R in this framework
(Risk Management Standard 2002) is expressed as the product of the probability Pa
2 http://www.cbd.int/doc/meetings/mar/ebsa-sio-01/other/ebsa-sio-01-imo-en.pdf.
4 T. Soomere
(of failure or accident) and the properly quantified severity Ca (cost or consequence)
of this disaster:
R = Pa × Ca . (1.1)
This notion of risk, called the risk equation below, is commonly used in natural
sciences, engineering and industry. The dawn of this concept stems from coastal en-
gineering and management efforts during the planning of the immense Delta Works
infrastructure. After the devastating storm surge in February 1953 that killed 1,835
people in the Netherlands, the Dutch government triggered the construction of an
impressive bulwark megastructure, the Delta Works, to reduce the country’s flood
vulnerabilities.
The most visionary aspect of the Delta Works in our context is the statistical
approach that guided the designs. The novelty of the Dutch decision was to include
both the properties of storms and the economics of the Netherlands. With the help of
renowned mathematician David van Dantzig, the safety levels were calculated using
the risk equation (1.1) to produce a sequence of economically rational public-safety
decisions. This kind of risk analysis is common today in many fields of science and
engineering but back in the 1950s, accounting for the projected cost of damage was
novel. Dutch law now requires this principle to be used to determine the strength of
flood defenses throughout the country (Wolman 2008).
While the majority of the research in marine traffic risk has been focused on
the probability (Fowler and Sørgård 2000; Soares and Teixeira 2001; Goerlandt and
Kujala 2011, to name a few), the use of this notion is gradually increasing in the
analysis and modelling of such risks (Montewka et al. 2011).
Both factors on the right-hand side of the risk equation have been massively
addressed in the scientific and technical literature. For example, the use of con-
temporary navigation devices and detailed charts, the overall improvement of the
construction of ships to withstand the forces of nature, the implementation of real-
time control through vessel traffic systems, etc., have considerably decreased the
probability Pa of ship accidents. In spite of all these developments, however, major
offshore accidents continue to happen with some frequency. Although our under-
standing of loads that may occur during severe storms or accompany an ice attack
(Kujala and Arughadhoss 2012) is continuously improving, it is economically un-
feasible to design all the ships to fully resist such forces on all occasions. Also, a
ship or offshore structure is such a complicated system that even minor defects in
its production or errors in its design may result in a significant failure (e.g., Collins
et al. 1997). Moreover, a human error or misbehaviour cannot be totally excluded.
As there is no way to stop ship traffic or to avoid extensive use of the marine
space for other purposes, society has to cope with the associated dangers. This im-
plies that, along with methods to properly quantify (Goerlandt and Kujala 2011) or
decrease the probability of misfortunes, an equally important task is to develop ways
for the mitigation of the potential damage that may occur. This can be achieved, for
example, by specific improvements in ship construction so that minor groundings
and collisions will not result in pollution. The key development here has been the
introduction of double-hull tankers (Paik 2003). This advance has considerably re-
duced the number of oil spills due to tanker traffic (Glen 2010).
The cost of damage can be defined in a variety of ways. Three categories of the
consequences of an accident are usually considered:
• safety of people (loss of life, limb and health);
• economic aspects (damage to the ship or structure, equipment and cargo, etc.);
• environmental damage (damage to the ecosystem, coastal pollution, deterioration
of fish stocks, influence on the tourism industry, decrease in the prices for coastal
property, etc.).
The methods used for the estimates of the resulting damage and for the quantifica-
tion of the risk R in the risk equation vary largely. The mathematical methods used
for minimizing the cost function are usually the same for all approaches. Conse-
quently, a solution to the problem of minimizing certain costs can be generalized in
a straightforward manner to any other cost category. For this reason, we shall only
address the environmental damage in this book. In other words, the ‘severity’ or
cost Ca of the potential consequences only accounts for the damage to the environ-
ment. For the same reason, we mostly perform the analysis from the viewpoint of
minimizing environmental damage caused directly or indirectly by ship traffic.
While the ‘price list’ for several economic costs is well-defined, there is con-
siderable freedom in the calculation of environmental losses. On the one hand, this
feature makes the estimates of environmental risks more complicated and ambigu-
ous. On the other hand, it provides substantial flexibility for the implementation of
different cost functions and thus for the pool of possible scenarios and estimates.
1.3 Changing Paradigms of Local and Remote Risks
The ever increasing density of marine transportation combined with a limited num-
ber of possible fairways results in an extreme concentration of ship traffic in certain
sea areas. Dangers are then traditionally associated with possible accidents (ship col-
lisions or groundings, technical and navigation problems caused by severe weather
or human errors, etc.) that may lead to loss of lives or property, or to environmen-
tal pollution. The basic assumption in the treatment of shipping-related risks has
typically been that they are localized within a small area around the ship. Indeed a
ship accident occurs at a certain point and the possible contamination is transported
relatively slowly due to winds and currents.
Recently, the overall progress in shipping technology and an increasing aware-
ness about the limits of the sustainability of marine ecosystems have led to a radical
change to this paradigm. The key alteration is that many by-products of the ship
traffic are actually not confined to a small vicinity of the vessel and even not lo-
cated in small areas. The relevant list includes (but is not limited to) the growth
of exhaust emissions (Hobbs et al. 2000) capable of creating substantial changes
in the atmosphere at a height of many hundreds of meters above the sea surface
(Durkee et al. 2000), a significant increase in external noise (Boyd et al. 2008) and
waves generated by large high-speed ships in certain areas (Wood 2000; Parnell and
6 T. Soomere
Kofoed-Hansen 2001; PIANC 2003; Soomere 2005). Some of these by-products

not only travel much faster than the vessels (e.g., the ship-generated noise) but also
constitute a part of global problems (e.g., exhaust emissions). Many of these risks
are not specific to ship traffic. Different potentially dangerous substances may be
released into the marine environment during various offshore and coastal activities.
This may occur during the construction, operation or dismantling of oil platforms,
breakwaters, quays, pipelines or cables. Even operations at open quays or the treat-
ment or storage of waste in the reach of large waves or high storm surges are not
completely safe in this respect.
1.4 Managing Propagation of Contaminants and Oil Spills
Typically, the potentially dangerous by-products are represented by several non-

local agents carrying acute danger over substantial distances and persisting for a
long time. A classical example of such a threat to the environment is a large oil
spill. Although it usually propagates much more slowly than the ship, under certain
conditions it may result in extensive remote impact. The accompanying oil pollution
is definitely one of the most serious consequences of coastal and offshore accidents.
To the general public, the sight of an oil spillage in coastal waters, particularly when
effecting wildlife, is the most dramatic example of environmental pollution. Partic-
ularly hazardous is the situation if a major fairway is located close to or crosses
vulnerable areas. A hit by an oil spill for certain regions (e.g., wintering areas of
globally endangered species, Elken et al. 2001) is not only economically expensive
but also environmentally unacceptable. Therefore, a key challenge is to arrange ship
traffic (and offshore and nearshore activities) so that the chances for the pollution of
vulnerable areas are minimized if an accident happens somewhere.
Although the relative numbers of dangerous oil spills have been decreasing over
the last decades (Burgherr 2007), the ever increasing capacity to unearth, handle
and transport large quantities of oil extends the radius of the potential impact of
ship traffic. The damaging potential of oil and other adverse impacts, accidentally
or intentionally released to the sea, has today increased to a level that may pose
acute dangers to the ecosystem or society in seemingly remote and perfectly safe
locations. As demonstrated by the Prestige or Deepwater Horizon events, an oil
spill may easily spread over hundreds and even thousands of kilometers from its
source (Fig. 1.1). This impact, carried to vulnerable sites jointly by wind, waves and
currents, may create extremely large risks for some regions (especially nearshore
regions and other vulnerable areas) at quite large distances. The long lifetime of
several adverse impacts in the marine environment substantially increases the level
of risk for remote areas and extends it over a substantial time interval.
The commonly used approach to manage potential pollution at sea is to develop
quick remedial action plans for the event of an accident (e.g., Keramitsoglou et al.
2003; Kostianoy et al. 2008, among others). Another, rapidly developing approach
is the preventive maritime planning strategy; for instance, the optimization of the
Fig. 1.1 The Prestige oil

spill in Spain demonstrated
how far oil can spread on the
sea surface. The vessel sank
about 100 miles offshore.
Courtesy of ESA
shipping routes (Schwehr and McGillivary 2007), or proper designation of marine

protected areas or particularly sensitive areas, and possible policies and regulations
(Ko and Chang 2010; Hassler 2011; Rusli 2012) to account for the effect that a
pollution accident would incur before it actually happens.
The challenge addressed in this book is how to minimize the follow-up damage to
the environment if an accident were to happen at some (previously undefined) point
of the journey. This is a standard procedure in the transport of dangerous goods over
both rail and road (Kara and Verter 2004). The relevant missions are commonly
limited to a specific itinerary such that sensitive (e.g., urban) areas and/or driving at
specific time intervals are systematically avoided. This basically static approach can
only be conditionally applied to the shipping and offshore industry. As described
above, a major marine accident is usually associated with a release of oil that indi-
rectly affects an area at a long distance from the accident site. The endangered area
may thus extend to tens, and even hundreds of kilometres from the accident site.3
Therefore restrictions for ship traffic to avoid e.g., spawning areas do not guarantee
that the oil pollution will not actually reach these areas.
The central goal of the work described in this book was to demonstrate that it
is possible and feasible to treat the potential influence of various offshore activities
(ship traffic, oil platforms, pipelines, mining, dumping, etc.) so that such non-local
effects are systematically accounted for. The key idea is, as in the rail or road trans-
3 There is, however, a clear parallel with the transport of adverse impacts by air flow after the release
of some gaseous substances in mainland accidents. These aspects have been studied in detail in the
context of acid rains and accidents in nuclear power stations (e.g., Vach and Duong 2011).
8 T. Soomere
port of dangerous goods, to minimize the severity (impact) of the potential disaster
beforehand, by means of methodically optimizing the possible accident site (loca-
tion of the offshore activities or, more generally, the sailing line). The problem is
relatively simple for stationary objects such as oil platforms. In the case of fairways
the challenge is to provide a non-stationary (ideally, real-time) smart choice of a
vessel’s trajectory that minimizes environmental risks.
An adverse impact from harmful human activity may be released at any depth.
While oil spills from ship traffic released to the surface layer occur relatively fre-
quently, much larger quantities of oil may be released into near-bottom waters after
failures of offshore oil platforms (Burgherr 2007). The failure of a nuclear sub-
marine may pollute any layer in the sea. Although there exist successful attempts
to follow the three-dimensional (3D) propagation of oil spills (Chang et al. 2011),
establishing a systematic link between the locations of release and the regions of
impact for substances in a 3D current field is far too complicated even for relatively
small sea domains. For this reason, the approach presented in this book is limited to
the analysis of substances released to and propagating within the surface layer.
Although this framework, strictly speaking, is only directly applicable for per-
sistent substances that are dissolved in the near-surface water in strongly stratified
environments under calm conditions or below ice cover, and when the contaminants
(e.g., dissolved radioactive substances) largely remain in the uppermost layer and
are carried exclusively by surface currents (e.g., Periáñez 2004), it is still a suitable
approximation for the vast majority of releases of potentially dangerous substances
to the marine environment. Most importantly, it adequately reproduces a large frac-
tion of the fate of the lighter components of oil that are usually thought to pose a
large danger to valuable areas.
In order to keep the analysis and presentation as transparent as possible, a greatly
simplified description of the (cost of the) consequences will be used. The ideas and
the implementation steps of the technology are described only in terms of one spe-
cific adverse by-product of marine surface transport and some offshore activities—
oil pollution and its potential hits on a vulnerable or high-cost coastal area. The rel-
evant risk categories are widely known and the benefits of the proposed new tech-
nology can be compared with the existing concepts and results.
1.5 Ship Routing Systems
Regulations concerning ship traffic and various offshore activities have already de-
creased the probability of marine accidents considerably. This development has,
however, had virtually no effect on their severity. The basic risk for the transport of
oil products is a coastal crash and the related pollution (Iakovou et al. 1999). Most
accidents, both with small and large consequences, happen in coastal waters (Romer
et al. 1994). Such events are, however, usually only the final stage of a sequence of
incidents that can be traced back to a failure, damage or misbehaviour within a cer-
tain section of the offshore sailing line. The key idea is to select this line so that a
potential accident would have minimum consequences. The end result is therefore a
variation of the ship routing.
A classical example of ship routing practice is the calculation of an optimal route
for crossing the North Atlantic (Calvert et al. 1991) under a particular distribution of
the sea state. Optimizing of routes with respect to persistent current patterns (Lo and
McCord 1995; McCord et al. 1999) is attempted frequently. The benefit of smart
ship routing is normally formulated in economical categories or from the safety
viewpoint.4
Methods for the optimization of fairways have been addressed in numerous stud-
ies over the last decades (Calvert et al. 1991; McCord et al. 1999; Iakovou et al.
1999, among others). An overview of the state of the art of ship routing is given
in Christiansen et al. (2004). Efforts have mostly been directed towards developing
general algorithms for decision support systems. Their contemporary realizations
are based on the optimization of a large set of parameters such as the time of transit,
fuel consumption, exhaust emissions, forces on the ship and ship motions, and last
but not least, the overall cost of the total logistic chain. A description of selected
recent efforts is provided by the SEAROUTES5 project. The outcome is a group of
alternatives of optimized routes based on the forecast of the sea state (waves, cur-
rents, winds, in particular), assimilation of hydrometeorological (in what follows
called metocean) data, and a specific optimization method. The methods and tech-
nology of (both long-term and operational) optimum ship routing systems, given
adequate metocean data and a relevant definition of risk, have been successfully
implemented in many parts of the world. They have, inter alia, also substantially
contributed to reducing the risk of collisions and groundings, and thus to an overall
decrease in the risk of accidents.
1.6 Environmental Management of Ship Routes
The problem of ship routing and/or fairway design is highly asymmetric. On the one
hand, the benefit from the smart use of metocean conditions is quite limited. For ex-
ample, the potential annual cost savings of the world fleet that can be achieved by
smart routing through exploiting ocean currents was estimated to be on the order
of 70 million USD/yr already two decades ago (Lo et al. 1991). On the other hand,
the adverse consequences from sea transport can be quite large, for instance, in case
of malfunction due to severe weather. Kite-Powell (2011) estimated that average
expected annual losses to container shipping in the absence of good information
about extratropical storm conditions would be on the order of 250 million USD/yr
in the North Pacific and 120 million USD/yr in the North Atlantic. On top of that,
4 Note that in industrial shipping, ship routing is frequently interpreted in the narrower context of
scheduling, where the objective is to minimize the cost of a fixed fleet of ships.
5 http://www.tu-berlin.de/fb10/MAT/searoutes/.
10 T. Soomere
estimated average expected annual losses to bulk shipping operations from extrat-
ropical storm exposure in these regions is on the order of 150 million USD/yr.
This asymmetry leaves quite a limited space for an environmentally friendly ap-
proach. The predominance of economic and safety aspects is clearly expressed in
recent applications of the potential use of currents. The driving force is the economic
benefit and optimum ship routing is defined as “the selection of an optimum track
for a transoceanic crossing by the application of long-range predictions of wind,
waves and currents to the knowledge of how the routed vessel reacts to these vari-
ables” (Lee et al. 2002). Even when the term ‘environmental routing’ is used, it is
frequently interpreted as a variation of methods of accounting for the potential ef-
fects of metocean factors such as ocean currents and heavy weather. In many cases
they are simply considered as potential sources of economic risk that may influ-
ence cruising speed and fuel consumption, and are usually attached to “other related
shipping problems” (Christiansen et al. 2004). Recent examples of mathematical re-
search on operational weather routing (Azaron and Kianfar 2003) use methods from
stochastic dynamic programming to find the best path under given environmental
conditions; this approach, however, leads to a fast increase in the complexity of the
algorithm.
The importance of environmentally friendly shipping was widely recognized in
the mid-1990s. It became evident that in addition to ship types and cargoes, also
the geography of shipping routes is an intrinsic major component of the environ-
mental management of shipping, in particular concerning risk and environmental
impact (Lo and McCord 1995; Judson 1997). There exist, however, very few studies
in the international scientific literature (Eide et al. 2007) of a fairway choice based
on environmental arguments or aimed at the identification of the potential influence
of existing fairways in terms of environmental criteria (Schwehr and McGillivary
2007). An early attempt to address this problem was made in terms of the environ-
mental impact assessments (Smith 1995). Since then, mainly the issue of whales
(who are potentially hit by ships in the North Atlantic and in the Mediterranean
Sea) has been considered in-depth in regard to environment/ecosystem-based risk
categories (Ward-Geiger et al. 2005; Panigada et al. 2006). An urgent need “for a
comprehensive, basin-wide conservation strategy . . . to re-locate ferry routes to ar-
eas of lower cetacean density” was recognized in the Mediterranean (Panigada et al.
2006).
To reduce the threat to whales, a mandatory ship reporting system was estab-
lished in July 1999 in the waters off the north-eastern and the south-eastern United
States. Reporting ships received an automated message indicating precautionary
steps to be taken to avoid hitting whales. One of the results was the identification of
the most frequently used ship tracks. Such a ‘portrait’ of ship traffic was then used
to develop measures to reduce the threat of ship strikes. A substantial by-product
was the identification of a systematic spatial variation in the pattern of the presence
of whales in different open sea regions (Fig. 1.2). An analysis of these two patterns
led to a breakthrough: the decision to relocate the fairway entering Boston Harbour
to the area of minimum probability of the presence of whales (Stokstad 2009). Do-
ing so increased the sailing time by only about 15 minutes and did not involve any
noticeable infrastructure costs.
Fig. 1.2 Relocation of the fairway entering Boston Harbour. From Stokstad (2009) with kind
permission of Mike Thompson, Dave Wiley and Richard Merrick. © M. Thompson, D. Wiley,
R. Merrick
The basic lesson from this pioneering initiative is that such decisions are feasi-
ble and justified when the sea area in question contains some (usually concealed)
inhomogeneous features or patterns. In this case the key pattern was revealed by
long-term observations of whales. The decision was then based on the existence
of a favourable feature—a clear minimum of the (pointwise) 2D probability distri-
bution of the whale population. The outcome was a considerable decrease in the
probability of damage to the environment.
1.7 Minimizing the Remote Impact of Ship Traffic
The relocation of the fairway entering Boston Harbour led, in essence, to a decrease
in the probability Pa in the risk equation (1.1) and did not affect the cost of conse-
quences of a particular hit. This case, however, exemplifies one of the central ideas
of the technology described in this book. Namely, practical use can be made of the
presence of a favourable feature (e.g., a well-defined minimum or maximum) in a
2D map characterizing certain properties of an offshore region. While the map in
Fig. 1.2 quantifies the local danger, our goal is to produce maps that quantify the
12 T. Soomere
remote danger to some other domain. This viewpoint reflects the above-discussed
paradigm that marine anthropogenic activities with high risks usually have a sub-
stantial remote influence. The situation is again essentially asymmetric: the risks
caused by metocean factors or by the coasts or obstacles affect the ship only at a
fixed point but the outcome of an accident may have much wider impact. Therefore,
a consistent implementation of methods and technology for an optimum choice of
the fairway in terms of environmental risks stemming from a ship has to cope with
this paradigm.
The first investigations systematically accounting for this aspect were probably
carried out for the analysis and mitigation of hazards associated with wake waves of
high-speed vessels. It is well known that large vessels sailing at near-critical speeds
in relatively shallow water may produce high wave groups that remain compact for
a long time and may severely impact coastal regions located many kilometres from
the ship lane (Parnell and Kofoed-Hansen 2001; Soomere 2005; Parnell et al. 2007).
The U.S. Patents Nos. 6171021 and 7082355 provide technically and economically
feasible solutions for avoiding the remote influence of dangerous waves.6
As mentioned above, a complicated set of restrictions and regulations exist for
the rail and road transport of hazardous goods. There have been very few attempts
so far to address these questions in marine conditions. Even in the most compre-
hensive overviews (Christiansen et al. 2004) this problem is simply mentioned but
no perspectives for this field are outlined. It is addressed at a strategic level in the
framework of transportation of oil products in the Gulf of Mexico (Iakovou et al.
1999). The solution is sought as a selection of paths that minimize a weighted sum
of transport costs and expected risk costs (Fagerholt et al. 2000). The basic risk here
is again a coastal crash, the coastline represents the obstacles and the potential ben-
efit is formulated in terms of economic categories and operational planning of ship
schedules.
An early attempt to address the more general problem of the assessment of risk
levels along fairways has been made in Judson (1997). The objective was to provide
timely risk assessment information to a mariner or decision-maker in a system capa-
ble of integration of, among other factors, also historical and metocean information
such as accident location, frequency and type, ice, wind, visibility, environmental
sensitivity and other factors. The resulting collision risk was assessed for each track
in a route plan by applying a predictive accident model patterned after the naviga-
tion and collision avoidance process. In essence, this method again addresses the
probability of the accident in the equation of risk.
6 The solutions are targeted at smart fairway design in the context of contemporary fast ferry traffic,
where as a by-product long and high waves are created by ships moving with a specific (critical)
speed, creating a large problem in shallow areas (see Soomere 2009 and references therein). U.S.
Pat. No. 6171021 proposed to design an underwater ramp (equivalently, the fairway crossing such
ramp) so that the water depth abruptly changes. By doing so the ship avoids as much as possible
sailing with critical speed and generating dangerous waves. U.S. Pat. No. 7082355 consists of
a combination of devices that warn the captain when the ship is entering the critical regime in
relatively shallow water. A practical use of this device consists in adjusting the sailing line so that
the danger to the environment caused by vessel waves is minimized.
The first operational model targeted at a preventive reduction of environmental

risks is the system of dynamical relocation of tugboats along the Norwegian Atlantic
coast (Eide et al. 2007). Its basic function is the forecast of risks caused by the po-
tential loss of propulsion or steering of vessels along the fairway, and the subsequent
drift and grounding.
1.8 Site-Specific Cost of the Consequences

The developments of the last decades in physical oceanography and biological ma-
rine sciences have not only demonstrated the richness of the internal structure of
different sea properties but have also led to establishing a number of marine pro-
tected areas. This practice reflects the natural assumption that different sea areas
may have different value and/or vulnerability.
There exists a natural division of confined seas into open sea and coastal areas.
The generic problem in shipping consists in avoiding a coastal crash (or grounding)
of vessels. The coastal areas are generally the major life reproduction areas and thus
play a major role in restoring the fish stock. The map of marine protected areas of
the Baltic Sea (Fig. 1.3) signifies that they are mostly located in the nearshore, with
a few exceptions to the south and north of the island of Gotland and in the northern
part of the Pomeranian Bay. While marine pollution on the open sea kills a number
of creatures, coastal pollution additionally destroys the possibility of birth of new
life. Combating oil pollution on the open sea is generally much more effective and
less expensive than in shallow coastal areas; thus keeping the oil spill or the damaged
ship in the open sea causes the least damage.
This division is also obvious from the viewpoint of oil pollution or the release
of other dangerous substances to the marine environment: hits on a major spawn-
ing area, a nesting or wintering region of birds, a beach of a national park, a large
tourist industry location, or important infrastructures operating at some sections of
the coast or offshore are highly undesirable. For the listed reasons the nearshore
areas are used in this book as a generic example of ‘high-value’ regions. They are
better to be avoided by ships and their hit by oil pollution is highly undesirable.
Such an approximation allows presenting the basic ideas in a simple and transparent
framework. In essence, its use causes no loss of generality, because the developed
method is applicable for a wide range of descriptions of consequences as well as in
many other marine-transport-related problems.
An obvious solution for waterborne transport for the protection of coastal and
nearshore areas consists in a shift of the fairway to the open sea, far enough from
any valuable or vulnerable region. This is only possible for very large water bodies
such as the Atlantic Ocean and cannot be used in smaller seas such as the Black Sea,
the Baltic Sea, or the Mediterranean Sea.
The above example of whales, however, vividly demonstrates that some offshore
sea areas are clearly more valuable than others. This line of thought is reflected
in terms of legal formulations by the International Marine Organization concept of
particularly sensitive sea areas (Lefebvre-Chalain 2007).
14 T. Soomere
Fig. 1.3 The Baltic Sea marine protected areas (BSPA). Courtesy of HELCOM. From
http://www.helcom.fi/environment2/biodiv/en_GB/bspas/
We further extend this observation by noting that in the context of remote impact
of some accidents, the cost of the consequences (e.g., of a ship accident) and accom-
panying oil pollution also depends at which offshore point the oil has been released.
This cost is realized later, when it becomes evident whether or not the pollution
actually hits some valuable or vulnerable area. Therefore, tagging some sea areas
(e.g., the nearshore) with a price label naturally yields an associated distribution of
costs of otherwise similar accidents but occurring at different locations. The cost of
an accident will then depend on the point where it happens. In case of oil pollution,
this point is linked with potential vulnerable spots by oil drift properties; in case of
exciting high wake waves by rules governing the wave propagation (Torsvik et al.
2009).
If vulnerable spots are selected based on environmental criteria, a smart use of
the distribution in question provides a natural way to mitigate or minimize envi-
ronmental damage. Doing so is feasible through a systematic quantification of the
relevant costs, or, equivalently, through systematically accounting for the transport
of pollution from the release (or accident) site to vulnerable areas. The challenge,
therefore, is the quantification of the offshore areas in terms of their potential for the
transport of adverse impacts from these areas to valuable ones. This quantification is
the corner stone of the method described in this book. It makes it possible to find op-
timum solutions for semi-enclosed sea areas and also partially answers the question
how far from the open ocean coast an economically feasible and environmentally
friendly sailing line should be located.
1.9 Separation of the Impact of Wind, Waves and Currents

The propagation of different substances within the water column is governed by the
3D system of currents. The substances in the surface layer are additionally affected
by wind and waves. The understanding of the exact role and contribution of each of
these factors still has large gaps in both scientific and operational models today and
definitely needs an independent study. While the wind and wave conditions in the
area of an oil spill can be forecast with an acceptable accuracy today (Ardhuin et al.
2009), the reconstruction of the role of in the transport of tracers (or adverse im-
pacts) is much less satisfactory (Vandenbulcke et al. 2009). This deficiency reflects
the extreme complexity of motions in the seas and oceans. Marine science has only
recently reached a stage where the development of mathematical models, the ac-
curacy and reliability of circulation modelling, the computational facilities and the
quality of information about forcing factors allow addressing the problem of drift
prediction in a dependable way.
Both wind- and wave-induced transport are highly anisotropic and mostly mimic
the behaviour of the wind and wave patterns. To a first approximation, the relevant
transport velocities can be simply added to each other. In the absence of currents, the
probability of hitting a vulnerable area is determined by the joint transport velocity
and the time it takes until the hit occurs is roughly proportional to the resulting joint
downwind/downwave distance from the location of the oil spill. This observation
once more signifies that the remotely affected area and thus the cost of the con-
sequences of a ship accident substantially depend on where and when exactly the
accident happens. Although both these factors have large spatio-temporal variability
and the knowledge of the properties of their interaction with each other and with the
oil spill is far from being perfect, the solution to the problem of finding the optimum
fairway (ship location) is straightforward. The optimal location is just as far in the
combined upwind/upwave direction from the susceptible spots as possible.
16 T. Soomere
The situation for the current-induced transport is fundamentally different. The in-
stantaneous field of currents is an integral reaction of water masses to a variety of
forcing factors such as river discharge, various air–sea interaction processes (wind
forcing, precipitation, evaporation, freezing and melting), dynamics of water masses
in remote sea areas, local mesoscale dynamics, optionally tidal forcing or long
waves, etc. Sea currents form a highly nontrivial system of motions. It is usually
highly anisotropic, inhomogeneous, non-stationary and exhibits large spatial varia-
tions even for practically stationary wind events. In semi-enclosed sea areas such as
the Baltic Sea the currents are frequently even in antiphase with wave- and wind-
induced transport features (Andrejev et al. 2004; Gästgifvars et al. 2006). Conse-
quently, the probability of a vulnerable spot being hit by an oil spill (or a damaged
and drifting ship) carried by surface currents may vary largely for different areas,
even if they are located at an equal distance from this point.
In spite of fast developments concerning the 3D modelling of the marine envi-
ronment, reconstruction of even surface transport of tracers is today still an extreme
challenge. There is not yet a deterministic method capable of adequately repro-
ducing the floating object drift (Vandenbulcke et al. 2009) and even small errors
in its estimates can drastically change the calculated particle trajectories (Griffa
et al. 2004). In particular, the inadequate representation of current patterns is one
of the main reasons why the forecast of current-induced transport is usually much
less reliable compared to the description of wind- and wave-induced transport. For
this reason most of our contributions concentrate on the analysis of the role of the
current-induced transport where we see the largest unused potential for the use of
intrinsic features of marine dynamics for environmental management. Yet we still
stress the importance of the wind- and wave-induced drift and highlight their joint
impact on the seasonal variation of the location of the optimum fairway in Chap. 11.
1.10 Dynamical and Statistical Forecast
A consequence of the intrinsic complexity of the reproduction of the details of ocean

circulation is that current patterns cannot be exactly forecast today. This limits using
the simulation results in many direct applications such as the exact forecast of the
dynamics of an oil spill or of the drift of an item (ship, rescue boat, lost container,
iceberg, etc.) at sea. Yet making use of certain favourable statistical features of cur-
rents is a promising approach. It has been used, for example, in minimizing fuel
consumption along Atlantic-crossing ship routes (Lo and McCord 1998), where for
the eastbound voyages the statistical approach consistently outperformed the deter-
ministic approach.
The importance of statistical methods and their consequences in marine design
and operation are now widely acknowledged. They are most extensively used in
studies of wave properties. Although the outcome of the statistical approach is not
always explicit, it has been extensively used to quantify and mitigate the probabil-
ities of accidents and failures. For example, the traditional format of classification
societies’ rules is mainly prescriptive, without any transparent link to an overall

safety objective. The IMO has developed guidelines for the use of the formal safety
assessment methodology7 in rule development, which will provide risk-based goal-
oriented regulations.
Another example is the use of the gradually adopted first-principle-based ap-
proaches in design that can provide—with the help of structural reliability anal-
ysis—more detailed failure probabilities than those represented by ship accident
databases. Still sufficient information is necessary about the physical and statistical
models of the ocean environment and related uncertainties. Thus, one of the major
challenges (partially addressed in this book) consists in the further development of
methods and technology for the use of this type of information in solving dynamical
problems.
An intrinsic property of the statistical approach is that the outcome only ex-
presses certain significant features of the underlying pool of data. In other words,
the use of such an outcome provides a clear benefit if a large pool of cases is consid-
ered while for single particular cases the damage can still be substantial. However,
this approach is one of the few feasible ways to reach an approximate solution to
the inverse problem of pollution propagation as discussed below.
1.11 Favourable Patterns of Currents and Current-Induced

Transport
As discussed above, a key prerequisite for the reduction of risk for ship traffic is the
existence of certain favourable patterns in the sea domain in question. In the above-
discussed case of the fairway to Boston Harbour (Stokstad 2009) such a pattern
became available in the density of the presence of whales. In principle, any similar
favourable pattern, including patterns of currents or current-induced transport, may
be used for similar purposes. The line of thinking here is the same as for the use of
jet currents saving some fuel, finding sea areas with lower waves and decreasing the
risk of wave damage, or for systems of currents transporting potential pollution to
less vulnerable sea areas (Soomere and Quak 2007).
The most impressive and well-known patterns in the dynamics of the World
Ocean are persistent currents such as the Gulf Stream, Kuroshio, Agulhas or many
coastal currents. Their potential in the reduction of coastal pollution is similar to
the impact of wind and waves: the safest location is the one that is located maxi-
mally upstream from the vulnerable domain. Further progress in the practical use
of current patterns has recently been achieved because of major findings in semi-
enclosed sea areas. Namely, certain sea areas (in which the systems of currents were
previously believed to be highly stochastic and practically impossible to forecast)
were shown to possess persistent patterns of mesoscale currents at certain depths
on a multi-year scale (Andrejev et al. 2004). In addition, semi-persistent patterns of
7 http://www.safedor.org/resources/1023-MEPC392.pdf.
18 T. Soomere
currents, with a typical lifetime from weeks up to a few months have been recently
identified for different areas of the Baltic Sea in terms of Eulerian transport of water
masses (Lehmann et al. 2002; Meier 2007).
These patterns are normally concealed, usually hidden behind the enormous com-
plexity of the instantaneous fields of currents and their short-term variability. Their
existence may essentially modify the transport of potential adverse effects compared
with the classical understanding of the local circulation. Moreover, in certain cases
they predominantly carry water (and pollution ) in a preferred direction that not nec-
essarily matches the downwind or downwave direction (Soomere et al. 2011). On
the one hand, such phenomena obviously have a high potential for the rapid and
systematic transport of both water masses and adverse impacts such as nutrients,
toxic substances or oil pollution between specific sea areas. On the other hand,
their smart use offers a particular solution to the problem of reduction of anthro-
pogenic impacts to vulnerable areas. Namely, if the resulting patterns of transport
are systematic enough, they can be used for placing human activities (such as marine
traffic) in specific regions (areas of reduced risk), from which the pollution transport
to vulnerable or high-cost areas is unlikely (Soomere and Quak 2007).
1.12 Quantification of Offshore Domains

It is heuristically obvious that such current patterns and associated features of (Eu-
lerian or Lagrangian)8 transport may provide a clear environmental benefit. The op-
timal solution is a fairway (or location where to tow the damaged ship, etc.) chosen
so that an accident, if it happens, takes place in the most favourable location. To
specify such a location, it is necessary to quantify the potential of different offshore
sites to serve as a source of risk to the vulnerable regions if pollution were to be
released at this site. For ideally homogeneous currents patterns, the most favourable
locations would be the ‘most offshore’ sites.
This picture needs to be modified if some current patterns systematically carry
adverse impacts in a certain direction. Such a transport reflects certain features
of Lagrangian flow and not necessarily coincides with the long-term patterns of
currents calculated from Eulerian velocities. Therefore, in certain areas of seas or
oceans, which host semi-persistent Lagrangian flow patterns, the potential for re-
mote coastal damage may be smaller than it would be expected from purely geo-
metrical arguments (such as the distance to the nearest coast). These areas are called
8 The Eulerian specification of the flow field is a way of treating, measuring or calculation the mo-
tion properties for each fixed location in space as time passes. In marine conditions one has to use
fixed moorings in order to properly measure the Eulerian velocities. This framework is mostly used
in circulation modelling. The Lagrangian specification is a way of looking at fluid motion where the
observer follows an individual water particle as it moves through space and time. The transport of
different substances and items in the marine environment is obviously Lagrangian. The Lagrangian
velocity and transport can be directly measured using drifting buoys. An overview of these speci-
fications and their implementation in the developed technology is presented in Chaps. 3, 4 and 7.
areas of reduced risk. It is very likely that such areas are located at a considerable
distance from the coast and thus are suitable for shipping. They are the best candi-
dates for fairways and locations of high-risk offshore structures. In semi-sheltered
domains such areas are generally located within a reasonable distance from the ex-
isting fairways and therefore the re-direction of activities to these areas incurs very
limited additional costs.
There exists a variety of meaningful quantifications of the offshore areas de-
pending on the choice of parameters to be optimized. The ‘fair way’ of dividing the
costs equally between the opposite coasts (Soomere et al. 2010) is a local solution
that does not normally provide the minimum level of risk for the entire water body.
A global optimum could be determined in terms of the probability of vulnerable
areas being hit by potential adverse effects released at different offshore sites (An-
drejev et al. 2011). Note that here the reduction in the cost of the consequences is
obtained by means of reducing the probability of hitting valuable areas.
Alternatively, the time it takes for the impact to reach a vulnerable area character-
izes well the cost of the consequences: the longer time the pollution remains in the
open sea, the larger part of it may be removed before it hits a vulnerable spot. This
quantity (called particle age in what follows, Andrejev et al. 2011) may be decisive
in estimates of the capacity of oil combating services. Once a map of the probabili-
ties of coastal hit (or the map of the time it takes for the potential pollution to reach
the coast) has been constructed, the optimum fairway roughly follows the minima
for the probabilities or the maxima for the particle age. In both cases the decrease in
the risk will be obtained through a decrease in the cost of the consequences on the
right-hand side of the risk equation.
1.13 A Preventive Technology of Environmental Management
The crucial point is the comparison of the existing fairways (or locations of other
potentially dangerous activities) with the maps of the level of potential risk associ-
ated with different offshore sites. The situation is perfect if the fairway coincides
with the areas of reduced risk (the minima of the above probabilities or the maxima
of the time). It is, however, much more likely that the major shipping lanes, at least
partially, cross regions that may provide larger danger to the coastal environment
than the adjacent domains. In this case, the risk to the environment can be reduced
by shifting the fairway to the areas hosting lower probabilities for the coastal hit or
larger particle age.
The key idea is therefore the adjustment of the sailing line (and potential acci-
dent site) beforehand, not waiting until an accident happens. If there is no accident,
we have lost nothing (except for minor costs related to the lengthening of the sailing
line). If, however, an accident occurs, the related environmental costs are minimized
for this particular journey. The aim is therefore a method for the systematic preven-
tive reduction of the consequences. The developed method can thus be called a
preventive technology of environmental management.
20 T. Soomere
The classical problem of oil spill forecast is solved for the given location of the
accident, properties of the spill and current metocean conditions. Contemporary oil
spill models forecast the trajectory, fate and impact areas with a high accuracy. In
our work we employ several modifications of such models for modelling the fate
and drift of single pollution particles and oil spills (clusters of particles). We then
make use of them for solving a more complicated problem. The key challenge is to
establish for which release site the particular oil spill will cause the least damage
under the same metocean conditions.
Generally speaking, we are faced with an example of so-called inverse problems:
the identification of the ‘best’ location of the oil release presumes inverse tracking of
the pollution propagation. A straightforward solution to this class of problems is not
possible.9 Moreover, no universal method exists for their analysis. This of course
does not mean that they should not be addressed at all. For the particular problem of
oil spill transport one can use, for example, a sequence of simulations with the same
release time but with slightly shifted release site. This variation of the ‘trial and
error’ method may give some hints about the fate of oil spills released in adjacent
locations. If applied systematically, one could estimate certain locations from where
the oil is less likely to be transported to some vulnerable spots, or from where the
transport will take a longer time. Furthermore, the least dangerous offshore areas
for a particular example of oil spill released at different locations can be estimated
based on a large enough number of trials.
This book describes how such a ‘trial and error’ method, to some extent resem-
bling the Monte Carlo approach, is applied systematically to gather information
about the properties of the inverse problem of oil spill propagation. Its approximate
solution is sought by means of statistical analysis of a large number of trajectories
of single pollution particles released to the sea at different time instants and loca-
tions. Each trajectory is, in essence, an approximate solution of the associated direct
problem of propagation of a very small but persistent oil spill. The central idea of
the technology and its applications is to make use of the properties of Lagrangian
transport of such particles by surface currents.
A major technical problem is how to extract rational information from the vast
amount of numerically simulated data and how to build a reasonable implementa-
tion for the shipping industry with its specific needs and restrictions. Since the drift
of pollution or any other items in the sea is also affected by wind and waves, its
practical implementation generally presumes models integrating all potential fac-
tors. The most widespread state-of-the-art models of this type are operational oil
spill models. Their output will be used in comparisons of the results obtained from
circulation models. The importance of the local wind and wave fields and their im-
pact to the resulting optimum fairways is established for the Gulf of Finland using
an advanced operational oil spill drift and fate model in Chap. 11.
9 Although several hydrodynamic or oil spill models are formally invertible (Ambjörn 2008), realis-
tic 3D fluid dynamics is, in principle, non-invertible. Moreover, a unique solution to the governing
Navier–Stokes equations only exists during a finite time. Therefore, solving even the direct prob-
lem of oil spill propagation overrides this fundamental feature of fluid dynamics.
The entire technology thus combines a number of components into one entity.
Its corner stone is a high-resolution 3D circulation model that adequately resolves
the majority of mesoscale features in the sea area of interest. The model output pro-
vides the entire technology with (Eulerian) velocities at certain grid points and with
a certain temporal resolution. This data set makes it possible to evaluate Lagrangian
trajectories of single particles. For simplicity, we only consider the particular case
when these trajectories are locked to the uppermost layer and thus reflect the be-
haviour of the lighter fractions of oil pollution.
A proper statistical analysis of the trajectories leads to 2D maps of various quan-
tities characterizing the damaging potential of different offshore sites in case of a
release of oil spill. Among those, the probability of hitting vulnerable areas and
the time it takes until the hit evidently have the largest value for decision-making.
The final step of the sequence of operations is decision-making. This is relatively
simple if a location of a single site is to be defined but may be highly nontrivial
when an optimal fairway must be constructed.
1.14 From Classical Physical Oceanography to Environmental

Management
The entire method essentially relies on the state-of-the-art knowledge of the physical
oceanography, meteorological factors driving the currents and processes on the sea
surface and the finest circulation model. A fundamental precondition of receiving a
certain benefit is the presence of (semi-)persistent patterns of currents or winds that
modify the otherwise homogeneous and isotropic statistics of surface-layer trans-
port. As the nature of internal dynamics and the properties of such patterns vary
largely in different seas and oceans, it is natural to assume that the parameters of
the used models as well as the outcome of the entire technology are extremely site-
specific.
For the wide variety of seas surrounding Europe and suffering from extensive
anthropogenic impact (such as the Black Sea, Sea of Marmara, Adriatic Sea, North
Sea, Norwegian Sea, or Barents Sea), the situation is the most desperate in the Baltic
Sea. In the Baltic Sea (incl. the Kattegat) about 76 ports handle more than 1 million
tonnes of cargo per year. On the one hand, it is a relatively small brackish sub-
basin of the Atlantic Ocean with an extremely sensitive ecosystem (Leppäranta and
Myrberg 2009). This feature has been legally recognized by including this sea on
the list of particularly sensitive sea areas by the IMO (Lefebvre-Chalain 2007). On
the other hand, its extremely rich internal dynamics (incl. the absence of quasi-
stationary current systems) makes it a perfect test area of the method for highlighting
the impact of usually concealed semi-persistent current patterns. On top of that,
the Baltic Sea is subject to enormous anthropogenic pressure. The number of ship
operations (voyages, excluding ferry traffic) in the Baltic Sea is estimated at 150,000
per year (Gollasch and Leppäkoski 2007), and it is assumed to increase considerably
in the future.
22 T. Soomere
Since the entire Baltic Sea is too large to reproduce the currents with the neces-
sary accuracy, we specifically concentrate on two target regions: the Gulf of Finland
and the south-western Baltic Sea. The former is the most vulnerable area hosting
extremely heavy ship and tanker traffic (Kujala et al. 2009), and the busiest port,
namely Saint Petersburg with >14,500 ship visits per year. Moreover, the spatial
distribution of the probability of ship accidents has been quantified in great detail
for this area (Goerlandt and Kujala 2011). The Belt Sea and the Arkona Basin in the
south-western Baltic Sea are the areas where all the ships travelling into/from the
North Sea have to pass through.
Stressing the generic character of the reasoning, we shall use the Baltic Sea not
only as an example for explaining the concept but also as an area where innovative
solutions are urgently needed, and where the proposed technology is expected to
work. For this reason, we start with a detailed overview of the Baltic Sea as the
target area for which the technique has been developed (Chap. 2). This overview
focuses on the basic features of the dynamics of currents and the requirements for
their modelling.
A successful implementation of the entire technology requires excellent compe-
tence in circulation modelling for the target area. There exist a variety of different
models, from greatly simplified toy models reproducing the basic flow features up to
highly sophisticated modelling systems. For making a good choice among different
models for a particular target area it is advisable to have an extended understanding
about how such models work, what their strong and weak points are and, most im-
portantly, what they can and what they cannot describe. This subject is covered by a
generic introduction into principles of ocean modelling (Chap. 3) and by an in-depth
insight into the challenges of circulation modelling in the Baltic Sea (Chap. 4). In
particular, Chap. 4 describes the outcome of a complementary application of the Eu-
lerian transport for the optimization of the fairways with respect to coastal pollution
in the Baltic Proper.
This material is complemented by a comprehensive overview of specific features
of other major European seas and inside information about the requirements and
potential pitfalls in their modelling (Chap. 5). The potential of the use of this method
for both ‘ideal’ persistent pollution particles and for more realistic behaviour of oil
spills will be demonstrated in the context of the optimization of ship routes in the
Gulf of Finland in terms of minimizing the risk of coastal pollution. Chapter 6 gives
an insight into the richness of the dynamics of this basin and a number of interesting
features that make this area challenging for modellers.
The second key component of the technology, the art of modelling of Lagrangian
trajectories of selected particles based on precomputed Eulerian velocity fields, is in-
troduced in Chap. 7. The modelling procedure is exemplified using the open-source
TRACMASS code. Our research has highlighted an essential difference between the
Lagrangian and Eulerian transport in the Baltic Sea (Chap. 8). This feature leads to
the necessity of a specific tuning of the modelled Lagrangian trajectories for certain
sea areas in order to properly represent the statistics of oil spill propagation. Along
with an essential collection of very recent experimentally measured subsurface drift
properties in the Baltic Proper and surface drift patterns in the Gulf of Finland, rec-
ommendations are given in Chap. 8 how to adjust the models using the existing
Eulerian velocity data in such conditions.
The following three chapters exploit the central idea of the entire technique,
namely, that a considerable decrease in environmental risk can be obtained by an
optimum design of the fairway (or the location of an installation), based on the de-
tailed knowledge of certain underlying properties of sea currents. The essence is that
certain areas of semi-enclosed seas have a low probability of transport of released
substances towards high-cost areas, and should be used for potentially dangerous
activities.
The first outcome from the statistical analysis of Lagrangian trajectories on the
sea surface is the presence of a very rich, usually concealed internal structure of
semi-persistent transport patterns (Chap. 9). As described above, it is natural to as-
sume that these patterns make the probability of transport of dangerous substances
or undesired items from different open sea areas to vulnerable sections (such as
spawning, nursing or also tourist areas) highly variable.
This assumption is found to be true for the targets areas (Gulf of Finland and
south-western Baltic Sea). It is shown that such patterns can be identified in a num-
ber of different characteristics of offshore waters such as the probability of current-
induced transport of an adverse impact released in a particular offshore domain to
the nearshore, or the time it takes for the impact to reach the coast (Chap. 10).
An substantial extension of the results based on ‘ideal’ pollution particles solely
advected by surface currents derived from the circulation modelling exercises to
much more realistic modelling of of the drift and fate of oil spills is presented in
Chap. 11 using the advanced oil spill model model currently in use in the Danish
Meteorological Institute. The results are used for the design of fairways in the Gulf
of Finland that also accounts for the direct impact of wind on the propagation of oil
in the marine environment. Finally, Chap. 11 presents some insight into the potential
impact of waves on the oil spill propagation in the Gulf of Finland.
We conclude the book with a short discussion of several aspects of potential
applications of the technology such as (i) in which sea areas solutions of this type
are feasible, (ii) how large the environmental benefit could be, (iii) how the new
solutions can be integrated with the existing operational oceanographic services,
and (iv) how such solutions can be used within (or additionally to) the existing
fairway planning and ship routing practice.
1.15 Concluding Remarks

Putting it short, the following chapters present both the foundations and details of
a technology to construct the fairway so that if an accident happens, the resulting
current-driven (oil) pollution will occur in the most favourable location in terms of
a simplified notion of the environmental risk. The described technology thus serves
as a natural extension of advisory on-board devices for safe routing and for the
operation of marine structures.
24 T. Soomere
The presented technology only answers a very particular question. After retriev-
ing the solution, one has to combine the extracted knowledge with other important
factors that play a role in deciding an optimal ship route (such as potential extreme
loads along the route, the probability of fatigue damage, the fuel consumption along
the route, and the time to reach a harbour), etc., and that can be estimated using
well-known methods. Although the questions of (oil) pollution and of the accompa-
nying environmental risks are of immense importance, generally a number of factors
with possibly different weights have to be taken into account in an industrial fairway
design or ship routing system.
There may be even several environmental objectives, not necessarily converging
but to be optimized jointly. For instance, the shore may be protected, but the fishing
grounds or some other sensitive, offshore, important and/or vulnerable areas may
suffer at the same time. The change of the fairways from the existing relatively long,
straight legs, with wide room for manoeuvring to short, narrow legs with numerous
bends may substantially increase the number of conflicting ship-ship encounters as
discussed in Chap. 11. Moreover if the fairways fluctuate seasonally, this may bring
even more confusion for mariners. This in turn may increase the probability of an
accident and finally the aggregated risk.
The wider importance of the technology is that, differently from various direct
methods of estimates of environmental risks, the resulting quantities and decisions
are associated with the points of release of adverse impacts. The key contribution
from such methods based on approximate solving of an inverse problem is the possi-
bility of extracting and visualizing important information that usually remains con-
cealed (and unused) in classical methods for the analysis of ocean currents.
Acknowledgements The underlying studies were performed in the framework of the BalticWay
project, which was jointly supported by the funding from the Estonian Science Foundation
and the European Commission’s Seventh Framework Programme (FP 2007–2013) under grant
agreement No. 217246 made with the joint Baltic Sea research and development programme
BONUS. The follow-up research was partially supported by the Estonian Science Foundation
(grant No. 9125), targeted financing by the Estonian Ministry of Education and Research (grant
SF0140007s11), and by the European Regional Development Fund via support to the Centre of
Excellence for Non-linear Studies CENS.
References
Ambjörn C (2008) Seatrack web forecasts and backtracking of oil spills—an efficient tool to find
illegal spills using AIS. In: IEEE/OES US/EU-Baltic International Symposium, Tallinn, Esto-
nia, May 27–29, 2008. IEEE Press, New York, pp 168–176
Andrejev O, Myrberg K, Alenius P, Lundberg PA (2004) Mean circulation and water exchange
in the Gulf of Finland—a study based on three-dimensional modelling. Boreal Environ Res
9:1–16
Andrejev O, Soomere T, Sokolov A, Myrberg K (2011) The role of spatial resolution of a three-
dimensional hydrodynamic model for marine transport risk assessment. Oceanologia 53:309–
334
Anonymous (2002) Results from the reader challenge: which MPA is the oldest? MPA News 3:6
Ardhuin F, Marie L, Rascle N, Forget P, Roland A (2009) Observation and estimation of La-
grangian, Stokes, and Eulerian currents induced by wind and waves at the sea surface. J Phys
Oceanogr 39:2820–2838
Azaron A, Kianfar F (2003) Dynamic shortest path in stochastic dynamic networks: ship routing
problem. Eur J Oper Res 144:138–156
Banister D (2007) Sustainable transport: challenges and opportunities. Transportmetrica 3:91–106
Boissonnas J, Connolly N, Mantoura F, d’Ozouville L (eds) (2002) Integrating marine science in
Europe. ESF Marine Board, Strasbourg
Bowden-Kerby A (2001) The original MPAs (Marine Protected Areas). MPA News 3:5–6
Boyd I, Brownell B, Cato D, Clarke C, Costa D, Evans P, Gedanke J, Gentry R, Gisiner B, Gor-
don J, Jepson P, Miller P, Rendell L, Tasker M, Tyack P, Vos E, Whitehead H, Wartzok D,
Zimmer W (2008) The effects of anthropogenic sound on marine mammals. A draft research
strategy. In: Connolly N, Calewaert J-B (eds) ESF Marine Board Position Paper, Oostende,
Belgium, 92 pp
Burgherr P (2007) In-depth analysis of accidental oil spills from tankers in the context of global
spill trends from all sources. J Hazard Mater 140:245–256
Calvert S, Deakins E, Motte R (1991) A dynamic system for fuel optimization trans-ocean. J Navig
44:233–265
Chang YL, Oey L, Xu FH, Lu HF, Fujisaki A (2011) 2010 oil spill: trajectory projections based
on ensemble drifter analyses. Ocean Dyn 61:829–839
Christiansen M, Fagerholt K, Ronen D (2004) Ship routing and scheduling: status and perspectives.
Transp Sci 38:1–18
Collins MP, Vecchio FJ, Selby RG, Gupta PR (1997) The failure of an offshore platform. ASI
Concr Int 19:28–35
Durkee PA, Chartier RE, Brown A, Trehubenko EJ, Rogerson SD, Skupniewicz C, Nielsen KE
(2000) Composite ship track characteristics. J Atmos Sci 57:2542–2553
Eide MS, Endresen Ø, Brett PE, Ervik JL, Røang K (2007) Intelligent ship traffic monitoring for
oil spill prevention: risk based decision support building on AIS. Mar Pollut Bull 54:145–148
Elken J, Kask J, Kõuts T, Liiv U, Perens R, Soomere T (2001) Hydrodynamical and geological
investigations of possible deep harbour sites in north-western Saaremaa Island: overview and
conclusions. Proc Est Acad Sci, Eng 7:85–98
Fagerholt K, Heimdal SI, Loktu A (2000) Shortest path in the presence of obstacles—an applica-
tion to ocean shipping. J Oper Res Soc 51:683–688
Fowler TG, Sørgård E (2000) Modeling ship transportation risk. Risk Anal 20:225–245
Gästgifvars M, Lauri H, Sarkanen A-K, Myrberg K, Andrejev O, Ambjörn C (2006) Modelling
surface drifting of buoys during a rapidly-moving weather front in the Gulf of Finland, Baltic
Sea. Estuar Coast Shelf Sci 70:567–576
Glen D (2010) Modelling the impact of double hull technology on oil spill numbers. Marit Policy
Manag 37:475–487
Goerlandt F, Kujala P (2011) Traffic simulation based on ship collision probability modeling.
Reliab Eng Syst Saf 96:91–107
Gollasch S, Leppäkoski E (2007) Risk assessment and management scenarios for ballast water
mediated species introduction into the Baltic Sea. Aquat Invasions 2:313–340
Griffa A, Piterbarg LI, Özgökmen T (2004) Predictability of Lagrangian particle trajectories: ef-
fects of smoothing of the underlying Eulerian flow. J Mar Res 62:1–35
Hassler B (2011) Accidental versus operational oil spills from shipping in the Baltic Sea: risk
governance and management strategies. Ambio 40:170–178
Hobbs PV, Garrett TJ, Ferek RJ, Strader SR, Hegg DA, Frick GM, Hoppel WA, Gasparovic RF,
Russell LM, Johnson DW, O’Dowd C, Durkee PA, Nielsen KE, Innis G (2000) Emissions from
ships with respect to their effects on clouds. J Atmos Sci 57:2570–2590
Iakovou E, Douligeris C, Li H, Ip C, Yudhbir L (1999) A maritime global route planning model
for hazardous materials transportation. Transp Sci 33:34–48
Judson B (1997) A tanker navigation safety system. J Navig 50:97–108
26 T. Soomere
Kachel MJ (2008) Particularly sensitive sea areas. Hamburg studies on maritime affairs, vol 13.
Springer, Berlin, 376 pp
Kara BY, Verter V (2004) Designing a road network for hazardous materials transportation. Transp
Sci 38:188–196
Keramitsoglou I, Cartalis C, Kassomenos P (2003) Decision support system for managing oil spill
events. Environ Manag 32:290–298
Kite-Powell H (2011) The value of ocean surface wind information for maritime commerce. Mar
Technol Soc J 45:75–84
Ko TT, Chang Y-C (2010) Integrated marine pollution management: a new model of marine pol-
lution prevention and control in Kaohsiung, Taiwan. Ocean Coast Manag 53:624–635
Kostianoy AG, Ambjörn C, Soloviev DM (2008) Seatrack Web: a numerical tool to protect the
Baltic Sea marine protected areas. In: IEEE/OES US/EU-Baltic International Symposium,
Tallinn, Estonia, May 27–29, 2008. IEEE Press, New York, pp 7–12
Kujala P, Arughadhoss S (2012) Statistical analysis of ice crushing pressures on a ship’s hull during
hull-ice interaction. Cold Reg Sci Technol 70:1–11
Kujala P, Hanninen M, Arola T, Ylitalo J (2009) Analysis of the marine traffic safety in the Gulf
of Finland. Reliab Eng Syst Saf 94:1349–1357
Lee H, Kong G, Kim S, Kim C, Lee J (2002) Optimum ship routing and it’s implementation on the
web. In: Chang W (ed) Advanced Internet services and applications. Lecture notes in computer
science, vol 2402. Springer, Berlin, pp 125–136
Lefebvre-Chalain H (2007) Fifteen years of particularly sensitive sea areas: a concept in develop-
ment. Ocean Coast Law J 13:47–69
Lehmann A, Krauss W, Hinrichsen H-H (2002) Effects of remote and local atmospheric forcing
on circulation and upwelling in the Baltic Sea. Tellus A 54:299–316
Leppäranta M, Myrberg K (2009) Physical oceanography of the Baltic Sea. Springer, Berlin,
378 pp
Lo HK, McCord MR (1995) Routing through dynamic ocean currents—general heuristics and
empirical results in the Gulf-stream region. Transp Res, Part B, Methodol 29:109–124
Lo HK, McCord MR (1998) Adaptive ship routing through stochastic ocean currents: general
formulations and empirical results. Transp Res, Part A, Policy Pract 32:547–561
Lo HK, McCord MR, Wall CK (1991) Value of ocean current information for strategic routing.
Eur J Oper Res 55:124–135
McCord MR, Lee YK, Lo HK (1999) Ship routing through altimetry-derived ocean currents.
Meier HEM (2007) Modeling the pathways and ages of inflowing salt- and freshwater in the Baltic
Montewka J, Krata P, Goerlandt F, Mazaheri A, Kujala P (2011) Marine traffic risk modelling—an
innovative approach and a case study. Proc Inst Mech Eng, Part O, J Risk Reliab 225:307–322
Paik JK (2003) Innovative structural designs of tankers against ship collisions and grounding:
a recent state-of-the-art review. Mar Technol SNAME News 40:25–33
Panigada S, Pesante G, Zanardelli M, Capoulade F, Gannier A, Weinrich MT (2006) Mediterranean
fin whales at risk from fatal ship strikes. Mar Pollut Bull 52:1287–1298
Parnell KE, Kofoed-Hansen H (2001) Wakes from large high-speed ferries in confined coastal wa-
ters: management approaches with examples from New Zealand and Denmark. Coast Manage
29:217–237
Parnell KE, McDonald SC, Burke AE (2007) Shoreline effects of vessel wakes, Marlborough
Sounds, New Zealand. J Coast Res 50(Special Issue):502–506
Periáñez R (2004) A particle-tracking model for simulating pollutant dispersion in the Strait of
Gibraltar. Mar Pollut Bull 49:613–623
[PIANC] (2003) Guidelines for managing wake wash from high-speed vessels. Report of the Work-
ing Group 41 of the Maritime Navigation Commission. International Navigation Association
(PIANC), Brussels, 32 pp
Risk Management Standard (2002) The Institute of Risk Management, London, 14 pp. http://www.
theirm.org/publications/documents/Risk_Management_Standard_030820.pdf
Romer H, Brockhoff L, Haastrup P, Petersen HJS (1994) Marine transport of dangerous goods—
risk assessment based on historical accident data. J Loss Prev Process Ind 6(4):219–225
Rusli MHB (2012) Protecting vital sea lines of communication: a study of the proposed designation
of the Straits of Malacca and Singapore as a particularly sensitive sea area. Ocean Coast Manag
57:79–94
Schwehr KD, McGillivary PA (2007) Marine ship Automatic Identification System (AIS) for en-
hanced coastal security capabilities: an oil spill tracking application. In: Proceedings of the 2007
OCEANS conference, Vancouver, Canada, September 29–October 04, 2007. IEEE Press, New
York, pp 1131–1139
Sherman K, Duda AM (1999) An ecosystem approach to global assessment and management of
coastal waters. Mar Ecol Prog Ser 190:271–287
Smith HD (1995) The environmental-management of shipping. Mar Policy 19:503–508
Soares CG, Teixeira AP (2001) Risk assessment in maritime transportation. Reliab Eng Syst Saf
74:299–309
Soomere T (2005) Fast ferry traffic as a qualitatively new forcing factor of environmental processes
in non-tidal sea areas: a case study in Tallinn Bay, Baltic Sea. Environmental Fluid Mechanics
5:293–323
Soomere T (2009) Long ship waves in shallow water bodies. In: Quak E, Soomere T (eds) Ap-
plied wave mathematics: selected topics in solids, fluids, and mathematical methods. Springer,
Heidelberg, pp 193–228
Soomere T, Quak E (2007) On the potential of reducing coastal pollution by a proper choice of the
fairway. J Coast Res 50(Special Issue):678–682
Soomere T, Viikmäe B, Delpeche N, Myrberg K (2010) Towards identification of areas of reduced
risk in the Gulf of Finland, the Baltic Sea. Proc Est Acad Sci 59:156–165
Soomere T, Delpeche N, Viikmäe B, Quak E, Meier HEM, Döös K (2011) Patterns of
current-induced transport in the surface layer of the Gulf of Finland. Boreal Environ Res
16(Suppl A):49–63
Stokstad E (2009) US poised to adopt national ocean policy. Science 326:1618
Torsvik T, Didenkulova I, Soomere T, Parnell KE (2009) Variability in spatial patterns of long
nonlinear waves from fast ferries in Tallinn Bay. Nonlinear Process Geophys 16:351–363
Vach M, Duong VM (2011) Numerical modeling of flow fields and dispersion of passive pollutants
in the vicinity of the Temelin nuclear power plant. Environ Model Assess 16:135–143
Vandenbulcke L, Beckers J-M, Lenartz F, Barth A, Poulain P-M, Aidonidis M, Meyrat J, Ard-
huin F, Tonani M, Fratianni C, Torrisi L, Pallela D, Chiggiato J, Tudor M, Book JW, Martin P,
Peggion G, Rixen M (2009) Super-ensemble techniques: application to surface drift prediction.
Prog Oceanogr 82:149–167
Verta M, Salo S, Korhonen M, Assmuth T, Kiviranta H, Koistinen J, Ruokojärvi P, Isosaari P,
Bergqvist P-A, Tysklind M, Cato I, Vikelsøe J, Larsen MM (2007) Dioxin concentrations in
sediments of the Baltic Sea—a survey of existing data. Chemosphere 67:1762–1775
Ward-Geiger LI, Silber GK, Baumstark RD, Pulfer TL (2005) Characterization of ship traffic in
right whale critical habitat. Coast Manage 33:263–278
Wolman D (2008) Before the levees break: a plan to save the Netherlands. Wired Magazine
17.01.2008
Wood WA (2000) High-speed ferry issues for operators and designers. Mar Technol SNAME News
37:230–237
Part I
Modelling the Underlying Dynamics
Chapter 2
Topography, Hydrography, Circulation
and Modelling of the Baltic Sea
Kai Myrberg and Andreas Lehmann
Abstract The hydrography and dynamics of the Baltic Sea, although ruled by the
same principles and forcing factors as any part of the World Ocean, contain several
distinguishing features. Apart from the complicated geometry and bathymetry of
the basin, two major factors contribute to the complexity of the processes here. The
interplay between inflowing saline, dense waters from the North Sea in the bottom
layer with the excess of light, and fresh riverine waters coming into the system in
the upper layer leads to the formation of a permanent two-layer structure of density
separated by a sharp jump layer (halocline). Due to the layered structure, the direct
atmospheric forcing is restricted to the upper layer with a typical thickness of 40–
80 m, while in the bottom layer advection and mixing processes govern the patterns
of the hydrographic fields. On the top of the upper layer, a well-mixed surface layer,
with a typical thickness of 15–20 m, is formed due to summer-time heating, whereas
at the bottom of this layer a rather sharp jump layer of temperature (thermocline)
exists. During autumn the vertical temperature gradient vanishes due to thermal
convection and turbulent mixing. There are four mechanisms which induce currents
in the Baltic Sea: the wind stress at the sea surface, the surface pressure gradient,
the thermohaline horizontal gradient of density and the tidal forces. The currents are
steered furthermore by the Coriolis acceleration, topography and friction, forming
a general (cyclonic) circulation in this stratified system with positive fresh water
budget. Due to the shallowness of the Baltic Sea, bottom friction damps the currents
remarkably. Voluminous river runoffs can produce local changes in the sea level
height and consequently also in currents. Inflowing waters penetrate at depths where
the density of the ambient water matches the inflowing water masses. Due to the
K. Myrberg (B)
Finnish Environment Institute, Marine Research Centre, Mechelininkatu 34 a, P.O. Box 140,
00251, Helsinki, Finland
e-mail: kai.myrberg@ymparisto.fi
K. Myrberg
Department of Geophysics, Klaipėda University, Herkaus Manto Str. 84, Klaipėda, Lithuania
A. Lehmann
Helmholtz Centre for Ocean Research Kiel (GEOMAR), Düsternbooker Weg 20, Kiel, Germany
e-mail: alehmann@geomar.de

DOI 10.1007/978-3-319-00440-2_2,
32 K. Myrberg and A. Lehmann
Fig. 2.1 The Baltic Sea and its sub-basins (Leppäranta and Myrberg 2009, modified from Fon-
selius 1995)
small baroclinic Rossby radius (2–10 km), the proper descriptions of mesoscale
eddies, fronts and mixing processes need high-resolution modelling.
2.1 Introduction to the Baltic Sea Geometry and Topography
The Baltic Sea belongs to the shelf seas connected to the Atlantic Ocean. It is also
sometimes called a small intra-continental sea (Leppäranta and Myrberg 2009).
When arriving from the Atlantic Ocean via the North Sea, the outermost part of
the Baltic Sea is located in the narrow region between mainland Denmark and Swe-
den (see Leppäranta and Myrberg 2009 for details). The traditional definition sets
its boundary between the Kattegat and the Danish Straits (Fig. 2.1), which consist
of the Øresund and the Belts. This definition is also employed here: The Kattegat is
not considered a part of the Baltic Sea.
Land uplift has been ongoing in the Baltic area during the Holocene (Ekman
1996). Its magnitude ranges from zero in the south up to 9 mm/yr in the northern
Gulf of Bothnia. The location of the shoreline is thus influenced by the budget of
the eustatic sea level rise, the hydrography of the Baltic waters and the land uplift
pattern. At present the eustatic sea level rise is 2–3 mm/yr (IPCC 2007). When all
2 Topography, Hydrography, Circulation and Modelling of the Baltic Sea 33
Fig. 2.2 The main features of the geometry and bathymetry of the Baltic Sea and related important
geographical locations (Leppäranta and Myrberg 2009 according to Fonselius 1995)
the effects are put together as one number representing the apparent land uplift in
relation to the mean sea level, the result is largest in the Bay of Bothnia. In Kemi
it equals 7.20 ± 0.27 mm/yr while in the eastern Gulf of Finland (Hamina) it is
1.62 ± 0.26 mm/yr only. The zero line of the apparent uplift goes across Estonia,
and in the southern Baltic Sea the land is apparently sinking. The apparent uplift will
evidently vanish in a long time perspective in most of the Gulf of Finland (Johansson
et al. 2004). This is due to the increase of the eustatic sea level rise, which is in turn
caused by the recent warming trend in the global climate. However, on top of the
eustatic sea level rise, the mean annual sea level of the Baltic Sea is also positively
correlated with the strength of the westerlies.
The present Baltic Sea consists of different basins, for which the character of the
coastline and the bottom topography vary considerably (Fig. 2.2). This variability
is one of the key factors in the background of the complex physics and dynamics
of the Baltic Sea. Geometry, morphology, deeps and depressions of each sub-basin
Table 2.1 The dimensions of the Baltic Sea basins and the Kattegat (modified from Fonselius
1995)
Area km2 Mean Maximum Volume
depth, m depth, m km3
Kattegat 22,287 23 130 515

Baltic Sea 392,978 54 459 21,205
South-western Baltic Sea

Danish Straits 20,121 14 81 287
Belt Sea 17,821 15 81 260
Øresund 2,300 12 53 27
Arkona Basin 19,068 23 53 442
Bornholm Basin 38,942 46 105 1,780
Gotland Sea 151,920 71 459 10,824
Gdańsk Bay 25,234 57 114 1,439
Eastern Gotland Basin 63,478 77 249 4,911
Northern Gotland Basin 28,976 71 150 2,056
Western Gotland Basin 34,232 71 459 2,418
Eastern and northern gulfs
Gulf of Finland 29,498 37 123 1,098
Gulf of Riga 17,913 23 51 405
Gulf of Bothnia 115,516 55 293 6,369
Åland Sea 5,477 75 301 411
Archipelago Sea 8,893 19 104 169
Sea of Bothnia 64,886 66 293 4,308
Bay of Bothnia 36,260 41 146 1,481
(Fig. 2.1) have their own specific features. The coastal areas are often characterized
by archipelagos.
A characteristic feature of the Baltic Sea is its shallowness. The mean depth is
54 m and even the largest depth is 459 m only, in Landsort Deep between Stock-
holm and the island of Gotland. The area of the Baltic Sea is also small, namely
392,978 km2 , while its volume equals 21,205 km3 .
An overall view of the major features of the Baltic Sea can be obtained based on
the areas, volumes and depths in the different basins (Table 2.1, Fig. 2.2). The largest
basin is the Gotland Sea1 with an area of 151,920 km2 or 39 % of the total area of
the Baltic Sea, whereas its volume of 10,824 km3 is 51 % of the total volume. Not
surprisingly the mean depth of the Gotland Sea (71 m) is larger than the mean depth
of the entire Baltic Sea. The second largest basin is the Gulf of Bothnia, which has
1 An often used notion is Baltic Proper to denote the Eastern, Northern and Western Gotland Basin
(Table 2.1), Bornholm Basin and Gdańsk Bay.

an area of 115,516 km2 , a volume of 6,369 km3 and a mean depth of 55 m. The next
one in size is the Bornholm Basin. Its area is 38,942 km2 , its volume 1780 km3 and
its mean depth 46 m. The Gulf of Finland covers an area of 29,498 km2 (8 % of the
Baltic Sea) and has a volume of 1,098 km3 (5 % of the Baltic Sea). Its mean depth is
37 m, clearly less than that of the entire Baltic Sea. The Danish Straits, the Arkona
Basin, and the Gulf of Riga have relatively small volumes.
The shallowness of the sea becomes clearly visible if the depth distribution is
presented as a hypsographic curve (not shown). Only about 12 % of the total area
has a depth of more than 100 m, and only 2.7 % of more than 150 m. Consequently,
a big part of the Baltic Sea belongs to a shallow, coastal-like area. As a rule of thumb
it can be said that about 50 % of the area of the Baltic has a depth of 50 m or less.
2.2 Basic Hydrography
2.2.1 Salinity
In the Baltic Sea the salinity mostly determines the stratification of the water masses.
When the inflowing saltier water masses from the North Sea enter the brackish
Baltic Sea, they sink to the bottom, move further as dense bottom currents and fill
the deep water pools (see for details Chap. 6 for the Gulf of Finland). As a result,
the Baltic Sea water body has a permanent two-layer structure (Fig. 2.3): the up-
per layer and the bottom layer (or lower layer), separated by a halocline (see also
Fig. 2.4). The depth of the location of the halocline is usually 40–80 m, but in the
shallow south-western basins it is even less. The upper layer is homohaline, while
the bottom layer is continuously stratified. The salinity stratification is regularly
modified by the formation of the seasonal surface layer in summer. This is a warm,
well-mixed layer, isolated from the deeper water due to lower density. The spring
and summer runoff additionally lower its salinity.
The thickness of the permanent halocline is 10–20 m. Its depth is determined
by advection, wind-induced and convective mixing and the sill depths. This depth
changes very little in time. An exception is the area of the Danish Straits, where
the haline stratification is different from that, for example, in the Gotland Sea. Its
wedge-shaped structure moves back and forth in reaction to prevailing wind condi-
tions. In areas where the vertical stratification of salinity is weak, the halocline can
disappear in certain specific conditions. For example in the western Gulf of Finland
the predominant south-westerly winds work against the standard estuarine circula-
tion. Long-lasting and strong winds push a large amount of relatively fresh surface
water into the gulf. The resulting increase in the hydrostatic pressure may lead to
a gradual export of the salt wedge at the bottom layer of the basin (Elken et al.
2003). The reversal may occur if the average speed of south-westerly winds exceeds
4–5.5 m/s.
Another specific region is the easternmost Gulf of Finland, where the fresh wa-
ter from the River Neva has a strong effect on the stratification (see Chap. 6). In
Fig. 2.3 A schematic diagram of vertical stratification of Baltic Sea water masses; S stands for
salinity and T for temperature (Leppäranta and Myrberg 2009)
this shallow domain the salinity increases approximately linearly with depth and no
clearly defined halocline exists. In the Gulf of Bothnia the halocline is relatively
weak but still the salinity2 difference between the upper and bottom layers is more
than 0.5 ‰ in the deeper areas of this basin.
The waters below the halocline are not affected by wind-induced and convective
mixing, which is restricted to the homohaline upper layer. The stratification in the
bottom layer is maintained by advection and turbulence, and the salinity increases
almost linearly with depth. In the deepest areas sometimes another clearly defined
near-bottom layer of saltier water may give rise to a three-layer structure in density.
There is some evidence that a transient secondary halocline may also be formed be-
tween the ‘old’ water in the bottom layer and recently advected saltier bottom water
at a depth of some 125 m (Mälkki and Tamsalu 1985). However, such a secondary
halocline is usually not evident in climatological patterns and may be found only in
the deepest basins of the Baltic Sea, like in the Gotland Basin.
The seasonal variability of the salinity stratification is much weaker than the
corresponding changes in temperature. The salinity minimum is observed in spring
in the surface layer due to the fresh water flux from the rivers. The formation of the
thermocline and the weak vertical mixing in the upper layer through the thermocline
keeps the fresh waters in the surface layer and thus further reduces the salinity there.
This salinity minimum in spring is usually by 0.5 ‰ lower than the maximum in
winter. There is a lag between the salinity minimum and the maximum of river
runoff. For example, the minimum is observed near the island of Utö about 2.5
months after the runoff peak. This corresponds to an average speed of current-driven
2 Although the salinity unit ‰ (per mill) is discouraged since 1978, it has been customary in a large
part of the oceanographic and popular literature to use this notion. We only use this unit in data
and estimates extracted from older sources. The new international standard TEOS-10 uses absolute
salinity values in g/kg (Millero et al. 2008).
transport of a few cm/s. The stable summer stratification with weak vertical mixing
induces a salinity maximum in the bottom layer. In fall and winter mixing is strong
and the salinity difference between surface and bottom layer is reduced again. This
development is most pronounced near the coast where the halocline might not exist.
The salinity in the Baltic Sea decreases from the south to north and towards east.
This pattern is governed by the inflowing saline waters from the North Sea, the vo-
luminous river water inflow mainly from the north and east and the overall cyclonic
circulation. The surface layer water at the entrance of the Baltic Sea (in the Kattegat
and at Danish Straits; at times also in the Belt Sea and Arkona Basin) is usually
Baltic Sea water (salinity 8–12 ‰), whereas the bottom layer water originates from
the Kattegat with salinities up to 32–33 ‰. Towards east and north the salinity de-
creases both in surface and bottom layers. In the Gotland Basin the salinity of the
upper layer is 7–8 ‰. Below the halocline the salinity increases approximately lin-
early with depth to 9–12 ‰ at a depth of 100 m and to 11.5–13 ‰ at 200 m. In
the northern basins surface salinity varies between 0 and 6 ‰ and bottom salinity is
slightly higher, i.e., a weak halocline exists in part of the basins (see Chap. 6).
2.2.2 Temperature
In the Baltic Sea the temperature first follows the two-layer structure determined
by the salinity-driven density changes (Figs. 2.3, 2.4). The temperature of the up-
per layer experiences a remarkable annual cycle forced by the radiation budget and
air–sea interaction, while the bottom layer below the halocline is decoupled from
atmospheric forcing and largely influenced by the advection from the North Sea.
A warm seasonal mixed layer develops during summer above a seasonal thermo-
cline in the upper layer. In the northern part of the Baltic Sea an inverse thermocline
is formed in wintertime: the temperature is at the freezing point at the surface and
increases with depth to the temperature of maximum density (2–3 ◦ C). A dicother-
mal layer (see below) forms in summer at the bottom of the upper layer because the
water is heated from the top and cold water remains further down. This cold layer is
also called winter water in the Baltic Sea. The layered structure is clearly evident in
the mean temperature gradient.
The seasonal evolution of temperature in the Baltic Sea is very different in the
upper and lower layers. Due to large seasonal variations of the energy budget at
the sea surface, the surface water temperature reaches its maximum in summer and
minimum in winter, and in winter at least a part of the sea freezes. The seasonal
variability of the temperature in the lower layer is weak. Especially in the southern
Baltic Sea it mostly depends on advection from the North Sea through the Danish
Straits. When the water masses flow northwards they sink down to the bottom due
to their higher salinity compared with the ambient water. This is why the bottom
waters are relatively warm, in the Gotland Basin the bottom temperature is 4–6 ◦ C,
and in the Gulf of Finland, the Gulf of Riga and the Gulf of Bothnia it is 2–4 ◦ C.
In spring, after the melting of the ice, a thin upper layer is heated due to solar ra-
diation. The surface waters quickly reach the temperature of maximum density Tm
Table 2.2 The main characteristics of the Baltic Sea stratification

Layer Thickness Maintenance Occurrence
Upper layer 40–80 m Wind, convection All year

Surface layer 10–20 m Heating Summer
Summer thermocline 5–10 m Wind, Sun Summer
Dicothermal layer 5–10 m Cold winter Summer
Halocline (HC) 10–20 m Advection, water All year
budget, wind, convection
Lower layer HC–bottom Advection All year
Secondary halocline 125 m–bottom Advection Transient
(1.5–3 ◦ C). In the southern Baltic the wintertime surface temperature may remain
above Tm . After the temperature has become higher than Tm , the surface waters be-
come lighter than the water masses below, the convection stops and the thermocline
is formed separating the warm surface layer from the remarkably colder waters.
Wind and solar radiation absorbed in the surface layer affect the deepening of the
thermocline. The thickness of the thermocline is typically 5–10 m, and its shape
may vary substantially.
During summer the warm surface layer and the thermocline are at the top of the
upper layer. The surface layer is not always homogeneous but can possess a tran-
sient echelon (micro)structure consisting of minor thermoclines. The well-mixed
surface layer is often defined to be such a layer where the vertical change of temper-
ature does not exceed some prescribed value, say 0.1 ◦ C/m (Table 2.2). Usually the
thermocline can be easily determined from a single profile, but there are also cases
where the interpretation is complicated. A statistical analysis of the distribution of
the thermocline depth has been presented by Alenius and Leppäranta (1982).
The seasonal thermocline in summertime is located at a depth of 15–30 m in
all basins of the Baltic Sea. Its climatic conditions exhibit large variability in the
Baltic region. Its formation starts in the southern Baltic Sea at the beginning of May
but in the Bay of Bothnia only one month later. The deepening of the thermocline
due to autumn cooling starts in the north already in August, whereas in the south it
happens one month later. The sea surface temperature follows the air temperature
with a certain lag because the heat capacity of the sea is much larger than that of the
atmosphere. In other words the sea reacts relatively slowly to the heat input from
the atmosphere because of its big thermal inertia.
During summer the thickness of the surface mixed layer deepens to some ex-
tent, mainly due to the mechanical mixing caused by the wind forcing. The summer
thermocline is strong, with a temperature drop by up to about 10 ◦ C across a few
meters distance. It prevents to a large extent the wind-induced mixing from affect-
ing the layer below. The strongly weakened exchange of both material and heat
has important effects on biogeochemical processes. The relatively warm and calm
summertime weather restricts the deepening of the surface mixed layer to a depth
of 10–20 m only. The thermocline suppresses vertical mixing, favouring the fresh
waters from rivers to stay in the surface layer. This in turn strengthens the density
difference across the thermocline. Thus both the warming and fresh water content of
the surface layer enhance the stability of this layer. In turn, the bottom layer salinity
increases in summer.
A specific feature of the Baltic Sea and many other ice-covered seas is a di-
cothermal layer at the bottom of the upper layer. The formation of such a layer
takes place in the following way. During the spring the upper layer starts to warm
from the surface while the temperature in the lower part is close to the temperature
of the density maximum or to the freezing point in the north. On the other hand the
bottom layer water is throughout the year at temperatures between 3 ◦ C and 6 ◦ C
but has a much higher density than the upper layer water due to higher salinity.
The dicothermal layer remains throughout the summer and disappears only in the
autumn convection. Consequently, the temperature may have a four-layer structure
during the summer season: surface layer, thermocline, dicothermal layer and lower
(bottom) layer.
Usually in late August or early September, depending on the latitude, the energy
budget of the sea surface becomes negative, the surface waters cool, become heav-
ier and sink due to convection. Simultaneously the sea surface releases heat to the
atmosphere through turbulent fluxes. This all leads to a deepening of the surface
mixed layer and to a weakening of the thermocline, i.e., a weakening of the verti-
cal temperature gradient3 (Fig. 2.4). This process continues during autumn. At the
same time the lower parts of the upper layer are still warming due to slow thermal
diffusion. For example, at a depth of 30 m the temperature maximum is reached
only in October while on the surface the highest values are usually measured during
late July–early August. During late autumn or early winter the upper layer becomes
isothermal as a consequence of thermohaline convection and wind-induced mixing.
However, it is important to recognize that the mixing does not destroy the halocline,
which remains throughout the year in the central basins.
After the surface temperature has dropped below the temperature of maximum
density, transient weak winter thermoclines may form close to the surface but in
general mixing continues due to forced, mechanical convection, if the necessary
cooling is allowed by the surface heat flux. It is worth noting that as the temper-
ature of maximum density decreases with increasing salinity, the strength of this
maximum decreases at the same time. In other words the difference between the
maximum density and the density at the freezing point decreases with increasing
salinity. The inverse winter thermocline is therefore more stable in fresh waters than
in brackish waters.
In fact the permanent winter thermocline, where temperature increases approx-
imately from the freezing point to the temperature of maximum density, is located
3 The temperature difference between the surface layer and the lower part of the upper layer (above
the halocline) gradually weakens because not all the heat energy is mixed downwards due to con-
vection. A part of the heat energy is released to the atmosphere by turbulent heat fluxes and thus
the surface water cools down and its temperature difference in comparison to the waters below the
thermocline is reduced.
Fig. 2.4 Typical profiles of temperature T , salinity S and σT = density − 1000 kg/m3 in the
Gotland Deep in 2003: (A) Winter; (B) Spring; (C) Summer; (D) Autumn. Source: Finnish Institute
of Marine Research; prepared by Riikka Hietala. From Leppäranta and Myrberg (2009)
usually at the halocline. But due to the effect of horizontal advection the water tem-
perature beneath the winter thermocline can be much higher than the temperature of
maximum density. Ice forms in the Baltic Sea annually for 5–7 months. Landfast ice
occurs in coastal and archipelago areas; significant salinity stratification may form
even in shallow waters in the neighbourhood of river mouths. Further offshore drift
ice fields exist. Mechanically forced mixing continues due to the motion of the ice
with similar strength as in the ice-free winter season.
2.3 Circulation Dynamics
2.3.1 Basic Principles
The system of ocean currents is three-dimensional (3D). Their dynamics is gov-

erned by the Navier–Stokes equations and the continuity equation. These equations
express the fundamental laws for the conservation of momentum and mass in dif-
ferential form for a continuous medium. In ocean dynamics the water is assumed to
be incompressible and the Boussinesq approximation4 is used, stating that density
differences are sufficiently small to be neglected, except where they appear in terms
multiplied by the acceleration due to gravity. The basic equations of ocean dynamics
are (see, e.g., Cushman-Roisin and Beckers 2011):
∂u 1
+ u · ∇u + 2 × u = − ∇p + ∇·τ + ν∇ 2 u, (2.1)
∂t ρ
∇ · u = 0, (2.2)
where u = (u, v, w) is the current velocity, is the Earth’s rotation rate (Ω =

0.7292 × 10−4 1/s), ρ is density, p is pressure, τ is the Reynolds stress tensor and ν
is the molecular viscosity. Traditionally, the x-axis is directed to the east, the y-axis
to the north and the z-axis upwards. Equations (2.1), (2.2) are purely dynamical. For
the complete circulation system the equation of state for sea water and equations
describing the budget of heat and salt must be added. A more detailed treatment
of the resulting system of equations in the rotating reference frame is presented in
Chap. 3, Sect. 3.3.2, and in Chap. 4.
The molecular viscosity is generally neglected in the circulation theory and the
horizontal part of the Coriolis acceleration is included. There is a strong distinction
between the horizontal and vertical directions due to the fact that gravity strongly
limits vertical motions. The horizontal currents in the Baltic Sea have a typical mag-
nitude of 10 cm/s, whereas the vertical velocities are typically less than 0.1 mm/s.
This feature, a much more prominent property of the velocity field in the shallow
4 After Joseph Valentin Boussinesq (1842–1929).

Table 2.3 Typical values for different terms of the equations of motion (units 10−6 m/s2 )
Inertia Advection Coriolis Pressure gradient Internal friction (horizontal and vertical)
U/T U 2 /L fU ρ −1 H p/L AH U/L2 Av U/H 2

0.2 2 10 10 4 10
Baltic Sea compared to that in the open ocean, implies that the equation for the ver-
tical component of motion can be normally simplified with the use of the hydrostatic
approximation.
Let us denote the horizontal and vertical velocity with U = (u, v) and w, respec-
tively, and let ∇H be the horizontal gradient operator. The Reynolds stress is taken
in the first-order approximation as τ = 2A · ε̇, where A is the eddy viscosity tensor
and ε̇ is the strain rate tensor. Due to the anisotropy of the ocean dynamics the mix-
ing length is much larger in the horizontal than in the vertical direction. Therefore
the eddy viscosity tensor components can be taken as Axx = Ayy = AH , Azz = Av ,
and Apq = 0 when p = q. The equations of ocean dynamics can be written in the
following form:

∂U ∂U 1 ∂ ∂U
+ U · ∇H U + w + f k × U = − ∇H p + AH ∇H 2
U+ Av ,
∂t ∂z ρ ∂z ∂z
(2.3)
∂w ∂p
+ ∇H · U = 0, = −ρg, (2.4)
∂z ∂z
where f = 2Ω sin φ is the Coriolis parameter, φ is latitude, k is the unit vector in
the vertical direction, U = |U| and g is acceleration due to gravity.
In the Baltic Sea dynamics the typical scales are: U = 10 cm/s, T = 5 days (syn-
optic scale), L = 50 km and H = 25 m. Representative eddy viscosity coefficients
are AH = 105 m2 /s and Av = 0.05 m2 /s. The sea level measurements in the Baltic
Sea show that the inclination of the sea surface is typically β ∼ 1 mm/(1 km). Then
the horizontal pressure gradient in the surface layer is ρ −1 ∇p = −gβ ∼ 10−5 m/s2 .
These estimates lead to the characteristic magnitudes of the horizontal equation of
motion in Eq. (2.3), presented in Table 2.3 and often used in the scaling of the dy-
namical equations.
The vertical velocity scale W is obtained from the continuity equation: W U/H =
U 2 /L, and thus all the advection terms have equal magnitudes. The dominating
terms are the ones representing the contributions of the Coriolis acceleration, the
pressure gradient and the vertical friction. The last term includes the transfer of
wind stress to the sea surface and the damping of motion by bottom friction. If the
pressure gradient vanishes, the Coriolis acceleration and vertical friction provide
the leading balance, but in deep waters the influence of the vertical friction becomes
small. Advection and horizontal friction are smaller than the Coriolis acceleration
by almost one order of magnitude. However, advection plays an important role in
intensive dynamics, i.e., when U becomes large. Horizontal friction becomes impor-
tant near the coasts when L decreases. The inertia term is significant in relatively
fast processes with the time scale T of only some hours and in so-called inertial
oscillations.
Given the small size of the Baltic Sea, it is acceptable to assume the Coriolis
parameter to be constant. Its frequently used value for the particular geographical
location is f = 1.26 × 10−4 1/s and corresponds to φ = 60◦ N. The use of this value
is particularly suitable for the description of the dynamics in the Gulf of Finland
(Chap. 6). A further simplification is to consider motions on a local plane projec-
tion, which rotates around the local vertical axis (so-called f -plane approximation).
The beta-plane approximation, in which the Coriolis parameter varies in the north–
south direction, is infrequently used in the Baltic Sea context because of its small
size. The equations would be much more complicated in spherical coordinates. The
boundary conditions are rather simple in the Baltic Sea. Nearly everywhere there is
a passive solid boundary. Inflow of fresh water takes place in the river mouths, and
a dynamic inflow-outflow system exists in the Danish Straits. The water exchange
in the Danish Straits has a dominant role for the circulation of deep water and in-
directly for the general circulation of the Baltic Sea through regulating the general
sea level elevation.
2.3.2 Barotropic and Baroclinic Flows and Rossby Radii
The main dimensionless numbers in the Baltic Sea dynamics are the (external
or barotropic) Rossby number Ro = U/f L, the Froude number Fr = U/(Lg)1/2
and the (vertical and horizontal) Ekman numbers Ekv = Av /(U H ) and EkH =
AH /(U L). They describe the significance of, respectively, the Coriolis acceleration,
gravity acceleration, vertical friction and horizontal friction (see, e.g., Leppäranta
and Myrberg 2009, for details).
There are two basic types of large-scale flow fields in the marine environment:
barotropic and baroclinic circulation. The motions are barotropic, for example,
when the density of sea water is constant. More generally, in the barotropic mo-
tions isopycnals (isolines of density) are parallel to isobars. In the baroclinic case
isopycnals and isobars are inclined with respect to each other (the inclination angle
usually varies with depth). Consequently, the pressure gradient and the resulting cur-
rents vary also in depth. The baroclinic and barotropic components of motions can
be formally separated in a linear framework and are usually called baroclinic and
barotropic modes. In practice, in homogeneous waters the circulation is barotropic
and in stratified waters the baroclinic circulation needs to be considered.
A number of case studies based on current measurements have shown that the
response of the Baltic Sea is baroclinic to wind events with a duration of more than
50 hours, whereas in short-term wind events, of duration 10–40 hours, the response
is barotropic. The response to wind events shorter than 10 hours quickly vanishes.
Differently from the open ocean, where the energy of the two modes is comparable,
the energy in the baroclinic mode exceeds that of the barotropic mode by one order
of magnitude in the Baltic Sea. Therefore, the Baltic Sea cannot be treated as a
homogeneous water body.
Many properties of geophysical flows are governed by the Rossby radius (of
deformation). This fundamental length scale of planetary-scale flows for a particular
mode of motion is, in essence, the ratio of the phase speed of the waves associated
with this mode to the Coriolis parameter. As the barotropic mode is associated with
long shallow-water5 waves (which propagate very fast) and the baroclinic mode(s)
with internal wave(s) (that are much slower, see Sect. 3.3.1.1 in Chap. 3 for more
detailed information), the relevant wave speeds and the resulting Rossby radii are
drastically different.
The baroclinic (also called internal) Rossby radius R1 of deformation (under-
stood here as the first eigenvalue of the baroclinic Rossby radius) is one of the
fundamental length scales in geophysical fluid dynamics that inter alia defines to
a large extent the typical scale of mesoscale dynamic features such as (synoptic)
eddies, fronts and local jets. The above-discussed specific features of the Baltic Sea,
especially the strong stratification, have substantial implications on its dynamics al-
ready at the level of this fundamental scale. Owing to strong gradients, R1 is very
small in the Baltic. Its typical values are between 3 and 10 km in the open Baltic
(Fennel et al. 1991). Differently from the barotropic Rossby radius, the baroclinic
Rossby radius depends not only on changes in local stratification conditions but also
on changes in longer time scales. Osiński et al. (2010) found that the major inflow
in the winter 2002/2003 increased the value of R1 in the southern Baltic from about
4 km (during the pre-inflow period) to more than 9 km.
Such spatial scales given by the Rossby radius for the open Baltic Sea can easily
be resolved by contemporary numerical models. Namely, to properly describe the
small-scale eddies, fronts and jets, it has been suggested that in numerical modelling
the grid size should be 1/2–1/3 of R1 (Drijfhout 1989; Lindow 1997). The situa-
tion is more complicated in shallow and strongly stratified sub-basins of the Baltic
Sea such as the Gulf of Finland (see Chap. 6) or the south-western Baltic Sea, at
which most of the efforts of the developed technology for the preventive manage-
ment of pollution have been targeted. In these basins the requirements for numerical
modelling are even higher than for the open Baltic Sea. Based on a dataset of about
1800 observations, the baroclinic Rossby radius was estimated to be on average in
the range of only 2–4 km in the Gulf of Finland (Alenius et al. 2003) whereas it
has also extensive spatial and temporal variability. Similar and even smaller val-
ues, down to 1 km, have been recently obtained for the south-western part of the
Baltic Sea (Osiński et al. 2010). In shallow coastal regions and in the eastern Gulf
of Finland R1 is sometimes as small as about 500 m (Nekrasov 1999; Nekrasov and
Lebedeva 2002). In such a case the numerical models need very high resolution,
down to about 0.25 nautical miles (Andrejev et al. 2010).
The meteorological conditions govern to a large extent the changes in the flow
field, and, not surprisingly, there are similarities in the spectra of wind and current
velocities. Especially in the surface layer the variability of the currents is strongly
5 The term ‘shallow water’ is used here and on some occasions below to distinguish the situation
where the typical length of waves of a particular class (Rossby waves, internal waves, surface
waves, etc.) considerably exceeds the water depth.
determined by the wind forcing, but, however, inertial oscillations (see below) and
long shallow-water waves are at times excited by the internal dynamics of the sea.
The wind-driven transport of surface waters builds up pressure gradients, which
provide forcing to deeper waters. The resulting motion is called wind-generated
secondary circulation.
2.3.3 Dynamics of Surface Currents
There are four mechanisms to induce currents in the Baltic Sea: wind stress at the
sea surface, surface pressure gradient, thermohaline horizontal gradient of density
and tidal forces. Furthermore, the currents are steered by the Coriolis acceleration,
topography and friction, all of which play a relatively large role in the Baltic Sea ow-
ing to its location at high latitudes, its shallowness and to its complicated bathymetry
and geometry. Voluminous river runoffs can produce relatively strong local changes
in the sea level height and consequently contribute substantially to the formation of
currents. Due to the small size of the Baltic Sea basins, friction caused by the bottom
and the shores damp the currents remarkably. The general circulation is typical for
a stratified system. Inflowing waters into a basin interleave at the depth where the
ambient water has an equal density. While the fresher water from rivers goes into
the upper layer, the more salty water masses from the North Sea, optionally partly
mixed with the existing waters in the Baltic Sea, go into a certain intermediate or
lower layer.
In the longest time scale—from several months to years—a baroclinic basic cir-
culation appears, independent of short-term variations in the wind properties. This
pattern is mainly driven by the positive fresh water budget and the resulting large
horizontal gradient of salinity. The fresh waters leave the Baltic Sea in the near-
surface layers, whereas the inflow of saline water masses takes place in the lower
layer.
In short time scales (1–10 days) the currents are caused by the wind stress. Due
to the large variability of the winds, the resulting long-term wind-driven mean circu-
lation is weak, and transient currents are larger than the average ones by one order
of magnitude. Drift currents produce in coastal areas upwelling and downwelling
features that are affected by Kelvin-type waves. The water is laterally mixed by
mesoscale eddies and deep-water circulation (see, e.g., Fennel and Sturm 1992; Lass
and Talpsepp 1993; Raudsepp 1998; Stigebrandt et al. 2002; Elken and Matthäus
2008). In the time scale from 1 hour to 1–2 days, there are several periodic dy-
namical processes. The most important are seiches (less than 40 hours) and inertial
oscillations (13.2–14.5 hours, discussed in the next section). Another phenomenon
of specific interest in the Baltic Sea conditions is the Ekman drift in the upper layer
(Sect. 2.3.5) that is not only responsible for frequent up- and downwellings (Myr-
berg and Andrejev 2003; Lehmann and Myrberg 2008; Lehmann et al. 2012) but
also for large surface transport.
In conclusion, the observed features of the long-term mean surface circulation
in the Baltic Sea are created by a non-linear combination of the wind-independent
Fig. 2.5 Inertial oscillations

in the Baltic Sea in August
17–24, 1933 according to
Gustafsson and Kullenberg
(1936). The circular diagram
shows the current velocity on
August 21. From Leppäranta
and Myrberg (2009)
baroclinic mean circulation and the mean wind-driven circulation. Which one is
more important is difficult to determine in the case of such a strongly non-linear
system. The predominant component depends on the case studied and on the time
scale under investigation.
2.3.4 Inertial Oscillations
In physical oceanography inertial oscillations refer to the circular motions of water

in which the inertial and Coriolis acceleration balance each other. This situation oc-
curs often, e.g., when wind ceases to drive a surface current. The observation of in-
ertial oscillations in the Gotland Sea (Fig. 2.5) by Gustafsson and Kullenberg (1936)
was the first time when this phenomenon was documented in physical oceanogra-
phy. Their study was based on current measurements collected during the summer
period in 1931–1933. Later on it became clear that the inertial motion is one of the
dominating features in the dynamics of the Baltic Sea.
Energy spectra of Baltic Sea current velocities show that the motions with the
inertial period of 13.2–14.5 hours are, on average, the strongest motions with a
more or less fixed period. At periods between 10 and 12 hours there are indications
of interaction between tidal waves and inertial oscillations. Also relatively intense
motions with shorter periods (8, 7.5, 6.5 and 4 hours) have been found.
When the external forcing ceases, the rotational effects take over and the inertial
oscillation becomes visible. Mathematical analysis of this situation is simple to
√per-
form using complex variables. Denote the velocity as q = u + iv, where i = −1.
The basic equation for the inertial motion is obtained from the horizontal equations
of motion. The external forcing and friction terms are zero and the flow field is
assumed to be spatially homogeneous. Thus the equations of motion are reduced to
dq
= −if q. (2.5)
dt
The multiplication of q by the imaginary unit −i on the right-hand side of Eq. (2.5)
(the operation −iq) means that the presence of the Coriolis acceleration results
in a permanent inclination of the velocity vector q by 90◦ to the right from its
instantaneous direction. This first-order equation can be directly integrated to reveal
that
q = q0 e−if t . (2.6)
Therefore, the current speed |q| = |q0 | is constant and the direction of the velocity
rotates by an angular velocity −f (clockwise in the northern hemisphere where
f > 0). Each resulting cycle draws the so-called inertial circle with a radius RI and
period TI :
|q0 | 2π
RI = , TI = . (2.7)
f f
In the Baltic Sea the inertial period is 13.2–14.5 hours (13.8 hours at latitude
60◦ ). If the initial velocity is 10 cm/s, the radius of the circle is about 800 m. In
reality the size of the circle reduces in time due to frictional forces. This can be
modelled in a simple way by adding a damping term −rq on the right hand side of
Eq. (2.5) where r is a friction coefficient. The solution is then q = q0 e−rt e−if t . The
quantity 1/r is obviously the relaxation time6 of the system. The data of Gustafsson
and Kullenberg (1936) suggest that 1/r ≈ 1 week in summer conditions. The iner-
tial motion is in different phases above and below the thermocline. Near the coasts
it is quickly damped due to friction.
2.3.5 Ekman Drift
The Swedish oceanographer V.W. Ekman explained theoretically the formation of

wind-driven surface-layer currents already more than 100 years ago (Ekman 1905).
The wind induces a shear stress onto the sea surface. The vertical turbulent friction
transfers the momentum of the wind downwards while the Coriolis acceleration
turns the flow direction to the right in the northern hemisphere. The Ekman equa-
tions describe the steady state of this process in a horizontally homogeneous ocean.
6 In a spin-down or spin-up process, during the relaxation time (also called e-folding time scale)
the speed (or any other suitable measure of the process) changes e times compared to its original
value.
They can be written in the following form for a sea with finite depth:
d 2q
Av − if q = 0. (2.8)
dz2
Their general solution is

1+i f
q = C1 eλz + C2 e−λz , λ= . (2.9)
2 2Av
√
In the classical Ekman layer theory the length D = π 2Av /f is defined as the
Ekman depth. When the depth of the sea H D, the velocity profile draws the fa-
mous Ekman spiral. At a depth of z = D the current speed is e−π times smaller than
the surface speed (that is, ≈4 % of the surface speed) and the direction has rotated
clockwise by the angle of π ; consequently, the current at this depth is opposite to
the surface current. At the sea bottom there is also an Ekman layer where the veloc-
ity decreases to zero at the bottom with a spiral profile. For a representative eddy
viscosity in the whole Ekman layer, Av ∼ 10−3 –10−2 m2 /s, we have D ∼ 10–40 m.
In the Baltic Sea the eddy viscosity is large and H ∼ D. This means that the speed
decays faster with depth and the rotation is less evident than in the deep ocean, and
the surface and bottom Ekman layers may merge together. The boundary conditions
for Eq. (2.8) are:

∂q τa
qz=0 = 0, Av = = , (2.10)
∂z z=H ρ
where z = 0 is the sea bottom, z = H is the sea surface and τa is the (scalar) wind
stress. The relevant particular solution of Eq. (2.8) is
sinh λz τa
q = qs = 0, qs = tanh λH, (2.11)
sinh λH ρAv λ
where qs is the surface velocity. The resulting ‘truncated’ spiral approaches the
Ekman spiral as λ−1 H (Fig. 2.6). The vertically integrated velocity of the flow
in the Ekman layer (called Ekman transport) is
H τa − τ0
Q= q dz = i , (2.12)
0 ρf
where τ0 is the bottom drag. If τ0 = 0, Eq. (2.12) expresses the classical result of
Ekman transport that is perpendicular to the wind stress. In the Baltic Sea this comes
true for the upper layer in deep areas where the halocline exists. In shallow areas
the bottom friction reduces both the transport volume and its veering from the wind
direction. Using the velocity solution (2.11) we have

τa 1
Q = −i 1− . (2.13)
ρf cosh λH
Fig. 2.6 The Ekman profile

of current velocity in shallow
seas. Here d is the sea depth
and D is the Ekman depth.
From Leppäranta and
Myrberg (2009)
Thus the Ekman transport for a finite sea depth and zero bottom friction depends on
the eddy viscosity via the parameter λ.
The Ekman solution holds qualitatively in the Baltic Sea: The current speed re-
duces and the direction turns to the right with increasing depth. However, the ap-
proximation that the vertical eddy viscosity is constant is not always adequate here
and that is the most probable reason why the profiles observed in nature differ from
the theoretical one. For large sea depths the observed surface currents are about 2–
3 % of the wind speed, but the turning angle is 20–30◦ . This simple surface drift
model is still used in many practical applications like in forecasting the drifting of
floating objects. Figure 2.7 presents observations of freely moving sea ice and sur-
face layer water velocity. In a qualitative sense the Ekman spiral is observed beneath
the ice7 (Leppäranta 2010). The ice drifts by about 15◦ to the right from the wind
speed, with a drift speed about 2 % of the wind speed.
Probably the most widely known application of the Ekman theory explains the
phenomenon of wind-induced upwelling. If the wind blows parallel to the coast so
that the coast remains on the left on the northern hemisphere, the Ekman transport
is directed offshore. The drift of water to offshore is compensated by a vertical
movement of water from deeper layers towards the sea surface.
It is debatable whether, or how exactly, the structure of the Ekman layer is rep-
resented in the ocean modelling efforts. On the one hand, this phenomenon is obvi-
ously not replicated in two-dimensional (2D) models of circulation and in models
of the deep ocean with the thickness of the upper layer comparable with the Ek-
man layer depth. In cases where specifically the drift of substances or objects in the
uppermost layer is of importance (e.g., studies of pollution or oil propagation) it
has been customary to either insert artificially a similar drift of the uppermost layer
(e.g., Periáñez 2004) or to enhance the direct wind impact on the objects at the sea
surface (Verjovkina et al. 2010).
7 The Ekman spiral was first observed under the ice by Fridtjof Nansen in the 1890s. The spiral was
documented much later for the open ocean.

Fig. 2.7 The wind-driven

motion of ice and the currents
at different depths in the
Ekman layer in the Bay of
Bothnia, April 1975
(Leppäranta 1990)
On the other hand, contemporary 3D models of shallow shelf seas with very
high vertical resolution (the thickness of the uppermost model layers about 10 %
of the Ekman layer depth) apparently are capable of replicating, at least qualita-
tively, the Ekman spiral. Such models have been extensively used, for example, for
the modelling of the intensity and statistics of upwelling phenomena in the Baltic
Sea (Myrberg and Andrejev 2003; Lehmann and Myrberg 2008). It has been ob-
served in various modelling examples during the last decade (from Andrejev et al.
2004 onwards) that a very thin upper layer (of only a few metres) can even move
in the opposite direction to the layers below it, and that the overall surface circu-
lation pattern may, counter-intuitively, contain anticyclonic gyres in the northern
hemisphere (Beletsky et al. 2006; Soomere et al. 2011). This means in practice that
the actual Ekman-type layered flow may contain surprisingly thin layers and/or its
structure may substantially deviate from the theoretical predictions. These matters
of current dynamics are discussed to some extent in Chap. 9. This is a field where
intense research is also ongoing to understand the reasons behind different kinds of
behaviour and to quantify the actual dynamics more exactly (Heinloo and Toompuu
2011, 2012).
2.3.6 Geostrophic Flow
The geostrophic flow—the current driven by the pressure gradient on a rotating

planet—is a steady, frictionless current in which the Coriolis acceleration is bal-
anced by the pressure gradient. In the vertical direction hydrostatic balance is as-
sumed. The momentum equation is written for the geostrophic balance as
1 ∂p
if q = − ∇p, = −ρg. (2.14)
ρ ∂z
The solution is q = i∇p/(ρf ), thus the flow direction is perpendicular to the pres-
sure gradient (i.e., parallel with the isobars) with higher pressure on the right when
facing the direction of the motion in the northern hemisphere. In the barotropic case
the geostrophic flow is depth-independent, whereas in the baroclinic case the flow
varies with depth. The geostrophic solution is horizontally non-divergent and there-
fore w = 0 everywhere. Consequently, the flow conserves its forcing, the pressure
field. Taking the derivative with respect to the depth gives the relationship
∂p g
= −i ∇ρ, (2.15)
∂z ρf
known as the thermal wind law as it was first found in meteorological research
(Holton 1979; Cushman-Roisin and Beckers 2011). If the density changes in the
horizontal direction, then the geostrophic current will vary in the vertical direc-
tion. Such a situation is present in baroclinic situations; in barotropic situations the
geostrophic current is independent of the depth.
The geostrophic flow in the surface layer results from the inclination of the sea
surface. If the tilt of the sea surface towards the east is β, the northward directed
flow speed is v = gβ/f . In practice β ∼ 10−6 (∼1 mm/1 km), thus resulting in
v ∼ 10 cm/s. In a two-layer system, the flow in the surface layer v1 is obtained
from the tilt of the sea surface, and the difference between the flow in the surface
and lower layer is obtained from the density difference between these two layers
according to the thermal wind law:
g ρ1 − ρ2 ∂H1
v1 − v2 = , (2.16)
f ρ2 ∂x
where v1 and v2 are the flow speeds in the surface and bottom layers, ρ1 and ρ2 are
the corresponding densities, and H1 is the thickness of the surface layer.
The determination of the geostrophic flow includes one of the fundamental prob-
lems of physical oceanography—the problem of a reference level where the pressure
gradient is known. In the deep ocean it is usually assumed that below the permanent
thermocline layer, at a depth of 1–2 km, the isobars are horizontal and thus the
geostrophic flow vanishes (‘level of no motion’). In the Baltic Sea no such depth
can be assumed. An approximate solution to this problem has been to consider the
deep geostrophic flow as small (≈zero). This assumption is inaccurate because the
near-bottom frictional layer is located at a depth where the geostrophic flow is still
significant. But even if the absolute flow speeds were inaccurate, the relative currents
inside the water body as obtained from the thermal wind law (2.15) are adequate.
The exact determination of the surface tilt would solve the problem concerning a
reference level. However the tilts of isobars are usually very small (10−6 ) and their
adequate evaluation is a highly nontrivial measurement problem. One of the first
proofs for the existence of a geostrophic flow are the measurements of gradients of
sea level in the Great Belt and the comparisons of the results to measured flows in
the strait (Fig. 2.8).
Fig. 2.8 The tilt of the sea

surface across the Great Belt
calculated according to the
sea level differences between
Korsör and Slipshavn and the
relation between the tilt and
the surface currents measured
at the lightship Halskov Rev
according to Dietrich et al.
(1963). From Leppäranta and
Myrberg (2009)
Consider the full horizontal equation of motion (2.3) for the combination of the
Ekman theory and the geostrophic theory. It is customary to assume that the result-
ing current velocity q = qE + qG is a linear superposition of the Ekman solution
qE and the geostrophic current qG . In a stationary regime and assuming that the
advection and horizontal friction can be omitted, Eq. (2.3) is reduced to
1 ∂ 2 (qE + qG )
if (qE + qG ) = − ∇p + Av . (2.17)
ρ ∂z2
If it is furthermore assumed that in the last term the influence of the geostrophic
flow is negligible, the Ekman equation and the geostrophic equation can simply be
summed up. As both the equations are linear, the solution of the resulting equation
is the sum of the Ekman and geostrophic solutions. Especially in the barotropic case
qG = const and thus ∂ 2 uG /∂z2 = 0.
2.3.7 Surface Circulation of the Baltic Sea and the Gulf of Finland
First studies of the Baltic Sea circulation were based on observations collected on-
board lightships already since the beginning of the 1900s (Witting 1912; Palmén
1930). The measurements were based on quite simple instrumentation (surface
drifters, ‘flow crosses’, deployed and followed from lightships) but they provided
the basic knowledge, and hence already 100 years ago an overall picture of the sur-
face circulation was available for the northern Baltic Sea. These early measurements
showed that the mean circulation in the Baltic Sea main basins (Gotland Sea, Gulf
of Finland, Gulf of Riga, Sea of Bothnia and Bay of Bothnia) is cyclonic (counter-
clockwise, Fig. 2.9). This feature, although heuristically obvious, has important re-
flections in the transport processes in the Baltic Sea. The mean circulation transports
salt and heat and therefore the water is warmer and saltier in the eastern sides of the
basins than in the western parts. In the narrow, west-east oriented Gulf of Finland
such a difference induced by mean circulation is observed between the northern and
southern side (Fig. 2.9).
The Baltic Sea does not host permanent, stable current structures such as the Gulf
Stream or the Kuroshio in the Atlantic and Pacific Oceans. Even though the long-
term mean currents are weak, average speeds are some 5 cm/s and the persistency of
the circulation system is in some areas relatively strong (Andrejev et al. 2004); see
also Chap. 9 for the discussion of semi-persistent transport patterns). During storms
the wind drift currents can reach 50 cm/s, in straits up to 100 cm/s; on average the
speed of the surface current is 2–3 % of the wind speed and the direction is 20–30◦
to the right from the wind direction. This is a realization of the Ekman spiral on the
surface.
Palmén (1930) defined the persistency R of the direction of the mean circulation
as the ratio
| U
|
R= . (2.18)
U
The vector | U
| is the mean current velocity, while U
is the mean current speed.
If the direction is constant the persistency is 100 %. When the persistency reaches
zero also the mean flow and net transport of water are, on average, equal to zero. In
an asymmetric, bimodal coastal current, the persistency equals the difference of the
modes; for instance, if the flow is 75 % to west and 25 % to east with the same speed,
the persistency is 50 %. In practice the persistency of the surface circulation is 20–
40 %, and can be even higher in specific locations such as north of the longitudinal
axis of the Gulf of Finland (Andrejev et al. 2004).
The described general current pattern also becomes visible from the directional
distribution of currents. Figure 2.10 shows the distributions for current velocities
(‘current rose’) in the Sea of Bothnia. Near the coast the currents are parallel to
the shoreline and thus have generally a bimodal structure. In the open sea the steer-
ing effect of the coast is negligible, the current system is more isotropic and thus
the persistency of currents is lower. This is partially because the offshore current
field contains many mesoscale vortices that manifest themselves through short-term
fluctuating currents.
Fig. 2.9 Long-term mean surface circulation (cm/s) in the Baltic Sea according to lightship mea-
surements. The persistency of currents is marked as percentages (Palmén 1930; Sjöberg 1992).
From Leppäranta and Myrberg (2009)
2.3.8 Three-Dimensional Water Circulation
The above quasi-permanent cyclonic circulation takes place in the upper layer of
the Baltic Sea. In the lower layer the water dynamics is rather different from that
Fig. 2.10 ‘Current roses’ based on measurements outside Rauma, Sea of Bothnia, during the In-
ternational Gulf of Bothnia Year 1991 (Murthy et al. 1993). The graphs describe the cumulative
probability of current speeds at 5 cm/s spacing in different directions. The station CM1 is closest
to the coast (10 km) whereas the station CM7 is 55 km from the coast. The x-axis is directed north-
ward and the y-axis is directed eastward. The measurement depth is 8 metres. From Leppäranta
and Myrberg (2009)
in the upper layer and the flow is steered by channels and restricted by various sills
(Fig. 2.2). The ‘deep’ water that penetrates from the North Sea to the Baltic Sea
moves first southwards through the Belt Sea and Øresund to the Arkona Basin. From
there it proceeds further through the Bornholm Channel to the Bornholm Deep.
Then the deep water moves through the Stolpe Channel to the Gotland Deep. Note
that the sill between the Bornholm Basin and the Western Gotland Basin is shallow
and the water cannot reach directly the Western Gotland Basin. From the Eastern
Gotland Basin a small part of the water flows into the Gdańsk Deep because there is
no sill between these basins (Fig. 2.2). However, most of the water flows northwards
to the Gotland Deep. After that the water movement continues to the Fårö Deep and
further on to the Northern Gotland Basin. Due to the shallow connection to the
Eastern Gotland Basin the deep water cannot enter the Gulf of Riga.
In the north the Gotland Sea is bounded by the Archipelago Sea and by the Åland
Sea, preventing the saline deep waters of the Gotland Sea from flowing into the Gulf
of Bothnia. The deep waters of the Northern Gotland Basin flow partly southwest
to the Western Gotland Basin and produce the deep water for the Landsort Deep
and the Norrköping Deep. Due to the absence of a sill towards the Gulf of Finland
the waters from the Northern Gotland Basin can enter there and contribute to the
variation of extremely complex estuarine dynamics in this basin. These aspects are
described in more detail in Chap. 6.
The interaction between the upper and lower layers is quite restricted in the Baltic
Sea due to the strong stratification. In the Kattegat the dense waters originating
from the North Sea form a deep-water pool whereas the fresher Baltic waters are
located in the surface layer. The deep water circulation is characterized by dense
bottom currents in the inflowing saline water at the mouth area of the Baltic Sea.
Convection and mechanical mixing, entrainment and vertical advection of waters
lead to interactions between the upper and lower layers in other parts of the Baltic
Sea (Fig. 2.11).
The water is effectively recirculating in the Baltic Sea in spite of the existing
low-permeable halocline. Döös et al. (2004) coined the term ‘haline conveyor belt’
to describe at a general level the overturning circulation of the Baltic Sea in anal-
ogy to the deep-water conveyor belt of the World Ocean. The vertical overturning
circulation consists of many important factors: the gravity-driven dense bottom cur-
rents of the inflowing waters from the North Sea, the entrainment of ambient surface
Fig. 2.11 A schematic of the

large-scale internal water
cycle in the Baltic Sea. The
deep layer below the
halocline is given in the lower
part of the figure. Green and
red arrows denote the surface
and bottom layer circulation,
respectively. The light green
and beige arrows show
entrainment, the grey arrow
denotes diffusion (Elken and
Matthäus 2008)
waters, mixing due to diffusion, interleaving of the inflowing water masses into a
depth corresponding to the level of neutral buoyancy, vertical advection due to the
conservation of mass, and upward entrainment of deep water into moving surface
water in the Northern Gotland Basin.
The presented classical division between the upper and lower layer, where the
halocline is the border between the less dense upper layer and the denser bottom
layer, might be too idealized even if the haline conveyor belt is accounted for in a
greater detail. The modelling and measurement efforts during the last decade have
revealed that even the dynamics of the uppermost layer, well-mixed in temperature,
can be very complicated and fine-scale, long-lived layered structures with a thick-
ness of a few metres exist there. This is a welcome and evident development in
oceanography, where numerical models with extremely high resolution in the verti-
cal direction are available today. These results, which are introduced in a more detail
in Chaps. 4, 5 and 9–11, are of vital importance whenever detailed analysis of the
drift of various substances in surface and near-surface layers of strongly stratified
environments is necessary.
2.4 Numerical Modelling of the Baltic Sea
The start of 3D modelling of the Baltic Sea took place already in the 1970s. The first
attempts in this direction were not able to properly describe the complete 3D baro-
clinic structure of the flow field in this basin with realistic bottom topography and
forcing. Due to the low computational power several simplifying assumptions were
made. Kuzin and Tamsalu (1974) studied the baroclinic circulation of the Baltic Sea
using a flat bottom. Sarkisyan et al. (1975) developed a diagnostic model to study
the interaction between circulation and bathymetry. Pure wind-driven circulation in-
teracting with realistic (albeit strongly generalized) bottom topography was studied
in Simons (1981), Kielmann (1981). Lehmann (1995) was the first to address the
full problem of 3D dynamics of the Baltic Sea in an eddy-permitting resolution, in
Fig. 2.12 Baltic Sea circulation as viewed from modelling results. Average transport per unit
length (m2 /s) for 1981–2004 above (left panel) and below (right panel) the halocline (Meier 2007)
which both wind-driven and baroclinic circulation, interacting with realistic bottom
topography, were taken into account under real atmospheric forcing.
Three-dimensional circulation modelling is today an important tool to improve
our understanding of the circulation dynamics in the seas and oceans. Recent Baltic
Sea models have been applied, for example, to study the mean thermohaline and
wind-driven circulations.
The mean circulation of the entire Baltic Sea was modelled recently by Meier
(2007). The results (Fig. 2.12), not unexpectedly, confirm the main characteristics
of the early findings of Palmén (1930) about the general cyclonic circulation, and
the outcome of earlier attempts of numerical modelling (Lehmann and Hinrichsen
2000; Lehmann et al. 2002) (Fig. 2.13), but also provide new fine-scale character-
istics. The possible relationship between the NAO index and Baltic Sea circulation
is interesting at a climatological time scale. The local wind field over the Baltic Sea
can be related to the large-scale atmospheric circulation via the Baltic Sea Index
(BSI), which is the difference in normalized sea level pressures between Oslo in
Norway and Szczecin in Poland. The BSI is significantly related to the NAO index
and furthermore highly correlated with the mean sea level of the Baltic Sea and the
water exchange through the Baltic Sea (Lehmann et al. 2002).
The mean circulation is likely variable over longer periods, with changes in the
character of wind forcing, heat fluxes and ice extent, fresh water budget and inflow
activity. A hindcast for the period of 1958–2001 showed that yearly averaged surface
velocities (mean over the whole sea area) have increased by 0.21 cm/s per decade
(Jȩdrasik et al. 2008). Based on the presented time series of annually mean current
speeds, one may also interpret this increase as a regime shift that occurred in the
Fig. 2.13 Average barotropic currents for 1992–1995 (cm/s) with the flow stability contours
(Lehmann and Hinrichsen 2000)
late 1980s. Over shorter periods (months), there is evidence that water movements
may take alternative paths as compared to the regular spreading patterns.
As seen even from this very brief summary, the number and variety of present
models and applications for the Baltic Sea is large. The reader is referred to the
review (Omstedt et al. 2004) and books (Feistel et al. 2008; BACC Author Team
2008; Leppäranta and Myrberg 2009) for further details in Baltic Sea circulation
modelling.
Today’s 3D numerical modelling in the Baltic Sea (incl. the examples presented
before) is typically based on the horizontal resolution of 1–5 km and a vertical struc-
ture described by 20–100 layers. In local applications (Zhurbas et al. 2008a, 2008b;
Andrejev et al. 2010; Lu et al. 2012) even higher resolutions are often needed. For
relatively low-resolution simulations targeted to model validation and intercompar-
isons (Myrberg et al. 2010) and to the identification of long-term changes (e.g., Myr-
berg and Andrejev 2006; Meier 2006, 2007; Lehmann et al. 2011), the properly ad-
justed geostrophic wind data (e.g., the gridded meteorological data by the Swedish
Meteorological and Hydrological Institute) have been a popular choice. For higher-
resolution simulations the output of local atmospheric models such as different ver-
sions of the High Resolution Limited Area Model (HIRLAM) (www.hirlam.org)
or the DWD model (www.dwd.de), ERA-40 reanalysis (Uppala et al. 2005) or its
downscalings (Höglund et al. 2009; Samuelsson et al. 2011) are typically used as
meteorological forcing. As the resolution of such analyses is often too coarse for
local hindcast studies and several systematic features of air flow in some basins
are not captured (Keevallik and Soomere 2010), new, high-resolution reanalyses are
urgently needed.
One of the most important inputs in the Baltic Sea modelling is the volumi-
nous river discharge. The relevant data are available, e.g., from the BALTEX Hy-
drological Data Centre in terms of monthly mean values (Bergström and Carlsson
1994). The initial conditions for hydrographic parameters are available, e.g., from
the Baltic Environment Database (BED), which can be used through the Data As-
similation System (DAS, Sokolov et al. 1997). Observations from buoys and tide
gauges can be used for model verification and data assimilation (Myrberg et al.
2010). A big challenge is still the availability and accurate description of the open
boundary conditions in the Danish Straits where only a limited number of data are
available. These problems are often avoided through coupling the Baltic Sea models
with North Sea models.
2.5 Summary: How the Baltic Sea Can Be Replicated

by Numerical Models of Today
This Chapter has introduced the reader to the very specific environment of the Baltic
Sea, especially its topography, hydrography and circulation dynamics. Keeping in
mind the extensive ongoing modelling activities and the ever increasing use of vari-
ous models to operationally forecast important features of the Baltic Sea dynamics,
a question arises: how accurately can the present 3D numerical models describe the
Baltic Sea physics? What kind of problems might still exist?
Proper numerical modelling of the physics and dynamics of the Baltic Sea is
a great challenge and a very demanding task even for the modern 3D hydrody-
namic models because of a number of reasons. These problems and potential short-
comings in the model results should be kept in mind whenever studying particular
model-based analyses or conclusions, especially when interpreting the outcome of
the modelling-oriented chapters of this book. The first task for any 3D model to
solve is the proper description of the stratification. Regrettably, in the Baltic Sea
this has, strictly speaking, not exactly been solved to a satisfactory level. Even the
very best models tend to often exhibit problems to replicate the right position of the
halocline or a realistic vertical gradient of density. There is no unique opinion about
the reasons behind this feature that may substantially undermine the accuracy and
reliability of the simulations of water circulation and spreading of substances over
the entire water column. Fortunately this shortage only insignificantly affects the
results of modelling the velocities and transport in the uppermost layer.
The reasoning behind this problem is manifold. The existing knowledge of the
water budget (in- and outflowing waters, precipitation, evaporation and river runoff),
albeit based on quite a large data set, is not very accurate. An implication from
the listed uncertainties is that the water cycle is not ‘closed’. There is always a
source of errors, which at times become evident as systematic bias in the water level
in operational models (Lagemaa et al. 2011). The performed measurements of the
different constituents of the water budget are not uniform. Moreover, they are highly
unsynchronized and biased, and have very different temporal and spatial resolution.
The meteorological data used for model forcing are often inaccurate and have a
low horizontal resolution, which does not allow a proper description of open sea
meteorological conditions and the land-sea differences due to too few grid points.
This problem still remains today even if the present operational models have high
resolutions, but there is a lack of reanalyses, which are necessary for long-term
hindcasts of spatial distributions of specific fields generated by the current driven
transport as described in Chaps. 4, 5 and 9–11.
A more generic problem is the description of vertical turbulence. Usually the ex-
isting and widely used parameterizations are tuned based on the data from the open
oceans (where the stratification conditions greatly differ from those in the Baltic) or
extracted from laboratory experiments that only partially replicate the complexity
of the marine environment. The choice of the proper (potentially spatially and ver-
tically varying) parameters of the numerical scheme still remains a major challenge
for the Baltic Sea modelling. For these reasons also the closure of the energy budget
is a complicated task. This is not only due to the inaccuracies in the wind forc-
ing but also due to challenges in the parameterizations of heat and moisture fluxes
in the rapidly varying atmospheric stability conditions, abrupt changes in surface
roughness conditions in the complicated coastal areas and due to local, instanta-
neous changes in sea surface temperature due to upwelling. Additional problems for
closing the energy budget appear because of the sparse network of radiation mea-
surements and the locally varying surface albedos. Frankly speaking, the present
forcing data for models still suffers from several uncertainties, which are reflected
in the model results as well.
High demands exist for the simulation of the dynamics of currents—flow fields,
current-driven transport and spreading processes—which is from the pure modelling
perspective the main challenge of the technology presented in this book. In addition
to the previously mentioned problems with meteorological forcing, even more se-
rious problems arise. The small baroclinic Rossby radius of deformation requires
high-resolution modelling in both horizontal and vertical directions. Together with
the request to perform calculations over at least several years, this leads to high com-
putational requirements. In order to fulfil the two latter requirements, a nested grid
approach is often used, where a high-resolution local model get its open boundary
conditions from a large-scale model (the whole Baltic, Andrejev et al. 2011, or the
Baltic Sea–North Sea, Lu et al. 2012). These technical difficulties apparently can be
partially resolved thanks to parallelized model codes and ever faster computers.
There is, however, room for many more issues. As the Baltic Sea is one of the
most studied basins of the World Ocean, one might expect that the relevant ba-
sic information such as high-resolution bottom topography is readily available. The
reality is different. For example, the best available gridded Baltic Sea bathymetry
with a resolution of 1 nautical mile (Seifert et al. 2001) is not up-to-date for several
sub-basins and there is an urgent need to construct a higher-resolution bathymetry.
There are several attempts to increase the formal resolution of this dataset to 0.5
nautical miles by means of interpolation of the existing information (Zhurbas et al.
2008a, 2008b, Laanemets et al. 2009). Nevertheless, much of the dynamics of the
Baltic Sea is apparently significantly affected by the local topography. As a first ap-
proximation, this problem was solved for the Gulf of Finland by means of manually
re-digitizing information from navigational maps (Andrejev et al. 2010), which ob-
viously is not the proper form of information technology in the 21st century but still
quite adequately reflects the pool of problems faced by the Baltic Sea oceanogra-
phers.
Even if there are still problems which are partly unsolved, 3D modelling efforts
of the Baltic Sea in Chaps. 4, 5 and 9–11 show their great ability to simulate various
circulation processes with a high accuracy.
Acknowledgements This overview is a contribution to the BalticWay project, supported by the

funding from the Finnish Academy (KM), Federal Ministry of Education and Research (BMBF),
Germany, and the European Commission’s Seventh Framework Programme (FP7 2007–2013) un-
der grant agreement no. 217246 with the joint Baltic Sea research and development programme
BONUS. The authors gratefully acknowledge the contribution of Prof. Matti Leppäranta to the
material used here from the book (Leppäranta and Myrberg 2009).
References
Alenius P, Leppäranta M (1982) Statistical features of hydrography in the northern Baltic Sea. In:
Proceedings of the 13th conference of Baltic oceanographers, Helsinki, August 24–27, 1982,
vol 1, pp 95–104
Alenius P, Nekrasov A, Myrberg K (2003) The baroclinic Rossby-radius in the Gulf of Finland.
Cont Shelf Res 23:563–573
9:1–16
Andrejev O, Sokolov A, Soomere T, Värv R, Viikmäe B (2010) The use of high-resolution
bathymetry for circulation modelling in the Gulf of Finland. Est J Eng 16:187–210
334
BACC Author Team (2008) Assessment of climate change for the Baltic Sea basin. Springer,
Berlin, 473 pp
Beletsky D, Schwab D, McCormick M (2006) Modeling the 1998–2003 summer circulation and
thermal structure in Lake Michigan. J Geophys Res Oceans 111:C10010
Bergström S, Carlsson B (1994) River runoff to the Baltic Sea: 1950–1990. Ambio 23:280–287
Cushman-Roisin BJ, Beckers J-M (2011) Introduction to geophysical fluid dynamics: physical and
numerical aspects. Elsevier/Academic Press, Amsterdam/San Diego, 828 pp
Dietrich G, Kalle K, Ostapoff F (1963) General oceanography: an introduction. Wiley, New York.
German original: Dietrich, G, Kalle K, (1957), Allegemeinde Meereskunde. Gebrüder Born-
träger, Berlin, 492 pp
Döös K, Meier HEM, Döscher R (2004) The Baltic haline conveyor belt or the overturning circu-
lation and mixing in the Baltic. Ambio 33:261–266
Drijfhout SS (1989) Eddy-genesis and the related heat transport: a parameter study. In: Nihoul JCJ,
Jamart BM (eds) Mesoscale/synoptic coherent structures in geophysical turbulence. Elsevier
oceanography series, vol 50, pp 245–263
Ekman VW (1905) On the influence of the earth’s rotation on ocean currents. Ark. Mat. Astron.
Fys. 2:1–52
Ekman M (1996) A consistent map of the postglacial uplift of Fennoscandia. Terra Nova 8:158–
165
Elken J, Matthäus W (2008) Physical system description. In: The BACC author team, assessment
of climate change for the Baltic Sea basin. Springer, Berlin, pp 379–386
Elken J, Raudsepp U, Lips U (2003) On the estuarine transport reversal in deep layers of the Gulf
of Finland. J Sea Res 49:267–274
Feistel G, Nausch G, Wasmund N (2008) State and evolution of the Baltic Sea, 1952–2005. Wiley,
Hoboken, 703 pp
Fennel W, Seifert T, Kayser B (1991) Rossby radii and phase speeds in the Baltic Sea. Cont Shelf
Res 11:23–36
Fennel W, Sturm M (1992) Dynamics of the western Baltic. J Mar Syst 3:183–205
Fonselius S (1995) Västerhavets och østersjöns oceanografi (Oceanography of the Baltic Sea, Kat-
tegat and the Skagerrak). SMHI, Norrköping, 200 pp
Gustafsson T, Kullenberg B (1936) Untersuchungen vor Trägheitströmungen in der Ostsee (Inves-
tigations of inertial currents in the Baltic Sea). Sven Hydrogr-Biol Komm Skr, Ny Ser Hydr
13:1–28
Heinloo J, Toompuu A (2011) A modified Ekman layer model. Est J Earth Sci 60:123–129
Heinloo J, Toompuu A (2012) A modification of the classical Ekman model accounting for the
Stokes drifts and stratification effects. Environ Fluid Mech 12:101–113
Höglund A, Meier HEM, Broman B, Kriezi E (2009) Validation and correction of regionalised
ERA-40 wind fields over the Baltic Sea using the Rossby Centre Atmosphere Model RCA3.0.
Rapport Oceanografi No 97, Swedish Meteorological and Hydrological Institute, Norrköping,
Sweden, 29 pp
Holton JR (1979) An introduction to dynamic meteorology, 2nd edn. Academic Press, New York.
391 pp
IPCC (2007) Climate change 2007: the physical science basis. Contribution of working group I
to the fourth assessment report of the intergovernmental panel on climate change Solomon S,
Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (eds) Cambridge
University Press, Cambridge, 996 pp
Jȩdrasik J, Cieślikiewicz W, Kowalewski M, Bradtke K, Jankowski A (2008) 44 years hindcast of
the sea level and circulation in the Baltic Sea. Coast Eng 55:849–860
Johansson M, Kahma KK, Boman H, Launiainen J (2004) Scenarios for sea level on the Finnish
coast. Boreal Environ Res 9:153–166
Keevallik S, Soomere T (2010) Towards quantifying variations in wind parameters across the Gulf
of Finland. Est J Earth Sci 59:288–297
Kielmann J (1981) Grundlagen und Anwendung eines numerischen Modells der geschichteten
Ostsee (Principles and applications of a numerical model for the stratified Baltic Sea). PhD
thesis, Berichte aus dem, Institut für Meereskunde der Universität Kiel 87
Kuzin VI, Tamsalu R (1974) Vetrovye techeniya v baroklinnom more postoyannoy glubiny
(Wind-driven currents in a baroclinic sea of constant depth). In: Chislennye metody rascheta
okeanologicheskikh techeniy (Numerical methods for simulation of ocean currents), Novosi-
birsk, USSR, pp 103–114 (in Russian)
Laanemets J, Zhurbas V, Elken J, Vahtera E (2009) Dependence of upwelling-mediated nutrient
transport on wind forcing, bottom topography and stratification in the Gulf of Finland: model
experiments. Boreal Environ Res 14:213–225
Lagemaa P, Elken J, Kõuts T (2011) Operational sea level forecasting in Estonia. Est J Eng 17:301–
331
Lass H-U, Talpsepp L (1993) Observations of coastal jets in the Southern Baltic. Cont Shelf Res
13:189–203
Lehmann A (1995) A three-dimensional baroclinic eddy-resolving model of the Baltic Sea. Tellus
A 47:1013–1031
Lehmann A, Hinrichsen H-H (2000) On the thermohaline variability of the Baltic Sea. J Mar Syst
25:333–357
Lehmann A, Myrberg K (2008) Upwelling in the Baltic Sea—a review. J Mar Syst 74:S3–S12
Lehmann A, Getzlaff K, Harlass J (2011) Detailed assessment of climate variability in the Baltic
Sea area for the period 1958 to 2009. Clim Res 46:185–196
Lehmann A, Myrberg K, Höflich K (2012) A statistical approach to coastal upwelling in the Baltic
Sea based on the analysis of satellite data for 1990–2009. Oceanologia 54:369–393
Leppäranta M (1990) Observations of free ice drift and currents in the Bay of Bothnia. Acta regiae
societatis scientiarum et litterarum Gothoburgensis. Geophysica 3:84–98
Leppäranta M (2010) The drift of sea ice. Springer Praxis, Berlin, 266 pp
Leppäranta M, Myrberg K (2009) Physical oceanography of the Baltic Sea. Springer Praxis, Berlin,
378 pp
Lindow H (1997) Experimentelle Simulationen windangeregter dynamischer Muster in hochauf-
lösenden numerischen Modellen. Meereswissenschaftliche Berichte, No 22. Institut für Ost-
seeforschung, Warnemünde (in German)
Lu X, Soomere T, Stanev EV, Murawski J (2012) Identification of the environmentally safe fairway
in the South-Western Baltic Sea and Kattegat. Ocean Dyn 62:815–829
Mälkki P, Tamsalu R (1985) Physical features of the Baltic Sea. Finnish Marine Research, vol 252,
110 pp
Meier HEM (2006) Baltic Sea climate in late 21st century: a dynamical downscaling approach
using two global models and two emission scenarios. Clim Dyn 27:39–68
Millero FJ, Feistel R, Wright DG, McDougall TJ (2008) The composition of standard seawater and
the definition of the reference-composition salinity scale. Deep-Sea Res I 55:50–72
Myrberg K, Andrejev O (2003) Main upwelling regions in the Baltic Sea—a statistical analysis
based on three-dimensional modelling. Boreal Environ Res 8:97–112
Myrberg K, Andrejev O (2006) Modelling of the circulation, water exchange and water age prop-
erties of the Gulf of Bothnia. Oceanologia 48(S):55–74
Myrberg K, Ryabchenko V, Isaev A, Vankevich R, Andrejev O, Bendtsen J, Erichsen A,
Funkquist L, Inkala A, Neelov I, Rasmus K, Rodriguez Medina M, Raudsepp U, Passenko J,
Söderkvist J, Sokolov A, Kuosa H, Anderson TR, Lehmann A, Skogen MD (2010) Validation
of three-dimensional hydrodynamic models in the Gulf of Finland based on a statistical analysis
of a six-model ensemble. Boreal Environ Res 15:453–479
Murthy R, Håkansson B, Alenius P (1993) The Gulf of Bothnia year 1991: physical transport
experiments. SMHI reports oceanography RO, vol 15. SMHI, Norrköping, 127 pp
Nekrasov AV (1999) Sharp contrasts in summer oceanographic conditions observed in Luga-
Koporye region in 1997 and 1998. BFU Res Bull 3:28–36
Nekrasov AV, Lebedeva IK (2002) Estimation of baroclinic Rossby radii in Luga-Koporye region.
BFU Res Bull 4–5:89–93
Omstedt A, Elken J, Lehmann A, Piechura J (2004) Knowledge of the Baltic Sea physics gained
during the BALTEX and related programmes. Prog Oceanogr 63:1–28
Osiński R, Rak D, Walczowski W, Piechura J (2010) Baroclinic Rossby radius of deformation in
the southern Baltic Sea. Oceanologia 52:417–429
Palmén E (1930) Untersuchungen über die Strömungen in den Finnland umgebenden Meeren.
Commentationes physico-mathematicae, vol 12. Societas Scientarium Fennica, Helsinki
(in German)
Raudsepp U (1998) Current dynamics of estuarine circulation in the lateral boundary layer. Estuar
Coast Shelf Sci 47:715–730
Samuelsson P, Jones CG, Willén U, Ullerstig A, Gollvik S, Hansson U, Jansson C, Kjellström E,
Nikulin G, Wyser K (2011) The Rossby centre regional climate model RCA3: model description
and performance. Tellus A 63:4–23
Sarkisyan AS, Staśkiewicz A, Kowalik Z (1975) Diagnostic calculations of summer circulation in
the Baltic Sea. Okeanologiya 15:1002–1009 (in Russian)
Seifert T, Tauber F, Kayser B (2001) A high resolution spherical grid topography of the Baltic Sea,
2nd edition. In: Baltic Sea Science Congress, Stockholm, 25–29 November 2001, Poster #147.
www.io-warnemuende.de/iowtopo
Simons TS (1981) Wind-driven circulations in the southwest Baltic. Tellus 30:272–283
Sjöberg B (ed) (1992) Hav och Kust. Sveriges Nationalatlas Förlag. Almquist & Wiksell Interna-
tional, Stockholm
Sokolov A, Andrejev O, Wulff F, Rodriguez Medina M (1997) The data assimilation system for
data analysis in the Baltic Sea. System ecology contributions, vol 3. Stockholm University,
Sweden, 66 pp
Soomere T, Delpeche N, Viikmäe B, Quak E, Meier HEM, Döös K (2011) Patterns of current-
induced transport in the surface layer of the Gulf of Finland. Boreal Environ Res 16(Suppl
A):49–63
Stigebrandt A, Lass H-U, Liljebladh B, Alenius P, Piechura J, Hietala R, Beszczynska A (2002)
DIAMIX: an experimental study of diapycnal deepwater mixing in the virtually tideless Baltic
Sea. Boreal Environ Res 7:363–369
Uppala SM, Kållberg PW, Simmons AJ, Andrae U, da Costa Bechtold V, Fiorino M, Gibson JK,
Haseler J, Hernandez A, Kelly GA, Li X, Onogi K, Saarinen S, Sokka N, Allan RP,
Andersson E, Arpe K, Balmaseda MA, Beljaars ACM, van de Berg L, Bidlot J, Bormann N,
Caires S, Chevallier F, Dethof A, Dragosavac M, Fisher M, Fuentes M, Hagemann S, Hólm E,
Hoskins BJ, Isaksen L, Janssen PAEM, Jenne R, McNally AP, Mahfouf J-F, Morcrette J-J,
Rayner NA, Saunders RW, Simon P, Sterl A, Trenberth KE, Untch A, Vasiljevic D, Viterbo P,
Woollen J (2005) The ERA-40 re-analysis. Q J R Meteorol Soc 131:2961–3012
Verjovkina S, Raudsepp U, Kõuts T, Vahter K (2010) Validation of seatrack web using surface
drifters in the Gulf of Finland and Baltic Proper. In: 2010 IEEE/OES US/EU Baltic International
Symposium, Riga, Latvia, August 25–27, 2010. IEEE Press, New York, 7 pp
Zhurbas VM, Laanemets J, Kuzmina NP, Muraviev SS, Elken J (2008a) Direct estimates of the
lateral eddy diffusivity in the Gulf of Finland of the Baltic Sea (based on the results of numerical
experiments with an eddy resolving model). Oceanology 48:175–181
Zhurbas V, Laanemets J, Vahtera E (2008b) Modeling of the mesoscale structure of coupled up-
welling/downwelling events and the related input of nutrients to the upper mixed layer in the
Gulf of Finland, Baltic Sea. J Geophys Res—Oceans 113:C05004
Witting R (1912) Zusammenfassende Übersicht der Hydrographie des Bottnischen und Finnischen
Meerbusens und der nördlichen Ostsee nach den Untersuchungen bis Ende 1910. Finnländische
hydrographisch-biologische Untersuchungen, vol 7, 82 pp (in German)
Chapter 3
Introduction to Computational Fluid Dynamics
and Ocean Modelling
Tomas Torsvik
Abstract This chapter provides a short introduction to the most commonly used
numerical methods in computational fluid dynamics, and some specific aspects of
numerical modelling of ocean currents. When working with numerical models, it
is essential to have a basic understanding of the possibilities and limitations of the
modelling toolbox. The first part gives a demonstration on how the model differen-
tial equations for fluid motions can be transformed into computer code instructions.
Important concepts such as consistency, stability and convergence are discussed.
The second part of the chapter provides an introduction to ocean modelling, with
emphasis on the approximations that are commonly used in general circulation mod-
els.
3.1 Introduction
Computational Fluid Dynamics (CFD) is a branch of fluid mechanics where fluid

flow problems are analysed by numerical methods. As a scientific field of study,
CFD is intimately related to the physics of fluids and the mathematical formulation
of fluid mechanics, as well as the theory of computation and algorithms and the
constant evolution of computer architecture. The complexity of CFD models varies
greatly, depending on the geometry of the flow problem, the physical properties
of the fluid and number of fluid components, and restrictions imposed by the flow
regime (e.g., turbulent vs. laminar flow, subsonic vs. supersonic flow, etc.). Some
models also incorporate chemical or biological effects in the flow problem.
While many CFD codes can run and produce useful results on simple desktop
computers, the development of large and complex computer models, and the fact
that even simple flow problems often display a high degree of complexity, have
pushed the demand for computational power. The CFD community is a prominent
part within High Performance Computing (HPC), and the evolution of CFD codes
T. Torsvik (B)
e-mail: tomas.torsvik@ioc.ee

DOI 10.1007/978-3-319-00440-2_3,
66 T. Torsvik
often runs in parallel with the evolution of supercomputers. Hence CFD is a field in
constant development and is likely to remain so in the years to come.
There is great diversity in the application of CFD, including industrial areas such
as the aerospace and automotive industry, energy production (nuclear power, hy-
dro power, wind power, petroleum industry), chemical and process manufacturing,
physiological applications related to cardiovascular flow and the respiratory system,
and environmental applications concerning atmospheric and ocean motion, water
and wastewater management, and gravity flows such as avalanches and pyroclastic
flows, to name just a few. In this review we focus on the motion of ocean currents
and waves.
The motion of ocean currents is usually studied with the use of a General Cir-
culation Model (GCM). GCMs constitute a subset of CFD models, and are used
both for atmospheric and ocean modelling. These models are at the core of both
weather forecast models and climate models. Large scale wave motions, such as
tidal waves, are usually well reproduced within the GCM framework, whereas other
wave phenomena, such as wind generated surface waves, usually require dedicated
wave models in order to obtain a realistic representation of the physical processes.
Although we will mainly focus on ocean models in what follows, it is important to
have in mind that ocean processes, and particularly motions near the ocean surface,
cannot be regarded in isolation from what is happening in the atmosphere.
Despite the capabilities of CFD, it is important to emphasize that our ability to
simulate fluid motion does not diminish the importance of more traditional methods
involving theoretical and experimental work, and in situ and remote data recording.
Theoretical and experimental results are required to provide benchmark test cases,
results that a CFD simulation must be able to reproduce to demonstrate that physical
properties of the model are handled correctly. Such exercises are a part of model
validation, and are essential in order to identify errors in the model and calibrate
model parameters.
Once a model is deemed to produce results with acceptable accuracy and reliabil-
ity, simulations can be used to provide information that is impractical or impossible
to get through the other methods, e.g., filling out the gaps in field measurement data
records due to sparse deployment of measurement instruments, analysing flow sce-
narios that are too complicated for theoretical study, or providing forecasts of likely
future events. Many scientists and engineers will not need to acquire the skills to
develop CFD models, but as the use of CFD methods expands into new fields of sci-
ence and engineering,more people will likely find that they need to work with data
from simulations or even run CFD simulations themselves. It is therefore important
also for non-specialists to have some idea about what goes on inside a CFD code,
and what the capabilities and limitations are for these models.
The following review is intended for readers with no prior experience with nu-
merical modelling, and consists of two main parts; (1) a general introduction to nu-
merical modelling, followed by (2) an introduction to ocean modelling. Given the
limited space, this review will only scratch at the surface of each subject to provide
the reader with a picture of what goes on “behind the scene” when we run numerical
models. Several books are available on each of these subjects, and a list of suggested
reading will be provided in connection with each section.
3 Introduction to Computational Fluid Dynamics and Ocean Modelling 67
3.2 Numerical Models
Since the time of Sir Isaac Newton physicists and mathematicians have expressed
the relations governing mechanical systems, the “Laws of Nature”, in terms of dif-
ferential and integral equations. In general, the collection of a suitable subset of
these equations can be regarded as a mathematical model when applied to predict
the behaviour of a specific mechanical problem. Continuous model equations can be
applied to model certain aspects of phenomena that appear continuous at a macro-
scopic level, and the deviation between modelled and measured results are usually
considered to be model errors. Since there are no independent phenomena in na-
ture, and in principle no independent way to measure parameters governing natural
processes, a comparison between models and measurements always includes some
probabilistic element. Hence the model error should not be regarded as an absolute
measure of how well the model describes nature.
The classical mathematical models formulated as differential equations are not
well suited for numerical computations. Although computers are capable of per-
forming a huge number of calculations in a short time, they are limited by the finite
precision in number representation and a finite storage space. While the mathe-
matical equations usually define a function for all real number coordinate values
within the model domain (except at discontinuities) and for continuous time, a com-
puter can only assign values to a finite number of distinct points, surface elements
or volumes at discrete time intervals. The process of transferring continuous model
equations to discrete counterparts is called discretization. The choice of a discretiza-
tion method is intimately connected with the choice of a numerical method used to
solve the problem. The discretization and computational limitations are the origin
of numerical errors that add to the original model errors.
3.2.1 Methods Used in CFD Modelling
In the classical theory of fluid mechanics there are two main branches providing
different view points on how to specify the flow field. The Lagrangian type of spec-
ification tracks the motion of material elements of fluid in time, and the history of
each material element constitutes a single path line. The Eulerian type of specifi-
cation defines flow quantities in terms of functions for the velocity vector of the
fluid throughout the fluid domain at each time instance, which change according to
flow quantities such as density and pressure. Corresponding to these branches, nu-
merical methods for CFD applications can also be categorized in two main classes;
grid-based methods corresponding to the Eulerian viewpoint, and grid-free meth-
ods corresponding to the Lagrangian viewpoint. Grid-free methods, in particular
smoothed-particle hydrodynamics (SPH), have been successfully applied to study
highly dynamic problems such as wave breaking and the Rayleigh-Taylor instabil-
ity. A large number of fluid elements are usually needed to obtain a satisfactory
resolution, in which case the computational cost is large compared to equivalent
68 T. Torsvik
Fig. 3.1 Four grid configurations. Source: Wikimedia Commons
grid-based simulations. SPH is a subject of active research, and the range of applica-
tions for these methods may expand as new algorithms are developed and computer
power increases (see Monaghan 2012 for a recent review).
Still, most CFD models apply grid-based methods, in particular models for geo-
physical fluid dynamics problems, and these methods will be the subject for the rest
of this review. The grids used by grid-based methods can be further divided into two
categories; structured and unstructured. Examples of different grid configurations,
three structured grids and one unstructured, are shown in Fig. 3.1. The choice of grid
configuration is a trade-off between simplicity of the numerical algorithms, and by
extension the efficiency of the code, and the accuracy of geometric representation.
The simplest grid configuration is generated by applying equidistant partitioning
along each principal axis, resulting in a regular, Cartesian grid as shown in Fig. 3.1a.
In this case all grid cells are equal in shape and size, so the numerical method can be
formulated without taking into account local grid variations, and the interconnection
between grid cells can be accounted for by simple indexing methods. Structured
grids may include local grid refinement, as shown in Fig. 3.1b, or include curved grid
lines to better fit a specific geometrical shape, as shown in Fig. 3.1c. In particular,
the geographical coordinate system where a position is determined by longitude,
latitude and elevation is often used in geophysical models. For these structured grids
the numerical method must take into account the shape of each individual grid cell,
but the interconnection between grid cells can still be accounted for by indexing
methods. Unstructured grids, shown in Fig. 3.1d, provide the largest flexibility in
terms of grid refinement and geometry fitting, but in this case a record must be
kept stating explicitly how the grid cells are connected by node points and surface
elements.
CFD models are often required to combine simulation of large spatial domains
with a detailed description of the flow at specific locations, hence the need for local
grid refinement. We may for instance need high resolution at a river estuary to get
correct mixing of river and sea water, or be interested in the local effect at a spe-
cific coast section during a storm surge. The obvious choice in such a case would
be to use unstructured grids, but an alternative method which is often applied in
both ocean and atmospheric models is to use a system of nested grids. In this case
the model domain with coarse grid resolution contains a sub-region with higher
grid resolution, which in turn may contain further sub-regions with even higher res-
olution. An example is shown in Fig. 3.2. A separate model is used within each
Fig. 3.2 Nested grid system
sub-domain, and the system may be one-way (information travels only from the
low resolution model domain to the high resolution model domain, or vice versa)
or two-way nested. It is usually more computationally costly to run several models
with different grid resolutions than a single model with local grid refinement, but the
nested grid system allows flexible local grid refinement with structured grids. The
nested grid system also allows a model to run with different model configurations,
or even the use of entirely different models, for each nested level in order to improve
the representation of flow properties.
There is no unique way to represent a continuous function on a discrete grid. Con-
sider the continuous function (blue line) displayed in Fig. 3.3a, and how this can be
approximated by a single value representation on equidistant intervals. Two possi-
ble representations are demonstrated in the figure; the red points represent the blue
curve by the function value at the middle point of each interval, and the green bars
indicate the average value of the function on the interval, in which case the height of
the bar can be used to represent the curve. Clearly, the two different approximation
methods do not provide the same representation of the curve. Increasing the grid
resolution by increasing the number of grid cells, thereby making each cell smaller,
provides a better representation of the curve, as shown in Fig. 3.3b.
There are several numerical methods available for numerical modelling of fluid
dynamic problems, but most CFD models in use today apply either the finite differ-
ence scheme (FD) method, finite volume (FV) method or finite element (FE) method.
The FD method is the oldest and conceptually simplest method, which uses a point-
wise approximation method as shown in Fig. 3.3. This method will be discussed
in the next section in more detail. For the purpose of comparison with FV and FE
methods it is sufficient to mention that the FD method discretizes the partial differ-
ential equations directly, requires a structured grid and approximates the solution
locally at each grid point.
The FV method is conceptually similar to the FD method, but instead of discretiz-
ing the differential equations, FV works with the integral form of the equations of
motion. Essentially, the FV method involves calculation of cell average values in
such a way that the sum of fluxes across cell boundaries are conserved or match the
70 T. Torsvik
Fig. 3.3 Discretization of a continuous function
forcing field at the boundary of the model domain. The FV method is applicable
for unstructured as well as structured grids. The main strength of the FV method is
to maintain global conservation of mass, momentum and energy within the model
domain. Maintaining these conservation properties is often vital for an accurate de-
scription of the flow problem. This method is mainly used when the equations of
motion can be written in the form of conservation equations. In principle, the FV
approximation is discontinuous at cell borders, but it is usually possible to visual-
ize the solution by specifying the cell average values at the cell centres and make a
linear interpolation between these centre points. However, the interpolated solution
constructed in this way will in general not have the same conservation properties as
the underlying cell averaged FV solution.
The FE method has its origin within the field of complex elasticity and structural
analysis problems, but has also found its use within CFD applications. The method
works by reformulating the given differential equations in an equivalent variational
form, and then solves a minimization problem in terms of a given set of basis func-
tions. The main advantages of the FE method are that it is easy to apply for com-
plicated geometry as it works nicely with unstructured grids, and that it is based
on a mathematical theory that facilitates error estimation for the numerical solution.
When comparing the FE and FV methods, which can both work with unstructured
grids, the traditional view has been that the FE method is more computationally de-
manding, but the difference in performance is highly dependent on the particular
implementation of each method.
It is not easy to give general advice on which method to choose. Different ap-
plications have different requirements in terms of computational efficiency, geomet-
rical complexity, conservation properties and error sensitivity, and no single tool
will be best for all applications. GCMs have usually used the FD method for dis-
cretization, but there is a trend towards increasing use of the FV method within this
particular field of research.
There is an extensive literature available on numerical methods and CFD, both
as introduction and reference level textbooks. As introductions to CFD, the books
Anderson (1995) with focus on the FD method, and Ferziger and Perić (2002)
which includes discussions on the FD and FV methods, can both be recom-
mended. A useful resource for CFD-related topics is the web site CFD Online
(http://www.cfd-online.com), which includes overviews of CFD software and lit-
erature, discussion forums and an overview of conferences and workshops. For
introduction level books on numerical methods, Thomas (1995), LeVeque (2002)
and Johnson (2009) can be recommended for the FD method, FV method and FE
method, respectively.
3.2.2 The Finite Difference Method
This short introduction is not intended as an exposition of fundamental theory or

advanced CFD methods. We will instead focus on simple methods that demonstrate
some general concepts of numerical modelling. For this purpose we will now do a
step by step demonstration of how to construct a simple numerical model, and leave
the discussion of more general topics for later. This part is intended for readers with
no previous experience with numerical models.
The problem we will consider for this demonstration is the 1D transport equation
∂f (x, t) ∂f (x, t)
+c = 0, x ∈ (0, L), t ∈ (0, ∞), (3.1)
∂t ∂x
where f (x, t) is the unknown function we wish to find, x is a variable for one hori-
zontal dimension, t is time, and c is a constant parameter. In order for the problem
to be well-posed we need to specify an initial condition
f (x, 0) = f0 (x) for x ∈ [0, L], (3.2)
and boundary conditions
f (0, t) = g0 (t) and f (L, t) = gL (t) for t > 0. (3.3)
The solution to this problem is usually studied in an introductory level course for
partial differential equations (PDEs). If we simplify the problem by specifying the
boundary conditions
g0 (t) = gL (t) = 0,
the problem defined by Eqs. (3.1)–(3.3) has the general solution
f (x, t) = f0 (x − ct), x ∈ (0, L), t ∈ (0, ∞). (3.4)
The solution is constant along characteristics
ξ = x − ct,
72 T. Torsvik
Fig. 3.4 Solution of the

transport equation. Lines of
characteristics are indicated
in red
Fig. 3.5 The Finite

Difference method
which shows that the parameter c is the propagation speed of the solution. When
c is constant the characteristics are lines with constant slope in the (x, t)-plane, as
shown in Fig. 3.4. The solution propagates to the right if c > 0 and to the left if
c < 0. When c is constant, we only need information from one of the boundary
conditions.
We will use the FD method to solve the 1D problem numerically. The first step
of the solution process is to construct a point distribution or a lattice graph within
the (spatial) model domain, usually called the (numerical) grid, defining the discrete
points where the numerical solution will be specified. The simplest form of the FD
method is obtained with a uniform grid, i.e., the spatial domain is divided into equal
length segments along each principal axis. For a 1D problem the discretization will
be as shown in Fig. 3.5, where x represents the length of the constant step. For
our example we may construct a uniform grid with M points, including points at
the domain boundaries at x = 0 and x = L, if the distance between the points is
x = L/(M − 1).
We will also need to discretize the solution in time, and may for this purpose
introduce a finite time step t, analogous to the constant space step x. Given
initial conditions that define the function f (x, 0) = f0 (x) for all grid points in the
domain, and boundary conditions g0 and gL , our task is now to define a procedure
that will give us f (x, t). Once the method is defined, we may proceed to find
f (x, 2t), f (x, 3t), etc., by iterations.
Fig. 3.6 Notation for the

Finite Difference method
At this point we will introduce a new notation, in order to keep the formulas in
the following paragraphs as short as possible. Consider a uniform grid in the (x, t)-
plane, as shown in Fig. 3.6, and that the spatial grid points are labelled as
x1 < x2 < · · · < xm−1 < xm < xm+1 < · · · < xM−1 < xM .
For the function value at (x = xm , t = nt), we will use the notation
f (xm , nt) = fm(n) .
The neighbouring grid points in space and time are labelled as

(n) (n)
f (xm−1 , nt) = fm−1 ; f (xm+1 , nt) = fm+1
and

f xm , (n − 1)t = fm(n−1) ; f xm , (n + 1)t = fm(n+1) ,
respectively.
The FD method works directly with the PDE form of the problem, and approx-
imates the solution by polynomial approximations at discrete points. The Taylor
series approximation1
∞
f (j ) (x0 )
f (x0 + x) = x j
j!
j =0
f (x0 ) 2 f (x0 ) 3
= f (x0 ) + f (x0 )x + x + x + · · · (3.5)
2 6
is applicable for analytical functions, and can be used to approximate the function
value at a point x = x0 + x, provided the function value and all derivatives are
known at the point x = x0 . Figure 3.7 shows the Taylor series approximation for the
1 InEq. (3.5) the expression f (j ) (x0 ) indicates the j -th derivative of the function f , not to be
confused with the discretization notation f (n) .
74 T. Torsvik
Fig. 3.7 Taylor series

approximation for the sine
function (black line). Lines in
other colour show how the
approximation improves with
increasing number of terms in
the series
sine function, which is given by

∞
(−1)j x 2j +1 x3 x5 x7
sin(x) = =x− + − + ··· .
(2j + 1)! 3! 5! 7!
j =0
The polynomial approximation to the sine function is poor if only the first term in
the series is used (blue line in Fig. 3.7), but improves as the order of the polynomial
approximation increases. The error in the approximation which is associated with
truncation of the series approximation is called the truncation error.
In our case we know the function values at the grid points, but the derivatives
in the PDEs are unknown. We can use the Taylor series approximation to estimate
the derivatives using finite differences. Ignoring the time dimension for the moment,
and looking only at the spatial derivative, we can estimate the value of the derivative
(n)
∂fm /∂x, by the Taylor series
(n) (n) (n)
(n) ∂fm x 2 ∂ 2 fm x 3 ∂ 3 fm
fm+1 = fm(n) + x + + + ··· , (3.6)
∂x 2 ∂x 2 6 ∂x 3
which gives us the Forward Difference method
(n) (n)
(n)
∂fm f − fm
≈ m+1 . (3.7)
∂x x
The accuracy of the finite difference approximation is determined by the local trun-
cation error (LTE), which for small values of x is dominated by the leading trun-
cated term. For the Forward Difference method the LTE is

(n)
x ∂ 2 fm
= O(x),
2 ∂x 2
where the O(·) notation2 is used to characterize the error in terms of the spatial
discretization step x. We may similarly get an estimate of the same differential
term by using the Taylor series
(n) (n) (n)
(n) ∂fm x 2 ∂ 2 fm x 3 ∂ 3 fm
fm−1 = fm(n) − x + − + ··· , (3.8)
∂x 2 ∂x 2 6 ∂x 3
which gives us the Backward Difference method
(n) (n)
(n)
∂fm fm − fm−1
≈ (3.9)
∂x x
with the LTE
(n)
x ∂ 2 fm
= O(x).
2 ∂x 2
The order of the FD method is given by the LTE, so both the Forward Difference and
Backward Difference methods are 1st order. A 2nd order method can be constructed
by combining Eqs. (3.6) and (3.8) to yield the Central Difference method,
(n) (n)
(n)
∂fm f − fm−1
≈ m+1 , (3.10)
∂x 2x
where the 2nd order derivative terms cancel, giving the LTE
x 2 ∂ 3 fm
(n)

3
= O x 2 .
6 ∂x
Higher order approximations may be obtained by including terms from the Taylor
(n) (n)
series approximations for fm+2 , fm−2 , or points even further away from xm .
A similar cancellation strategy can be applied to obtain approximations for higher
order derivatives. For instance, we can obtain an estimate for the 2nd order spatial
2 The ‘Big O’ notation describes the limiting behaviour of a function f by comparing it to the
behaviour of a simpler function g. Therefore, if

f (x)
lim =C
x→x0 g(x)
where C is some finite number, then f (x) = O(g(x)) as x → x0 . When discussing finite difference
schemes it is implicitly understood that the approximation is valid for x → 0, and therefore this
condition is usually not stated explicitly in the formulas.
76 T. Torsvik
derivative by combining Eq. (3.6) and Eq. (3.8) in a way that eliminates the 1st order
derivative term
(n) (n) (n−1)
(n)
∂ 2 fm f − 2fm + fm
2
≈ m+1 , (3.11)
∂x x 2
where the LTE is given by
x 2 ∂ 4 fm (n)

4
= O x 2 .
24 ∂x
Finite differences for time derivatives can be constructed in the same way as for
the space derivatives, by replacing derivatives in space with derivatives in time, and
replacing x by t. For instance, the Forward Euler method
(n) (n+1) (n)
∂fm fm − fm
≈ (3.12)
∂t t
and Backward Euler method
(n) (n) (n−1)
∂fm fm − fm
≈ (3.13)
∂t t
are direct time derivative analogs to the forward difference and backward difference
methods, respectively. Although the principles for obtaining the finite difference
approximations are the same in space and time, time differentiation usually involves
unknown function values, i.e., we may know the function values for all times up
(n) (n+1)
to and including the terms fm , whereas the function values fm are unknown.
Higher order methods and approximations for higher order time derivatives may
therefore involve several terms backward in time but will as a rule not involve terms
beyond the first unknown time step.
Returning to our 1D transport equation example, we are now ready to discretize
the equation by replacing derivative terms with finite difference approximations. As
we have seen, the FD method does not provide a unique approximation of deriva-
tives. In fact, this is one of several choices that the modeller will have to make,
guided by experience and the expected behaviour of the problem at hand. If we dis-
cretize the transport equation (3.1) with the Backward Difference method in space
and Forward Euler method in time, we obtain the difference equation
(n) (n)
fm
(n+1) (n)
− fm fm − fm−1
+c = 0. (3.14)
t x
Assuming the function values are known for t = nt, we only need to solve the
(n+1)
difference equation for fm
(n) (n)
fm − fm−1
fm(n+1) = fm(n) − c t. (3.15)
x
Fig. 3.8 Stencil for the

Forward Euler–Backward
Difference method
The interdependence between grid points in the numerical scheme can be visualized
as a specific geometrical arrangement, called a stencil. The stencil for the numerical
scheme defined by Eq. (3.15) is illustrated in Fig. 3.8, where filled circles indicate
known function values and open circles indicate unknown function values.
In order to test the numerical model for a practical case study, we need to specify
some parameters. We will use the boundary conditions
g0 (t) = gL (t) = 0,
and specify the initial condition as

⎧
⎨ 1 − cos 2πx
when 0 < x < 1,
f0 (x) = 2
⎩0 when 1 ≤ x < L.
For the following example we will use the non-dimensional parameters c = 1.0 and
L = 10.0 for the initial-boundary value problem defined by Eqs. (3.1)–(3.3), and
use M = 201 spatial grid points, resulting in a spatial grid resolution of x = 0.05.
The only parameter that remains undefined at this point is the time step t, where
we will experiment by using the values t ∈ [0.5x, x, 1.06x]. For practical
reasons we also need to specify the maximum number of time steps to make sure
the computer program terminates, so we will set this to N = 100.
A computer program that solves the 1D transport problem with the parame-
ters specified above can be written with a few lines of code in a modern script-
ing language (see frame below). The example included here has been written as a
MATLAB© script, with in-line comments describing the purpose of the different
command lines.
%====================================
% Matlab script for calculation of
% df/dt + c*df/dx = 0
%====================================
M = 201; % number of gridpoints
N = 100; % number of timesteps
L = 10.0; % length of domain
78 T. Torsvik
c = 1.0; % wave speed

h = L/(M-1); % grid-spacing
k = 0.05; % timestep increment
for i=1:M % Initial condition

if (i*h < 1.0)
f(i) = 0.5*(1.0 - cos(2*pi*i*h));
else
f(i) = 0.0;
end
end
x = 0.0:h:L; % x-vector
plot(x,f); % plot initial state

pause(0.1);
for j=1:N % time step

f_new(1) = 0.0; % boundary condition:
for i=2:M % f(0,t) = 0.0
dfdx = (f(i) - f(i-1))/h; % spatial derivative
f_new(i) = f(i) - k*c*dfdx; % time derivative
end
f = f_new; % update old f from f_new
plot(x,f); % plot each time step

pause(0.1);
end
There are basically two parts of the script; the first part defines the parameter
values and initial conditions (the boundary conditions are defined implicitly as zero
values), the second part solves the finite difference scheme by iterative steps in time
and space. Since the stencil of the method we use only require values at the (n) and
(n + 1) time levels we only need two arrays to define the function values for all grid
points; f stores the known function values at the (n) time step level, and f_new
stores the calculated values for the (n + 1) time step level. After each completed
sweep updating the f_new array, the old f array is replaced by the updated values.
Figure 3.9 shows the results of the numerical simulations for different values of
the time step t. These results are remarkably different from each other. In this
case we know what the result should be from the analytical solution of the PDE.
The initial profile defined by f0 should simply be transported along characteristics
without any change in shape, which is what we see in Fig. 3.9b. Figure 3.9a shows a
transport process, but the initial profile is being smoothed out with time, which is an
effect of numerical diffusion. Figure 3.9c also shows a transport process initially, but
at the end of the simulation the picture is dominated by oscillations on the length
scale of the spatial grid resolution x which grow in amplitude with increasing
Fig. 3.9 Simulation of the 1D transport equation problem. Panels (a)–(c) show results using the
Backward Euler–Backward Difference scheme, whereas a Backward Euler–Forward Difference
scheme is used in panel (d)
time. If the simulation was allowed to progress for more time steps, the amplitude
of these oscillations would eventually grow beyond the register capacity for floating
point numbers, causing the simulation to crash due to an arithmetic overflow error.
This behavior is an example of numerical instability.
The lesson to learn from this exercise is that the discretization process and nu-
merical calculation with limited accuracy in the representation of numbers lead to
errors that may cause the simulation to fail in several unexpected ways. This may
prompt the question of whether or not we can trust the results of any of our simula-
tions. Fortunately, at least for the finite difference methods and linear PDEs, we can
get some assurance through the Lax equivalence theorem (Thomas 1995).
Lax Equivalence Theorem A consistent difference scheme for a well-posed linear

initial-value problem is convergent if and only if it is stable.
Convergence in this context means that the solution of the numerical scheme
actually converges towards the solution of the original PDEs, which is a property
that is difficult to prove in general. Consistency means that the finite difference
80 T. Torsvik

Forward Euler–Forward
Difference method
equations converge towards the PDEs in the limit when x, t → 0, which is a
property that can easily be verified. Provided the scheme is stable, we can trust
the model to obtain a reasonable approximation for the solution of the PDEs we
wish to solve if we integrate with small enough steps in time and space. Numerical
stability in this context means that numerical errors should not grow significantly
during the calculation, preferably remaining at levels close to the unit roundoff used
in the computation. There exist methods that can be used to analyse the stability of
schemes under certain conditions, but even a superficial discussion of these methods
is beyond the scope of this introduction.
We have already seen one example of a numerical scheme for our 1D transport
equation problem. We may repeat this exercise with a minor modification, using
the Forward Difference method instead of the Backward Difference method. This
results in the numerical scheme
(n) (n)
fm+1 − fm
fm(n+1) = fm(n) − tc (3.16)
x
with the stencil as shown in Fig. 3.10. This is quite similar to the scheme used
earlier, but in this case numerical instabilities occur regardless of what time step
value is used. An example using the time step t = 0.05x is shown in Fig. 3.9d.
In order to understand why we get such different results with different numerical
schemes and parameter choices, we should consider again properties of the mathe-
matical problem as formulated by Eqs. (3.1)–(3.3). Information is transported along
the lines of characteristics, which defines the only physical speed of propagation
for the problem. In order to have a convergent numerical solution it is obviously
necessary that the numerical scheme captures how information is transported along
characteristics, i.e., the lines of characteristics going through a grid point for which
we wish to estimate the function value must lie within a region defined by the sten-
cil of the numerical method, defining the domain of dependence for the method.
Figure 3.11 shows how the lines of characteristics relate to the numerical examples
shown in Fig. 3.9. The figures are a bit misleading, since the slope of the character-
istics is the same in all cases, defined by c = 1.0. The spatial discretization is also
the same, but the time step changes so that the distances between the points along
the t-axis are different in each case.
For Fig. 3.11a the lines of characteristics lie within the domain of dependence de-
fined by the stencil, which result in a stable but highly diffusive numerical solution.
Fig. 3.11 Lines of characteristics for solution of the 1D transport equation superimposed on the
stencil of the numerical scheme
In Fig. 3.11b the lines of characteristics go exactly along the border of the domain
of dependence, which in this case leads to a highly accurate numerical solution for
the transport problem. By choosing the spatial resolution x and time step t so
that x/t = c, the Forward Euler–Backward Difference scheme becomes
t (n) (n) (n)

fm(n+1) = fm(n) − c fm − fm−1 = fm−1 ,
x
so we see that the information at grid point (xm−1 , nt) is transferred to grid point
(xm , (n + 1)t) after one time step with the accuracy of machine precision. Note
that this high accuracy of the solution is only achievable due to the linear characteris-
tics, and would not be achievable with a static, equidistant grid if the characteristic
was non-linear or if we were simulating the transport of several waves moving at
different velocities.
For the two last cases, Figs. 3.11c and 3.11d, the lines of characteristics lie out-
side the domain of dependence defined by the stencil, and both these cases are unsta-
ble. The method shown in Fig. 3.11c is unstable because we have made the time step
too large compared to the spatial discretization, but the method works, as we have
seen, with a different choice for the time step. The method shown in Fig. 3.11d is
unconditionally unstable for the problem at hand, because no possible combination
of numerical parameters will cause the lines of characteristics to intersect the do-
main of dependence or the stencil. However, if we changed the physical parameter c
to a negative value the preference of method would change: the Forward Difference
82 T. Torsvik

Backward Euler–Backward
Difference scheme
method would be appropriate, and the Backward Difference method would in that
case be unconditionally unstable.
As this little exercise has shown it is of vital importance to understand the phys-
ical and mathematical properties of a problem before undertaking the task to con-
struct a numerical model. Without such insight, it is very likely that the numerical
model does not correspond to the problem we wish to solve. This usually leads
to unstable schemes, which are obviously wrong, but may also lead to reasonably
looking results that are completely determined by artificial numerical effects.
So far we have only seen the use of explicit time stepping schemes, which are
schemes that can be written on the form

finite
(n )
fm(n+1) = fm , (3.17)
n ≤n
(n+1)
i.e., the calculation of fm does not depend on any other (n + 1) terms. A scheme
that does not satisfy the property given by Eq. (3.17) is called an implicit scheme.
As an example we can consider the 1D transport equation with a Backward Euler–
Backward Difference scheme
(n+1) (n+1)
fm
(n+1)
− fm
(n)
fm − fm−1
+c =0
t x
with the stencil as shown in Fig. 3.12. Implicit schemes often have better stabil-
ity properties than explicit schemes. However, the use of implicit methods makes
the solution process more complicated because now we need to solve a system of
coupled linear equations for each time iteration, which in general requires the cal-
culation of a matrix inverse or the solution of a system of linear equations. For the
1D problem we get a coefficient matrix with non-zero values restricted to a band
along the diagonal, which can easily be inverted through Gauss elimination. For
more elaborate schemes we may combine an explicit and implicit scheme to create
a Predictor–Corrector method, where the explicit scheme is used to make a first
estimate for the time integration, the predictor step, which is then improved by one
or more implicit iteration steps, the corrector.
As we have seen in the examples, the stability of a numerical scheme is closely
related to the ratio between the physical speed of information c and the speed of
information as dictated by the spatial and temporal resolution of the numerical
scheme. This relationship is characterized by the Courant number

ct
CCFL = . (3.18)
x
For an explicit scheme for the hyperbolic equation
∂f/∂t + c∂f/∂x = 0
of the form:
(n) (n)
fm(n+1) = αfm−1 + βfm(n) + γfm+1 ; α, β, γ = const
with a constant ratio between time step and spatial resolution t/x, a necessary
condition for numerical stability is the Courant–Friedrichs–Lewy (CFL) condition
CCFL ≤ 1. The CFL condition guarantees that the FD method can propagate infor-
mation (energy) at least as fast as the PDE.
3.3 Numerical Ocean Modelling

We now move away from the general discussion on numerical models and focus
on numerical ocean modelling within the framework of GCMs. It is essential that
a numerical model captures the prominent features of the physical problem, so we
will start the discussion by considering what these features are in the case of fluid
motion in ocean and large sea areas.
3.3.1 Physical Characteristics of Ocean Dynamics
The ocean is a dynamic system where water is in constant motion. Fluid motion
is driven by local pressure differences which are induced by several forcing mech-
anisms. The gravitational force acting on the Earth from the Moon and the Sun
creates ocean tides. Solar radiation heats the surface water which results in the wa-
ter cycle of evaporation and precipitation. Local and regional temperature variations
in the atmosphere create horizontal pressure gradients, giving rise to wind which in
turn generates waves through shear stress. Rivers and glaciers provide sources for
fresh water input, and the formation of sea ice contributes to changes in salinity and
the reduction of wind effects. At the sea floor the bottom friction can be a significant
factor, as well as ground water seepage and sources of geothermal heat. In addition
to these physical processes, there are a number of chemical and biological factors
that may influence the properties of water masses.
The main function of ocean currents on the global scale is to transport warm
surface water from the equator to the poles and cold bottom water to the equator.
This global motion is called the thermohaline circulation and is often pictured as a
84 T. Torsvik
Fig. 3.13 The thermohaline

circulation visualized as a
global ocean conveyor. Blue
paths represent deep water
currents and red paths
represent surface currents.
Source: Wikimedia Commons
conveyor belt system of water transport as shown in Fig. 3.13, although this greatly
oversimplifies the complexity of global water transport. The large scale horizontal
motion we observe in ocean and sea basins is mainly a product of the Earth’s rota-
tion and the restrictions imposed by the coast line and bottom topography.3 In most
sea areas the water temperature decreases and the salinity increases with increasing
depth, in which case the water basin will have a persistent vertical density strat-
ification. As a result, flow motion is predominantly horizontal, specifically along
isopycnal surfaces. Strong vertical motion requires the breakdown of stratification,
which is common in brackish water during winter conditions, or a vertical slope of
pycnoclines, which is usually connected with specific flow regimes and topographic
or coastal features, as discussed in detail in Chaps. 2, 5 and 6.
3.3.1.1 Time Scales and Space Scales for Ocean Dynamics
One of the challenges connected with the modelling of ocean dynamics is the large
range of scales on which relevant processes occur. If we consider the range of length
scales, as shown in Fig. 3.14, the largest relevant scale is connected to the thermo-
haline circulation and large ocean gyres with a typical length scale on the order
L ∼ 106 m. The smallest relevant scale is usually considered to be the length scale
that is connected with turbulent energy dissipation, which can be characterized by
the Kolmogorov length scale and is usually of the order L ∼ 10−2 m or less in the
ocean (see Thorpe 2007 for a discussion on ocean turbulence). In between these
extremes are length scales associated with geostrophic eddies (L ∼ 105 m), vertical
convection (L ∼ 104 m), internal waves (L ∼ 102 m) and wind generated surface
waves (L ∼ 1 m).
These length scales are important when we consider how to discretize the prob-
lem we wish to model. The spatial grid scale is the distance between two grid points
in space, and can be used as a measure for the spatial resolution of the numerical
3 Note that the rotation of the Earth is not a driving force in itself. When Newton’s laws of motion
are applied in a rotating frame of reference, motion along a straight line appears to be deflected to
the left or right. The Coriolis and centrifugal forces are introduced to account for such deflections.
These forces are often called inertial forces or fictitious forces in order to emphasize that they are
not connected to any physical force, but to the acceleration of the rotating system itself.
Fig. 3.14 Length scales in the ocean
model. In theory, the shortest wavelength that can be represented on a numerical

grid is λ = 2x, hence 3 or 4 grid points per wave period should be sufficient to
represent a particular wave mode. However, such accuracy can seldom be realized in
practical GCM applications. Small scale processes are strongly influenced by trun-
cation errors and non-linear interactions are poorly represented at grid scale level
because of the damping imposed by dissipation terms. Climate modellers who are
concerned about extracting regional climate predictions from global climate mod-
els, so-called downscaling of global models, have introduced the term skillful scale
to indicate the scale for which local model predictions start to resemble phenomena
as observed in nature. For climate model applications, the skillful scale is usually
considered to be 8 grid points or more (Benestad et al. 2008). The skillful scale
may vary with different applications, but it is clear that modellers should be care-
ful when interpreting small scale effects and run models with a set of different grid
resolutions whenever possible to keep control of truncation errors.
In order to determine relevant time scales, we should look at the velocity of
propagations for various forms of energy. The speed of sound in water is 1497 m/s.
If we were required to include sound waves in GCMs with a horizontal resolution of
x ∼ 103 m, this would require a time step of less than one second. Fortunately we
can ignore sound waves if we assume that the fluid is incompressible. We are then
left with surface gravity waves as the phenomena that define the maximum speed
of propagation for the model. These waves attain their maximum velocity when
the wave length λ is large compared to the water depth√ H , so-called shallow water
waves, in which case the velocity is given by c2D = gH , where g = 9.81 m/s2 is
the acceleration of gravity. Typical examples of shallow water waves in the ocean
are tidal waves and tsunamis. Since the speed of these waves is entirely determined
by the local depth it is easy to calculate typical wave speeds for different water
basins (provided a typical water depth can be defined). For the Atlantic Ocean the
typical water depth is H = 4000 m, which gives a wave speed of c2D = 198 m/s, but
for a shallow sea area, such as the Baltic Sea with an average depth of H = 54 m,
the wave speed is reduced to c2D = 23 m/s.
The motion of water masses induced by shallow water waves is almost uniform
along the vertical axis, which means that these waves can be approximated as 2D
phenomena where only horizontal motion is of interest. Vertical motion becomes
important for internal gravity waves, which propagate at the speed of

ρ ρ
c3D = gH = c2D ,
ρ0 ρ0
86 T. Torsvik
where ρ0 is a reference density, typically 1000 kg/m3 for fresh water and around
1030 kg/m3 for salt water, and ρ is the deviation from the reference level for a
particular volume of water. For internal waves on the interface between fresh and
salt water, which is the strongest density gradient likely to occur in the ocean, we
will have a density difference of ρ = 0.03ρ0 , which gives us the maximum wave
speed of c3D = 0.17c2D for internal gravity waves. Normally such strong density
differences are not maintained for very long in the ocean, so for practical applica-
tions we may assume c3D < 0.1c2D .
From the numerical modelling point of view we see that the CFL condition im-
posed by surface gravity waves is stronger than the CFL condition associated with
internal gravity waves. Since the surface gravity waves can be calculated with the
use of 2D models, many ocean models apply a mode-split method where separate
model equations are used for the fast barotropic motion (2D) and the slow baro-
clinic motion (3D). This allows the model to use different time steps for the differ-
ent modes, integrating the barotropic mode at a higher frequency than the baroclinic
mode but with a simpler set of model equations. This feature is built into many
ocean models, including the Rossby Center Ocean Model RCO that is thoroughly
discussed in Chap. 4 and the results from which are extensively used in Chaps. 9
and 10.
3.3.2 Model Equations on a Rotating Frame of Reference
The equations of motion in fluid dynamics express the rate of change of some quan-
tity F (x, y, z, t) at each point in space following a fluid volume of fixed identity. The
Lagrangian description provides a natural frame of reference for such motions, but
the equations we wish to use are formulated within the Eulerian frame of reference.
For arbitrary increments dx, dy, dz, and dt the increment in F is
∂F ∂F ∂F ∂F
dF = dt + dx + dy + dz. (3.19)
∂t ∂x ∂y ∂z
If, however, the increments are not arbitrary but are associated with the movement
of a small volume of fluid, the space and time increments are linked by the local
velocity components
dx = u(x, y, z, t)dt, dy = v(x, y, z, t)dt, dz = w(x, y, z, t)dt,
and we can write Eq. (3.19) as
dF ∂F ∂F ∂F ∂F
= +u +v +w . (3.20)
dt ∂t ∂x ∂y ∂z
In order to emphasize that the operation of time derivative is carried out while fol-
lowing a volume of fluid, the notation D/Dt is often used as replacement for d/dt
in fluid mechanics. The time derivative
DF ∂F
= + u · ∇F, (3.21)
Dt ∂t
where u = (u, v, w) is a velocity vector in three dimensions, is called the material

derivative.
The motion of Newtonian fluid in a fixed frame of reference is governed by equa-
tions expressing the conservation of mass
Dρ
+ ρ∇ · u = 0 (3.22)
Dt
and the conservation of momentum given by the Navier–Stokes equations
Du
ρ = −∇p + μu + ρgN , (3.23)
Dt
where μ is the molecular viscosity and gN is the Newtonian gravity acceleration.

The acceleration on a fixed (F) and rotating (R) frame of reference are related by
aF = aR − Ω 2 R + 2Ω × uR ,
where Ω is the angular velocity of the rotating reference frame, R is the radius of the
rotation and uR is the velocity (Fig. 3.15). The two additional terms on the right hand
side of the equation represent the centripetal acceleration −Ω 2 R, and the Coriolis
acceleration 2Ω × uR . The conservation of mass (3.22) remains unchanged when
we move to the rotating frame of reference, but the Navier–Stokes equations (3.23)
become
Du
ρ = −∇p + μu + ρg − 2ρΩ × u, (3.24)
dt
where we have dropped the subscripts R for convenience, and defined the effective
gravity acceleration as
g = gN + Ω 2 R.
The angular velocity for the rotation of the Earth is
Ω = 2π rad/day ≈ 0.73 × 10−4 1/s.
A local Cartesian system on a tangent plane can be applied if the length scale of the
model domain is small compared to the Earth’s radius (6371 km) (Fig. 3.15). The
angular velocity components in the Cartesian system of reference are
Ωx = 0, Ωy = Ω cos θ, Ωz = Ω sin θ,
88 T. Torsvik
Fig. 3.15 Rotating frame of

reference
where θ is the latitude. We evaluate the vector cross product 2Ω × u as the determi-
nant of a matrix
⎛ ⎞
i j k 2Ωw cos θ − 2Ωv sin θ

2Ω × u = 0 2Ω cos θ 2Ω sin θ = ⎝ 2Ωu sin θ ⎠,
u v w −2Ωu cos θ
where i, j and k are the unity vectors in the direction of the x-, y- and z-axis,
respectively. Since w v, the term w cos θ can be ignored in the i-direction. We
may now define the Coriolis frequency f = 2Ω sin θ and write the three components
of the Coriolis acceleration as
(2Ω × u)x = −f v,
(2Ω × u)y = f u,
(2Ω × u)z = (−2)Ωu cos θ.
The Coriolis frequency takes values within the range

f ∈ −0.73 × 10−4 , 0.73 × 10−4 ,
attaining negative values on the Southern hemisphere and positive values on the
Northern hemisphere. The effect is strong near the poles and negligible near the
equator. On the Northern hemisphere fluid elements moving from high to low pres-
sure are deflected to the right (Fig. 3.16a). Due to persistence in long term averaged
winds, large subtropical gyres tend to form at around 30◦ latitude in the northern and
southern hemisphere. The effect of the wind is to push surface water into a broad
mound at the centre of the gyre (through Ekman transport as described in Chap. 2,
Sect. 2.3.5), creating a high pressure centre.
Fig. 3.16 Coriolis effect and geostrophic currents. Blue arrows indicate the pressure gradient force
and red arrows indicate the Coriolis force
As water moves from high to low pressure areas, this will induce a current below
the surface boundary layer where water moves away from the centre of the gyre
(Fig. 3.16b). The Coriolis effect deflects this current to the right. This process even-
tually contributes to maintain geostrophic currents, where the pressure gradient is
balanced by the Coriolis effect, and the current direction is parallel to the isobars
(see Chap. 2, Sect. 2.3.6). The major currents in the world oceans, such as the Gulf
Stream and the Agulhas Current, are examples of such persistent systems that are
approximately in geostrophic balance at all time.
3.3.2.1 The Boussinesq Approximation
Since density variations are relatively small in the ocean, i.e., ρ/ρ 1, we intro-
duce a constant reference density ρ0 :
ρ(x, y, z, t) = ρ0 + ρ(x, y, z, t), where ρ0 ρ.
The fact that density fluctuations are relatively small justifies the Boussinesq ap-
proximation, where the variable density ρ(x, y, z, t) is replaced by the constant
reference density ρ0 in all terms except where gravitational forces or spatial and
temporal variations in density are essential. For instance, applying the Boussinesq
approximation on the horizontal pressure gradient term gives us
1 1 1
∇p = ∇p ≈ ∇p.
ρ ρ0 (1 + ρ/ρ0 ) ρ0
Another way of stating the justification for the Boussinesq approximation is that in
a ‘Boussinesq ocean’, the difference in inertia between water masses is negligible,
but gravity is sufficiently strong so that the difference in specific weight between
water masses may still be significant. This approximation is good as long as acoustic
waves can safely be ignored.
90 T. Torsvik
3.3.2.2 The Hydrostatic Approximation
The vertical component of the momentum equation (3.24) is given as
Dw ∂p
ρ =− + μw − ρg + 2Ωu cos θ,
Dt ∂z
where the gravity term is negative since g is directed downwards. In many cases the
vertical momentum equation will be dominated by the hydrostatic balance between
the gravity force and the vertical pressure gradient, in which case the momentum
equation can be replaced by the hydrostatic equation
∂p
0=− − ρg.
∂z
Essentially, this equation simply states that the pressure at a given depth level is
equal to the weight of the water above it. This approximation is reasonable for mod-
els that do not resolve convective processes. However, for models that resolve pro-
cesses on a length scale L 10 km, the validity of the hydrostatic approximation
is questionable. Convective motion such as internal waves or gravity currents over
sills, cannot be represented correctly if we use the hydrostatic equation.
3.3.2.3 The Hydrostatic Primitive Equations
As a summary of the discussion above, we present here the hydrostatic primitive

equations, consisting of the equation for incompressibility
Dρ
= 0,
Dt
the continuity equation
∇ · u = 0,
the horizontal momentum equations given by the Navier–Stokes equations
Du 1 ∂p μ
−fv = + u,
Dt ρ0 ∂x ρ0
Dv 1 ∂p μ
+fu= + v,
Dt ρ0 ∂y ρ0
and the vertical momentum equation given by the hydrostatic equation
1 ∂p ρg
0=− − .
ρ0 ∂z ρ0
In addition to these equations we need balance equations for temperature T and

salinity S
DT
= heat flux sources and sinks,
Dt
DS
= salinity sources and sinks
Dt
and finally an equation of state
ρ = ρ(T , S, p)
to specify the dependence between density, pressure, temperature and salinity. The
equation of state is an internationally agreed standard for sea water, published by the
Joint Panel on Oceanographic Tables and Standards (JPOTS), derived from curve
fitting through laboratory measurements of temperature, salinity, and pressure. The
currently accepted equation of state consists of three polynomials with 41 constants
(JPOTS 1991).
3.3.3 Ocean Turbulence
Turbulence is an essential feature of ocean dynamics. It is difficult to give a pre-

cise definition of the term ‘turbulence’, but it is generally accepted that turbulence
involves a state of highly energetic, rotational and irregular motion that results in
a rapid transfer of momentum, heat and dissolved matter greatly exceeding trans-
fer rates achievable by molecular processes alone. By generating relatively large
gradients of velocity at small scales (typically 1 mm to 1 cm), turbulence facili-
tates the transfer of kinetic energy into heat through viscous dissipation. The first
substantial scientific study of turbulent motion was published by Reynolds (1883),
who demonstrated that the transition between a smooth, laminar flow and turbulent
motion depends on what is now known as the Reynolds number
ρU d Ud
Re = = ,
μ ν
where U is the flow speed, d is a length scale (in Reynolds’ experiments the diam-
eter of a tube), and ν = μ/ρ is the kinematic viscosity, which is about 106 m2 /s
for water. The transition Reynolds number, also called the critical Reynolds num-
ber, is highly dependent on fluid properties, geometric features constraining the
flow and the level of small scale background disturbances influencing the flow. For
instance, the critical Reynolds number for Reynolds’ tube experiment can range
from Re = 1 × 103 to 4.5 × 104 depending on the permitted background noise level
(Thorpe 2007). Fluid motion is usually considered to be turbulent if the Reynolds
number exceeds 104 .
92 T. Torsvik
Turbulent motion enhances mixing and dispersion. This is achieved by a combi-

nation of stirring at large and intermediate length scale, and diffusion at the molec-
ular level. Stirring extends the contact surface between fluid volumes, thereby in-
creasing the area across which diffusive transfer can occur. As a result, regions of
intense turbulent motion tend to produce well mixed volumes of fluid. While stir-
ring is usually considered to be a reversible process, mixing is irreversible. If for
instance a dye is inserted as a point source in a turbulent flow, the dye will even-
tually be mixed in the entire fluid volume and will not be accessible for extraction
except by some additional, artificial process that sorts the different fluid compo-
nents.
Since the mean depth of the world ocean is 3795 m and the average flow velocity
usually exceeds 0.1 m/s, the characteristic Reynolds number for the ocean appears to
be about 4 × 108 , which is far greater than the critical Reynolds number of 104 . By
this estimate we would expect the world ocean to be dominated by large overturning
eddies maintaining a well mixed condition. However, density variation due to sur-
face water heating and the distribution of salinity maintains a vertical stratification
in deep ocean and sea areas. This stratification is usually too strong to permit large
overturning eddies, so turbulent motion is usually found in patches with vertical
length scales of 1–100 m.
Due to the predominant effect of vertical stratification it is common to separate
between mixing that involves transfer of fluid across isopycnal surfaces, called di-
apycnal mixing, and mixing of water masses parallel to isopycnal surfaces, called
isopycnal mixing. Diapycnal mixing involves the mixing of water masses with dif-
ferent densities, which requires an overturning motion where dense water is lifted
above less dense water, hence involving a transfer from kinetic to potential energy.
The end result is a patch of water with intermediate density which will spread along
an isopycnal surface and contribute to weaken the vertical stratification. Isopycnal
mixing involves the redistribution of volumes of water with equal density, often
through the formation of mesoscale eddies with diameter of 50–300 km and a life-
time extending from a few days to months. Although the interior of the eddy is well
mixed, such mesoscale eddies may drift far from their origin, and a significant tem-
perature difference can often be observed between the interior and exterior of such
eddies.
Turbulence is prominent in patches and localized regions in the ocean. Strong
turbulent motion is usually present in the upper boundary layer near the sea surface
and in the bottom boundary layer. In the upper boundary layer the wind stress gener-
ates waves which can induce turbulent motion, and the heat flux may create patches
of variable buoyancy. At the bottom boundary layer the frictional forces between the
bottom current and the sea floor are usually the origin of turbulent motion. Patches
of turbulent motion may also occur away from boundaries. Internal waves create
vertical gradients of current velocities, and breaking internal waves are well known
sources of diapycnal mixing. In regions where warm and salty water lies beneath
cold and less salty water, such as in Arctic regions, the difference in diffusion rate
of heat and salinity may lead to instabilities and create what is called double diffu-
sive convection. The source of this instability is the fact that the diffusion of salt is
much smaller than the diffusion of heat. For instance, if a volume of cold surface
water is displaced downwards, after being heated by the warmer water below, it will
attain a lower density than in its initial state, therefore rise into the colder water
above to be cooled down, and in this way create a persistent oscillating motion.
3.3.4 Subgrid-Scale Parameterization
Due to the finite resolution of spatial grids and the finite time integration step, and
practical requirements that these should not be too small, there will always be un-
resolved processes that are taking place on the subgrid scale. Specifically, we have
seen that viscous effects work on a typical length scale of L < 10−2 m, and the same
is true for thermodiffusivity and salt diffusivity. All this would be fine from a mod-
elling point of view if these small scale effects did not influence the large scale ocean
dynamics. Unfortunately, this is not the case. Viscous and diffusive effects cannot
be ignored for several reasons. The global inputs of momentum, heat and vorticity
require a sink in order to avoid unphysical accumulation of energy. Important driv-
ing forces, such as wind stress, are frictional in origin, and bottom friction may be a
significant factor in localized regions, in particular for coastal water. Since it is clear
that the models cannot resolve these processes, such effects must be modelled with
the use of subgrid-scale parameterization.
One approach to obtain such parameterization dates back to the 19th century
and is due to Reynolds, who suggested to write all hydrodynamic variables as a
sum of a large-scale (or long-period) component and a small-scale (or short-period)
component
p = p̄ + p̂, u = ū + û, etc.,
where ( ¯· ) represents some averaging operator, e.g., time averaging, and (ˆ·) repre-
sents the deviation from the average. Ideally, these scales should be clearly separated
in spectral space, a requirement that unfortunately is not easily satisfied in the ocean
where normally the spectrum is continuous. Starting from the inviscid momentum
equation written in tensor notation
∂ui ∂ui 1 ∂p ρg
+ uj − f ij 3 uj = − − δi3 ,
∂t ∂xj ρ0 ∂xi ρ0
where i, j ∈ {1, 2, 3}, and we use the Kronecker’s delta

1 for i = j,
δij =
0 for i = j,
define ij k as
⎧
⎨1 for cyclic order of indices
ij k = −1 for anticyclic order of indices
⎩
0 for two or more identical indices
94 T. Torsvik
and consider only the averaged variables, we obtain the Reynolds–Averaged Navier–
Stokes (RANS) equation
∂ ūi ∂ ūi 1 ∂ p̄ ρ̄g ∂ ûi ûj

+ ūj − f ij 3 ūj = − − δi3 − .
∂t ∂xj ρ0 ∂xi ρ0 ∂xj
The RANS equation is similar to the standard Navier–Stokes equation, but contains
one additional term which is due to averaging over small-scale components. The
term ûi ûj is called the Reynolds stress tensor. It is a symmetric tensor where the
diagonal elements ûi ûi represent normal stresses and the off-diagonal components
are shear stresses. The turbulent momentum equation
∂ ûi ∂ ûi ∂ ūi ∂ ûi ûi ûj 1 ∂ p̂ ρ̂g

+ ūj + ûj + ûj − − f ij 3 ûj = − − δi3
∂t ∂xj ∂xj ∂xj ∂xj ρ0 ∂xi ρ0
provides a way to specify the Reynolds stress tensor, but only in terms of unknown
small-scale variable components. This introduces the so-called closure problem of
the parameterization. In order to specify the values of the unknown variables in the
turbulent momentum equation we need yet more equations with yet more unknown
variables.
A large number of subgrid-scale mixing schemes have been proposed. Differ-
ent parameterization schemes are typically used to represent horizontal (isopycnal)
and vertical (diapycnal) mixing, and advanced model systems may include several
different mixing schemes. The simplest approach is to parameterize the Reynolds
stresses by assuming a linear dependence on the gradients of the large scale flow
field

∂ ûi ûj ∂ ∂ ūi
=− A ,
∂xj ∂x ∂xj
where A is a 3 × 3 matrix. Further simplification is obtained by specifying that only
the diagonal elements of A should be non-zero
⎛ ⎞
AM 0 0
A=⎝ 0 AM 0 ⎠,
0 0 Av
where AM and Av are the horizontal and vertical eddy viscosity coefficients, re-
spectively. Typically, AM Av due to the difference in typical length scales in the
ocean.
A more advanced method relies on the evolution equation for the turbulent kinetic
energy
1 1
TKE = ûi ûi = (ûû + v̂ v̂ + ŵ ŵ),
2 2
which is given by
∂TKE ∂TKE ∂ ūi ∂ ûi p̂ g ∂ ûi ûi ûj

+ ūj + ûi ûj =− − ûi ρ̂δi3 − .
∂t ∂xj ∂xj ∂xi ρ0 ∂xj
A closed set of equations results from parameterization of the small scale compo-
nents in the equation, e.g., by setting
∂ ūi ∂ ρ̄
ûi ûj = AM , ûi ρ̂ = AT ,
∂xj ∂xi
where AT is a horizontal eddy diffusivity coefficient, and parameterizing the triple

product term by some length scale l:

∂ ûi ûi ûj (TKE)3
= ε = cε .
∂xj l
This approach forms the basis for popular schemes such as the k–l and k–ε schemes.
For high resolution simulations it may be feasible to simulate the large scale isopy-
cnal turbulent motion and only parameterize the small scale motion. In this case
a large-eddy simulation (LES) scheme, such as the Smagorinsky model, may be
applied.
The procedure outlined above shows how parameterizations can be introduced
in the model equations. A simple parameterization may replace a subgrid-scale
variable with a constant value believed to be representative, whereas a complicated
scheme may involve a set of coupled differential equations. The requirements placed
upon the parameterization often depend on what processes are resolved in the nu-
merical model, i.e., the parameterization is usually scale dependent, depending on
what physics needs to be represented. As modellers push for higher resolution mod-
els, the need for subgrid-scale parameterization does not diminish, but the formula-
tions used for parameterization may very well need to be changed.
3.3.5 Classification of Ocean Models
Most of the ocean models in use today have their origin in the university or meteo-
rological institute sector, and are developed and maintained by the user community.
Ocean models have traditionally been based on FD methods, where the main differ-
ences in model design have been related to the vertical grid. This picture is chang-
ing, as both FV and FE methods have started to appear in ocean models, and hybrid
methods for vertical grids are being developed. However, it is instructive to follow
the traditional approach and classify the models primarily in terms of the vertical
discretization method.
Figure 3.17 shows the three main categories of vertical grid configurations used
for ocean models. The z-coordinate system (also called geopotential coordinates) in
96 T. Torsvik
Fig. 3.17 Vertical coordinate systems
Fig. 3.17a uses horizontal surfaces to divide grid cells in the vertical. This is an easy
grid to work with from a design point of view, and has the advantage that the hy-
drostatic condition is represented exactly. Sloping bottom topography is a challenge
when working with this grid. The simple approach where grid cells are defined as
‘dry land’, i.e., inactive cells, whenever the bottom topography intersects the cell
may result in a staircase representation at steep slopes. This problem can be miti-
gated by allowing for partial cells near the bottom boundary, at the cost of making
the model more complicated. In practical applications with significant bathymetry
variations, the use of the z-coordinate system may result in a large number of inac-
tive cells because the grid must extend globally to the maximum depth of the basin.
The z-coordinate system is also problematic for locations where the isopycnals are
not horizontal surfaces, in which case pressure gradient errors may occur. The rigid
structure of the grid also makes it difficult to combine with a free surface represen-
tation at the interface between ocean and atmosphere. Examples of models using
z-coordinates for vertical discretization include MOM4 and MITgcm.5
The s-coordinate system (also called the sigma-coordinate or terrain-following
coordinate system) shown in Fig. 3.17b, defines the vertical coordinates as fractions
of the total water depth, which may be static or dynamic depending on whether
the mean sea level or the free surface level is used as reference point at the ocean-
atmosphere interface. This coordinate system allows the model to use a high vertical
grid resolution near the bottom and free surface boundaries, thereby improving the
representation of turbulent boundary layers. However, the s-coordinate system has
potentially more challenges related to pressure gradient errors than the z-coordinate
system, in particular at the location of steep slopes in the bottom topography. Ex-
amples of models using the s-coordinate system include POM6 and ROMS.7
The isopycnal coordinate system shown in Fig. 3.17c defines vertical coordinates
as isosurfaces of equal density. This solves the problem of artificial pressure gradi-
4 GFDL Modular Ocean Model, http://www.gfdl.noaa.gov/ocean-model.

5 M.I.T. general circulation model, http://mitgcm.org/.
6 Princeton Ocean Model, http://aos.princeton.edu/WWWPUBLIC/htdocs.pom/.
7 Regional Ocean Modeling System, http://www.myroms.org/.
ents which are associated with the z-coordinate and s-coordinate systems. Another
advantage is that this coordinate system makes it easy to obtain high grid resolu-
tion where density gradients are strong. A challenge with this method is that the
vertical layers may become very thin or even vanish in parts of the model domain.
For instance, brackish coastal water usually has a lower density than any water mass
found in the middle of the ocean. Thin layers are also a source of numerical instabil-
ities. Another potential problem is that the isopycnal coordinate system consolidates
isopycnals as material surfaces, thereby restricting the cross-isopycnal mixing. Ex-
amples of models using isopycnal coordinates include MICOM8 and NLOM.9
Each of the vertical grids presented above have specific strengths and weak-
nesses. Hybrid models, such as HYCOM,10 allow modellers to apply different ver-
tical grid systems for different regions in the model domain, using each system
according to its strengths. At the same time as hybrid systems are being developed,
improved methods are being developed for traditional coordinate systems in order
to overcome the perceived weaknesses. The introduction of FE and FV methods al-
low models to use unstructured grids. One of the popular FV models, FVCOM,11
uses unstructured grids in the horizontal dimensions only, but still uses generalized
terrain-following coordinates for vertical discretization.
3.3.6 Initial and Boundary Conditions
In order to have a working numerical model, we need to specify boundary condi-

tions, initial conditions and forcing mechanisms. These conditions will obviously
depend to some extent on the particular application, so only a very general descrip-
tion of these features is presented here.
An ocean model typically has boundary conditions at the sea floor and at the sea
surface. In addition, a numerical model covers a limited computational domain, and
boundary conditions must also be specified for the lateral model boundaries.
For the sea floor it is most common to specify a no-flux condition. The kinematic
boundary condition for the bottom is
∂H ∂H
wb = −ub − vb at z = −H (x, y),
∂x ∂y
where H (x, y) is the water depth and ub = (ub , vb , wb ) is the bottom current ve-
locity vector. This is a free-slip boundary condition if the horizontal velocity com-
ponents u, v are allowed to be non-zero near the bottom. Alternatively, one may
8 Miami Isopycnic Coordinate Ocean Model.

9 Navy Layered Ocean Model, http://www7320.nrlssc.navy.mil/global_nlom/.
10 HYbrid Coordinate Ocean Model, http://hycom.org/.
11 Finite Volume Coastal Ocean Model, http://fvcom.smast.umassd.edu/FVCOM/.
98 T. Torsvik
specify a no-slip boundary condition
ub = vb = wb = 0 at z = −H (x, y).
In either case, additional parameterization terms are usually needed to account for
the influence of the bottom friction and the bottom boundary layer on the general
circulation. The simplest parameterization is the linear bottom stress formula
τ B = rb ub ,
which enters as an additional term in the momentum equation for the bottom veloc-
ity. A typical value for the bottom resistance coefficient rb is 0.0002 m/s. A more
realistic treatment is provided by the quadratic stress formula

τ B = −cB ub u2b + vb2 ,
where the non-dimensional quadratic bottom stress coefficient cB usually is set to a

constant value between 0.001 and 0.003.
The simplest boundary condition for the sea surface interface is the rigid lid
approximation, where we specify
w = 0 at z = 0,
which is usually used in combination with z-coordinate models. When terrain-

following vertical coordinates are used, it is more common to use the kinematic
boundary condition
∂η ∂η ∂η
w= +u +v at z = η(x, y, t).
∂t ∂x ∂y
To simplify the calculation of the kinematic boundary condition it is often assumed
that the sea surface displacement is small and therefore that the condition can be
computed for z = 0 instead of for z = η. The wind stress acting on the free sur-
face interface can be accounted for in a similar manner as the bottom friction term.
A typical wind stress parameterization is provided by the quadratic stress formula
τ w = ρa cD |U10 |U10 ,
where ρa is the density of the air and U10 is the wind speed 10 meters above the
sea surface (so-called 10 m wind). A typical value for the non-dimensional drag
coefficient cD is 0.003.
The wind stress depends on wind velocity data that is not provided by the ocean
model itself. Wind data from a climatological dataset is required to perform this
calculation. Atlases of climatological data have been built based on observations
from weather stations and satellites, and by using output from global or large scale
weather or climate models. Meteorological reanalysis is a data assimilation tech-
nique where numerical weather models are fitted to historical data. Notable exam-
ples include the European Centre for Medium-Range Weather Forecasts (ECMWF)
reanalysis projects, ERA-40 and ERA-Interim, and the National Centres for En-
vironmental Prediction and for Atmospheric Research (NCEP/NCAR) reanalysis,
which provide data for the state of the Earth’s atmosphere from mid-20th century
until present time. In addition to wind data, ocean models require data for atmo-
spheric temperature in order to prescribe a realistic heat flux at the free surface
interface.
The lateral boundary conditions are determined by the specific application. If the
sidewall of the domain can be considered as a solid wall, as is the case when the
sidewall is located on dry land, we can use the no-flux condition
u · n = 0,
where n is the outward normal vector for the sidewall. In many applications at least
some of the lateral boundaries will intersect sea areas. In the case of global mod-
els, where the lateral boundary may be located along a line of longitude, periodic
boundary conditions can be used at the east and west boundaries. The location of
Antarctica at the South Pole allows the use of a no-flux wall condition, but the Arc-
tic Ocean is difficult to represent in such a model. One solution to this problem is
to model the Arctic Ocean and North Atlantic with a separate model using a rotated
grid with poles at the equator, as is done in the OCCAM (Ocean Circulation and
Climate Advanced Modelling Project) global ocean model. An alternative method
is to apply a non-spherical coordinate system where polar singularities are hidden
over land.
The sea intersecting lateral boundary condition becomes more complicated for
regional models which should allow flow exchange with sea areas not covered by
the numerical model. Such open boundary conditions should be able to eliminate
any reflection from waves and currents leaving the computational domain, and al-
low inflow of water masses according to parameters describing conditions in the
exterior sea areas. The parameters for exterior sea areas may reflect transport due to
major ocean currents and seasonal variability in average temperature and salinity, or
may be extracted from a global ocean model in a similar way as the atmospheric cli-
matology datasets. There is no universally accepted ‘correct’ way to prescribe such
open boundary conditions, and a single model may use a combination of several
methods (zero gradient, radiation conditions, flow relaxation) to achieve the desired
result.
Climatological datasets, including hydrographic data for potential temperature
and salinity, are important for the initialization of ocean models. If a global ocean
model is initiated with a homogeneous state, it will take thousands of simulated
years for diffusive processes to generate realistic deep ocean water masses. Even if
the model is initialized with hydrographic data close to the long term equilibrium
state, the model will usually require some spin-up time before a realistic circula-
tion pattern emerges. At the start of a simulation the initial density field is adjusted
towards an equilibrium state. For a global circulation model it takes from a few
weeks to a few months to set up the barotropic wind generated currents, whereas
the baroclinic response is set up after a few years, and the thermohaline circulation
100 T. Torsvik
is established after about a decade of simulated time. For regional scale models the
spin-up time is much shorter, particularly if the model domain is influenced by large
water exchange across open lateral boundaries.
For readers who wish to learn more about numerical ocean modelling, Haidvogel
and Beckmann (1999) provide a good overview of the basic concepts. For further
information on how to run ocean models, the best and most up-to-date information
is usually available on the web sites and in the user manuals for each model, so no
references on this subject will be included here.
Acknowledgements This work, originally motivated by the BONUS project BalticWay, was
supported by the European Union through the Mobilitas grant MTT63 and through support to
the Estonian Centre of Excellence for Non-linear Studies (CENS) from the European Regional
Development Fund, and by targeted financing by the Estonian Ministry of Education and Research
(grant SF0140007s11).
References
Anderson JD Jr (1995) Computational fluid dynamics: the basics with applications. McGraw-Hill,
New York
Benestad RE, Hanssen-Bauer I, Chen D (2008) Empirical–statistical downscaling. World Scien-
tific, Singapore
Ferziger JF, Perić M (2002) Computational methods for fluid dynamics, 3rd edn. Springer, Berlin
Haidvogel DB, Beckmann A (1999) Numerical ocean circulation modeling. Imperial College
Press, London
Johnson C (2009) Numerical solution of partial differential equations by the finite element method.
Dover, New York
Joint Panel on Oceanographic Tables and Standards [JPOTS] (1991) Processing of oceanographic
station data. Unesco, Paris. ISBN 9231027565/9789231027567
LeVeque RJ (2002) Finite volume methods for hyperbolic problems. Cambridge University Press,
Cambridge
Monaghan JJ (2012) Smoothed particle hydrodynamics and its diverse applications. Annu Rev
Fluid Mech 44:323–346
Reynolds O (1883) An experimental investigation of the circumstances which determine whether
the motion of water shall be direct or sinuous, and the law of resistance in parallel channels.
Philos Trans R Soc Lond Ser A 174:935–982
Thomas JW (1995) Numerical partial differential equations: finite difference methods. Springer,
Berlin
Thorpe SA (2007) An introduction to ocean turbulence. Cambridge University Press, Cambridge
Chapter 4
Studying the Baltic Sea Circulation
with Eulerian Tracers
H.E. Markus Meier and Anders Höglund
Abstract As shipping of environmentally hazardous cargo, like oil, has increased

considerably in the Baltic in recent years, methods are needed to calculate the fair-
ways between two harbours such that hazardous substances from a hypothetical
accident will stay as long as possible away from ecologically sensitive areas like the
coastal zone. For this purpose an ensemble approach based upon Eulerian tracer sim-
ulations is presented which has the potential to be further developed to become op-
erational for the optimization of fairways. First, we introduce and compare Eulerian
and Lagrangian descriptions of any fluid in general. Second, a three-dimensional
circulation model of the Baltic Sea is presented from which currents are used to cal-
culate the evolution of the Eulerian tracers in time that obey traditional advection-
diffusion equations. The model set-up is presented in detail to illustrate the potential
of ocean circulation models for our purposes but also their shortcomings. Third, ex-
amples of studies using Eulerian tracers are presented that analyse the characteris-
tics of the circulation, like ventilation time scales and age of water masses. Finally,
we focus on three selected examples of oil spill modelling using Eulerian methods.
Although oil spill modelling very often utilizes a Lagrangian particle approach, we
show that even Eulerian methods can be used that might under certain circumstances
have some advantages compared to the Lagrangian approach.
H.E.M. Meier (B) · A. Höglund

Swedish Meteorological and Hydrological Institute, 60176 Norrköping, Sweden
e-mail: markus.meier@smhi.se
A. Höglund
e-mail: anders.hoglund@smhi.se
H.E.M. Meier
Department of Meteorology, Stockholm University, 10691 Stockholm, Sweden

DOI 10.1007/978-3-319-00440-2_4,
102 H.E.M. Meier and A. Höglund
Fig. 4.1 Persistency (in %) of the mean transport above (left panel) and below (right panel) the
halocline for 1981–2004. Adopted from Meier (2007)
4.1 Introduction
4.1.1 Motivation
The approach presented in this chapter is based upon the idea to use surface currents
simulated with an ocean circulation model for the calculation of fairways with min-
imized risk that an oil spill will harm ecologically sensitive areas, like the coastal
zone (cf. Chap. 1). For this purpose both long and spatially highly resolved records
of currents are needed because in the Baltic Sea their temporal and spatial variability
are large. For instance, in the Baltic Sea inertial oscillations (caused by the Earth’s
rotation) have periods of about 14 hours (Gustafsson and Kullenberg 1936; see also
Krauss 1973) and in the Baltic Proper1 the baroclinic Rossby radius (a spatial scale
of mesoscale eddies in the ocean) has values in the range between 3 and 10 km
(Fennel et al. 1991).
Despite the large variability of currents, model studies by, e.g., Lehmann and
Hinrichsen (2000) and Meier (2007), showed that a relatively persistent mean sur-
face circulation consisting of sub-basin wide, cyclonic gyres exists (Fig. 4.1). Here,
the persistency of the direction of the mean circulation is defined as the ratio be-
1 This notion is used here to denote the Eastern, Northern and Western Gotland Basin, Bornholm
Basin and Gdańsk Bay (cf. Chap. 2, Fig. 2.1).

4 Studying the Baltic Sea Circulation with Eulerian Tracers 103
Fig. 4.2 Examples of real

maritime routes for hazardous
cargo (oil, gas and
chemicals). The routes are
optimized for fuel
consumption by the GAC
SMHI Weather Solution
service. Adopted from
Höglund and Meier (2012)
tween the mean current velocity and the mean current speed2 (Palmén 1930, see also
Chaps. 2 and 6). Below it will be shown that the knowledge about the anisotropic
nature of the current field can be used to calculate optimized shipping routes.
Today only fuel consumption and travel time are taken into account when ship-
ping routes are set which result in routes that basically follow the shortest distance
between two harbours (see Chap. 11 for more information). Examples of real mar-
itime routes for hazardous cargo between the Kattegat and some selected destina-
tions in Russia and Estonia are shown in Fig. 4.2. The question is how these routes
will change if both the ecological risk of potential oil spills and fuel consumption
are minimized simultaneously.
However, the problem is very complex because the spreading of oil is con-
trolled not only by currents but also by wind and surface waves. Earlier studies
showed that in offshore areas oil slicks drift, in light winds without breaking waves,
with about 3–3.5 % of the wind speed in the direction of the wind (ASCE 1996;
Reed et al. 1999). According to our results over the Baltic Sea the median wind
speed amounts to 6.8 m/s which may correspond to an oil drift of 20.5 cm/s if
only the wind effect is taken into account. As the median current speed amounts
to 6.7 cm/s, in most wind conditions the impact of the wind cannot be neglected
(Fig. 4.3).
In addition, the oil spill is not only restricted to the sea surface but vertically
mixed in the water column due to ocean turbulence, especially under strong wind
conditions, and the released oil is transformed into various phases by physical,
chemical and biological processes as described in more detail in Chap. 11. To il-
lustrate one possible method how to calculate the dynamics of oil spills and the
subsequent optimization of fairways, in this chapter a simplified approach based
upon the so-called Eulerian description (see the next section) is presented taking
only the impact of currents on the oil spill into account.
2 A current vector is defined by its magnitude (or current speed) and its direction in three dimen-
sions.
Fig. 4.3 Occurrence (in %)

of simulated currents (solid
line) and currents calculated
from 3 % of the wind speed
(dashed line) (in cm/s) during
a 8-year period (1961–1969)
4.1.2 Eulerian Versus Lagrangian Approaches
The motion of fluids can be described in several ways. Most common are Eule-
rian and Lagrangian descriptions. Within the Eulerian description the observer is
stationary and observes what passes by the observer’s location. The motion is de-
scribed by the speed and direction with which the matter passes by this location. In
a Lagrangian description the observer follows along with the observed matter. The
motion is described by the time-dependent position of the matter.
The two approaches for describing the motion make the mathematical descrip-
tions very different. There is not merely a change of coordinates. Such a change
would not turn a velocity into a position. In fact, both descriptions can be used in
any coordinate system.
A traditional ocean circulation model describes the motion and other physical
properties of the ocean in a fixed three-dimensional (3D) grid of calculation points
or boxes. In other words, traditional ocean circulation models (that are discussed in
some detail in Chap. 3) are Eulerian (Bryan 1969). While this is a proper description
of the horizontal calculations, there is a larger variety of alternatives for the vertical
part. In many ocean models the vertical positions are chosen differently for each
time step in order to take the varying sea level into account or to allow the different
levels to follow the density of the water as it changes (see Chap. 3). These models
are still regarded to be Eulerian.
Note that an Eulerian description does not necessarily imply a regular grid. There
are, for instance, ocean models based on an unstructured mesh and finite element
methods. The mesh can be adaptive to catch small scale dynamics or fixed with a
denser mesh in areas where important small scale phenomena are known to occur.
The Eulerian and Lagrangian descriptions are special cases of a more general
scheme with a moving observer that does not necessarily move with the matter.
Such an observer would measure a relative speed and direction of the matter. For
the special case when this relative speed is constantly zero the location of the ob-
server would be the same as the location of a parcel of matter and we would get the
Fig. 4.4 Left: The tracer model is run on-line. Currents from the ocean circulation model are
directly used by the tracer model. Right: The tracer model is run off-line. Currents from the ocean
model are stored on disk and the tracer model reads the currents from disk. The tracer model can
be run several times with varying setups using the same currents
Lagrangian description. For the other special case when the observer is stationary
the relative velocity would be the same as the absolute velocity of the matter and we
would get the Eulerian description.
The Lagrangian method based upon trajectories of particles has the disadvan-
tage that subgrid scale mixing used for the calculation of currents is not included.3
For the results of the analysis, especially vertical turbulent mixing might be impor-
tant on longer time scales. As the circulation models utilized so far for long runs,
like climate simulations, are only eddy-permitting (see below), the results might
also be affected by the horizontal mixing parameterizing mesoscale eddies. Anal-
ysis methods using Eulerian tracers do not have this limitation because the same
parameterizations are used for the calculation of currents and particle spreading.
Rather frequently one wants to utilize the simulated currents for the calculation of
drifting matter in the ocean. For instance, it can be the tracking of (1) a person who
has fallen overboard from a boat, (2) drifting algae, or (3) oil spills. Henceforth, the
simulated variable, which indicates drifting matter, is called tracer. A tracer model
may use stored currents from an ocean circulation model and it may be run at a later
occasion than the ocean circulation model. This is called to run the tracer model off-
line (Fig. 4.4). Alternatively, the tracer model may be built together with the ocean
circulation model and it may get the currents directly from the ocean model. This
is called coupling the models or running the tracer model on-line. One advantage
with the latter choice is that the currents can be updated with the same time step as
the ocean model is using, while in the off-line case the currents are typically stored
sparser in time compared to the time step of the ocean model. An advantage with the
off-line method, especially if the tracer model requires less computer power than the
ocean circulation model, is that it can more easily be repeated with small changes
3 The formal parameterizations of sub-grid processes described in Chaps. 3, 7 and 10 only improve
statistics of spreading of trajectories but do not improve the match of trajectories with reality.
than if the models were coupled. The choice of off-line or on-line is independent of
whether the model is Eulerian or Lagrangian.
In a typical ocean circulation model there is code to transport tracers that drift
with the currents. This process is described using a simple advection-diffusion equa-
tion which is used for temperature and salinity and possibly other tracers as well
(Krauss 1973). In general, it is not very complicated to add more such fields, typi-
cally just to extend some loops in the source code. These extra tracers which are not
active in the dynamics of the ocean model are called passive tracers. In this chapter
applications of passive tracers with focus on the idealized modelling of spreading
oil are discussed.
4.2 Ocean Circulation Modelling
4.2.1 Model Dynamics
4.2.1.1 Basic Equations
Following the basic principles and approximations as outlined in Chaps. 2 and 3,

general circulation models are used to calculate the Baltic Sea circulation in three
dimensions. An example of such a model is the Rossby Centre Ocean model (RCO)
which has open boundary conditions in the northern Kattegat (Meier et al. 2003;
Meier 2007). The RCO model is a traditional Bryan–Cox–Semtner type, primitive
equation model (Bryan 1969; Semtner 1974; Cox 1984) based upon the z-level
Ocean Circulation and Climate Advanced Model (OCCAM) (Stevens 1991; Kill-
worth et al. 1991; Webb et al. 1997). The primitive equations are derived from the
Navier–Stokes equations using the Boussinesq, the shallow water, the so-called tra-
ditional and the hydrostatic approximation (e.g., Krauss 1973; Müller and Wille-
brand 1989). The traditional approximation neglects the component of the Coriolis
force due to vertical current velocity. The equations expressing the conservation of
momentum, mass, potential temperature and salinity are
∂u 1 ∂p
+ Γ (u) + f v = − + Fu , (4.1)
∂t ρ0 R cos φ ∂λ
∂v 1 ∂p
+ Γ (v) − f u = − + Fv , (4.2)
∂t ρ0 R ∂φ
Γ (1) = 0, (4.3)
∂p
= −gρ, (4.4)
∂z
∂T 1 ∂I
+ Γ (T ) = FT + , (4.5)
∂t ρ0 cpw ∂z
∂S
+ Γ (S) = FS , (4.6)
∂t
ρ = ρ(T , S, p), (4.7)
with the advection operator

1 ∂ ∂ ∂
Γ (μ) = (uμ) + (vμ cos φ) + (wμ), (4.8)
R cos φ ∂λ ∂φ ∂z
with μ = (T , S, u, v, 1), the Coriolis parameter f = 2Ω sin φ, Earth’s radius
R = 6370 km, angular speed of rotation of the Earth Ω = 2π/86164 rad/s, acceler-
ation of gravity g = 9.81 m/s2 , reference density of water ρ0 = 103 kg/m3 , specific
heat capacity of water cpw = 4.186 × 103 J/(kg K) and solar insolation I .
This system of partial differential equations (4.1)–(4.7) can be solved for the
seven dependent variables: velocity components u, v, w, pressure p, potential tem-
perature T , salinity S and density ρ as a function of time t, latitude φ and longitude
λ as well as water depth z (usually interpreted as <0) if an equation of state (4.7)
and boundary conditions are prescribed. The vertical coordinate z is positive up-
ward and zero at the sea surface. Fu , Fv , FT and FS denote divergences of turbulent
fluxes which are parameterized according to

∂ ∂u
Fu = νt (z) + AM ∇ 2 u, (4.9)
∂z ∂z

∂ ∂v
Fv = νt (z) + AM ∇ 2 v, (4.10)
∂z ∂z

∂ νt ∂T
FT = (z) + AT ∇ 2 T , (4.11)
∂z σt ∂z

∂ νt ∂S
FS = (z) + AT ∇ 2 S, (4.12)
∂z σt ∂z
with

1 ∂ 2μ 1 ∂ ∂μ
∇ 2μ = + cos φ . (4.13)
R 2 cos2 φ ∂λ2 R 2 cos φ ∂φ ∂φ
Additional metric terms in Fu and Fv are neglected, AM and AT denote horizontal
viscosity and diffusivity coefficients, respectively, νt is the turbulent vertical friction
coefficient and σt is the turbulent Prandtl number. As timescales of barotropic and
baroclinic processes are different, it is more efficient rather than integrating Eqs.
(4.1)–(4.7) to introduce an external and internal mode with different time steps. For
details the reader is referred to Killworth et al. (1991), Meier et al. (1999). The
prognostic equations are discretized on an Arakawa B grid (Mesinger and Arakawa
1976).
The lateral boundary conditions are ‘no slip’ for momentum and isolation for
tracer:
∂T ∂S
u=v= = = 0, (4.14)
∂n ∂n
where n is a normal vector to the coast. At the ocean surface (z = 0) the boundary
conditions are
∂u ∂v
ρ0 νt = τλ , ρ0 νt = τφ , (4.15)
∂z ∂z
νt ∂T νt ∂S
ρ0 cpw = QT , = SF , (4.16)
σt ∂z σt ∂z
∂ζ
w= . (4.17)
∂t
Here QT and SF are total heat flux at the sea surface without solar insolation and
salt flux, respectively, ζ is the sea level elevation, and τλ and τφ are the longitudinal
and latitudinal components of the wind stress at the sea surface.
At the sea bottom (z = −H ) the boundary conditions are
∂u ∂v
ρ0 νt = τλB , ρ0 νt = τφB , (4.18)
∂z ∂z
νt ∂T νt ∂S
ρ0 cpw = 0, = 0, (4.19)
σt ∂z σt ∂z
u ∂H v ∂H
w=− − , (4.20)
R cos φ ∂λ R ∂φ
where τλB and τφB are the longitudinal and latitudinal components of the bottom
shear stress.
4.2.1.2 Equation of State
The equation of state of sea water (4.7) is calculated using a third order polynomial
approximation in form of sigma anomalies. The method follows Bryan and Cox
(1972). The coefficients compute density as a function of temperature and salinity
at pre-determined depths as used in the model. The equation of state is set by the
Joint Panel on Oceanographic Tables and Standards (UNESCO 1981) as described
in Gill (1982). An iterative least-squares polynomial fitting for the over-determined
system is performed.
4.2.1.3 Sea Surface Boundary Conditions
To calculate sea surface fluxes of momentum, heat and matter from parameters of the
atmospheric planetary boundary layer like 10 m wind (that is, the wind velocity at a
height of 10 m), 2 m air temperature, 2 m specific humidity, sea surface atmospheric
pressure, total cloudiness and precipitation bulk formulae are used.
Wind Stress Surface wind stress in Eq. (4.15) is parameterized according to

Large and Pond (1981):
τ = caw
d
ρa |U10 |U10 (4.21)
with

−3 1.2 if 0 < |U10 | ≤ 11 m/s
d
caw = 10 × , (4.22)
0.49 + 0.065|U10 | if 11 m/s < |U10 |
where U10 and ρa denote the wind vector at the height of 10 m and air density,
respectively.
Heat Flux The total heat flux through thesea surface QTOT is given by the sum
of short-wave radiation QSW , long-wave incoming radiation QLW↓ , long-wave out-
going radiation QLW↑ , sensible QS and latent heat fluxes QL :
QTOT = QSW + QLW↓ − QLW↑ + QS + QL . (4.23)
The short-wave radiation QSW penetrates into deeper ocean layers and is dis-
cussed in Sect. 4.2.1.4. The total heat flux without solar insolation is given by
QT = QTOT − QSW (4.16). The long-wave incoming radiation QLW↓ is calculated
according to Bodin (1979). The long-wave outgoing radiation QLW↑ is calculated
from Stefan Boltzmann’s law:
QLW↑ = εw σs T 4 (4.24)
with the emissivity of the water surface εw = 0.97, Stefan Boltzmann’s constant
σs = 5.68 × 10−8 W/(m2 K4 ) and water temperature T . Sensible and latent heat
fluxes are determined by temperature and moisture differences between water and
air, respectively. The parameterizations follow (Large and Pond 1982) including
a stability dependence of the transfer coefficients. The stability is estimated using
the difference between 2 m air temperature and sea surface temperature, i.e., water
temperature of the first model layer.
Fresh Water Flux According to Krauss (1973), in the sea surface boundary con-
dition for salinity (partial mass, see Eq. (4.16)), the salt flux SF is given as
SF = −HS S (4.25)
with the fresh water flux HS
HS = R + P − E, (4.26)
governed by river runoff R, precipitation P and evaporation E. This traditional
approach is used here. Steinhorn (1991) discussed an inaccuracy in (4.25) and sug-
gested another relationship:
S
SF = −HS . (4.27)
1−S
As the error for the Baltic Sea surface water is smaller than one percent in (4.25),
the traditional boundary condition is suitable. However, for the hypersaline Dead
Sea the revised formula should be used instead.
4.2.1.4 Insolation
The divergence of absorbed intensity I of the penetrated short-wave radiation is

heating the water column (Eq. (4.5)). This effect can contribute to summertime
warming of up to 2 ◦ C in 10 m depth. The solar intensity is parameterized according
to Paulson and Simpson (1977) with two extinction lengths

z z
I = QSW RSW exp + (1 − RSW ) exp (4.28)
ζ1 ζ2
with RSW = 0.64, ζ1 = 1.78 m and ζ2 = 3.26 m. The short-wave incoming radiation
QSW is calculated according to Bodin (1979). Climatological data from Dera (1992,
see his Table 5.3.1) have been used to optimize the unknown constant RSW and
the extinction lengths ζ1 and ζ2 utilizing a least-squares fit. The available data are
average monthly means of solar energy over the entire spectrum reaching particular
depths in the southern Baltic. The optimization procedure has been carried out as
described in Paulson and Simpson (1977). For comparison, the most turbid optical
water type III for oceans according to Jerlov (1968) uses the values RSW = 0.78,
ζ1 = 1.14 m and ζ2 = 7.9 m.
4.2.1.5 Horizontal Mixing
Horizontal viscosity and diffusivity are parameterized using a harmonic approach

(4.9)–(4.12). More advanced but not necessarily better parameterizations are scale-
dependent approaches like biharmonic friction or the Smagorinsky (1963) scheme.
For long-term simulations it is important that diffusivity is chosen as small as
possible. Hence, the coefficients for harmonic viscosity and diffusivity amount to
AM = 2 × 102 m2 /s and AH = 0, respectively. Minimum values for horizontal fric-
tion are necessary to ensure numerical stability of the discretization schemes.
4.2.1.6 Turbulence Model
Subgrid-scale vertical mixing is parameterized using a turbulence closure scheme

(Rodi 1980) with flux boundary conditions to include the effect of turbulence en-
hanced in the uppermost layer due to breaking surface gravity waves and a param-
eterization for breaking internal waves (Meier 2001). The approach follows the so-
called k–ε model with two additional prognostic equations for turbulent kinetic en-
ergy, TKE (k), and dissipation of TKE (ε) that need to be solved at every grid point
of the 3D model:

∂k ∂ νt ∂k
− = P + G − ε, (4.29)
∂t ∂z σk ∂z

∂ε ∂ νt ∂ε ε ε2
− = cε1 (P + cε3 G) − cε2 (4.30)
∂t ∂z σε ∂z k k
with
2 2
∂u ∂v νt 2
P = νt + , G=− N , (4.31)
∂z ∂z σt
g ∂ρ k2
N2 = − , νt = cμ . (4.32)
ρ0 ∂z ε
Here, N denotes the Brunt–Väisälä frequency. The constants cμ , cε1 , cε2 , cε3 , σk and
σε are given in Table 4.1 according to Rodi (1980). The turbulence model gives no
information about the turbulent Prandtl number σt . We used an empirical formula
following (Blanke and Delecluse 1993). Both, the turbulent friction coefficient νt
and σt close in (4.9)–(4.12) the system of partial differential equations for tempera-
ture, salinity, density, pressure and current velocity (4.1)–(4.13).
4.2.1.7 Bottom Friction
Unresolved bottom roughness is parameterized according to Cox (1984) using a

second order law for bottom friction (4.18):

−−−−→
τ B = ρ0 cB u2b + vb2 · (ub , vb ) (4.33)
with ub = uz=−H , vb = vz=−H . As during some major inflows, like the one in Jan-
uary 1993, the accumulated transport through the Danish Straits into the Baltic Sea
is known from observations (e.g., Matthäus et al. 1993; Jakobsen 1995), the bot-
tom friction coefficient cB can be determined comparing modelled and observed
transports. Provided that the critical strait cross sections are correct the bottom drag
coefficient is estimated to be cB = 0.5 × 10−3 .
4.2.2 Model Setup
4.2.2.1 Introduction
Ocean circulation models like the RCO model integrate the primitive equations for-
ward in time (cf. Chap. 3). Output fields are water temperature, salinity, currents,
sea level, and sea ice parameters like sea ice concentration and sea ice thickness
(Mårtensson et al. 2012). In the applications presented below, the RCO model was
used with a horizontal resolution of 3.7 km (2 nautical miles) and with equally thick,
vertical levels with layer thicknesses of 3 m (Höglund and Meier 2012). For the in-
tegration of the primitive equations the following information is needed:
1. water depth at all model grid points,
2. initial conditions of 3D fields of water temperature and salinity,
3. water temperature and salinity profiles at the lateral boundary that are prescribed
here in the case of inflow into the model domain,
4. records of sea level elevations at the lateral boundary,
Table 4.1 Model parameters and forcing data sets used in the RCO model. Details are explained
in the text
Abbreviation Parameter Value
φ, λ Horizontal grid resolution 2 , 4 (about 2 nautical miles)

z Vertical grid resolution 3m
t Baroclinic time step 150 s
Hmax Maximum depth (83 levels) 249 m
ρ = ρ(T , S, p) Equation of state Gill (1982)
AM Horizontal viscosity 2 × 102 m2 /s
AH Horizontal diffusivity 0
cμ Turbulence parameter 0.09
cε1 Turbulence parameter 1.44
cε2 Turbulence parameter 1.92
cε3 Turbulence parameter 0 (stable) and 1 (unstable)
σk Prandtl number 1
σε Prandtl number 1.3
σt Richardson number dependent Blanke and Delecluse (1993)
Prandtl number
RSW Penetration of short-wave 0.64
radiation
ζ1 Extinction length 1.78 m
ζ2 Extinction length 3.26 m
cB Bottom drag coefficient 0.5 × 10−3
τ Wind stress Large and Pond (1981)
QS Sensible heat flux Large and Pond (1982)
QL Latent heat flux Large and Pond (1982)
QLW↓ Long-wave incoming radiation Bodin (1979)
QLW↑ Long-wave outgoing radiation Stefan Boltzmann’s law
QSW Incoming solar radiation Bodin (1979)
T, S Initial conditions Observed profiles at monitoring stations
(Meier et al. 2003)
Ta , U10 , P , . . . Atmospheric forcing Three-hourly atmosphere variables
calculated with a regional atmosphere
model (Höglund et al. 2009; Meier et al.
2011)
R Runoff Monthly mean runoff from 29 rivers
(Bergström and Carlsson 1994)
ζ Sea level in the Kattegat Daily mean sea level at Smögen (Meier
and Kauker 2003)
5. two-dimensional (2D) forcing fields of atmospheric parameters to calculate the

surface fluxes of heat, momentum and matter, and river discharge representing
the total runoff from the Baltic Sea catchment area.
4.2.2.2 Bottom Topography
The Baltic Sea is a semi-enclosed sea consisting of a number of sub-basins that are
connected by narrow and shallow channels (Chap. 2). Hence, there is a large impact
of the bottom topography on currents. In the RCO model the topography by Seifert
et al. (2001) is used which is based upon a gridded data set of digitized charts of the
bathymetry with a horizontal resolution of 1 nautical mile.
4.2.2.3 Initial Conditions and Spin-Up
For the calculation of initial temperature and salinity fields observations are assim-
ilated into available model results from previous experiments. Initial currents and
sea level height are assumed to be zero.
A spin-up integration has to be performed to guarantee that current and density
fields fit dynamically together. Two time scales are important. The first is the so-
called advective time scale which describes the time needed to develop density gra-
dients and to adjust currents. This time scale amounts to approximately a couple of
weeks to months. However, such a spin-up would be insufficient because not only
advection affects the horizontal and vertical density distributions. Even diffusion
(and here especially diapycnal mixing) with a much longer time scale contributes to
changes of density gradients. Consequently, a proper spin-up should at least cover
the overturning time scale of the Baltic Sea which amounts to 30 years, approxi-
mately (Meier 2002). The latter is the second important time scale which should be
taken into account.
4.2.2.4 Lateral Boundary Conditions
The open boundary conditions are chosen to be located in the northern Kattegat
because available computational resources allow only a limited model domain. In
addition, the water exchange between North Sea and Baltic Sea should be calculated
explicitly. The boundary conditions require water temperature and salinity profiles
in case of inflow. Sensitivity experiments showed that water temperature and salin-
ity in the Baltic Sea domain limited by the Danish Straits depend only on deep water
salinity in the Kattegat (Meier et al. 2003). Boundary conditions for water tempera-
ture in the whole water column and salinity of the surface layer impact the Kattegat
hydrography but not the Baltic Sea interior. The reason is that even during major
inflow events the inflowing water volume is smaller than the volume of the Kattegat
surface layer in the model. Hence, usually no water from outside the model domain
is advected through the Danish Straits into the Baltic Sea. As measurements sug-
gest that the variability of the deep water salinity is negligible, sufficient boundary
conditions are built upon climatological mean profiles for water temperature and
salinity in the northern Kattegat.
4.2.2.5 Sea Level in the Kattegat
The situation is different for the sea level boundary conditions because the daily
variations affect the water exchange between the Kattegat and the western Baltic
Sea. Barotropic flows generated by the sea level difference along the Danish Straits
are important for the volume and salt exchange between the Kattegat and Baltic Sea.
Hence, daily sea level observations from a station close to the northern boundary,
e.g., the Swedish stations Smögen, Ringhals or Varberg, are prescribed. Although
these boundary conditions are very much simplified, they ensure a reasonable sim-
ulation of the water exchange variability on time scales from days during inflow
events to decades during high and low (stagnation) saline phases (Meier et al. 2003).
4.2.2.6 Atmospheric Forcing
In climate simulations the integration period is long compared to the overturning

time scale. Hence, the numerical integration of the primitive equations is a bound-
ary rather than an initial value problem. Usually, empirical bulk formulae are used
to calculate the air–sea, air–ice and ice–ocean fluxes from atmospheric and oceanic
state variables. In the RCO model the surface fluxes are calculated from 2 m air
temperature, 2 m specific humidity, sea level pressure, 10 m wind, precipitation and
total cloudiness. The surface fluxes that need to be specified are sensible and latent
heat fluxes, incoming short- and long-wave radiation, precipitation minus evapora-
tion and momentum fluxes as outlined in Sects. 4.2.1.3 and 4.2.1.4. Alternatively,
the surface fluxes from an atmospheric model might be used as forcing.
There are two alternatives of atmospheric forcing data sets for long-term simu-
lations with high-resolution ocean models. Firstly, the Swedish Meteorological and
Hydrological Institute (SMHI) provided three-hourly gridded observations of sea
level pressure, 2 m air temperature, 2 m relative humidity, and total cloud cover
since 1980 (e.g., Kauker and Meier 2003). In addition, 12-hourly accumulated pre-
cipitation fields are available at 06 and 18 UTC. Geostrophic wind is calculated and
reduced to 10 m wind by using a varying factor for the geostrophic wind speed in
the range between 0.5 and 0.6, depending on the distance to the coast (Bumke and
Hasse 1989). In addition, a constant ageostrophic angle of 17◦ following (Bumke
and Hasse 1989) was used. Data from all available synoptic stations (about 700 to
800) covering the entire Baltic Sea drainage basin are interpolated with the help of
a 2D univariate optimum interpolation on a 1◦ × 1◦ regular horizontal grid with re-
spective latitude and longitude ranges of 50◦ N to 72◦ N and 8◦ E to 40◦ E. These
atmospheric data are also available since 1970 but with only six-hourly temporal
resolution and calculated with fewer meteorological stations.
Some caution is necessary when applying these data as forcing for Baltic Sea
models. Firstly, the horizontal grid resolution is very coarse compared to the geo-
graphical dimensions of the Baltic Sea. Secondly, as most of the stations are located
on land (only a few of them are located on islands), the data set is obviously biased
over sea, for instance, with slightly too warm air temperatures during summer and
slightly too cold temperatures during winter.
Nevertheless, many modelling studies for the Baltic Sea have applied this data
set successfully as atmospheric forcing (e.g., Lehmann 1995; Neumann et al. 2002;
Meier et al. 2003; Meier 2007; Andrejev et al. 2004a, 2004b; Myrberg and Andrejev
2003, 2006). Unfortunately, SMHI stopped its production in 2011.
The second alternative for the atmospheric forcing is the output of a regional
high-resolution atmosphere model. As the resolution of global reanalysis products
like ERA-40 (Uppala et al. 2005) or ERA-Interim (Dee et al. 2011) is too coarse
for Baltic Sea applications (Omstedt et al. 2005), surface fields of a regional high-
resolution model are needed. Samuelsson et al. (2011) evaluated surface fields of a
regional atmosphere model driven with ERA-40 data at the lateral boundary. Fur-
thermore, Meier et al. (2011) assessed the quality of atmospheric surface fields over
the Baltic Sea of a regional coupled atmosphere-ocean model with respect to ocean
dynamics. In both studies it was concluded that the modelled atmospheric surface
fields have sufficiently good quality and the data might be used as forcing for ocean
models. This approach was applied, for instance, by Höglund and Meier (2012).
However, there are also disadvantages. The model domain of the regional atmo-
sphere model by Samuelsson et al. (2011) is rather large allowing a regional weather
development that differs from the observed weather due to the chaotic nature of the
atmospheric circulation. Hence, the tracks of low and high pressure anomalies might
be shifted in time or space compared to observations. Due to this internal variabil-
ity the quality of the atmospheric forcing might be limited and data assimilation is
needed. At present regional reanalyses utilizing the assimilation of observations into
regional high-resolution atmosphere models are under development (Luhamaa et al.
2011) and will soon replace earlier downscaling products that lack data assimilation.
The use of proper atmospheric data is crucial in Baltic Sea applications because
this water body is rather shallow with a mean water depth of 54 m (Seifert et al.
2001) and the variability of its dynamics is largely controlled by atmospheric vari-
ability. Further, as mixing in the ocean depends on the cubed wind speed, it is im-
portant for the stratification in the Baltic Sea that wind speed extremes are cor-
rectly represented. Otherwise the simulated vertical stratification might be biased.
Höglund et al. (2009) showed that wind speed extremes in a state-of-the-art regional
atmosphere model are underestimated. To compensate for this shortcoming they de-
veloped a bias-correction based upon the simulated wind gustiness (Höglund et al.
2009). There are many studies devoted to this problem. However a more detailed
discussion on simulated wind speed extremes is beyond the scope of this chapter.
4.2.2.7 River Runoff
Finally, runoff from rivers and other fresh water sources are needed to simulate the
water cycle of the Baltic Sea. River runoff is the main contributor to the fresh water
surplus of the Baltic Sea. Note that the Baltic Sea catchment area is four times larger
than the surface of the Baltic Sea (Sjöberg 1992). In the RCO model the 29 largest
rivers have been implemented (Meier et al. 2003). These 29 rivers represent about
84 % of the total runoff. Monthly river runoff data were taken from the Baltic Sea
Experiment (BALTEX) Hydrological Data Center (BHDC) at SMHI (Bergström
and Carlsson 1994). Large-scale hydrological models are used to update the runoff
in recent years where the runoff data are not available (Graham 1999).
When the topography, initial and boundary conditions, and the atmospheric and
hydrological forcing are prepared the ocean model is ready to be integrated forward
in time with a baroclinic time step, which amounts in case of the RCO model to
150 s. Hence, the model resolves the advection of matter with the currents and fast
travelling baroclinic waves, as long as the baroclinic Rossby radius (see Chap. 2,
Sect. 2.3.2 and Chap. 3, Sect. 3.3.1.1) is larger than at least two grid cells of the
model. As the RCO model was designed to study past and future climate variability
of the Baltic Sea on a centennial time scale, a horizontal resolution higher than 2
nautical miles was technically impossible (Meier et al. 2003). As baroclinic Rossby
radii in the Baltic Proper amount to 3–10 km approximately (Fennel et al. 1991), the
RCO model and other models with similar set-up (e.g., Lehmann 1995; Neumann
et al. 2002; Jȩdrasik et al. 2008) are considered to be eddy-permitting but not eddy-
resolving.
4.2.2.8 Numerical Implementation and Shortcomings
All subgrid scale processes, like internal gravity waves and boundary layer turbu-
lence, are parameterized which means that the impact of subgrid scale processes
on the resolved motion is taken into account using empirical relationships. Some of
these parameterizations might be used to calibrate the models using available ob-
servations because many of the mechanisms are still unknown. Processes important
for the calculation of currents for that we lack detailed knowledge today are, for
instance, bottom friction, vertical mixing and surface fluxes of momentum.
However, even if the mathematical equations describing the relevant processes
are known, the numerical treatment may cause biases of the solution of the problem
under consideration. For instance, test experiments showed that in long simulations
the modified SPLIT-QUICK advection scheme (Webb et al. 1998), which was earlier
embedded within the RCO model, does not guarantee numerical stability of passive
tracers released at point sources. SPLIT-QUICK is of third order accuracy but does
not preserve monotonicity. At large gradients overshooting due to numerical dis-
persion is observed. Hence, for the study of tracers a flux-corrected, monotonicity
preserving transport (FCT) scheme following (Gerdes et al. 1991) was embedded
(Meier 2007). In the simulations by Meier (2007) no explicit horizontal diffusion
was applied.
Due to the lack of knowledge about proper parameterizations of subgrid-scale
processes, insufficient horizontal and vertical grid resolution, biases of atmospheric
and hydrological forcing fields, too coarse topographical data sets, and insufficient
numerical schemes, Baltic Sea models suffer from uncertainties. For instance, in
case of the RCO model these shortcomings are a too shallow halocline (Meier 2007),
artificial mixing during salt water inflows (Meier et al. 2004), too strong vertical
stratification in the Gulf of Finland (Meier 2007), and numerical noise affecting the
sea surface temperature occasionally (Löptien and Meier 2011). Other Baltic Sea
circulation models have similar shortcomings.
For further details of the RCO model the reader is referred to Meier (2001, 2007),
Meier et al. (1999, 2003). Table 4.1 summarizes the most important characteristics
of the RCO model.
4.3 Eulerian Tracer Methods

We define an Eulerian tracer as a field that obeys a classical advection-diffusion
equation driven with currents of the ocean circulation model. The tracer is said to be
passive if it has no impact on ocean dynamics. Sources, sinks and initial and bound-
ary conditions are specific for the substance under consideration. For instance, in
biogeochemical models dissolved nutrients are treated as passive Eulerian tracers
with riverine nutrient loads as the most important external sources and the internal
transformation to organic material and export, burial in the sediments and denitri-
fication as the most important sinks (e.g., Fennel and Neumann 2004). However,
Eulerian tracers could also be just concentrations of markers of water masses or of
their characteristics, defined by simplified sources, sinks, and initial and boundary
conditions. To analyse ocean dynamics with tracer methods either completely arti-
ficial or observed tracers, like colours or radioactive substances, were used in the
literature.
An example is the concept of age distribution and transit time (Bolin and Rodhe
1973) which has been applied, for instance, to Himmerfjärden Estuary (Engqvist
1996), the Gulf of Finland (Andrejev et al. 2004b), the Gulf of Bothnia (Myrberg
and Andrejev 2006), and the entire Baltic Sea (Meier 2005, 2007). In these studies
a passive tracer was added to the model variables to characterize the average age
of sea water in the reservoir with prescribed values at the lateral open boundaries
or at the sea surface. For instance, Meier (2005) found for the period 1903–1998
mean age of the bottom water of one year at Bornholm Deep, five years at Gotland
Deep, and seven years at Landsort Deep. For the whole Baltic Sea a maximum age
of about 11 years appeared in the bottom water at Landsort Deep. These numbers
are important information to characterize the climatological renewal process of the
Baltic Sea deep water which is not possible to derive just from observations (see
Chap. 2).
In the study by Meier (2005) the age of sea water is the time elapsed since a water
particle left the sea surface. If the age of the inflowing water at the source regions
(i.e., the lateral open boundaries towards the Baltic Proper and the river mouths)
is set to zero, Andrejev et al. (2004b) and Myrberg and Andrejev (2006) found
maximum water age of around two years in the Gulf of Finland and of around 7.4
years in the Gulf of Bothnia, respectively. These studies have in common that the age
of sea water is estimated from the equation of the age of pure water (Deleersnijder
et al. 2001). Hence, it is not possible to distinguish between different water masses.
Meier (2007) extended the approach to analyse different water masses of the
Baltic Sea. He calculated concentrations of passive tracers and associated age con-
centrations following (Delhez et al. 1999; Deleersnijder et al. 2001). At any time t
and location r, the concentration of a tracer C(t, r) would obey the following equa-
tion:
∂C
+ ∇ · (uC − K · ∇C) = 0, (4.34)
∂t
where u is the water velocity and K denotes the diffusivity tensor. The age con-
centration α(t, r) of the water mass under study is the solution of the following
equation:
∂α
+ ∇ · (uα − K · ∇α) = C, (4.35)
∂t
Finally, the age is then given as the ratio a(t, r) = α(t, r)/C(t, r). For further details
of the concept of age in marine modelling the reader is referred to Deleersnijder et al.
(2001).
Meier (2007) performed a simulation using the RCO model where various water
masses were marked with passive tracers and the associated age of the specific water
masses was calculated. The aim was to better understand the large-scale circulation
and related time scales.
In this approach the initial tracer and age concentrations were set to zero. At
the open boundary in the Kattegat the same radiation conditions as used within the
RCO model for temperature and salinity were utilized. In case of inflow, at the open
boundary the concentrations of the tracers and associated age were relaxed to zero,
with a time scale of one day. At the sea surface no-flux boundary conditions were
used.
Mean age (in years) for the last five years of a 96-year long simulation asso-
ciated to tracers either marking inflowing saltier water at the Darss and Drogden
Sills (Fig. 4.5) or inflowing fresh water from all rivers (Fig. 4.6) has been evaluated
by Meier (2007). At the sea surface of the Bornholm Basin, Gotland Basin, Sea
of Bothnia, and Bay of Bothnia the mean age associated to inflowing water from
the Kattegat amounts to 26–30, 28–34, 34–38, and 38–42 years, respectively. The
largest mean sea surface age of more than 30 years associated to the fresh water
of the rivers is found in the Eastern Gotland Basin and Belt Sea. At the bottom the
mean age is largest in the Western Gotland Basin and amounts to more than 36 years.
In the Baltic Proper vertical gradients of age associated to the fresh water inflow are
smaller than in the case of inflowing saltier water from the Kattegat indicating an
efficient recirculation of fresh water in the Baltic Sea. Thus, studies of passive trac-
ers and associated age help to understand the physical processes important for the
large-scale vertical circulation in the Baltic Sea.
The studies described above have in common that 3D artificial tracers (that is,
tracers that are free to move in any direction in the water column) with simplified
sources and sinks are utilized. In the following the tracer concept will be applied to
oil spills restricting the calculation of spread to only two dimensions in the surface
layer.
Fig. 4.5 Mean age (in years) for the last 5 years of the 96-year spin-up associated to the tracer
marking inflowing water at the Darss and Drogden Sills. The figures depict age at the sea surface
(upper left panel), at the halocline depth (upper right panel), at the surface with a salinity of 17
(lower left panel), and at the longitudinal section S3 across the Baltic Sea (lower right panel).
The location of the section is shown as the white line in the lower left panel. Note the different
colour bars. In the upper right panel mean age is not depicted in shallow areas without a perennial
halocline, where the conceptual two-layer model does not apply. Per arbitrary definition these areas
(in brown) are characterized by a water depth less than 22.5 m. Is the salinity in the water column
below 17 the associated surface is set equal to the water depth. Hence, the lower left panel shows
bottom age in most of the Baltic interior. Adopted from Meier (2007)
Fig. 4.6 As Fig. 4.5 but the age is associated to the tracer marking fresh water from all rivers.
Adopted from Meier (2007)
4.4 Oil Spill Modelling Using Eulerian Methods

Oil spills are traditionally simulated as a large number of particles following La-
grangian trajectories, as detailed in Chap. 11 of this book. This approach has the
advantage of resolving processes at small scales as well as large scales with a rea-
sonable amount of computer power. At the same time it is crucial to sample the oil
spill with a large enough number of particles to properly represent the complete oil
spill at later stages when it is more spread out. If too few particles are released, there
will be a risk of missing a potential coastal hit.
With an Eulerian approach the complete field is covered without the need of
initial choices. On the other hand, small-scale processes are missed if a fixed grid
size is used. It would be possible to use an unstructured grid with higher resolution
close to the initial release or even an adaptive grid. To our knowledge no model
for oil spills uses such an advanced method today. However, there are examples of
tracking oil spills where Eulerian principles have been used. Two examples using
purely Eulerian methods and one example with a hybrid approach are presented
below to illustrate the potential of the Eulerian approach.
4.4.1 Ensemble Approach
One use of Eulerian tracers to model oil spills was demonstrated by Höglund and
Meier (2012). They simulated a large number of oil spills in different seasons and in
different weather conditions in the Baltic Proper. From the ensemble of simulations,
maps of several statistical measures of the fate of the oil spills were drawn showing
how severe consequences are that one may expect from an oil spill at different loca-
tions (Fig. 4.7). Further, Höglund and Meier (2012) calculated maritime routes that
minimize the risks of an accidental oil spill demanding that both the chosen measure
and the length of the ship track are as small as possible. Examples of routes with
different weights for distance relative to the measure are shown in Fig. 4.8.
The oil spill simulations were done with a 2D Eulerian tracer advected with
the currents in the uppermost model layer of the ocean circulation model. This
approach considers both advection and diffusion of the oil slick and is similar to
the studies described in Chaps. 9–11. Explicitly prescribed horizontal diffusion pa-
rameterizes those subgrid-scale dispersive processes that are not resolved by the
ocean circulation model and that are not directly wind dependent, e.g., turbulent
motions.
The initial concentration of the tracer was set to zero except for one grid point
representing the location of the original release. Per definition, all grid points with
land next to them were perfect sinks of the tracer (just setting the concentration to
zero) to simulate that the oil sticks to the coast. As there were only sinks and no
sources, the total content of the tracer must be decreasing (not necessarily strictly
decreasing). The output of each simulation was a time series containing the evolu-
tion of the total remaining tracer content in time.
The use of ensembles to cover different seasons as well as many different weather
conditions implies that a fairly large number of simulations must be done to get good
statistics. This obviously requires considerable computer resources and forces some
other simplifying approximations like the restriction to the surface layer (see below).
However, the ensemble approach has the advantage to compensate non-systematic
Fig. 4.7 Various measures calculated from tracer releases (a–e) and distance to the nearest coast
according to the bathymetry of the model (f). The limits of the colour are chosen such that the area
of each colour is the same in all panels: 25 % each for blue, green, yellow and red (light + dark)
and 10 % for dark red. (a) Average of still-at-sea after 30 days. (b) Median of still-at-sea after
30 days. (c) Fifth percentile of still-at-sea after 30 days. (d) Average time for 90 % still-at-sea (see
text for details). (e) Fifth percentile of time for 90 % still-at-sea. For details see Höglund and Meier
(2012)
errors that might be introduced by approximations. Höglund and Meier (2012) used
100 simulations per release point, evenly distributed over the year and over several
years to cover different seasons and weather conditions.
Fig. 4.8 Routes with

different weights (various
colours) for the distance using
the ensemble average of the
tracer concentration which is
still-at-sea after 30 days. Red
has full weight for distance,
and blue has full weight for
the release of tracer
measurement. For details see
Höglund and Meier (2012)
Producing maps of statistical measures as well as constructing maritime routes

requires both a geographical distribution of release points as well as a sufficient
number of them. The geographical distribution was chosen evenly as there was no a
priori information about areas of particular interest. Positions very close to the coast
might seem reasonable to exclude but current patterns might be such that oil spills
usually would be transported away from the coast, thus reducing the potential risk.
Höglund and Meier (2012) chose to use every model grid point in the particular area
in the ocean circulation model as a release point, in total 15,652 locations.
In total more than 1.5 million tracer experiments were done. The restriction of
propagation of tracers only in the horizontal direction ignores vertical mixing of oil
within the water column and vertical advection with currents below the surface. This
restriction, however, was necessary to keep the usage of computational resources
reasonable. This enabled to track the evolution of 15,652 tracers, one for each re-
lease point, with one run of the ocean circulation model. In total, 100 different runs
formed the ensemble discussed below.
In the approach by Höglund and Meier (2012), the impact of missing processes
on individual oil spill forecasts might be considerable, e.g., for the planning of coun-
termeasures. However, in a statistical sense they appear to be almost unimportant.
The statistics of the results of the oil spill simulations were validated against the oil
spill model OSMS (Anomymous 2002). Although the results in Höglund and Meier
(2012) agree well with the OSMS modelling results, the time scale for an initial
coastal hit differs by a factor of two.
As already mentioned, the output from each Eulerian tracer was a time series
with the amount of remaining tracer representing the portion of oil that has not
hit any coast. Working with many tracers makes it practically impossible to save
complete fields at any reasonable frequency. In this case the complete fields would
have required more than 100 TB of disk to store (and, of course, much more for
3D fields). The disadvantage is less flexibility in the analysis and lost information,
e.g., which coasts were contaminated. It would of course be possible to save this
particular information without saving the complete fields. This illustrates the need
to plan the simulation and implement the right analysis already in the model.
4.4.2 Multi-tracer Approach
Until now we focused the discussion on the usage of individual tracers. However,
it is possible to use also several passive tracers representing different states of any
substance. The idea is to connect sinks and sources of different tracers. The trans-
formation between the tracers represents a physical, chemical or biological process,
and its calculation might be combined even with a third tracer. For instance, one
tracer is used for the description of the oil slick at the surface and another one for oil
vertically mixed within the water column. The transformation between the oil slick
and other oil compounds would represent vertical mixing due to surface waves. This
process will depend on the wave height. In the other direction buoyant oil will be
floated up to the surface.
This approach is realized within the Multiphase Oil Spill Model (MOSM)
(Tkalich et al. 2003). The MOSM model uses a regular grid and six fields: one
two-dimensional tracer for the oil slick thickness at the surface, three 3D tracers
for emulsified oil, particulate oil and dissolved oil in the water column, and finally
two 2D fields for particulate oil and dissolved oil buried in the bottom sediments
which are not advected. The advection of the surface tracer is not only done with
the surface currents but also with a fraction of the wind. A number of processes
can transform the oil compounds into each other. Oil can also leave the system by
evaporation. The composition of the oil, together with many other parameters, de-
termines coefficients for the rate of transfers and evaporation. For more details of
oil spill modelling the reader is referred to Chap. 11 of this book.
Tkalich et al. (2003) motivated the usage of Eulerian tracers with “Nowadays,
an oil slick dynamics model can afford to routinely use such accurate and physi-
cally relevant formulation as the Navier–Stokes equations.” Another motivation is
to make it easier to couple the oil spill model to other models. For instance, the
MOSM model was coupled with a food chain model to analyse the effects of an
oil spill on the food web (Gin et al. 2001). The food chain model contains phyto-
plankton, zooplankton, small fish, large fish and benthic invertebrates. Although the
coupling of the models would not have been impossible if the oil spill model was
Lagrangian, it would have been more complicated.
Finally the very interesting question arises whether the model performed better
than other more traditional models. Although Tkalich et al. (2003) presented some
validations, unfortunately no details were given. It was concluded that “test simu-
lations of oil slick dynamics, as well as transport and kinetics of the oil phases in
the water column, show good consistency with empirical data and other models”
(Tkalich et al. 2003).
4.4.3 Hybrid Approach
In the previous two sub-sections we have discussed two applications of Eulerian

tracers for tracking oil spills, which is traditionally done with the help of Lagrangian
methods. However, a pure Lagrangian or Eulerian method is not the only choice.
Hybrid methods may sometimes have advantages. An example where Eulerian and
Lagrangian methods are combined is the oil spill model presented in Guo and Wang
(2009).
The initial oil slick was simulated with Lagrangian particles using random walk
particle tracking. When the slick thickness locally dropped below a certain threshold
the calculation continued with modelling of the slick thickness at that place with an
Eulerian-Lagrangian method on a grid.
The Eulerian-Lagrangian method combined the Eulerian and the Lagrangian ap-
proaches for the advection-diffusion equation. The equation was split into two equa-
tions, one pure advection equation and one pure diffusion equation. The advection
equation was solved with Lagrangian methods like backtracking along stream lines
and the diffusion equation with Eulerian methods. The method is known to work
well in regions that are advection dominated. In regions where the advection is not
the dominating part, like close to point sources, the method is less good. By using
a purely Lagrangian method in the initial phase and by switching to an Eulerian-
Lagrangian method when the thickness of the slick is low, the strengths of both
methods were combined.
In a similar manner, the oil that is vertically mixed within the water column
by surface waves induced turbulence, first follows with a Lagrangian method as
droplets. Those droplets that do not resurface, i.e., droplets with buoyancy which
is not dominating, are recalculated to a concentration. The evolution of the passive
tracer in time is calculated with an Eulerian-Lagrangian method.
4.5 Outlook
In this overview we introduced the Eulerian tracer approach to study the charac-
teristics of the Baltic Sea circulation. Three examples of oil spill modelling using
Eulerian methods are presented. The method which allows the calculation of opti-
mized fairways following (Höglund and Meier 2012) is rather idealized because the
oil dynamics is not realistically treated. However, combining the various presented
methods will result in a powerful tool which may help to prevent the pollution of
ecologically sensitive regions after accidents of oil tankers. The knowledge gained
from the idealized simulations may be used to develop more advanced management
tools taking additional processes of spreading oil slicks into account.
Acknowledgements The research presented in this study is part of the project BalticWay (The
potential of currents for environmental management of the Baltic Sea maritime industry) and has
received funding from the European Commission’s Seventh Framework Programme (FP7 2007–
2013) under Grant agreement No. 217246 made with BONUS, the joint Baltic Sea research and
development program, and from the Swedish Research Council for Environment, Agricultural Sci-
ences and Spatial Planning (Formas, Ref. No. 2008–1898).
References
Anomymous (2002) An updated assessment of the risk for oil spills in the Baltic Sea area. Pre-
sented as a Status report on risk analyses for use in response to oil pollution in the Baltic
Sea by Dr S Ovsienko, Fifth meeting of the Sea-based Pollution Group HELCOM SEA,
Turku, Finland, 13–17 May 2002. Helsinki Commission, Helsinki, 77 pp. http://www.helcom.
fi/stc/files/shipping/RiskforOilSpillsReport2002.pdf
Andrejev O, Myrberg K, Alenius P, Lundberg PA (2004a) Mean circulation and water exchange
9:1–16
Andrejev O, Myrberg K, Lundberg PA (2004b) Age and renewal time of water masses in a semi-
enclosed basin—application to the Gulf of Finland. Tellus A 56:548–558
ASCE, Task Committee on modeling of oil spills of the Water Resources Engineering Divi-
sion (1996) State-of-the-art review of modeling transport and fate of oil spills. J Hydraul Eng
122:594–609
Bergström S, Carlsson B (1994). River runoff to the Baltic Sea: 1950–1990. Ambio 23:280–287
Blanke B, Delecluse P (1993) Variability of the Tropical Atlantic Ocean simulated by a general
circulation model with two different mixed layer physics. J Phys Oceanogr 23:1363–1388
Bodin S (1979) A predictive numerical model of the atmospheric boundarylayer based on the tur-
bulent energy equation. Rep Meteorol Climatol 13, Swedish Meteorological and Hydrological
Institute, Norrköping, 139 pp
Bolin B, Rodhe H (1973) A note on the concepts of age distribution and transit term in natural
reservoirs. Tellus 25:58–62
Bryan K (1969) A numerical method for the study of the circulation of the World Ocean. J Comput
Phys 4:347–376
Bryan K, Cox MD (1972) An approximate equation of state for numerical models of ocean circu-
lation. J Phys Oceanogr 2:510–514
Bumke K, Hasse L (1989) An analysis scheme for the determination of true surfacewinds at sea
from ship synoptic wind and pressure observations. Bound-Layer Meteorol 47:295–308
Cox MD (1984) A primitive equation 3-dimensional model of the ocean. Technical Report 1, Geo-
physical Fluid Dynamics Laboratory, Princeton University, Princeton, NJ, 141 pp
Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S, Andrae U, Balmaseda MA,
Balsamo G, Bauer P, Bechtold P, Beljaars ACM, van de Berg L, Bidlot J, Bormann N, Delsol C,
Dragani R, Fuentes M, Geer AJ, Haimberger L, Healy SB, Hersbach H, Holm EV, Isaksen L,
Kallberg P, Koehler M, Matricardi M, McNally AP, Monge-Sanz BM, Morcrette J-J, Park BK,
Peubey C, de Rosnay P, Tavolato C, Thepaut J-N, Vitart F (2011) The ERA-Interim reanalysis:
configuration and performance of the data assimilation system. Q J R Meteorol Soc 137:553–
597
Deleersnijder E, Campin J-M, Delhez EJM (2001) The concept of age in marine modelling. I.
Theory and preliminary model results. J Mar Syst 28:229–267
Delhez EJM, Campin J-M, Hirst AC, Deleersnijder E (1999) Towards a general theory of the age
in ocean modelling. Ocean Model 1:17–27
Dera J (1992) Marine physics. Elsevier, Amsterdam, 516 pp
Engqvist A (1996) Long-term nutrient balances in the eutrophication of the Himmerfjärden estu-
ary. Estuar Coast Shelf Sci 42:483–507
Fennel W, Neumann T (2004) Introduction to the modeling of marine ecosystems. Elsevier
oceanography series, vol 72. Elsevier, Boston, 89 pp
Res 11:23–36
Gerdes R, Köberle C, Willebrand J (1991) The influence of numerical advection schemes on the
results of ocean general circulation models. Clim Dyn 5:211–226
Gill AE (1982) Atmosphere–ocean dynamics. Academic Press, London, 662 pp
Gin K, Huda K, Lim WK, Tkalich P (2001) An oil spill-food chain interaction model for coastal
waters. Mar Pollut Bull 42:590–597
Graham PL (1999) Modeling runoff to the Baltic Sea. Ambio 27:328–334

Guo WJ, Wang YX (2009) A numerical oil spill model based on a hybrid method. Mar Pollut Bull
58:726–734
Gustafsson T, Kullenberg B (1936) Untersuchungen vor Trägheitströmungen in der Ostsee (Inves-
tigations of inertial currents in the Baltic Sea). Sven Hydrogr-Biol Komm Skr, Ny Ser Hydr
13:1–28
Höglund A, Meier HEM (2012) Environmentally safe areas and routes in the Baltic Proper using
Eulerian tracers. Mar Pollut Bull 64:1375–1385
Sweden 29 pp
Jakobsen F (1995) The major inflow to the Baltic Sea during January 1993. J Mar Syst 6:227–240
Jȩdrasik J, Cieślikiewicz W, Kowalewski M, Bradtke K, Jankowski A (2008) 44 years hindcast of
the sea level and circulation in the Baltic Sea. Coast Eng 55:849–860
Jerlov NG (1968) Optical oceanography. Elsevier, Amsterdam, 194 pp
Kauker F, Meier HEM (2003) Modeling decadal variability of the Baltic Sea: 1. Reconstructing
atmospheric surface data for the period 1902–1998. J Geophys Res—Oceans 108:3267
Krauss W (1973) Dynamics of the homogeneous and the quasihomogeneous ocean. Gebrüder
Bornträger, Berlin, 302 pp
Killworth P, Stainforth D, Webb D, Paterson S (1991) The development of a free-surface Bryan–
Cox–Semtner ocean model. J Phys Oceanogr 21:1333–1348
Large WG, Pond S (1981) Open ocean momentum flux measurements in moderate to strong winds.
J Phys Oceanogr 11:324–336
Large WG, Pond S (1982) Sensible and latent heat flux measurements over the ocean. J Phys
Oceanogr 12:464–482
Lehmann A (1995) A 3-dimensional baroclinic eddy-resolving model of the Baltic Sea. Tellus A
47:1013–1031
25:333–357
Löptien U, Meier HEM (2011) The influence of increasing water turbidity on the sea surface tem-
perature in the Baltic Sea: a model sensitivity study. J Mar Syst 88:323–331
Luhamaa A, Kimmel K, Männik A, Rõõm R (2011) High resolution re-analysis for the Baltic Sea
region during 1965–2005 period. Clim Dyn 36:727–738
Mårtensson S, Meier HEM, Pemberton P, Haapala J (2012) Ridged sea ice characteristics in the
Arctic from a coupled multicategory sea ice model. J Geophys Res—Oceans 117:C00D15
Matthäus W, Lass HU, Tiesel R (1993) The major Baltic inflow in January 1993. ICES Statutory
Meeting. ICES CM 1993/C:51
Meier HEM (2001) On the parameterization of mixing in three-dimensional Baltic Sea models.
J Geophys Res—Oceans 106:30997–31016
Meier HEM (2002) Regional ocean climate simulations with a 3D ice-ocean model for the Baltic
Sea. Part 1: model experiments and results for temperature and salinity. Clim Dyn 19:237–253
Meier HEM (2005) Modeling the age of Baltic Sea water masses: quantification and steady-state
sensitivity experiments. J Geophys Res—Oceans 110:C02006
Meier HEM, Kauker F (2003) Modeling decadal variability of the Baltic Sea: 2. The role of
freshwater inflow and large-scale atmospheric circulation for salinity. J Geophys Res—Oceans
108:3368
Meier HEM, Döscher R, Coward AC, Nycander J, Döös K (1999) RCO—Rossby Centre regional
Ocean climate model: Model description (version 1.0) and first results from the hindcast pe-
riod 1992/93. Rapport Oceanografi No 26, Swedish Meteorological and Hydrological Institute,
Norrköping, Sweden, 102 pp
Meier HEM, Döscher R, Faxén T (2003) A multiprocessor coupled ice-ocean model for the Baltic
Sea: application to salt inflow. J Geophys Res—Oceans 108:3273
Meier HEM, Döscher R, Broman B, Piechura J (2004) The major Baltic inflow in January 2003
and preconditioning by smaller inflows in summer/autumn 2002: a model study. Oceanologia
46:557–579
Meier HEM, Höglund A, Döscher R, Andersson H, Löptien U, Kjellström E (2011) Quality as-
sessment of atmospheric surface fields over the Baltic Sea of an ensemble of regional climate
model simulations with respect to ocean dynamics. Oceanologia 53:193–227
Mesinger F, Arakawa A (1976) Numerical methods used in atmospheric models. GARP publica-
tions series 17, vol I. WMO, Geneva, 64 pp
Müller P, Willebrand J (1989) Equations for oceanic motions. In: Sündermann J (ed) Landolt-
Börnstein, group V. Oceanography, vol 3b
Neumann T, Fennel W, Kremp C (2002) Experimental simulations with an ecosystem model of
the Baltic Sea: a nutrient load reduction experiment. Glob Biogeochem Cycles 16:1033
Omstedt A, Chen Y, Wesslander K (2005) A comparison between the ERA40 and the SMHI grid-
ded meteorological databases as applied to Baltic Sea modeling. Nord Hydrol 36:369–380
Commentationes physico-mathematicae, vol 12. Societas Scientarium Fennica, Helsinki
Paulson CA, Simpson JJ (1977) Irradiance measurements in the upper ocean. J Phys Oceanogr
7:952–956
Reed M, Johansen O, Brandvik PJ, Daling P, Lewis A, Fiocco R, Mackay D, Prentki R (1999) Oil
spill modeling towards the close of the 20th century: overview of the state of the art. Spill Sci
Technol Bull 5:3–16
Rodi W (1980) Turbulence models and their application in hydraulics—a state-of-the-art review.
International Association of Hydraulic Research, Delft, 104 pp
Nikulin G, Wyser K (2011) The Rossby Centre Regional Climate model RCA3: model descrip-
tion and performance. Tellus A 63:4–23
2nd edition. In: Baltic Sea science congress, Stockholm, 25–29 November 2001, Poster #147,
Semtner AJ (1974) A general circulation model for the World Ocean. Technical Report 9, Depart-
ment of Meteorology, University of California, Los Angeles, California, 99 pp
Sjöberg B (ed) (1992) Hav och Kust. Sveriges Nationalatlas Förlag. Almquist & Wiksell Interna-
tional, Stockholm
Smagorinsky J (1963) General circulation experiments with the primitive equations: I. The basic
experiment. Mon Weather Rev 91:99–164
Steinhorn I (1991) Notes and correspondence—salt flux and evaporation. J Phys Oceanogr
21:1681–1683
Stevens DP (1991) The open boundary condition in the United Kingdom fine-resolution Antarctic
model. J Phys Oceanogr 21:1494–1499
Tkalich P, Huda K, Gin KYH (2003) A multiphase oil spill model. J Hydraul Res 41:115–125
UNESCO (1981) United Nations Educational, Scientific, and Cultural Organization. Tenth report
of the joint panel on oceanographic tables and standards, UNESCO technical papers in marine
science 36, 26 pp
Webb DJ, Coward AC, de Cuevas BA, Gwilliam CS (1997) A multiprocessor ocean circulation
model using message passing. J Atmos Ocean Technol 14:175–183
Webb DJ, de Cuevas BA, Richmond CS (1998) Improved advection schemes for ocean models.
J Atmos Ocean Technol 15:1171–1187
Chapter 5
European Semi-enclosed Seas: Basic Physical
Processes and Their Numerical Modelling
Emil V. Stanev and Xi Lu
Abstract A parallel analysis of the physical state of European semi-enclosed seas,

namely the Black Sea, Baltic Sea and Mediterranean Sea, is presented based on
observations and numerical modelling, with the aim of revealing similarities and
differences between the dominating physical processes. Each of the three basins
is presented in more detail with respect to processes which are very specific and
relatively well studied: the formation of intermediate water masses in the Black
Sea, strait exchange and circulation in the transition area between the Baltic and
North Sea, and deep water formation and the Eastern Mediterranean Transient. For
the three basins extended discussions on the strait exchanges, the routing of deep
inflows by straits, and underwater passage channels and shallows are provided. It is
demonstrated that although the topographic routing has very different appearances
in the three basins, in general it strongly controls the formation of water masses
and thermohaline circulation. This review aims also at motivating parallel studies on
similar geophysical systems and sharing of specific or common experiences between
researchers dealing with the individual European seas.
5.1 Introduction
The difference in the net water budget of semi-enclosed seas P + R − E expressed

as precipitation P plus river inflow R minus evaporation E results in different water
regimes in individual basins. Depending on the sign of the water budget, basins are
known to be of dilution (estuarine) type if P + R − E > 0, and of concentration
(lagoon) type if P + R − E < 0. In Europe there are two typical semi-enclosed es-
tuarine basins, the Black Sea and the Baltic Sea, and one concentration basin, the
Mediterranean Sea (Fig. 5.1). These seas provide paradigms of semi-enclosed seas
E.V. Stanev (B) · X. Lu

Institute for Coastal Research, Helmholz Zentrum Geesthacht, Max-Planck-Strasse 1,
21502 Geesthacht, Germany
e-mail: stanev@hzg.de
X. Lu
e-mail: xi.lu@hzg.de

DOI 10.1007/978-3-319-00440-2_5,
132 E.V. Stanev and X. Lu
Fig. 5.1 European

semi-enclosed seas and their
catchment basins. Colours
give the annual mean water
availability for 1961–1990.
Courtesy Tim aus der Beek,
Center for Environmental
Systems Research, University
of Kassel
with very different characteristics. However, there are some similarities, the ma-
jor one being the dominance of the water exchange through the straits for their
hydrophysical and biogeochemical state. The existing huge amount of available
knowledge makes parallel consideration of the three basins a real challenge, which
motivated the authors to write the present review, which we hope is also relevant to
other similar ocean systems like the ones in the Red Sea, the Persian Gulf, the Sea
of Japan and many other estuaries.
5.1.1 Fresh Water Fluxes
The issue of the water exchange between the Black Sea and the Mediterranean
Sea triggered one of the first modern discoveries in oceanography due to Count
Luigi Ferdinando Marsili (1658–1730). With the help of laboratory experiments he
proved in 1679–1680 the existence of inter-basin exchange at the straits, consisting
of the surface current transporting Black Sea water into the Mediterranean Sea and
a counter-current transporting Mediterranean water below the surface current in the
5 European Semi-enclosed Seas 133
Table 5.1 Characteristics of the basins and straits. References: (1) Unluata et al. (1990), (2) Mar-
iotti et al. (2002), (3) Struglia et al. (2004), (4) HELCOM (1986), (5) Simonov and Altman
(1991), Stanev and Peneva (2002), (6) Baschek et al. (2001), (7) Liljebladh and Stigebrandt (1996),
(8) Döös et al. (2004), Meier et al. (2006). More estimates can be found in Tsimplis et al. (2006)
Black Sea Mediterranean Sea Baltic Sea
Basin Area, km2 436,400 2,500,000 415,000

characteristics Volume, km3 547,000 3,700,000 21,700
Mean depth, m 1,253 1,500 55
Maximum depth, m 2,212 5,267 459
Surface water Precipitation 300/687(1) 827–1192/331–477(2) 224/540(4)

Budget Evaporation 350/802(1) 2335–2940/934–1176(2) 184/443(4)
(km3 /yr)/
(mm/yr) River inflow 350/802(1) 250–500/100–200(3) 436/1050(4)
Strait Width (narrowest Bosporus 0.7 Gibraltar 14 Øresund 4,

section, km) Great Belt 16
Outflow, km3 /yr 600(5) 25,544(6) 946(7)
Mean salinity 18(5) 38(2) 6–8(8)
Inflow, km3 /yr 300(5) 23,967(6) 473(7)
Mean salinity 36(5) 36(2) 10–14(8)
opposite direction (Marsili 1681). These first scientifically documented measure-

ments of density currents provided indirect evidence for the slope of the sea level
associated with the density difference between the Black Sea and the Mediterranean
Sea. This discovery has motivated a number of challenging research efforts and is
still an active research topic because of the importance of strait exchanges for the
regional and global oceanography.
Runoff and precipitation over the Black Sea exceed evaporation resulting in a
net outflow of about 300 km3 /yr (Unluata et al. 1990). One very coarse estimate of
water fluxes proposed by these authors would assume for the annual precipitation,
evaporation and river runoff values of 300, 350 and 350 km3 , correspondingly (Ta-
ble 5.1). However, according to Vorosmarty et al. (1997, 1998) monthly mean river
discharge ranges from 126 to 284 km3 /yr (see also Kara et al. 2008). Furthermore
for the period 1923–1997, the mean values of evaporation and precipitation are esti-
mated as E = 239 km3 /yr and P = 378 km3 /yr (Simonov and Altman 1991; Stanev
and Peneva 2002), which does not fit well with the above coarse estimates. The ratio
between the inflowing and outflowing water in the Bosporus Straits1 scales as 18:36,
which is the rate between the salinities of the outflowing and inflowing waters.
Water exchange of the Baltic Sea (Table 5.1) presents approximately an annual
balance of an outflow of about 946 km3 /yr of Baltic water with salinity of 8.5 and an
inflow of 473 km3 /yr of water with salinity of 17 (Liljebladh and Stigebrandt 1996).
1 This notion is used here to denote the Bosporus (alternatively Bosphorus Strait, also known as the
Istanbul Strait) and the Dardanelles Strait.

The difference, which is the total annual outflow of 473 km3 /yr, is due to the posi-
tive fresh water budget consisting of annual runoff, precipitation and evaporation of
about 436 km3 , 224 km3 and 184 km3 , correspondingly (HELCOM 1986).
Estimates of the river runoff in the Mediterranean Sea (Table 5.1) are in the range
of 250–500 km3 /yr (Struglia et al. 2004), which is equivalent to 8.1–16 × 103 m3 /s.
Basin-averaged precipitation during the period 1979–1993, as estimated by Mariotti
et al. (2002), shows annual mean values ranging from 331 to 477 mm/yr. Evapora-
tion is in the range of 934–1176 mm/yr. Water loss to the atmosphere exceeds the
river input resulting in a Mediterranean fresh water deficit of about 500 mm/yr,
which for a total area of 2.5 million km2 gives 1250 km3 /yr. This number is only
four times larger than the net transport from the Black Sea. Keeping in mind that the
salinity difference between the surface and deep flows in the Bosporus Straits (36
against 18) is much larger than in Gibraltar (38 against 36), one could conclude that
the two straits control the salinity budget in the Mediterranean Sea almost equally.
5.1.2 The Role of Topography
The Black Sea (not including the Azov Sea) and the Baltic Sea (including the Kat-
tegat) are the largest brackish seas in the world with a total area of 436,400 km2
and 415,000 km2 , and volumes of 547,000 km3 and 21,700 km3 , correspondingly
(Table 5.1). The Mediterranean Sea covers an approximate area of 2.5 million km2
and has a volume of 3.7 million km3 . The complex bathymetry and geometry of the
Baltic and Mediterranean Seas largely shape the geophysical flows, making them
strongly dependent on the regional geometries (Fig. 5.2). However, the principal
difference between these two basins is that the Mediterranean Sea is a deep ocean-
type basin (mean and maximum depth of 1,500 m and 5,267 m, correspondingly),
while the Baltic Sea is very shallow (mean and maximum depth of 55 m and 459 m,
correspondingly, see Chap. 2 for more detailed information). On the contrary, in
the Black Sea, which is also a deep basin, both large-scale topography and coasts
are relatively simple (Fig. 5.2). However, an extreme topographic slope combined
with deep canyons along the southern and eastern coasts, a large shallow shelf in the
north and a mild continental slope in the north-western part make the topographic
controls in the Black Sea rather different from what is known for the Baltic Sea, see
for instance Stanev (2005).
The small depths of the straits exert an important control on the inter-basin ex-
changes. The latter depend largely on the variations in precipitation and evaporation
over the watersheds. Because the rate between runoff and basin volume is much
larger for the Black and Baltic Seas compared to the one for the Mediterranean Sea,
the water exchange between Black Sea and Mediterranean Sea and Baltic Sea and
Atlantic Ocean are more sensitive to changes in river runoff in their catchment ar-
eas. Thus by integrating the variations of global forcing over vast catchment areas
(Fig. 5.1), the Black and Baltic Seas make large-scale climatic signals very clear
(Stanev and Peneva 2002; Hünicke et al. 2010).
Fig. 5.2 Bottom topography

of the European
semi-enclosed seas from
ETOPO 1 data
Correlation between the North Atlantic Oscillation, sea level variability and ther-
mal state of the Black Sea, Baltic Sea and Mediterranean (Lehmann et al. 2002;
Stanev and Peneva 2002; Meier and Kauker 2003; Tsimplis et al. 2004; Rixen
et al. 2005; Hünicke et al. 2010) reveal pronounced regional responses to global
climatic variability. Their possible impact on the exchanges between the Black Sea
and Mediterranean Sea could be detected in the correlation between the salinity of
the upper water of the Aegean Sea and the one in the layer between 50 and 300 m
in the Black Sea (Tsimplis et al. 2004).
5.1.3 Wind-Driven and Thermohaline Circulation
Density and pressure differences in straits drive systems of layered exchange flows.
Barotropic pressure differences caused by changes in atmospheric pressure, wind
set-up and water fluxes from rivers and atmosphere exert the major control on the
regime of exchange flows. In the Baltic Sea and Black Sea the haline buoyancy
anomalies at the sea surface enhance the cyclonic circulation because most of the
fresh water (the part that is due to rivers) enters the seas in the coastal area. In
the Mediterranean the same effect is due to relatively low salinity water from the
Atlantic Ocean. The circulation is usually structured in a number of interconnected
gyre systems; the common feature among them is that large-scale currents follow
the continental slope with the coast on their right.
In the Black Sea and in the Mediterranean Sea meanders are generated along the
main currents. Associated with them, eddies entrain significant amounts of coastal
water (between the jet current and the coast) and transport them into the central part
of the sub-basins. Subsequently, eddies modify the intermediate and deep ocean
circulation and contribute to diapycnal mixing. In the three basins mesoscale pro-
cesses play a key role for the mixing of water masses, with important consequences
on the functioning of the whole ecosystem (e.g., Millot et al. 1990; Gregg and Öz-
soy 1999; Robinson et al. 2001; Lozovatsky and Fernando 2001; Lass et al. 2003;
Fig. 5.3 The difference

between the mean for winter
(December–February) and
annual mean sea level
anomaly from altimeter data
averaged from 1993 to 2011.
Satellite tracks over the
European seas are
over-imposed
Reissmann et al. 2009). Meanders, eddies and filaments, which can be regarded as
resulting from instabilities of larger-scale currents and density fronts, are charac-
terized by scales depending on the baroclinic properties of the ocean, e.g., density
stratification, which determine the value of the internal Rossby radius of deforma-
tion R1 = N H /f , where N is the Brunt–Väisälä frequency, H is the vertical scale
and f is the Coriolis parameter. This is the length scale at which rotational ef-
fects become important (see for more detailed explanations (Cushman-Roisin and
Beckers 2011) and Chaps. 2, 3 and 6). In the three basins R1 does not exceed
10–15 km, which is substantially smaller than the typical values found in the world
oceans.
The quantitative understanding of the gyre and eddy circulation has significantly
improved with the advent of satellite altimetry. Altimeter data (Fig. 5.3) capture well
basin-scale and eddy-scale dynamics. These data are provided by the Envisat (tracks
in red) and Jason-2 (tracks in blue) satellites. Envisat orbits the Earth in about 101
minutes with a repeat cycle of 35 days. The corresponding numbers for Jason-2 are
112.4 minutes and 9.9156 days. The longer repeat cycle of Envisat enables a larger
density of the tracks.
The winter sea level anomaly defined as the difference between Dec–Jan–Feb
mean and annual mean (Fig. 5.4) reveals a number of specific appearances of sea
level patterns in the three basins. The overall conclusion is that the winter anomalies
are positive in the shelf seas (Baltic Sea and shelf seas to the north-west of Europe)
and negative in the deep basins. The largest difference between winter anomalies
in the shelf seas and in the Atlantic Ocean reaches a maximum of about 10 cm in
the area of the continental slope. This result may indicate that the steric effect due
to temperature has very different appearances in the deep ocean and in shelf seas
(Fenoglio-Marc 2001). Furthermore, in the Black Sea and Mediterranean Sea, the
seasonal intensification of the circulation (stronger cyclonic circulation in winter)
is identified by the negative sea level anomaly in the basin interior; the zone of
Fig. 5.4 Temporal change in sea level in 1993–2011 from satellite altimetry in the Atlantic Ocean
(47◦ N, −9◦ E), Black Sea (43.5◦ N, 34◦ E), Mediterranean Sea (35◦ N, 20◦ E) and Baltic Sea
(58◦ N, 20◦ E). For the positions see the black dots in Fig. 5.3. Red dots in the graphs illustrate the
annual mean sea level
maximum sea level slope follows the continental slope (see also Stanev and Staneva
2000).
In the Mediterranean Sea mesoscale eddies along the Algerian Current dominate
the patterns in Fig. 5.4, while the appearances of the Batumi and Sevastopol eddies
(see also Sect. 5.3.2.2) are very clear in the Black Sea. As the pressure gradient at
sea surface, which determines the geostrophic transport, is very well recorded in
the altimeter data, the latter have a high value when studying circulation not only
globally, but also in the regional basins.
Obviously, European semi-enclosed seas (including the tidally dominated North
Sea) show a very different and specific temporal and spatial evolution of the sea level
in comparison to the adjacent ocean (Fig. 5.4). This reflects specific appearances of
the fresh water budget, surface heat fluxes, as well as control through the connecting
straits.
The basic difference between the temporal variation of the sea level in the three
basins considered here and the one in the Atlantic Ocean (where a location in the
Bay of Biscay has been taken for comparison) is that in the ocean the magnitude of
the temporal variability of 4.2 cm is low in comparison to the Mediterranean Sea
(6.3 cm), Black Sea (8.4 cm) and Baltic Sea (14.1 cm). While the Atlantic Ocean
and the Mediterranean Sea show a pronounced annual cycle (a clear maximum in
Fig. 5.5 Positions of vertical profiles from Argo floats in the Mediterranean Sea and the Black
Sea since 2005. The colour scale shows water depth (m)
the spectrum for 365 days is observed in the two basins), the temporal variability
in the Black Sea and Baltic Sea is very pronounced also at intra-seasonal (between
several months and one year) and interannual periods (Fig. 5.4).
The pronounced seasonal variability of the sea level in the Mediterranean Sea
(the location used here is in the Ionian basin) is indicative for the role of heat and
water fluxes at sea surface, the later have also a very clear seasonal variability. In
the two estuarine basins the temporal variability is not as ‘monochromatic’, which
results from the variations in the transports in the straits caused by the atmospheric
forcing (e.g., wind), as well as by the less periodic river runoff. This strait control
is a crucial factor shaping the variation of the sea level in the two estuarine basins
considered here.
An important element of ocean circulation that the satellite altimetry, even com-
bined with numerical modelling, could not resolve well was the subsurface circula-
tion. This only became possible with the help of profile data from gliders and Argo
floats (Fig. 5.5). Argo floats are usually programmed to execute several-day-long
cycles and to drift at a neutral parking depth, measuring temperature and salinity
profiles up to the surface and transmitting the data to the Argos satellite system
(Korotaev et al. 2006; Poulain et al. 2007; Menna and Poulain 2010). Using newly
available profile data overcomes the traditional drawback of the previous knowledge
on the motion of the intermediate and deep waters, which was almost entirely based
on sporadic direct observations. As shown recently by Taillandier et al. (2010), as-
similation of profile data together with satellite altimeter data of sea level anomaly
substantially improved the estimates and forecasts of the Mediterranean circula-
tion. Unfortunately, the use of such platforms in the Baltic Sea is prevented by the
dominating shallow depths, requiring alternative devices to resolve baroclinic pro-
cesses.
5.1.4 Numerical Modelling of the Three Major European Seas
Numerical models present an important complement to the existing sparse observa-

tions and provide important scientific guidance. Several types of numerical models
based on primitive equations have been used to simulate circulation and transport
of matter for European semi-enclosed seas. Because of their small size and semi-
enclosed nature, these seas provide optimal possibilities, using easily manageable
models and observational data to address a wide spectrum of ocean processes. Ex-
tended references on numerical modelling of these seas are provided in several re-
views: (Stanev 2005) for the Black Sea (Meier and Kauker 2003; Lehmann et al.
2004) for the Baltic Sea (Pinardi and Masetti 2000) for the Mediterranean. The
most widely applied models are the Modular Ocean Model (MOM, Bryan 1969),
Nucleous of European Modelling (NEMO, Madec 2008) and High Resolution
Oceanographic Model of the Baltic Sea (HIROMB community model, Funkquist
and Kleine 2007).
Numerical modelling (see also Chaps. 3, 4 and 6) does not only provide a pow-
erful tool for estimating fundamental ocean characteristics that cannot be directly
measured. It supports the generation of new knowledge and can also be used as a
basis to optimize data use and marine activities. In the present book the use of nu-
merical modelling in optimizing the fairways in the Baltic Sea is considered in detail
in Chaps. 10 and 11. The proposed solution to an inverse problem for the dynamics
of pollutants uses a statistical analysis of a large pool of Lagrangian trajectories of
water particles calculated based on velocity fields from numerical simulations (see
Chaps. 9 and 10) or, alternatively, using an advanced oil spill drift and fate model
(Chap. 11). Modelling efforts for the open Baltic Sea are presented in some detail
in Chap. 4. Therefore we present in the following an introduction to the numerical
modelling of the European semi-enclosed seas. As for the Baltic Sea, we focus on
the southern region of this water body and its water exchange with the North Sea.
In this chapter we analyse the development of numerical modelling for the three
basins with a focus on the thermohaline circulation (an idea about the thermohaline
circulation in the Mediterranean Sea is given in the schematic Fig. 79 of Tsimplis
et al. (2006). The presentation of modelling approaches will eventually facilitate the
understanding of some of the physical issues addressed in the present book, as well
as contribute to wide recognition (mostly in the field of physical oceanography) of
the state-of-the-art research in the European semi-enclosed seas.
It is not possible (and perhaps also not needed) to present in one review the
development in the three basins in parallel in a comprehensive manner and in a
perfect balance. The overall features of the Baltic Sea are presented in Chap. 2, and
a detailed overview of the relevant phenomena in the Gulf of Finland in Chap. 6. In
the following several dominating or characteristic processes and relevant areas are
addressed for each basin in some detail: the Black Sea with its intermediate water
mass formation, the Baltic Sea with its transition zone around the Danish Straits,
and the Mediterranean Sea with the Eastern Mediterranean Transient (EMT) (see
Sect. 5.3.5). In this way there is some complementarity between the knowledge
presented for the individual basins, which is relevant to the other two basins.
Fig. 5.6 Schematic

representation of inflows and
outflows. This figure gives an
idea of only some of the
processes tackled in this
review, such as clockwise
rotation of surface current,
deflection to the right of the
gravity current, generation of
instabilities on the interface
in the regions of sills.
Topography and coastline are
schematic and are composed
of elements characteristic for
the straits considered further
in text. Courtesy Gerhard
Gayer
The unifying feature for the three basins, which is water and salt exchanges
through straits (Fig. 5.6), is put in the focus of this review. A basic problem in this
area of research is the gap between fine resolution process modelling in the straits
and outflow areas (Özgökmen et al. 2004), theoretical understanding (Price et al.
1993; Price and Baringer 1994) and valuable new observations (Jarosz et al. 2011)
on one side, and advanced basin-scale numerical modelling on the other. The cou-
pling of detailed models of straits with basin-wide models is still underdeveloped
and demands future studies.
Another motivation for this review is to provide an extensive list of references
informing students and beginners in the oceanography of the European seas about
important publications and the state-of-the-art research. The chapter is structured in
the following way: In Sect. 5.2 theoretical, observational and numerical modelling
aspects of the straits’ dynamics are presented, followed in Sect. 5.3 by a presentation
of the basic development of numerical modelling of the thermohaline circulation in
the three basins. Brief conclusions and perspectives for future development are put
at the end.
5.2 Straits
5.2.1 Theoretical Considerations
The hydraulic processes which dominate straits’ dynamics can be easily understood
in terms of two-layer exchange. The theory of internal hydraulics assumes that the
flow is steady, inviscid, nonrotating, and hydrostatic. The dynamics in such systems
is controlled by the Froude numbers F1 and F2 :
u2i
Fi2 = , (5.1)
g hi
where the subscript i = 1, 2 is the layer number, ui is the layer speed, g = gρ/ρ is
the reduced gravity, ρ is the typical density, ρ is the density difference between the
layers and hi is the layer depth. The flow is called supercritical if a Froude number
exceeds 1 and subcritical if its value is <1. Supercritical flows become subcritical
by way of hydraulic jumps, associated with a large level of turbulence adjusting the
flow to a subcritical state. One special case is the so-called maximal exchange which
occurs (Armi 1986):
G2 = F12 + F22 = 1. (5.2)
One can understand the above balance and relate it to a continuously stratified fluid
in terms of the relative propagation speed of long internal waves: a control occurs
where a long internal wave is arrested. Thus subcritical flow propagates in both
directions, while changes from one basin cannot propagate in both directions when
the flow is supercritical.
Farmer and Armi (1988) and Armi and Farmer (1988) found two controls in
the Strait of Gibraltar supporting the concept of maximal exchange. In most of the
real straits friction drives the flow towards a critical state. This effect is similar to
decreasing the width or increasing the bottom height of the strait in pushing the flow
towards critical conditions (Garrett 2004). Under this type of dynamics a control
point is shifted towards the location where it would be without friction. This happens
in longer straits, e.g., the Dardanelles (Pratt 1986). According to Pratt (1986) the
ratio of friction to the inertial term is
Cd u2 / h2 Cd l
= , (5.3)
u2 / l h2
where l is the along-channel distance over which layer thickness and velocity
change significantly, and Cd is the drag coefficient. Obviously bottom friction
should be important in long, shallow straits but not in short and deep ones. Pratt
(1986) estimated Cd l/ h2 ≈ 1 in the Bosporus and ≈0.1 in Gibraltar. With very
large drag coefficient the flow cannot be hydraulically controlled and will be limited
by friction.
One important question concerning straits’ dynamics is to what extent straits and
sills regulate the thermohaline circulation in the inter-connected basins. This issue is
relevant to how the signals are routed from basin to basin and needs to be addressed
accounting for the rotational effects (a review can be found in Gill 1977; Whitehead
et al. 1974; Pratt and Lundberg 1991). The problem becomes ‘rotational’ when the
first mode baroclinic Rossby radius is of the same order or smaller than the strait
width.
An interesting case are the straits connecting the Baltic Sea with the North Sea
(see Fig. 5.7 and Fig. 2.2 in Chap. 2). Two of them, the Little Belt and Øresund (also
known as Sound) are narrow (at their narrowest sections the Øresund is 4 km and
Little Belt 1 km wide). However, the Great Belt (width between 16 and 32 km) and
the straits’ extensions in the adjacent seas (e.g., the Kattegat) could be rotationally
dominated, provided the friction does not dominate the inertial forces. In the Strait
of Gibraltar, which is 14 km wide at its narrowest section and has a sill depth of less
Fig. 5.7 The Danish Straits.

The bright blue in this
MODIS true-colour image
from July 5, 2001 may
indicate a coccolithophore
bloom. Credit to Jacques
Descloitres, MODIS Land
Rapid Response Team
(http://visibleearth.nasa.
gov/view.php?id=56331)
than 300 m, rotational effects could also be important. However, rotation seems to
be negligible in the narrow straits of the Black Sea (with a width of less than 1 km
in the narrowest section), a case which is more similar topographically to the one in
the Little Belt.
The short introduction into the theory of the two-layer exchange flows presented
above justifies the expectation that for the Danish Straits, Gibraltar and Bosporus
very different dynamics could be expected. The Great Belt and Gibraltar are rela-
tively wide, however, Gibraltar is about ten times deeper than the Danish Straits. The
Bosporus and the Danish Straits have similar depth, however, the former is about ten
times narrower than the Great Belt, which provides most of the water exchange be-
tween the Baltic and North Sea. Furthermore, not only the individual geometries of
the straits connecting European semi-enclosed seas with adjacent basins, but also
the vertical stratification and mixing differ largely. Variable hydrological states af-
fected by seasonality and other shorter-time forcing mechanisms could also sub-
stantially impact the dynamic controls. Of particular importance is that while the
two-layer exchange in the Gibraltar and Danish Straits is tidally dependent, in the
shallow Danish and Bosporus Straits changing wind conditions (including atmo-
spheric pressure) provide the major contribution as an external barotropic forcing.
5.2.2 The Danish Straits and Baltic Sea Underwater Passages
The Baltic Sea and the North Sea are connected by three narrow straits: the Great
Belt, the Øresund (Sound) and the Little Belt. These straits, the adjacent Kattegat
and the western Baltic Sea form a transition zone between the low saline Baltic Sea
and the North Sea (Fig. 5.7; see also Fig. 2.2 in Chap. 2). The transition zone is
characterized by a complex bathymetry, most of which has glacial origin: numerous
sills and shallows in the Kattegat and Western Baltic Sea, and deep and narrow
channels. Estuarine circulation in the transition zone is dominated by a rigorous
Fig. 5.8 Difference between the sea level anomaly in the Arkona Basin (13◦ E, 55◦ N) and Katte-
gat (11.5◦ E, 56.38◦ N) estimated from AVISO altimeter data (left), and observed salinity at 6 and
25 m in the Fehmarn Belt (right) for 1992–1994. For the location see the red dots in Fig. 5.20
dynamics of exchange flows and frontal systems. During the usual outflow situation
low saline Baltic water occupies the upper layer while relatively high saline water
flows from the Skagerrak through the Kattegat and enters the straits between the
Danish Islands and the mainland.
The dominating barotropic transport is mostly due to the sea level difference be-
tween the northern Kattegat and the western Baltic Sea forced by wind (Fig. 5.8a).
Instantaneous flow velocities reach values, which are about one order of magnitude
higher than the long-term averaged flow. Occasionally large inflows occur and re-
new the bottom water in the deepest parts of the Baltic Proper.2 One such event
is exemplified in Fig. 5.8 by the rapid change in the sea level difference between
the Baltic and North Sea. About 90 % of these inflow events take place during
the late autumn and winter bringing once every 4–5 years, within a few days only,
hundreds of cubic kilometres of well oxygenated saline waters into the Baltic Sea
(Matthäus and Franck 1992; Borenäs and Piechura 2007). On their way along the
chain of channels, sills and flats, inflowing waters replace old deep water. This pro-
cess is controlled by strong mixing and mesoscale eddy dynamics. Excursions of
the frontal system extend over the whole transition zone and salinity fluctuates up to
5–15, with daily or longer term periods associated with the atmospheric circulation.
The high rate of change of hydrophysical parameters in the transition zone
changes tremendously the transport of heat and salt, providing a mechanism for the
ventilation of the Kattegat and the Baltic Sea. In the straits areas bottom water is ven-
tilated with surface water, which is revealed by the very small difference between
the salinity at sea surface and deeper layers during the inflow events (Fig. 5.8b).
This process shapes the magnitudes of oxygen fluxes on the two sides of the straits.
Consequently, the ecological state (hypoxia and ventilation of the deep Baltic Sea)
becomes very sensitive to the extreme inflow events. It has been proved by Meier
2 We use here the notion Baltic Proper to denote the Eastern, Northern and Western Gotland Basin,
Bornholm Basin and Gdańsk Bay (Chap. 2, Fig. 2.2)

(2005) that the average age of water masses is most sensitive to perturbations of the
wind speed (decreased wind speed causes significantly increased age of the deep
water).
Gustafsson (2000) quantified the residence times of the bottom water to be about
15 days in the northern and central Kattegat and 34 days in the southern Kattegat.
Analysis of the distribution of age tracers demonstrated that the ventilation of bot-
tom water in the Great Belt significantly influenced oxygen conditions in the south-
ern and western Baltic Sea (Bendtsen et al. 2009). In contrast, the central Kattegat
was primarily ventilated by advection of bottom water from the Skagerrak. An age
tracer representing the ventilation rate of bottom water with either Skagerrak water
or surface water was shown to be inversely correlated to the observed oxygen distri-
bution in the region. Recently, Stedmon et al. (2010) claimed that exchange between
the North and Baltic Sea is so important that water from the German Bight was de-
tectable at salinities down to 12 in the Kattegat and Belt Sea. On average, 23 %
of the coloured dissolved organic matter (CDOM) in bottom waters of the Katte-
gat, Great Belt, Belt Sea, Arkona Basin (see Fig. 2.2 in Chap. 2) and the Øresund
originated from the German Bight.
The theory of rotating hydraulics appeared useful to understand the dynamics
of topographically constrained currents. To describe rotating hydraulics of outflow
from a wide and shallow basin into a channel, the circulation in the upstream basin
has to be known (Gill 1977; Borenäs and Lundberg 1986). However, the Baltic sub-
basins are relatively shallow and wide therefore the characteristics of the upstream
basin only are not sufficient to fully describe the dominating processes (Laanearu
and Lundberg 2003). Furthermore, Borenäs et al. (2007) came to the conclusion
that a hydraulic framework, although providing an upper bound of the transport,
is of limited use when dealing with the Stolpe Channel (see Fig. 2.2 in Chap. 2)
overflow because it was more likely that the transport is governed by the combined
effects of friction and wind forcing. Therefore more detailed numerical modelling
is needed to understand the processes in question (Meier et al. 2006).
The first important step in the Baltic Sea oceanography demonstrating the power
of numerical models to provide knowledge on the water exchange through the straits
was done by Sayin and Krauss (1996). One of the important questions they asked
was the following: provided that the cross-sectional areas of the Great Belt and
Øresund were 0.255 and 0.08 km2 , respectively, do total through-flows obey the
rate 25.5 : 8? With the help of academic and realistic numerical simulations they
proved that this rate was strongly controlled by dynamics. The proportionality be-
tween flows and cross-sectional areas was particularly clear during outflow cases
characterized by very strong density gradients. However, Sayin and Krauss (1996)
were not able to establish the dependence of rates of transport on the variability
of wind conditions. Therefore the issue of how wind fluctuations control inflow-
outflow regimes presents a challenge for further research.
Another fundamental issue addressed by Sayin and Krauss (1996) was the con-
trol of the Arkona Basin circulation on the propagation of inflowing water into the
Gotland Basin. Their results clearly demonstrated that hydrodynamics can strongly
impact the ventilation of the deep sea, an issue which has also been well understood
Fig. 5.9 MODIS ocean colour image of the Black Sea, Sea of Marmara and Aegean Sea, acquired
on June 25, 2003. Bloom of coccolithophorids trace the propagation pattern of Black Sea water
into the Sea of Marmara and further into the Aegean Sea. The data used were acquired as part
of the activities of NASA’s Science Mission Directorate, and are archived and distributed by the
Goddard Earth Sciences (GES) Data and Information Services Center (DISC). Inserts, which are
taken from Soffientino and Pilson (2005), show Luigi Ferdinando Marsili’s (1681) physical model
of gravity-driven, two-layer flow (upper left) and his map (bottom right, see the approximate area
in the red box on the satellite map) summarizing the currents of the Bosporus
in the Mediterranean Sea oceanography with respect to the dynamics of the Eastern
Mediterranean Transient (see Sect. 5.3.5).
5.2.3 The Bosporus Straits
A system of two straits (Bosporus and Dardanelles) and the small Sea of Marmara
in between connect the Black Sea and Mediterranean Sea (Fig. 5.9). The two straits
have lengths of 35 and 75 km and minimum widths of 0.7 and 1.3 km, correspond-
ingly. The depth of the Bosporus varies between 30 and 100 m. Two sills located
near both entrances control the flows within this strait.
The mean sea level in the Black Sea is about 0.55 m higher than in the Aegean
Sea (Alpar and Yuce 1998), supporting an upper-layer transport from the estuarine
(Black Sea) to the concentration (Mediterranean) basin. This large slope is balanced
Fig. 5.10 Comparison between basin mean sea level (MSL) anomaly from numerical simulations
with the NEMO model (see Sect. 5.3 for the model description) and basin mean salinity. From
Grayek et al. (2010)
by friction and inertia in the straits of the Dardanelles and the Bosporus, as well as
by the baroclinic and wind-driven dynamics in the Sea of Marmara. The water level
difference between both ends of the Bosporus of about 0.2–0.4 m varies seasonally
within a range of 0.2–0.6 m (Yüksel et al. 2008).
The long-term averaged vertical mean transport in the Bosporus Straits, which is
controlled by the water budget and sea level variation, can be reconstructed using
water budget considerations (Stanev and Peneva 2002). Problems associated with
the indirect estimations of the strait exchange are the availability and the quality
of data such as river runoff, evaporation, precipitation and sea level. Grayek et al.
(2010) demonstrated that using outputs from up-to-date atmospheric models, along
with statistics based on historical data, can indeed contribute to reconstructing the
individual components of the water budget in a consistent way. Having such data one
can estimate the transport through the Bosporus as a residual using the water budget
equation. Comparisons between simulated mean sea level constrained by the water
budget equation and basin mean salinity (Fig. 5.10) demonstrate a clear negative
correlation. However the two curves do not perfectly mirror each other, revealing
delayed responses associated with the dynamics of salt penetration and distribution.
The large magnitude of the basin mean salinity, which is largely determined by
the inflow through straits, illustrates how important it is to account for an accurate
salinity budget in basin-wide numerical modelling.
Short-period oscillations in the Bosporus and Dardanelles Straits with periods
of 2–5 hours are related to the eigen-oscillations (Alpar and Yuce 1998). Tidal os-
cillations (a mix between diurnal and semi-diurnal tides) are characterized by small
amplitudes and vary along the entire straits system. Sub-tidal oscillations are mainly
driven by meteorological processes. Their interdependence is revealed by the coher-
ence between the mean sea level and barometric pressure (Yuce 1993a, 1993b) at
3.5–6.7 days. A simple least-squares fitting of water level data and forcing consist-
ing of atmospheric pressure and wind stress gives a reasonably good reconstruction
of observations, with an average error of 5 cm (Andersen et al. 1997).
The system of currents in the Bosporus (Fig. 5.11) is a typical example of density
currents with the upper layer current flowing toward the Sea of Marmara (south-
ward) and the underlying one flowing toward the Black Sea (northward). Thick-
nesses and velocities of both layers display large changes in time and space. At the
Fig. 5.11 Vertical profiles from observations in the Bosporus. Left: flow patterns on 23.01.2005
(dotted line), 16.01.2005 (dashed line) and 27.01.2005 (solid line; positive values are currents to
the north; right: salinity at stations in the Black Sea entrance (solid line) and at the entrace to the
Sea of Marmara (dashed line). The images are schematically replotted from Yüksel et al. (2008)
northern entrance, the Black Sea water has 40–50 m thickness and 5–90 cm/s cur-
rent speed (Yüksel et al. 2003). At the southern entrance, the Mediterranean water
is observed below 20–30 m with speeds sometimes exceeding 100 cm/s.
The average salinity of the top layer is 18 at the Black Sea side (Fig. 5.11).
This value increases gradually reaching 25 at the entrance of the Sea of Marmara.
The average salinity of the lower layer is 38 at the southern end of the Bosporus,
dropping to 33 at the northern sill (Oguz et al. 1990). The thickness of the interface
is ∼10 m at the entrance of the Sea of Marmara and ∼2 m at the entrance of the
Black Sea (Güler et al. 2006).
The water level difference, which is sufficient to block the lower layer flow, is
about 45 to 50 cm, whereas the upper layer flow is blocked at a difference below
10 cm (e.g., Oguz et al. 1990). Blocking events are not regular but occur only during
periods of extreme winds (Unluata et al. 1990). Blocking of the lower layer flow
occurs during northerly wind events (Latif et al. 1991; Yuce 1996), blocking of the
surface current occurs during southerly winds (Gunnerson and Ozturgut 1974).
Recent advances in developing tools to estimate vertical current profiles are re-
ported by Aydogan et al. (2010) who used artificial neural network techniques. In
total, 7039 hours of current profiles observations in total, along with meteorological
data and observations of surface elevation, have been used in training the neural net-
work. Extremely high accuracy of the estimates with an average root-mean-square
error of 0.16 m/s has been achieved. Prediction models based on this technique for
1–12 h into the future showed good overall agreement with observations, paving the
way towards reliable forecasting of straits exchange.
Using density and ADCP profiles, Gregg et al. (1999) and Gregg and Özsoy
(2002) studied the hydraulic control of the dense outflow at the southern entrance.
However, dynamic regimes deviate greatly from the hydraulic assumptions due to
the channel shape and bends along the channel. Furthermore, because of the very
thin outflow, its path and mixing are strongly dependent on the bathymetry. Under-
standing this complex dynamics would necessitate a better integration of high qual-
ity observations and numerical modelling resolving the dynamics of both straits,
Fig. 5.12 Map of the Strait

of the Strait of Gibraltar
showing the main topographic
features: TN is Tarifa
Narrows (14 km), CS and ES
indicate the sills of Camarinal
(297 m) and Espartel (360 m)
evolution of the gravity currents on the shelf, as well as their plunging into the deep
ocean in the vicinity of an extremely sloped bottom.
5.2.4 Gibraltar
The Strait of Gibraltar (Fig. 5.12) connecting the Mediterranean Sea with the At-
lantic Ocean is a narrow passage of 14 km width at its narrowest section and a sill
less than 300 m deep. Two important features of the topography of this strait, which
exert important topographic control on the inter-basin exchange, are the 297 m deep
Camarinal Sill and the 300 m deep Espartel Sill.
In the Strait of Gibraltar the Mediterranean Sea circulation couples to the ther-
mohaline circulation of the Atlantic Ocean. The vertical thermohaline circulation
of the former is composed of two thermohaline cells in the eastern and west-
ern basin and one basin-wide open cell driven by intermediate water formation
in the Levantine sub-basin, and Atlantic water via the Strait of Gibraltar. In the
western part of the Mediterranean Sea the surface circulation is well presented
by the Atlantic water inflows into the Alborán Sea, forming the Alborán gyres
(Gascard and Richez 1985; Tintoré et al. 1988, 1991). The Atlantic water prop-
agates further to the east (Millot 1987) and forms the so-called Algerian Current
(Fig. 5.13). This coastal current meanders due to baroclinic instabilities (Beckers
and Nihoul 1992) and anticyclonic Algerian eddies recirculate in the Algerian basin
(Millot 1999). The deep waters in the Strait of Gibraltar showed a preference to
be banked against the African coast (Bryden and Stommel 1982; Parrilla et al.
1986).
Unlike the case of the Black Sea’s straits, tides propagating in the Atlantic Ocean
affect the two-layer exchange in the Strait of Gibraltar. The amplitude of the tidal
oscillations measured as an equivalent transport reached about 3 Sv (Bryden et al.
Fig. 5.13 Sea surface temperature in the western Mediterranean Sea and eastern Atlantic Ocean
on 21/09/2010. Focus is on the Alborán Sea, but on the Atlantic side, cold water upwellings can
be seen on the coast of Portugal. Source: MyOcean product, Atlantic-Iberian Biscay Irish-Ocean
Physics Analysis and Forecast, http://www.myocean.eu/web/18-product-showcase.php
1994), which exceeded largely the mean flow (of about 1 Sv). Hydraulic controls
reduce the tidal signal so that the oscillations at the Camarinal Sill are smoothed in
the strait.
Before surpassing the Camarinal Sill and flowing out through the Espartel
Sill into the open ocean, the Mediterranean water moves back and forth for sev-
eral tidal cycles. The spring-neap tidal cycle uplifts deeper waters in spring tides
(Kinder and Bryden 1990) and induces variations in the properties of the out-
flow. As demonstrated by Macías et al. (2007) who used a coupled circulation
and biogeochemical model, mixing over the Camarinal Sill causes an average of
16 % of the outflowing nutrients to be returned back to the Mediterranean Sea.
This fraction varies between 4 % and 35 % as a function of the tidal ampli-
tude.
The periodic formation of internal hydraulic jumps and lee waves (Armi and
Farmer 1988; Bruno et al. 2002) drives diapycnal vertical mixing that entrains the
Atlantic water into the Mediterranean layer and erodes typical water masses. Ad-
ditionally, the meteorological forcing over the Mediterranean basin (Candela et al.
1989) modifies the exchange flow substantially (Millot 2009). Positive correlation
between the near-bottom temperature in Espartel Sill and the fluctuations of the
outflow driven by the local meteorological forcing demonstrates that also for the ex-
change in this deep strait atmospheric fluctuations are of great importance (García-
Lafuente et al. 2009). Furthermore, outflow is also strongly coupled with the for-
mation of a deep water mass in the western Mediterranean Sea in winter (García-
Lafuente et al. 2007), uplifting ‘old’ deep waters to depths from which they could
enter the strait.
Baschek et al. (2001) estimated the volume of the transport through the Strait of
Gibraltar using observations from current meter moorings, shipboard measurements
and inverse modelling. It was shown that more than 92 % of the variance of the
lower layer flow was associated with tides. These results were used to remove the
tidal currents from the individual measurements and to calculate the mean flow.
The magnitude of the transport was estimated to be 0.81 Sv for the upper layer
and 0.76 Sv for the lower layer (given in Table 5.1 in km3 /yr). Another important
outcome of this work is that it resulted in a reconstruction of transport patterns along
meridional vertical sections. Three-dimensional (3D) model simulations of Skliris
and Beckers (2009) suggested that the water exchange through the Strait of Gibraltar
was sub-maximal with a mean value of about 1.1 Sv.
5.2.5 Modelling Strait Processes and Outflows
Numerical modelling and observations of exchange flows in straits have recently

shown numerous promising results. Winters and Seim (2000) investigated the ef-
fects of interfacial and bottom friction on an exchange flow through a contrac-
tion. They used a Smagorinsky-type parameterization based on a local balance
of turbulent kinetic energy, in which dissipation and the buoyancy flux balance
the production of shear. Adding interfacial friction produced shear instabilities
and larger mixing. Using the same numerical model, Hogg et al. (2001) con-
firmed the results of analytical theories in the sense that fluxes and volume fluxes
depend on three dimensionless parameters: the aspect ratio D/L, the turbulent
Grashof number GrT = g D 3 /Kν2 , which compares the effect of buoyancy forces
to the turbulent viscous forces, and the turbulent Prandtl number σt = Kν /Kρ ,
where Kν and Kρ are the eddy diffusivity coefficients for momentum and mass.
They demonstrated that for geophysical flows GrT varies in a wide range be-
tween 107 for the hydraulic limit to 102 for the viscous-advective-diffusive limit.
While the flow in the Strait of Gibraltar lies close to the hydraulic limit, the flow
in the Bosporus lies within the transitional region, therefore the exchange flow is
damped by internal mixing, which supports the earlier analytical estimates of Pratt
(1986).
Simulations in the transition zone between the North Sea and the Baltic Sea
have reached a quite good level of realism. The two-way strait exchange with the
Baltic Sea has been addressed with the help of numerical modelling since the 1990s
(Krauss and Brügge 1991; Kleine 1994; Lehmann 1994, 1995; Sayin and Krauss
1996; Andrejev et al. 2002; Meier 2007). From the very beginning the roles of wind
and barotropic pressure differences across the straits have been identified as the ma-
jor mechanisms controlling the transport in the straits (Krauss and Brügge 1991).
The next important step was the parallel analysis of idealistic and realistic mod-
els (Sayin and Krauss 1996) aiming to check the dominating controls (hydraulic or
geostrophic). They proved that cyclonic circulation in the Arkona Basin plays an
important role in controlling the outflow of water to the Bornholm Basin and fur-
ther to the Baltic Proper. Numerical modelling made clear that the ratio between
transports in the two straits is extremely variable. It can strongly differ for inflow
and outflow, and is very sensitive to large disturbances in the forcing during extreme
events.
She et al. (2007) used a high resolution operational ocean model of the North
Sea–Baltic Sea region to study the transports in the Danish Straits and the sensitivity
of the bottom water salinity in the Arkona Basin to the bathymetry. By deepening
the bathymetry in the straits the realism of simulated bottom water salinity in the
Fehmarn Belt was substantially improved.
The development of modelling efforts on the dynamics of the Bosporus Straits
evolved from simple models (e.g., Oguz et al. 1990) to process modelling. An ex-
ample is given by Simeonov et al. (1997) and Stanev et al. (2001). They used re-
sults from numerical simulations with a reduced gravity model of the Mediterranean
plume (active deep layer and motionless surface layer) intruding into the Black Sea.
The model, which had a horizontal resolution of 600 m, was used to study how the
water properties and thickness of the Mediterranean plume is modified by turbu-
lent entrainment, bottom friction and stratification. This model was coupled with a
simple chemical model simulating the oxidation of H2 S by O2 .
The characteristics of the simulated outflow (Fig. 5.14) were comparable with its
observed features. A good illustration of the dominating spatial patterns is provided
by the simulated bottom concentration of oxygen (that is high in the area of the
strait and decreases when the depth increases). The flow first spreads on the shelf
in a funnel-like form, then after reaching the shelf edge (100 m isobath) cascades
down the very steep continental slope (blue colour in Fig. 5.14). It was found that
the mixing of the Mediterranean and Black Sea water, as well as the termination
depth of the plume, are very sensitive to specific combinations of the governing pa-
rameters. In the shown case the plume reaches a depth of ∼250 m. The relatively
large entrainment rate of about 10–12 compared to the one in the outflow from the
Strait of Gibraltar of about 2–3 (Baringer and Price 1997), the small deflection of
the outflow to the right caused by the Earth’s rotation compared to the angle in other
outflows, and the shallow penetration depth are explained as a result of specific
combinations of governing parameters, topography routing and ambient stratifica-
tion.
Knowledge of the Bosporus outflow was further developed by Stashchuk and
Hutter (2003) who used a non-hydrostatic two-dimensional radial model of the
Mediterranean Sea dense plume with saline and warm water intruding into the less
dense Black Sea water. The model takes into account the typical vertical temper-
ature and salinity distribution as well as the bottom relief. Similarly to the case of
Stanev et al. (2001), the Mediterranean water moves down the shelf slope and mixes
with Black Sea water. However, this model was superior because it simulated well
the separation from the bottom when buoyancy became neutral (see Konovalov et al.
2003). The level of the lens separation from the slope depended on the intensity of
the mixing processes.
One fundamental problem in the numerical modelling of the basin-wide circula-
tion in the European semi-enclosed seas is the correct representation of the exchange
through the straits. For the Mediterranean Sea this is not so difficult to achieve be-
cause the Strait of Gibraltar is relatively wide and a reasonable resolution could
be applied. The case in the Danish Straits is similar. However, the situation in the
Black Sea is completely different. Models with structured grids can hardly resolve
Fig. 5.14 Simulated

distribution of O2 in the
Bosporus plume. The plotted
area extends between 28.8◦ E
and 29.51◦ E and 41.18◦ N
and 41.6◦ N. Depth contours
down to 100 m are plotted
with dashed lines with a
contour interval of 20 m, the
last dashed line is the isobath
of 500 m. From Stanev
(2005), see also Stanev et al.
(2001)
the strait exchanges and at the same time basin-scale dynamics, and nesting would
necessitate too many nesting levels. This led to using either prescribed fluxes in the
strait (Stanev 1990) or their estimates based on simplified approaches (Stanev et al.
1997). These authors closed the surface water balance by computing the correspond-
ing Bosporus inflow. In essence, they compensated for the fresh water gain at the
sea surface by some fresh water removal in the strait. The corresponding computa-
tions used hydraulic control theory. In practice the computation of the under-current
transport followed the equation
Q2 = (a − 1)Q0 + b, (5.4)
where Q0 is the barotropic transport in the strait, a = 0.5 and b = 13, 000 m3 /s. As
Q0 did not equal the sum of the current water flux from the atmosphere and rivers,
an optimal delay parameter has been used to avoid negative barotropic flows under
calm weather and making the shape of the seasonal variability (not amplitudes in
individual years) consistent with the basic knowledge about the variability of the
two-layer exchange (Oguz et al. 1990).
The above method did not give a full solution of the problem because even with
a resolution enabling to resolve baroclinic eddies, the numerical models cannot ap-
proximate gravity currents and interleaving well. This is a major problem because
mixing has an important role in the overturning circulation. Simulated distribution
of tracers also suffers from crudely parameterized mixing. As the intermediate and
deep water formation occurs not only through open-ocean convection, which is the
basic process generating water masses in the ocean, but also through shelf water
cascading, Stanev et al. (2004) developed a special parameterization for convection,
which is an alternative of the convective adjustment in ocean models (e.g., MOM,
see Marotzke 1991) and handles the penetration of the Bosporus plume into the
halocline.
Fig. 5.15 Vertical profiles of CFC-12 in the Black Sea (area mean and one standard deviation)
as simulated by the Black Sea Modular Ocean Model (MOM). Data from the R/V Knorr 1988
cruise are plotted by symbols. The legend gives the correspondence between symbols and station
coordinates. The outlier at 200 m is from measurements very close to the strait, thus, this value is
higher, displaying deep ventilation by the buoyant plume. From Stanev et al. (2004)
The formulated entrainment-detrainment parameterizations were very efficient

in realistically unifying the outflow from the strait and the vertical circulation (in-
cluding vertical turbulent mixing and Ekman upwelling). This enabled a consis-
tent simulation of temperature, salinity and concentration of Chlorofluorocarbons
(CFCs). The latter passive tracers provide valuable information about pathways
of water masses (Stanev et al. 2004) and present an important source of data for
model validation (Fig. 5.15). These tracer fields enable to identify dynamic con-
trols of ventilation, and the relative contributions of sources at the sea surface and
outflow from the Bosporus Straits in the formation of intermediate and deep wa-
ters.
The analysis of numerical simulations demonstrated that most of the CFC pene-
trating the deep layers has its source at the sea surface within the Black Sea rather
than from the Sea of Marmara via the Bosporus undercurrent. Under present-day
conditions, the surface CFC signals have reached only the upper halocline (Stanev
et al. 2004). Under the present stratification conditions intrusions below 600 m were
not simulated. These results could motivate further research aiming to reduce the
amount of used parameterizations and calibrations by straightforward accounting of
basic dynamics with the help of unstructured and adaptive grids (Blain et al. 2010;
Hofmeister et al. 2011).
5.3 Water Mass Formation
5.3.1 Numerical Modelling of Wind and Thermohaline Circulation
5.3.1.1 Models
The hydrodynamic equations solved by the ocean models (see Chaps. 2, 3 and 4)
are discretized vertically and horizontally in a finite number of grid elements as
described in Chap. 3 based on several simple examples. In order to resolve all of the
important thermo-hydrodynamic processes the grid resolution has to be sufficiently
fine. This would necessitate very big computer resources, which are not presently
available. Therefore optimal model set-ups are sought under which grid resolution
is fine enough to sufficiently resolve ocean eddies (‘eddy-resolving models’). With
respect to the demand to resolve the complicated geometry and topography one
promising approach is to use unstructured grids, however this type of modelling has
not yet been developed enough for simulating baroclinic processes in the European
semi-enclosed seas and will not be considered below.
In the vertical direction different types of grids are used when modelling different
ocean regimes (e.g., surface, coastal, or deep ocean, Stanev 2005, see also Chap. 3).
The early versions of the Black Sea, Baltic Sea and Mediterranean Sea models
(Stanev et al. 1989; Stanev 1990; Krauss and Brügge 1991) used the MOM, which
is a z-coordinate model based on the so-called ‘box-concept’ (Bryan 1969). These
studies have been followed by a few others using terrain following (σ -coordinates)
models, one example of which is the Princeton Ocean Model (POM, Blumberg and
Mellor 1983). Its application to the three basins under consideration has been first
documented by Oguz and Malanotte-Rizzoli (1996) for the Black Sea, Lascaratos
and Nittis (1998) for the Mediterranean Sea and Laanemets et al. (2011) for the
Baltic Sea. Of similar origin, but with two σ -layers, is the GeoHydrodynamics and
Environment Research (GHER) model, which has been extensively used by Stanev
and Beckers (1999) and Beckers (1991) to model the Black Sea and Mediterranean
Sea circulation.
The application of isopycnal models, one of the most widely used being the Hy-
brid Coordinate Ocean Model (HYCOM, Bleck 2002), to the Black Sea is described
by Kara et al. (2005). The hybrid coordinates of HYCOM allow the model to be-
have like a conventional sigma- (terrain-following) model in very shallow oceanic
regions, like a z-level coordinate model in the mixed layer or other unstratified re-
gions, and like an isopycnal coordinate model in stratified regions. One big chal-
lenge was to apply this model to resolving the Mediterranean outflow in the Black
Sea (Stanev 2005) and the Gulf of Cádiz (Xu et al. 2007). These applications demon-
strated the potential of this model to realistically simulate the spreading of the out-
flowing plumes and the downstream evolution of temperature and salinity. Of similar
type is the model recently developed by Dick and Kleine (2007) using general verti-
cal coordinates or the model of Hofmeister et al. (2011) using non-uniform adaptive
vertical grids.
The Dietrich Center for Air Sea Technology (DieCAST, Dietrich 1997) model
was set up for the Black Sea by Staneva et al. (2001) and for the Mediterranean
Sea by Fernández et al. (2005). The high order numerics employed in this model
made it possible, for the first time, to resolve all important mesoscale features of
the circulation in both seas. The recently developed NEMO model (Nucleus for
European Modelling of the Ocean, Madec 2008) has also been widely used for the
simulation of climatic variability in the Mediterranean Sea (Beuvier et al. 2010), the
Black Sea (Grayek et al. 2010) and for preoperational use (Oddo et al. 2009).
The Baltic Sea is well known for its advanced operational modelling, which
was originally initiated by the German Federal Maritime and Hydrographic Agency
(BSH) (Kleine 1994; Dick et al. 2001). Recently the DMI/BSHcmod model, which
is a three-dimensional (3D) primitive equation, hydrostatic, free-surface ocean
model has been developed by the Danish Meteorological Institute (DMI) (Buch and
She 2005; Larsen et al. 2007; She et al. 2007; Liu et al. 2009) based on the BSH
model. Efforts to develop a community modelling platform resulted in the HIROMB
model (Funkquist and Kleine 2007).
5.3.1.2 Parameterization of the Vertical Exchange in Estuarine Basins
Most of the above models use rich physical parameterizations (e.g., turbulence
schemes, bottom layers, overflows, upper layer physics, and radiation schemes). The
vertical exchange of momentum and heat is very specific for the estuarine basins due
to extremely strong vertical stratification, which makes the role of internal wave
breaking decisive. Therefore parameterizations frequently used for the Black and
Baltic Seas are presented below. Analysing observations of the deep oceans, and in
particular accounting for the internal wave breaking, Gargett (1984) found an in-
verse proportionality between deep water mixing and the Brunt–Väisälä frequency.
This idea was applied by Stigebrandt (1987) in a model of the vertical circulation of
the Baltic Sea, who used the following relationship between the vertical diffusion,
κ, and N

α
κ = min , κmax , (5.5)
N
where α = 1 × 10−7 m2 /s2 and κmax is the background diffusion coefficient. In
the model of Stanev et al. (1997), α was specified as α = 4 × 10−8 m2 /s2 , which
ensured consistency with the results of Lewis and Landing (1991) for the Black
Sea. In the upper mixed layer, where the above parameterization was not valid, a
maximum surface value of the vertical mixing coefficient of 10−4 m2 /s was used.
This made it possible to take into account, at least very roughly, increased diffusion
in the upper mixed layer. In the deep barotropic levels of the Black Sea a constant
value of 10−5 m2 /s was used for κ. In the case of convective instability, the standard
convective adjustment procedure in the MOM was activated. Further details on the
model sensitivity to different parameterizations are presented in Stanev et al. (1997).
In a similar way (Meier 2001) used α = 1 × 10−7 m2 /s2 , which is in agreement
Fig. 5.16 Vertically integrated transport (m2 /s) and surface currents (m/s) averaged during
1992–1994 for the whole Baltic Sea (a, b) and for the transition zone (c, d). The setup of the
numerical model used to produce these figures has been described by Lu et al. (2012)
with the results from dissipation measurements by Lass et al. (2003). Generally,
these parameterizations have to be calibrated to the mixing conditions in individual
basins, which are sometimes very specific (Lilover and Stips 2011).
5.3.2 Numerical Modelling of Thermohaline Circulation
5.3.2.1 Baltic Sea
The horizontal circulation in the Baltic Sea (Fig. 5.16) is shaped by a series of straits
(Little Belt, Great Belt, Øresund, Fehmarn Belt, Bornholm Channel and Stolpe
Channel), submarine passages and shallow sills (Darss Sill with a depth of 18 m;
Drogden Sill, 7 m; see Fig. 2.2 in Chap. 2). Most of the transformation of wa-
ter masses from the North Atlantic to the Baltic type occurs in the Belts and Øre-
sund. Additional transformation, in particular during inflow events, is observed in
the Arkona Basin (Lass and Mohrholz 2003; Mohrholz et al. 2006) and the Stolpe
Channel (Piechura and Beszczyńska-Möller 2004).
Surface circulation is strongly variable in the Baltic Sea and is, like in the Black
Sea and Mediterranean Sea, on average counter-clockwise (Figs. 5.16a, b). Numer-
ical simulations support the analysis provided by Leppäranta and Myrberg (2009)
Fig. 5.17 Time versus

distance diagram of bottom
salinity anomaly along the
section in Fig. 5.20 during
January 1993. Distance is
measured in the direction
from the Darss Sill to the
western Baltic Sea. The setup
of the numerical model has
been described in Lu et al.
(2012)
and demonstrate that surface currents follow the topographic features. Deep water
enters the Bornholm Basin through the Bornholm Channel and continues through
the Stolpe Channel to the Eastern Gotland Basin and further to the Gotland Deep.
The water in the central Baltic Sea is strongly stratified with upper layer salinity
of about 6–8 and a deeper layer salinity of about 10–14. Despite this strong stratifi-
cation reducing the vertical mixing, water is circulating in the vertical direction in a
shallow (compared to the ocean) conveyor belt (Döös et al. 2004; Meier et al. 2006).
As described by Leppäranta and Myrberg (2009) and in Chap. 2, a large cluster of
processes (such as gravity currents, entrainment of ambient water, vertical motion
controlled by topography and convection, interleaving at the neutral density level)
is behind the vertical overturning in the Baltic Sea, making a complete quantitative
description of water mass formation in this basin a very difficult problem.
The deepest part of the central (western) Baltic Sea is ventilated occasionally
during strong inflow events, as shown in Fig. 5.17. These events known as the Major
Baltic Inflows (MBI) can be identified by the sporadically increased salinity in the
subsurface layers. Some of the MBIs, e.g., the one in 1993, which have been well
documented in the Baltic Sea literature, were accompanied by an extremely strong
inflow of Atlantic water resulting from very specific atmospheric conditions. Such
events occur usually in winter and do not show clear periodicity.
A more detailed presentation of the evolution of the anomaly of bottom salinity
during the MBI event in 1993 as simulated by Lu et al. (2012) demonstrates that the
bottom signal propagates from the Kattegat into the direction of the western Baltic
Sea (Fig. 5.17). The slope of the contours gives a measure of the speed of propa-
gation (about 10 km per day, which is about 12 cm/s), reaching maximum values
of about 25 cm/s. From 6 January to about 20 January the signal propagated con-
sistently along the Great Belt, however the local increase of the salinity anomaly at
180–200 km observed during day 26 has another origin. A more detailed analysis
of the numerical simulations (see Sect. 5.3.3) demonstrated that this increase was
mostly due to the waters inflowing from the Øresund, which revealed the large im-
pact of processes in this strait on the variability of the bottom salinity in the Arkona
Basin.
The circulation pattern of the Baltic Sea contains several sub-basin gyres. One
important sub-system is the Gulf of Finland (see Chap. 6 where the absence of sill
makes the hydrodynamic connection between the Northern Gotland Basin and the
Gulf of Finland strong (Andrejev et al. 2004). The dynamics in this area sometimes
reveals counter-estuarine transport forced by winds (Elken et al. 2003, 2006), which
adds to the mixing of water masses in this area. Although as a whole the Baltic Sea
can be considered as a two-layer stratified system, the coupling between surface
and bottom circulation is at times relatively strong in the Gulf of Finland. More
details about the Gulf of Finland and the modelling of its circulation are available
in Chap. 6.
Another interesting sub-basin is the Sea of Bothnia. Its connection with the cen-
tral Baltic Sea via the Northern Gotland Basin is restricted by the Archipelago Sea,
which leads to the formation of contrasting water masses. There seems to be a sim-
ilarity between this basin and the Adriatic Sea in the sense that both of them are
separated from the deep part of the sea by sills (with different depths) and that they
have the potential to provide an important source for cold water masses penetrating
into the deeper layers of the main body of the basin. Myrberg and Andrejev (2006)
presented an extensive review of the circulation in the Sea of Bothnia.
5.3.2.2 Black Sea
The circulation in the Black Sea is usually structured in two connected gyre systems
(eastern and western gyres) encompassing the basin. The structure in Fig. 5.18 com-
pares well with the scheme of the Black Sea surface circulation based on a synthesis
of dynamic heights derived from hydrographic observations by Oguz et al. (1993).
The baroclinic Rossby radius R1 in the Black Sea is about 20 km. Between the Rim
Current and the coast a number of coastal eddies (most of them anticyclonic ones)
exist. The names of these eddies are associated with geographic names, for exam-
ple, capes and towns. The major sub-basin gyres in the Black Sea are known as the
Batumi and Sevastopol eddies.
Surface velocities in the Black Sea are typically 0.3–0.5 m/s, reaching some-
times 1 m/s in the area of the Rim Current (Staneva et al. 2001). The corresponding
difference between the sea level in the coastal and open sea is about 0.2 m, with
the largest values located along the continental slope. Currents in the deeper layers
reach several cm/s and are essentially barotropic.
The horizontal transport almost doubles in winter, which is the season of more
intense circulation. The transition between summer and winter circulation is con-
trolled by baroclinic eddies and is well revealed by the changing balance between
vorticity in the open ocean and the coastal sea (Stanev and Staneva 2000). The
transition between different circulation states (one-gyre circulation in winter and a
system of multiple sub-basin-scale eddies in summer) is enhanced by the large sea-
sonal stratification cycle above a relatively shallow and strong pycnocline (Staneva
et al. 2001).
Fig. 5.18 Snapshot of sea level and surface streamlines simulated by the DieCAST model. For
more details see Staneva et al. (2001)
The vertical circulation cell includes the coastward transport of surface waters
due to cyclonic wind stress and compensating inward transport in the deeper lay-
ers. This cell is closed in the vertical by upward motion in the basin interior and
downward motion in the coastal regions.
5.3.2.3 Mediterranean Sea
Although observations are crucial to understanding the thermohaline circulation,

they cannot give a comprehensive four-dimensional description of the ocean. Here
the development of numerical models helped greatly. From the very beginning of the
numerical modelling of the Mediterranean thermohaline circulation (Stanev et al.
1989) it became clear that the seasonal surface forcing was crucial for the formation
of deep and intermediate water triggered by winter convection. Numerical mod-
elling proved that deep water formation in the Gulf of Lions and the formation of
the Levantine Intermediate Water were extremely sensitive to atmospheric forcing.
As fully adequate simulation of water mass formation was difficult with the horizon-
tal resolution used at that time, data assimilation experiments (Stanev and Friedrich
1991) provided a useful tool to quantify the impact of deep water masses on the
circulation.
The first numerical simulations mentioned above and the ones which followed
(Pinardi and Navarra 1993; Roussenov et al. 1995; Pinardi et al. 1997) used the
Bryan–Cox primitive equation general circulation model (Bryan 1969) forced with
climatological atmospheric data. The analysis of the simulations demonstrated that
the seasonal variability was manifested not only by the change in amplitude and
location of the gyres but also by the appearance of seasonally recurrent gyres in
different parts of the basin, such as the Mersa–Matruh gyre. The interannual fluc-
tuations in the circulation of the upper layers appeared to be of particular relevance
Fig. 5.19 Sea surface height (coloured field, m) and currents at 35 m depth (vectors, m/s), averaged
for 40 years (1961–2000). Each fourth velocity vector is plotted. From Beuvier et al. (2010)
to water mass formation. It was demonstrated by Pinardi et al. (1997) that the East-
ern Mediterranean basin was the area where the interannual ocean response was
strongest.
The physical realism of simulations improved a lot in about fifteen years after
the pioneering modelling works and made it possible to address fundamental pro-
cesses of thermohaline circulation of the Mediterranean Sea. The simulations of
Beuvier et al. (2010) for the period 1961–2000 have been carried out with an eddy-
permitting circulation model, driven by realistic interannual high-resolution air–sea
fluxes, realistic river runoff and Black Sea inflow. The focus was on the role of the
interannual variability of the fresh water forcing in controlling the EMT precondi-
tioning and convective phase. Beuvier et al. (2010) used the NEMOMED8 model
(Beuvier et al. 2008; Sevault et al. 2009), which is a Mediterranean configuration of
the NEMO ocean model (Madec 2008) with a horizontal resolution of 1/8◦ × 1/8◦ ,
equivalently, 9–12 km.
The atmospheric forcing was specified through a dynamical downscaling of the
ERA-40 reanalysis from the ECMWF (resolution of 125 km) by the regional cli-
mate model ARPEGE-Climate (Déqué and Piedelievre 1995). The exchange with
the Atlantic Ocean was introduced by specifying a buffer zone in the Atlantic Ocean
where the 3D temperature and salinity of the model were relaxed towards the clima-
tological fields of temperature and salinity. The river inputs were explicitly added
as a fresh water flux. The Black Sea was not included in NEMOMED8 and the ex-
change between the Black Sea and the Aegean Sea was prescribed as a river runoff
into the Aegean Sea.
The dynamics of the Mediterranean Sea illustrated in Fig. 5.19 by the sea surface
height and currents at a 35 m depth reveal the well-known general cyclonic path of
the Atlantic water. The simulations replicated three intense cyclonic gyres (in the
Gulf of Lions, in the southern Adriatic Sea, and the Rhodes gyre in the Levantine
basin) in zones where winter convection generally occurs. The anticyclonic gyres
were most pronounced in the Alborán Sea providing the origin of the meandering
coastal current along the African coasts.
The typical Mediterranean thermohaline circulation (Fig. 5.19) determined by

the simulations of Beuvier et al. (2010) is quite complicated. At the surface the low
density Atlantic water enters the western Mediterranean Sea at the Strait of Gibraltar
and flows along the African coast towards the eastern Mediterranean Sea (Millot
1999). Winter convection forms dense waters in the following four main areas: (1)
the Western Mediterranean Deep Water in the Gulf of Lions, (2) the Adriatic Deep
Water (ADW) in the Adriatic Sea, (3), the Cretan Intermediate Water in the southern
part of the Aegean Sea, and (4) the Levantine Intermediate Water in the Levantine
basin (mainly in the Rhodes gyre). The Levantine Intermediate Water, which is the
most important intermediate water mass, flows at a depth of about 300 to 500 m
from the Levantine basin towards the western Mediterranean Sea.
5.3.3 Numerical Modelling of the Estuarine Circulation:

The Baltic Sea Case
The DMI/BSHcmod model is an appropriate tool to study estuarine dynamics be-

cause it has up-to-date physical parameterizations and allows a two-way nesting.
The setup of this model to study processes in the transition zone between the North
Sea and the Baltic Sea has been described by Lu et al. (2012). The coarse model
domain has a horizontal resolution of 3 nautical miles; the fine one covering the
area from Skagerrak to the Bornholm Basin (Fig. 5.20) has a horizontal resolution
of 0.5 nautical miles. The upper 30 m thick layer is resolved using 1 m thick lay-
ers to enable a precise replication of the topography of straits and sills, as well as
of the thermohaline stratification. Below the depth of 30 m the layer thickness is
2 m. The model is forced by 6-hourly meteorological forcing (10 m winds, 2 m
air temperature, mean sea level pressure, surface humidity and cloud cover) from
the DMI operational weather prediction model HIRLAM (High Resolution Limited
Area Model) with a horizontal resolution of 12 km. The surface heat flux is pa-
rameterized with bulk aerodynamic formulae using atmospheric data and simulated
sea surface temperature. River runoff is provided monthly for 31 rivers from the
BALTEX Hydrological Data Centre (BHDC, http://www.smhi.se/sgn0102/bhdc/).
The similarity of the Baltic Sea transition zone with the tidal estuaries is high-
lighted below using the profiles of the mean salinity and meridional velocity along a
zonal section line in the Great Belt (see Fig. 5.20 for the position), averaged for the
inflow and outflow conditions during December 1992–January 1993 (Figs. 5.21a, b).
Part of this period was known as one of the largest MBI events.
The following analysis, which is also relevant to the Bosporus Straits (Fig. 5.11),
aims to illustrate the basic characteristics of inflows and outflows. Obviously, the
outflow phase resembles the typical estuarine circulation: export of low salinity wa-
ter occurs in the surface layers while saltier water is imported in the bottom layer. As
the mean for the entire section could not be fully representative for the conditions
in the axis of the channel, Fig. 5.21 also provides the profiles for the deepest area
(dashed lines). The two profiles of each pair in Fig. 5.21 are similar with slightly
Fig. 5.20 The bathymetry

(depth in m) of the transition
area between the Baltic Sea
and North Sea (fine resolution
model domain). The black
line indicates the section
along which salinity is
presented in Fig. 5.17
smaller vertical gradients in the channel axis. During the inflow phase the time-
averaged sea surface salinity in the middle of the channel increased by about 6.
This means that the stability of stratification was reduced. During part of the inflow
period the water column became completely homogeneous in the vertical direction.
Velocity profiles (Fig. 5.21b) give further insight into the strait’s dynamics dur-
ing the two contrasting periods. In the water column of about 18 m from the surface
the velocity shear is very pronounced during outflow conditions (mean velocity in
the middle of the channel drops from 0.5 m/s at sea surface to zero at 18 m). There
is a velocity maximum of about 0.12 m/s at about 21 m below sea surface, revealing
the axis of the counter-current during the outflow period. In the same layer the in-
flowing velocity is almost constant, about 0.6 m/s; however the mean for the entire
section velocity shows an intermediate maximum, which reflects the fact that the
velocity was lower at the boundary of the channel. Below 18 m (9 m for the chan-
nel mean velocity) a linear decrease to zero was observed during inflow conditions
(Fig. 5.21b). The difference between the inflow and outflow situations supports the
idea about a straining of the profiles due to a changing predominant direction of the
transport through the straits.
Surface salinity patterns during outflow conditions (Fig. 5.22) reveal a clear
mesoscale structure with extremely large gradients whereas differences between
maximum and minimum salinity reach about 20 in the Kattegat. The largest salinity
values are in the north-west of the transition zone next to the Skagerrak. The low
salinity plume originating from the Øresund bifurcates. One branch is aligned to the
eastern coast and another turns to the west. The former can be detected during most
of the time as a narrow coastal jet. The outflow from the Great Belt turns to the right
mostly due to the topography and propagates towards the central Kattegat. During
this phase salinity gradients in the Baltic Sea are lower than in the Kattegat.
With the reversal of the flow the surface salinity values increased. At the be-
ginning salinity values of water originating from the Øresund were higher. With
time progressing (Fig. 5.22) two clear intrusions into the Arkona Basin, one from
the Øresund and one from the Great Belt, are easily detected in the model data.
Fig. 5.21 Mean vertical profiles for the outflow and inflow conditions: the average salinity (a) and
meridional velocity (b) along the section line in the Great Belt (see Fig. 5.20 for the position);
salinity and meridional component of velocity in the deepest part of the section are indicated with
dashed lines; the salt transport across the section line per unit depth (computed as the product of
mean salinity × mean velocity × width of the channel) (c); the contribution of the correlation
between salinity and velocity to the transport through the strait (d)
A funnel-like jet from the Skagerrak extending towards the Øresund dominated the
surface salinity pattern during 7–25 January. East of the Drogden and Darss Sills,
sea surface salinity remains low (maintaining typical Baltic Sea values) during the
entire inflow period.
5.3.4 Intermediate Water Mass Formation: The Black Sea

Example
Differential heating through the sea surface causes density differences, which in the
presence of the gravitational field drive the thermohaline circulation in the ocean.
This fundamental principle of ocean dynamics is responsible for the vertical over-
turning and water mass formation and transformation. Water masses in the Black
Sea are structured in three major types. Surface water (SW) is characterized by low
Fig. 5.22 Surface salinity distribution before inflow conditions have been fully developed (28 De-
cember 1992, left panel) and during inflow conditions (12 January 1993, right panel)
salinity due to large river runoff. This water is well pronounced on the western shelf
(salinity below 17) from where salinity gradually increases into the direction of the
open sea. Cold intermediate water (CIW) represents a layer below the seasonal ther-
mocline with low temperature, which is caused by the winter cooling. The main
pycnocline at depths of 70–120 m separates the intermediate water from the deep
water (DW). This layer has been created just after the reestablishment of the con-
nection between the Black Sea and Mediterranean Sea and is periodically refilled
(down to depths of 400–500 m) by gravity currents originating from the Bosporus
Straits.
Black Sea water mass formation is driven by a variety of factors including surface
buoyancy loss through heat and fresh water exchanges, and wind forced precondi-
tioning of surface layer density through changing depth of isopycnals, as well as
advective preconditioning caused by horizontal transport of heat and salt in surface
layers. Below we address the formation of the cold intermediate layer (CIL) as one
typical representative of the upper ocean physics in semi-enclosed seas. Because
this layer is mostly associated with the air–sea exchange, we will first give a brief
presentation on the fluxes of momentum and buoyancy at the sea surface.
The area-averaged curl of wind stress in the Black Sea (Fig. 5.23) is positive
throughout the year, maintaining an overall cyclonic circulation (Stanev et al. 1997;
Staneva et al. 2001). To compare the relative strengths of basin-wide thermohaline
and mechanical forcing, the former one is represented by the buoyancy flux, and the
latter one by the velocity of Ekman pumping
1
W= ∇ × τ. (5.6)
ρf
We multiply Eq. (5.6) by the acceleration due to gravity and scale this value by
ρs − ρm
∼ 10−3 , (5.7)
ρs
Fig. 5.23 Temporal

variability (mean year) of
wind stress, thermal
buoyancy flux (5.9), and
haline buoyancy flux (5.8).
The results are diagnosed
from simulations carried out
with the 5-minute-resolution
Black Sea Modular Ocean
Model (MOM) (Stanev et al.
2003). Also shown are the
corresponding curves
calculated from the
ECMWF-re-analysed data
(ECMWF stands for the
European Centre for
Medium-Range Weather
Forecasts) using aerodynamic
bulk formulae, air and sea
surface temperature, relative
humidity, and winds
where ρm is the mixed layer density and ρs is the density at the base of the sea-
sonal thermocline (see Marshall and Marshall 1995). This value corresponds to a
buoyancy flux of 2 × 10−9 m2 /s3 , which is almost equal to the haline buoyancy flux
Bsalt = gβQw S (5.8)
and not negligible in comparison to the thermal buoyancy flux
αQT
Bheat = −g . (5.9)
ρ 0 cw
In the above equations QT is the heat flux through the sea surface and QW
is the fresh water flux through the sea surface, α = 1.3 × 10−4 1/◦ C and β =
7.5 × 10−4 1/psu are the thermal expansion and saline contraction coefficients, re-
spectively. Comparable magnitudes of forcing terms led Stanev et al. (2003) to the
conclusion that Ekman pumping could contribute significantly to the formation of
the CIW.
The diagnostics of air–sea buoyancy exchange presented by Stanev et al. (2003),
which is computed through bulk aerodynamic parameterizations using a combi-
nation of data from atmospheric analyses and simulated sea surface temperature
(Stanev and Staneva 2001), demonstrated that both the largest heating (reached in
June) and the largest cooling (reached in early winter) amounted to about 200 W/m2 .
The amplitude of the seasonal thermal buoyancy flux exceeded that of the haline
buoyancy flux by an order of magnitude (Fig. 5.23). However, the net thermal buoy-
ancy flux was about four times smaller than the net haline flux. As buoyancy fluxes
shape the density field (Stanev et al. 2003), we conclude that the annual mean strat-
ification in the Black Sea is dominated by the dilution of surface waters by rivers,
while the seasonal variability is created mostly by air–sea heat exchange.
The specific combination of different forcing mechanisms tends to create a sta-
ble salinity stratification, which changes very little with time; however, there is an
extremely high variability in the upper ocean thermal structure. The formation and
character of the CIL is one of the main consequences of the present-day balances of
buoyancy (Fig. 5.23).
The stagnant conditions in the Black Sea are well illustrated by the fact that
temperature signals in the numerical simulations and observations cannot be traced
much deeper than 200 to 400 m. The low rate of vertical mixing (see Sect. 5.3.1.2)
is a unique feature of the Black Sea (and other estuarine seas) and is explained by
the fact that the intensity of vertical mixing in strongly stratified fluids tends to reach
the level of molecular mixing (Stanev 1990; Gregg and Özsoy 1999).
The ideas of Walin (1982) were applied to the Mediterranean Sea by Tziperman
and Speer (1994). Using the same concepts and analysing the evolution of tem-
perature in density coordinates, Stanev et al. (2003) demonstrated in a numerical
study of Black Sea water mass formation that ventilation occured only during cer-
tain parts of winter, unlike in the ocean, where seasonal variability was manifested
by the north-south excursions of outcropping isopycnal surfaces during the whole
year. The above considerations become clearer (Stanev et al. 2003) if we remem-
ber that in the Black Sea, the ratio of the net heat flux through the ocean surface
to its seasonal amplitude is a small number (∼10−2 ). Therefore, the ventilation of
the intermediate layer is maintained by the seasonal signal (not by the permanent
structures of the isopycnal surfaces), which pumps cold water periodically into the
CIL (Fig. 5.24a). The ventilation in the Black Sea is confined to a very thin sur-
face layer and is manifested by the outcropping isopycnal surfaces in winter, which
completely submerge in summer as they are overlain by light surface water. Note
also the haline dominated density (Fig. 5.24b) and the up-and-down excursions of
isopycnal surfaces (Fig. 5.24c).
Numerical simulations with the 5-minute-resolution Black Sea MOM helped to
estimate the rates of water mass formation. From the results of numerical simu-
lations one could expect that the area south of the Kerch Strait where the largest
heat fluxes occur in winter would have the largest contribution to the intermediate
water mass formation. It is, however, the area west of the Crimea Peninsula where
most of the cold surface water intrudes the pycnocline (Fig. 5.25). The explana-
tion of this ‘discrepancy’ is that the atmospheric cooling around Kerch is almost
completely compensated by the advection of warm water by the Rim Current, thus
the dynamics reduces the water mass formation rates. Just the opposite situation
is observed in the western Black Sea where the Rim Current transports cold water
Fig. 5.24 Simulated

temperature, salinity, and
depth of isopycnal surfaces
during 1991–1995 in
time-density coordinates at
42.5◦ N, 31◦ E in the interior
of the western basin. From
Stanev (2005)
originating from the northern shelves over the continental slope (preconditioning by
advection). Furthermore, the mixing of cold shelf water with saltier open seawater
Fig. 5.25 Convective cooling

estimated from numerical
simulations with the
5-minute-resolution Black
Sea Modular OceanModel
(MOM) and presented as the
thickness of the water column
(m) intruding the pycnocline
every year. From Stanev
(2005)
increases instability and triggers convection along the continental slope. The area of
most efficient cooling acts as a small (compared to the basin surface) ‘throat’ where
cold water penetrates the CIL (Stanev et al. 2003).
Mixing with more saline open ocean water enhances convection rates. As demon-
strated by the numerical experiments of Stanev and Staneva (2001), small-scale pro-
cesses govern the ventilation regime not only in the bottom layer, but also at the
fringe of the mesoscale eddies.
5.3.5 The Eastern Mediterranean Transient
The deep waters in the Mediterranean Sea are strongly constrained by straits and
sills (Cretan Arc straits, Cretan Passage, Straits of Otranto, Sicily, Corsica and Sar-
dinia). The eastern Mediterranean deep layers are occupied by the Eastern Mediter-
ranean Deep Water (EMDW). Before the 1990s the EMDW was of Adriatic origin,
however, the discovery of the Eastern Mediterranean Transient (EMT) by Roether
et al. (1996) identified the possible changes of thermohaline circulation (Klein et al.
1999; Lascaratos et al. 1999; Tsimplis et al. 2006; Roether et al. 2007). This major
climatic event in the circulation and water mass properties of the Mediterranean Sea
in the last century was recognized when it appeared that during the beginning of the
1990s, the EMDW was not formed anymore in the Adriatic Sea but in the Aegean
Sea. The new water mass (the Cretan Deep Water, CDW) overflowed into the Lev-
antine and Ionian basins through the Cretan Arc straits (Theocharis et al. 2002;
Kontoyiannis et al. 2005). It then continued to flow along the continental slope in
the Ionian basin (Roether et al. 2007).
It is fundamental that the dense water mass formation in the eastern Mediter-
ranean Sea shifted from the Adriatic Sea to the Aegean Sea. Roether et al. (1996)
showed that between 1987 and 1995 the CDW became saltier and cooler (from
14 ◦ C, 38.85 to 13.75 ◦ C, >39) and denser than EMDW (>29.2 kg/m3 ). Further-
more, Theocharis et al. (1999) showed that the EMT can be divided into two differ-
ent phases: a first one until 1992 dominated by salinity increase, and a second one
from the winters 1992 and 1993 dominated by an intense cooling.
A 3D primitive equation numerical model for the eastern Mediterranean Sea
with 20 km grid size was used by Lascaratos et al. (1999) to simulate the observed
changes and understand the basic mechanisms of the transient. These results indi-
cated that the observed changes can be, at least partially, explained as a response of
the eastern Mediterranean Sea, and more specifically of the Aegean, to the variabil-
ity in the atmospheric forcing. Deep water was found to be formed mainly through
open ocean convection in the central and northern Aegean Sea (Nittis et al. 2003).
Existing hypotheses concerning the preconditioning of the EMT were recently
summarized by Beuvier et al. (2010) as: (1) a change in the surface circulation,
which could have prevented the Atlantic water flow towards the far eastern Mediter-
ranean Sea; (2) modification of the Levantine Intermediate Water path towards the
Aegean Sea by the presence of three anticyclonic eddies south of Crete, (3) changes
in the net surface evaporation and internal salinity redistribution, (4) a decrease in
the Black Sea fresh water input due to a reduction of the river runoff, (5) intense
winter convection in 1987 in the northern Aegean Sea, and (6) the occurrence of
two successive winters (1991–1992 and 1992–1993) with strong and deep convec-
tion in all the Aegean Sea.
In the long-term simulations of the circulation of the Mediterranean Sea (Beuvier
et al. 2010) examined the contributions of the atmospheric and oceanic circulations
during the late 1980s and early 1990s, namely the atmospheric buoyancy loss, the
change of the Atlantic water circulation in the Levantine basin, the occurrence of
convection events in the Aegean Sea before 1992 and the Black Sea fresh water
discharge. Their results suggest that the key triggering factors of the EMT were the
surface heat and water losses, hence buoyancy loss, that occurred during the severe
winters 1991/92 and 1992/93, as already suggested by Josey (2003).
Buoyancy loss caused by the heat and the water loss (the resulting haline
and thermal buoyancy flux are described by Eqs. (5.8) and (5.9)) is shown in
Fig. 5.26. Winter (November–December–January–February, NDJF) mean values
over the Aegean Sea are 8.5 × 10−8 m2 /s3 for the thermal buoyancy loss (com-
pare with similar estimates for the Black Sea in Fig. 5.23) and −0.6 × 10−8 m2 /s3
for the water buoyancy loss. Further analyses showed that the heat loss was the ma-
jor component of the buoyancy loss both in terms of mean value as well as in terms
of time variations. The analysis of the evolution of the volume of dense waters in
the Aegean Sea (Beuvier et al. 2010) demonstrated that the newly formed waters
were very dense (≥29.2 kg/m3 ), in agreement with observations of Theocharis et al.
(1999).
The numerical simulations reproduced the sequence of the EMT events in agree-
ment with the observations. The heat and water losses at the surface of the Aegean
Sea during the winters 1991/92 and 1992/93 reached −73 W/m2 and −2.0 mm/day
in NDJF 1991/92, −65 W/m2 and −2.8 mm/day in NDJF 1992/93, respectively.
This triggered strong winter convection, which is followed by a huge outflow of
dense waters from the Aegean Sea to the Ionian and Levantine basins during the
following two years (Fig. 5.27). With respect to the preconditioning, it was found
in the simulations that the changing of the Atlantic water path in the Levantine
basin together with a period with net reduced fresh water inputs over the far eastern
Mediterranean Sea led to a salt increase of the Aegean Sea. The key triggering ele-
ments of the EMT are thus the surface heat and water losses which occurred during
Fig. 5.26 Anomalies of the winter (NDJF) mean of the surface buoyancy loss (in 10−7 m2 /s3 )
over the Aegean Sea of the heat loss (red line) and of the water loss (blue line). The total buoyancy
loss is plotted with the black line. From Beuvier et al. (2010)
Fig. 5.27 Monthly potential density (in kg/m3 ) simulated on the bottom of the Cretan Arc straits
(lines) and observed just south of the sill channels (circles): Karpathos (yellow), Kassos (green),
and Antikithira (purple). The depth of each sill in the model and of each measurement is indicated
in parentheses. From Beuvier et al. (2010)
the severe winters 1991/92 and 1992/93. The interannual variations of the Black
Sea discharge in the surface layer of the northern Aegean Sea did not influence the
trigger of the EMT, but rather impact the chronology of the convection event during
the previous decades, with intensification in the 1970s and weakening in the 1980s.

Recent developments in oceanography are characterized by an intertwining of re-
search carried out in the fields of observations and of numerical simulations. This
trend influenced the present review, which, although dealing with numerical mod-
elling, considered other aspects of theory and observations. This is a quite natural
choice by the authors, which is consistent with the present-day needs requiring us-
ing models not only as tools to develop conceptual knowledge, but also as tools to
solve practical problems. This can only happen if models are sufficiently validated
against observations and some of the presented results showed how close or how far
these two research fields are from another.
It is difficult to find worldwide regional-ocean areas with so long traditions in
oceanography, so many and profound research contributions and so many past and
present activities in numerical modelling as the three seas addressed above. Nev-
ertheless some important problems are waiting for progress to come. One of the
expected developments in the near future is in bringing together up-to-date-research
in straits’ dynamics, gravity currents and cascading, as well as mixing in the ocean
interior and at the bottom. All these aspects need the development of further mod-
elling capabilities with unstructured and adaptive grids, and corresponding efforts
consistent with new data emerging from radars, gliders and other sources of infor-
mation using new observational technologies.
There are many exciting issues of basic research interest which have not been
dealt with in this paper; one of them is biogeochemistry and ecosystem modelling.
In this field, European semi-enclosed seas with their rich variety present perfect
laboratories to develop and test new advances.
Another important direction of recent research, which has not been addressed
in the present paper, is associated with data assimilation and operational oceanog-
raphy. Bringing data and modelling together using dedicated analysis and statis-
tical techniques, has proven to be very productive. One example is the Europe-
wide research cooperation in the field of operational oceanography framed under
the MyOcean-2 (http://www.myocean.eu.org/) activities. The research supporting
operational oceanography develops capabilities to estimate the errors associated
with model predictions, something bringing numerical modelling and observations
in synergetic co-existence. We hope that further reviews addressing the physical
systems of European semi-enclosed seas will deal not only with the dynamics of
geophysical fields but also with the dynamics of the associated error propagation.
In this review we tried to gather the presented results under the concept of straits
and bottom outflow dominated dynamics. This aspect has not been consistently ad-
dressed in the past when using numerical modelling as a research tool. We do not
pretend to have been able to make a big step in this direction, but will be happy if
the reader starts to think of the basins presented in this paper as physical systems
dominated by bottom flows.
Acknowledgements We thank all co-authors of our past research for supporting our initiative to
write this review and in particular J.-M. Beckers for his extensive comments. The second author
was supported by the European Commission’s Seventh Framework Programme (FP7 2007–2013)
under Grant Agreement No. 217246 made with the joint Baltic Sea research and development pro-
gramme BONUS within the BalticWay activities co-funded by the Federal Ministry of Education
and Research (BMBF). This work has been partially motivated by the initiative in the MyOcean
project to improve the representation of the dynamics of straits in regional ocean models.
References
Alpar B, Yuce H (1998) Sea-level variations and their interactions between the Black Sea and the
Aegean Sea. Estuar Coast Shelf Sci 46:609–619
Andersen S, Jacobsen F, Alpar B (1997) The water level in the Bosporus Strait and its dependence
on atmospheric forcing. Dtsch Hydrogr Z 49:467–476
Andrejev O, Myrberg K, Mälkki P, Perttilä M (2002) Three-dimensional modeling of the main
Baltic inflow in 1993. Environ Chem Phys 24:156–161
9:1–16
Aydogan B, Ayat B, Öztürk MN, Çevik E, Yüksel Y (2010) Current velocity forecasting in straits
with artificial neural networks, a case study: Strait of Istanbul. Ocean Eng 37:443–453
Armi L (1986) The hydraulics of two flowing layers of different densities. J Fluid Mech 163:27–58
Armi L, Farmer DM (1988) The flow of Mediterranean Water through the Strait of Gibraltar. Prog
Oceanogr 21:1–105
Baringer MO, Price JF (1997) Mixing and spreading of the Mediterranean outflow. J Phys
Oceanogr 27:1654–1677
Baschek B, Send U, Lafuente JG, Candela J (2001) Transport estimates in the Strait of Gibraltar
with a tidal inverse model. J Geophys Res—Oceans 106:31,033–31,044
Beckers JM (1991) Application of a 3 D model to the Western Mediterranean. J Mar Syst 1:315–
332
Beckers JM, Nihoul JCJ (1992) Model of the Algerian Current’s instability. J Mar Syst 3:441–451
Bendtsen J, Karin E, Gustafsson J, Söderkvist J, Hansen LS (2009) Ventilation of bottom water in
the North Sea–Baltic Sea transition zone. J Mar Syst 75:138–149
Beuvier J, Sevault F, Somot S (2008) Modélisation de la variabilité interannuelle de la mer Méditer-
ranée sur la période 1960–2000 á l’aide de NEMOMED8, Note Cent 105, Groupe de Meteorol
de Grande Echelle et Clim, Cent Natl de Rech Meteorol, Toulouse, France (in French)
Beuvier J, Sevault F, Herrmann M, Kontoyiannis H, Ludwig W, Rixen M, Stanev E, Béranger K,
Somot S (2010) Modeling the Mediterranean Sea interannual variability during 1961–2000:
focus on the Eastern Mediterranean Transient. J Geophys Res—Oceans 115:C08017
Blain CA, Cambazoglu MK, Kourafalou VH (2010) Modeling the Dardanelles Strait outflow
plume using a coupled model system. In: Proceedings of the Marine Technology Society Con-
ference MTS/IEEE Oceans09: Marine Technology for our future: global and local challenges,
Biloxi, MS, 8 pp
Bleck R (2002) An oceanic general circulation model framed in hybrid isopycnic-cartesian coor-
dinates. Ocean Model 4:55–88
Blumberg AF, Mellor GL (1983) Diagnostic and prognostic numerical circulation studies of the
South Atlantic Bight. J Geophys Res—Oceans 88:4579–4592
Borenäs K, Lundberg P (1986) Rotating hydraulics of flow in a parabolic channel. J Fluid Mech
167:309–326
Borenäs K, Piechura J (2007) Baltic variability and exchanges. CLIVAR Exch 40:13–14
Borenäs K, Hietala R, Laanearu J, Lundberg P (2007) Some estimates of the Baltic deep water
transport through the Stolpe trench. Tellus A 59:238–248
Bruno M, Alonso JJ, Cózar A, Vidal J, Ruiz-Cañavate A, Echevaria F, Ruiz J (2002) The boiling-
water phenomena at Camarinal Sill, the Strait of Gibraltar. Deep-Sea Res, Part II 49:4097–4113
Bryan K (1969) A numerical method for the study of the circulation of the World Ocean. J Comput
Phys 4:347–376
Bryden HL, Stommel H (1982) Origin of the Mediterranean outflow. J Mar Res 40:55–71
Bryden HL, Candela J, Kinder TH (1994) Exchange through the Strait of Gibraltar. Prog Oceanogr
33:201–248
Buch E, She J (2005) Operational ocean forecasting at the Danish Meteorological Institute. Environ
Res Eng Manage 3:5–11
Candela J, Winant CD, Bryden HL (1989) Meteorologically forced subinertial flows through the
Strait of Gibraltar. J Geophys Res—Oceans 94:12,667–12,679
numerical aspects. Academic Press, San Diego, 828 pp
Déqué M, Piedelievre J (1995) High resolution climate simulation over Europe. Clim Dyn 11:321–
339
Dick S, Kleine E (2007) The BSH new operational circulation model using general vertical co-
ordinates. Environ Res Eng Manage 3:18–24
Dick S, Kleine E, Müller-Navarra S (2001) The operational circulation model of BSH (BSH cmod).
Model description and validation. Berichte des Bundesamtes für Seeschifffahrt und Hydrogra-
phie 29/2001. Hamburg, Germany, 48 pp
Dietrich DE (1997) Application of a modified Arakawa A grid ocean model having reduced nu-
merical dispersion to the Gulf of Mexico circulation. Dyn Atmos Ocean 27:201–217
Elken J, Mälkki P, Alenius P, Stipa T (2006) Large halocline variations in the northern Baltic
Proper and associated meso- and basin-scale processes. Oceanologia 48(S):91–117
Farmer DM, Armi L (1988) The flow of Atlantic Water through the Strait of Gibraltar. Prog
Oceanogr 21:1–105
Fenoglio-Marc L (2001) Analysis and representation of regional sea-level variability from altime-
try and atmospheric-oceanic data. Geophys J Int 145:1–18
Fernández V, Dietrich DE, Haney RL, Tintoré J (2005) Mesoscale, seasonal and interannual vari-
ability in the Mediterranean Sea using a numerical ocean model. Prog Oceanogr 66:321–340
Funkquist L, Kleine E (2007) An introduction to HIROMB, an operational baroclinic model for
the Baltic Sea. Technical report 37. SMHI, Norrköping, 47 pp
García-Lafuente J, Sánchez-Román A, del Río GD, Sannino G, Sánchez-Garrido J (2007) Re-
cent observations of seasonal variability of the Mediterranean outflow in the Strait of Gibraltar.
J Geophys Res—Oceans 112:C10005
García-Lafuente J, Delgado J, Sánchez-Román A, Soto J, Carracedo L, del Río GD (2009) Interan-
ual variability of the Mediterranean outflow observed in Espartel sill, western Strait of Gibraltar.
Gargett AE (1984) Vertical eddy diffusivity in the ocean interior. J Mar Res 42:359–393
Garrett C (2004) Frictional processes in straits. Deep-Sea Res, Part II 51:393–410
Gascard J, Richez C (1985) Water masses and circulation in the Western Alborán Sea and in the
Straits of Gibraltar. Prog Oceanogr 15:157–216
Gill AE (1977) The hydraulics of rotating-channel flow. J Fluid Mech 80:641–671
Grayek S, Stanev EV, Kandilarov R (2010) On the response of Black Sea level to external forcing:
altimeter data and numerical modelling. Ocean Dyn 60:123–140
Gregg MC, Özsoy E (1999) Mixing on the Black Sea shelf north of the Bosphorus. Geophys Res
Lett 26:1869–1872
Gregg MC, Özsoy E (2002) Flow, water mass changes, and hydraulics in the Bosphorus. J Geophys
Res—Oceans 107:3016
Gregg MC, Özsoy E, Latif MA (1999) Quasi-steady exchange flow in the Bosphorus. Geophys
Res Lett 26:83–86
Güler I, Yüksel Y, Yalciner AC, Cevik E, Ingerslev C (2006) Measurement and evaluation of the
hydrodynamics and secondary currents in and near a strait connecting large water bodies—a
field study. J Ocean Eng 33:1718–1748
Gunnerson CG, Ozturgut E (1974) The Bosphorus. In: Degens ET, Ross DA (eds) The Black
Sea—geology, chemistry and biology. American Association of Petroleum Geologists, Tulsa,
p 103
Gustafsson BG (2000) Time-dependent modeling of the Baltic Entrance Area. 1. Quantification
of circulation and residence times in the Kattegat and the straits of the Baltic sill. Estuaries
23:231–252
HELCOM (1986) Water balance of the Baltic Sea. Baltic Marine Environment Protection
Commission—Helsinki Commission, Helsinki, Finland. Baltic Sea environment proceedings,
vol 16, 174 pp
Hofmeister R, Beckers JM, Burchard H (2011) Realistic modelling of the exceptional inflows into
the central Baltic Sea in 2003 using terrain-following coordinates. Ocean Model 39:233–247
Hogg A, Ivey G, Winters K (2001) Hydraulics and mixing in controlled exchange flows. J Geophys
Res—Oceans 103:30,696–30,711
Hünicke B, Zorita E, Haeseler S (2010) Baltic Holocene Climate and regional sea-level change: a
statistical analysis of observations, reconstructions and simulations within present and past as
analogues for future changes. Final report of the DFG research unit SINCOS-2, GKSS Report
2010/2
Jarosz E, Teague WJ, Book JW, Beşiktepe S (2011) Observed volume fluxes in the Bosphorus
Strait. Geophys Res Lett 38:L21608
Josey SA (2003) Changes in the heat and freshwater forcing of the eastern Mediterranean and their
influence on deep water formation. J Geophys Res—Oceans 108:3237
Kara AB, Wallcraft AJ, Hurlburt H (2005) A new solar radiation penetration cheme for use in ocean
mixed layer studies: An application to the Black Sea using a fine resolution Hybrid Coordinate
Ocean Model (HYCOM). J Phys Oceanogr 35:13–32
Kara AB, Wallcraft AJ, Hurlburt HE, Stanev EV (2008) Air–sea fluxes and river discharges in the
Black Sea with a focus on the Danube and Bosphorus. J Mar Syst 74:74–95
Kinder TH, Bryden HL (1990) Aspiration of deep waters through straits. In: Pratt LJ (ed) The
physical oceanography of sea straits. Kluwer Academic, London, pp 295–319
Klein B, Roether W, Manca BB, Bregant D, Beitzel V, Kovacevic V, Luchetta A (1999) The large
deep water transient in the eastern Mediterranean. Deep-Sea Res, Part I 46:371–414
Kleine E (1994) Das operationelle Modell des BSH für Nordsee und Ostsee, Konzeption und Über-
sicht. Bundesamt für Seeschifffart und Hydrographie, 126 pp (in German)
Konovalov SK, Luther III GW, Friederich GE, Nuzzio DB, Tebo BM, Murray JW, Oguz T, Glazer
B, Trouwborst RE, Clement B, Murray KL, Romanov AS (2003) Lateral injection of oxygen
with the Bosporus plume–fingers of oxidizing potential in the Black Sea. Limnol Oceanogr
48:2369–2376
Kontoyiannis H, Balopoulos E, Gotsis-Skretas O, Pavlidou A, Assimakopoulou G, Papageori-
giou E (2005) The hydrology and biochemistry of the Cretan Straits (Antikithira and Kassos
Straits) revisited in the period June 1997–May 1998. J Mar Syst 53:37–57
Korotaev G, Oguz T, Riser S (2006) Intermediate and deep currents of the Black Sea obtained from
autonomous profiling floats. Deep-Sea Res, Part II 53:1901–1910
Krauss W, Brügge B (1991) Wind-produced water exchange between the deep basins of the Baltic
Sea. J Phys Oceanogr 21:373–384
Laanearu J, Lundberg P (2003) Topographically constrained deep-water flows in the Baltic Sea.
J Sea Res 49:257–265
Laanemets J, Väli G, Zhurbas V, Elken J, Lips I, Lips U (2011) Simulation of mesoscale structures
and nutrient transport during summer upwelling events in the Gulf of Finland in 2006. Boreal
Environ Res 16(Suppl A):15–26
Larsen J, Høyer JL, She J (2007) Validation of a hybrid optimal interpolation and Kalman filter
scheme of sea surface temperature assimilation. J Mar Syst 65:122–133
Lascaratos A, Nittis K (1998) A high-resolution three-dimensional numerical study of intermediate

water formation in the Levantine Sea. J Geophys Res—Oceans 103:18,497–18,511
Lascaratos A, Roether W, Nittis K, Klein B (1999) Recent changes in deep water formation and
spreading in the Eastern Mediterranean Sea: a review. Prog Oceanogr 44:5–36
Lass HU, Mohrholz V (2003) On the dynamics and mixing of inflowing salt-water in the Arkona
Sea. J Geophys Res—Oceans 108:3042
Lass HU, Prandke H, Liljebladh B (2003) Dissipation in the Baltic proper during winter stratifica-
tion. J Geophys Res—Oceans 108:3187
Latif MA, Özsoy E, Oguz T, Unluata U (1991) Observations of the Mediterranean inflow into the
Black Sea. Deep-Sea Res 38(Suppl 2):711–723
Lehmann A (1994) The major Baltic inflow in 1993—a numerical model simulation. In: ICES
annual science conference, St John’s, ICES, CM1994/Q, p 9
Lehmann A (1995) A three-dimensional baroclinic eddy-resolving model of the Baltic Sea. Tel-
lus A 47:1013–1031
Lehmann A, Lorenz P, Jacob D (2004) Modelling the exceptional Baltic Sea inflow events in
2002–2003. Geophys Res Lett 31:L21308
378 pp
Lewis BL, Landing WN (1991) The biogeochemistry of manganese and iron in the Black Sea.
Deep-Sea Res 38:S773–S805
Liljebladh B, Stigebrandt A (1996) Observations of the deepwater flow into the Baltic Sea. J Geo-
phys Res—Oceans 101:8895–8911
Lilover MJ, Stips AK (2011) An alternative parameterization of eddy diffusivity in the Gulf of
Finland based on the kinetic energy of high frequency internal wave band. Boreal Environ Res
16(Suppl A):103–116
Liu Y, Zhu J, She J, Zhuang S, Fu W, Gao J (2009) Assimilating temperature and salinity profile
observations using an anisotropic recursive filter in a coastal ocean model. Ocean Model 30:75–
87
Lozovatsky ID, Fernando HJS (2001) Mixing on a shallow shelf of the Black Sea. J Phys Oceanogr
32:945–956
Macías D, Martin AP, García-Lafuente J, García CM, Yool A, Bruno M, Vázquez-Escobar A,
Izquierdo A, Sein DV, Echevarría F (2007) Analysis of mixing and biogeochemical effects in-
duced by tides on the Atlantic-Mediterranean flow in the Strait of Gibraltar through a physical–
biological coupled model. Prog Oceanogr 74:252–272
Madec G (2008) NEMO ocean engine. Note du Pole de modélisation. Institut Pierre-Simon
Laplace (IPSL), France, No 27, ISSN No 1288-1619, 217 pp
Mariotti A, Struglia MV, Zeng N, Lau KM (2002) The hydrological cycle in the Mediterranean
region and implications for the water budget of the Mediterranean Sea. J Climate 15:1674–1690
Marotzke J (1991) Influence of convective adjustment on the stability of the thermohaline circula-
tion. J Phys Oceanogr 21:903–907
Marshall D, Marshall J (1995) On the thermodynamics of subduction. J Phys Oceanogr 25:138–
151
Marsili LF (1681) Osservazioni intorno al Bosforo Tracio overo Canale di Constantinopoli rappre-
sentate in lettera alla sacra real Maestá Cristina Regina di Svezia da Luigi Ferdinando Marsigli.
Nicoló Angelo Tinassi, Roma. (Translated into English by Soffientino, B, Pilson MEQ)
Matthäus W, Franck H (1992) Characteristics of major Baltic inflows—a statistical analysis. Cont
Shelf Res 12:1375–1400
J Geophys Res—Oceans 106:30,997–31,016
Meier HEM, Kauker F (2003) Modeling decadal variability of the Baltic Sea: 2. The role of
freshwater inflow and large-scale atmospheric circulation for salinity. J Geophys Res—Oceans
108:3368
Meier HEM, Feistel R, Piechura J, Arneborg L, Burchard H, Fiekas V, Golenko N, Kuzmina N,
Mohrholz V, Nohr C, Paka VT, Sellschopp J, Stips A, Zhurbas V (2006) Ventilation of the
Baltic Sea deep water: a brief review of present knowledge from observations and models.
Oceanologia 48:133–164
Menna M, Poulain PM (2010) Mediterranean intermediate circulation estimated from Argo data in
2003–2010. Ocean Sci 6:331–343. doi:10.5194/os-6-331-2010
Millot C (1987) Circulation in the West Mediterranean Sea. Oceanol Acta 10:143–149
Millot C (1999) Circulation in the western Mediterranean Sea. J Mar Syst 20:423–442
Millot C (2009) Another description of the Mediterranean Sea outflow. Prog Oceanogr 82:101–
124
Millot C, Taupier-Letage I, Benzohra M (1990) The Algerian eddies. Earth-Sci Rev 27:203–219
Mohrholz V, Dutz J, Kraus G (2006) The impact of exceptionally warm summer inflow events on
the environmental conditions in the Bornholm Basin. J Mar Syst 60:285–301
Nittis K, Tziavos C, Thanos I, Drakopoulos P, Cardin V, Gacic M, Petihakis G, Basana R (2003)
The Mediterranean Moored Multi-sensor Array (M3A): system development and initial results.
Ann Geophys 21:75–87
Oddo P, Adani M, Pinardi N, Fratianni C, Tonani M, Pettenuzzo D (2009) A nested Atlantic–
Mediterranean Sea general circulation model for operational forecasting. Ocean Sci 5:461–473
Oguz T, Özsoy E, Latif MA, Sur HI, Unluata U (1990) Modelling of hydraulically controlled
exchange flow in the Bosphorus Strait. J Phys Oceanogr 20:945–965
Oguz T, Latun V, Latif MA, Vladimirov V, Sur HI, Markov A, Ozsoy E, Kotovshchikov B, Ere-
meev V, Unluata U (1993) Circulation in the surface and intermediate layers of the Black Sea.
Deep-Sea Res, Part I 40:1597–1612
Oguz T, Malanotte-Rizzoli P (1996) Seasonal variability of wind and thermohaline driven circula-
tion in the Black Sea: modeling studies. J Geophys Res—Oceans 101:16,551–16,569
Özgökmen TM, Fischer PF, Duan J, Iliescu T (2004) Entrainment in bottom gravity currents over
complex topography from three-dimensional nonhydrostatic simulations. Geophys Res Lett
31:L13212
Parrilla G, Kinder TH, Preller R (1986) Deep and intermediate Mediterranean water in the western
Alborán Sea. Deep-Sea Res 33:55–88
Piechura J, Beszczyńska-Möller A (2004) Inflow waters in the deep regions of the southern Baltic
Sea—transport and transformations. Oceanologia 46:113
Pinardi N, Masetti E (2000) Variability of the large scale general circulation of the Mediter-
ranean Sea from observations and modelling: a review. Palaeogeogr Palaeoclimatol Palaeoecol
158:153–173
Pinardi N, Navarra A (1993) Baroclinic wind adjustment processes in the Mediterranean Sea.
Deep-Sea Res, Part II 40:1299–1326
Pinardi N, Korres G, Lascaratos A, Roussenov V, Stanev E (1997) Numerical simulation of the
interannual variability of the Mediterranean Sea upper ocean circulation. Geophys Res Lett
24:425–428
Poulain PM, Barbanti R, Font J, Cruzado A, Millot C, Gertman I, Griffa A, Molcard A, Rupolo V,
Le Bras S, Petit de la Villeon L (2007) MedArgo: a drifting profiler program in the Mediter-
ranean Sea. Ocean Sci 3:379–395
Pratt LJ (1986) Hydraulic control of sill flow with bottom topography. J Phys Oceanogr 18:1970–
1980
Pratt LJ, Lundberg PA (1991) Hydraulics of rotating strait and sill flow. Annu Rev Fluid Mech
23:81–106
Price JF, Baringer MO (1994) Outflows and deep water production by marginal seas. Prog
Oceanogr 33:161–200
Price JF, Baringer MO, Lueck RG, Johnson GC, Ambar I, Parrilla G, Cantos A, Kennelly MA,
Sanford TB (1993) Mediterranean outflow mixing and dynamics. Science 259:1277–1282
Reissmann J, Burchard H, Feistel R, Hagen E, Lass HU, Mohrholz V, Nausch G, Umlauf L, Wiec-
zorek G (2009) State-of-the-art review on vertical mixing in the Baltic Sea and consequences
for eutrophication. Prog Oceanogr 82:47–80
Rixen M, Beckers J, Levitus S, Antonov J, Boyer T, Maillard C, Fichaut M, Balopoulos E, Iona S,
Dooley H, Garcia M, Manca B, Giorgetti A, Manzella G, Mikhailov N, Pinardi N, Zavatarelli M
(2005) The Western Mediterranean Deep Water: a proxy for climate change. Geophys Res Lett
32:L12608
Robinson AR, Leslie WG, Theocharis A, Lascaratos A (2001) Mediterranean Sea circulation. In:
Encyclopedia of ocean sciences. Academic Press, San Diego, pp 1689–1706
Roether W, Manca BB, Klein B, Bregant D, Georgopoulos D, Beitzel V, Kovacevic V, Luchetta A
(1996) Recent changes in Eastern Mediterranean deep waters. Science 271:333–335
Roether W, Klein B, Manca BB, Theocharis A, Kioroglou S (2007) Transient Eastern Mediter-
ranean deep waters in response to the massive dense-water output of the Aegean Sea in the
1990s. Prog Oceanogr 74:540–571
Roussenov V, Stanev E, Artale V, Pinardi N (1995) A seasonal model of the Mediterranean Sea.
Sayin E, Krauss W (1996) A numerical study of the water exchange through the Danish Straits.
Tellus A 48:324–341
Sevault F, Somot S, Beuvier J (2009) A regional version of the NEMO ocean engine on the
Mediterranean Sea: NEMOMED8 user’s guide. Technical Note 107, CNRM, Météo-France,
Toulouse, France
She J, Berg P, Berg J (2007) Bathymetry impacts on water exchange modelling through the Danish
Straits. J Mar Syst 65:450–459
Simeonov J, Stanev E, Backhaus J, Jungclaus J, Roussenov V (1997) Heat and salt intrusions
in the pycnocline from sinking plumes. Test case for the entrainment in the Black Sea. In:
Özsoy E, Mikaelyan AŽ (eds) Sensitivity to change: Black Sea, Baltic Sea and North Sea.
Kluwer Academic, Dordrecht, pp 417–438
Simonov AI, Altman EN (eds) (1991) Hydrometeorology and hydrochemistry of the USSR Seas.
The Black Sea, vol IV. Gidrometeoizdat, Saint Petersburg, 430 pp
Skliris N, Beckers JM (2009) Modelling the Gibraltar Strait/Western Alborán Sea ecohydrody-
namics. Ocean Dyn 59:489–508
Soffientino B, Pilson MEQ (2005) The Bosporus Strait. A special place in the history of oceanog-
raphy. Oceanography 18:16–23
Stanev EV (1990) On the mechanisms of the Black Sea circulation. Earth-Sci Rev 28:285–319
Stanev EV (2005) Understanding Black Sea dynamics: overview of recent numerical modelling.
Oceanography 18:52–71
Stanev EV, Friedrich HJ (1991) On the assimilation of climatological data by means of numerical
circulation models, exemplified for the Mediterranean Sea. Oceanol Acta 14:97–114
Stanev EV, Beckers JM (1999) Barotropic and baroclinic oscillations in strongly stratifyed ocean
basins. Numerical study for the Black Sea. J Mar Syst 19:65–112
Stanev EV, Staneva JV (2000) The impact of the baroclinic eddies and basin oscillations on the
transitions between different quasi-stable states of the Black Sea circulation. J Mar Syst 24:3–
26
Stanev EV, Staneva JV (2001) The sensitivity of the heat exchange at sea surface to meso and
sub-basin scale eddies. Model study for the Black Sea. Dyn Atmos Ocean 33:163–189
Stanev EV, Peneva EL (2002) Regional sea level response to global climatic change: Black Sea
examples. Glob Planet Change 32:33–47
Stanev EV, Friedrich HJ, Botev SV (1989) On the seasonal response of intermediate and deep
water to surface forcing in the Mediterranean Sea. Oceanol Acta 12:141–149
Stanev EV, Staneva JV, Roussenov VM (1997) On the Black Sea water mass formation. Model
sensitivity study to atmospheric forcing and parameterization of physical processes. J Mar Syst
13:245–272
Stanev EV, Simeonov JA, Peneva EL (2001) Ventilation of the pycnocline in the Black Sea by the
Mediterranean Plume. J Mar Syst 31:77–97
Stanev EV, Bowman MJ, Peneva EL, Staneva JV (2003) Control of Black Sea intermediate water
mass formation by dynamics and topography: comparisons of numerical simulations, survey
and satellite data. J Mar Res 61:59–99
Stanev EV, Staneva J, Bullister JL, Murray JW (2004) Ventilation of the Black Sea pycnocline:
parameterization of convection, numerical simulations and validations against observed Chlo-
rofluorocarbon data. Deep-Sea Res 51:2137–2169
Staneva JV, Dietrich D, Stanev EV, Bowman M (2001) Rim current and coastal eddy mechanisms
in an eddy-resolving Black Sea general circulation model. J Mar Syst 3:137–157
Stashchuk N, Hutter K (2003) Modelling the gravity current flowing from the Bosphorus to the
Black Sea. Geophys Astrophys Fluid Dyn 97:1–24
Stedmon CA, Osburn CL, Kragh T (2010) Tracing water mass mixing in the Baltic–North Sea
transition zone using the optical properties of coloured dissolved organic matter. Estuar Coast
Shelf Sci 87:156–162
Stigebrandt A (1987) A model for the vertical circulation of the Baltic deep water. J Phys Oceanogr
17:1772–1785
Struglia MV, Mariotti A, Filograsso A (2004) River discharge into the Mediterranean Sea: clima-
tology and aspects of the observed variability. J Climate 17:4740–4751
Taillandier V, Dobricic S, Testor P, Pinardi N, Griffa A, Mortier L, Gasparini GP (2010) Integration
of Argo trajectories in the Mediterranean forecasting system and impact on the regional analysis
of the western Mediterranean circulation. J Geophys Res—Oceans 115:C03007
Theocharis A, Nittis K, Kontoyiannis H, Papageorgiou E, Balopoulos E (1999) Climatic changes
in the Agean Sea influence the eastern Mediterranean thermohaline circulation (1986–1997).
Geophys Res Lett 26:1617–1620
Theocharis A, Klein B, Nittis L, Roether W (2002) Evolution and status of the Eastern Mediter-
ranean Transient (1997–1999). J Mar Syst 33:91–116
Tintoré J, La Violette PE, Bladé I, Cruzado A (1988) A study of an intense density front in the
eastern Alborán Sea: the Almeria–Oran front. J Phys Oceanogr 18:1384–1397
Tintoré J, Gomis D, Alonso S, Parrilla G (1991) Mesoscale dynamics and vertical motion in the
Alborán Sea. J Phys Oceanogr 21:811–823
Tsimplis MN, Josey SA, Rixen M, Stanev EV (2004) On the forcing of sea level in the Black Sea.
Tsimplis MN, Zervakis V, Josey SA, Peneva E, Struglia MV, Stanev E, Lionello P, Malanotte-
Rizzoli P, Artale V, Theocharis A, Tragou E, Oguz T (2006) Changes in the oceanography of
the Mediterranean Sea and their link to climate variability. In: Lionello P, Malanotte-Rizzoli P,
Boscolo R (eds) Mediterranean climate variability. Developments in Earth and environmental
sciences, vol 4. Elsevier, Amsterdam, pp 227–282
Tziperman E, Speer K (1994) A study of water mass transformation in the Mediterranean Sea:
analysis of climatological data and a simple three-box model. Dyn Atmos Ocean 21:53–82
Unluata U, Oguz T, Latif MA, Özsoy E (1990) On the physical oceanography of the Turkish
Straits. In: Pratt LJ (ed) The physical oceanography of sea straits. NATO/ASI series. Kluwer
Academic, Dordrecht, pp 25–60
Vorosmarty CJ, Sharma K, Fekete BM, Copeland AH, Holden J, Marble J, Lough JA (1997) The
storage and aging of continental runoff in large reservoir systems of the world. Ambio 26:210–
219
Vorosmarty CJ, Fekete BM, Tucker BA (1998) Global River Discharge Database (RivDIS) V1.l.
http://www.daac.ornl.gov
Walin G (1977) A theoretical framework for the description of estuaries. Tellus 29:128–136
Walin G (1982) On the relation between sea surface heat flow and thermal circulation in the ocean.
Tellus 34:187–195
Whitehead JA, Leetmaa A, Knox RA (1974) Rotating hydraulics of strait and sill flows. Geophys
Fluid Dyn 6:101–125
Winters KB, Seim HE (2000) The role of dissipation and mixing in exchange flow through a
contracting channel. J Fluid Mech 407:265–290
Xu X, Chassignet EP, Price JF, Özgökmen TM, Peters H (2007) A regional modeling study of the
entraining Mediterranean outflow. J Geophys Res—Oceans 112:C12005
Yuce H (1993a) Water level variations in the Sea of Marmara. Oceanol Acta 16:335–340
Yuce H (1993b) Analysis of the water level variations in the eastern Black Sea. J Coast Res 1075–
1082
Yuce H (1996) Mediterranean water in the Strait of Istanbul (Bosphorus) and the Black Sea exit.
Estuar Coast Shelf Sci 43:597–616
Yüksel Y, Yalciner AC, Güler I, Mater B, Erturk F (2003) Hydrography of the Bosphorus Strait,
and the effects to submarine constructions. In: Proceedings of the 13th international offshore
and polar engineering conference, ISOPE, Honolulu, Hawaii, USA, May 25–30, pp 303–309
Yüksel Y, Ayat B, Öztürk MN, Aydogan B, Güler I, Cevik EO, Yalçıner AC (2008) Responses of
the stratified flows to their driving conditions—a field study. Ocean Eng 35:1304–1321
Chapter 6
The Gulf of Finland, Its Hydrography
and Circulation Dynamics
Kai Myrberg and Tarmo Soomere
Abstract The basic physical and dynamical features of the Gulf of Finland, a shal-
low elongated estuary-like water body in the north-eastern extremity of the Baltic
Sea, are introduced. The hydrography of this water body is characterized by a large
spatio-temporal variability in salinity and temperature, both in vertical and hori-
zontal directions. The gulf receives a large amount of fresh water due to the River
Neva and experiences intense water exchange through an open connection with the
main Baltic basin. The circulation dynamics has been shown to be very complex in
the gulf. Pronounced variations in forcing functions contribute to this complexity
together with rapidly changing hydrography and the complex bottom topography.
The small internal Rossby radius makes the gulf particularly challenging for numer-
ical modelling exercises—high resolution models are needed to adequately describe
its circulation patterns where mesoscale eddies, fronts, specific mixing conditions
and optional ice cover are the most important players.
6.1 Introduction
Chapters 2, 4 and 5 have shown that the complexity of the dynamics of the Baltic
Sea extends far beyond the typical features of water bodies of comparable size. Pro-
K. Myrberg (B)
Finnish Environment Institute, Marine Research Centre, Mechelininkatu 34 a, P.O. Box 140,
00251 Helsinki, Finland
e-mail: kai.myrberg@ymparisto.fi
K. Myrberg
Department of Geophysics, Klaipėda University, Herkaus Manto Str. 84, Klaipėda, Lithuania
T. Soomere
T. Soomere

DOI 10.1007/978-3-319-00440-2_6,
182 K. Myrberg and T. Soomere
nounced salinity gradients and rich mesoscale dynamics distinguish this basin from
large lakes and create a similarity of its basic processes with those occurring in the
open ocean (Feistel et al. 2008). The regular presence of sea ice and remarkably
anisotropic marine meteorological conditions play a considerable role in its func-
tioning.
The Gulf of Finland, an elongated sub-basin with a mean depth of only 37 m in
the north-eastern extremity of the Baltic Sea, probably hosts the most interesting
dynamical features among the parts of the Baltic Sea. Understanding of its basic
physics is of vital importance for many purposes, not only for the development
of different kinds of models of hydrodynamic and ecological processes but also,
perhaps even more importantly, for a number of tasks seemingly remote from the
viewpoint of physics, such as the assessment of the ecosystem health, the status
of the marine environment, or perspectives of biodiversity. Physics is particularly
needed for the investigation of the circulation, spreading, and mixing of its water
masses. Not unexpectedly, the basic physics and dynamics form the core of many
practical applications, like those connected to pollution control, marine safety or
search-and-rescue issues.
The gulf has been a part of the dawn of physical oceanography more than a cen-
tury ago. Like many other basins of the Baltic Sea, it has been shared between sev-
eral countries and basin-wide studies have required not only an interdisciplinary ap-
proach but also international cooperation. The first comprehensive research project
in the gulf after the fall of the ‘iron curtain’ (that separated scientists at its oppo-
site sides for more than a half-century) was the Estonian–Finnish–Russian Year of
the Gulf of Finland 1996. Intense studies were carried out during this thematic year
including an extensive analysis of historical and recently collected data sets, numer-
ical modelling exercises and the introduction of new theoretical concepts (Alenius
et al. 1998). Their output eventually contributed to another milestone—the designa-
tion of the Baltic Sea as a particularly sensitive sea area by the International Mar-
itime Organization at the end of 2005. The recent advances in the knowledge of the
physics and dynamics of the gulf for the decade 1997–2007 (Soomere et al. 2008)
following the Gulf of Finland Year 1996 have been reported in more than 200 peer-
reviewed publications. The next wide-ranging step is the upcoming Gulf of Finland
Year 2014.
Although the physics, hydrography and dynamics of water masses in all parts of
the World Ocean are ruled by the same principles and driven by almost the same set
of forcing factors, no two domains are perfectly identical. Even seemingly similar
areas may host completely different dynamics (and thus may require radically dif-
ferent solutions for their modelling) as has been vividly demonstrated in Chap. 5.
This peculiarity is particularly evident in the dynamics of the Baltic Sea where the
impact of different forcing factors (and the response of both the water masses and
the entire ecosystem) is hugely different, for example, for the south-western part of
the Northern Gotland Basin1 and for the Gulf of Bothnia in the far North.
1 From now on we follow the nomenclature recommended in Leppäranta and Myrberg (2009) and
use the Northern Gotland Basin (see Fig. 2.1 in Chap. 2) to denote the northern part of the Baltic
Proper.
6 The Gulf of Finland 183
The entire idea of the smart use of intrinsic dynamics of marine currents for en-
vironmental management of semi-enclosed seas, which is the major focus of this
book, first came up in the context of the Gulf of Finland (Soomere and Quak 2007).
Its starting point was not only the necessity of more detailed control of the possi-
ble pollution. Another even more important reason was a peculiar combination of
the complexity (seeming randomness) of the dynamics of this basin and frequently
occurring highly ordered transport patterns, at least in numerical simulations (An-
drejev et al. 2004a). The situation in this gulf thus is very much different from that
in the south-western Baltic Sea (where the inflow and outflow conditions occasion-
ally replace each other and in this way create a specific pattern of alternating flow
regimes) (Lu et al. 2012) or in the Gulf of Riga (that is much less affected by the
circulation in the other parts of the Baltic Sea and is separated from the saline bot-
tom waters by a sill) (Omstedt and Axell 2003). There is a wide range of impacts
stemming from the Northern Gotland Basin, from the general circulation, buoyancy-
based and mesoscale features, or internal and surface wave activity, down to a wedge
of saline water that extends deep into the gulf. The dynamics of the Gulf of Finland
represents a delicate balance between these impacts, the voluminous fresh water
river discharge into the eastern end of the gulf, the local bathymetry and specific
forcing.
A thorough description of the physics, circulation, topographic and hydrographic
features, the main dynamical scales and forcing functions affecting the observed
circulation patterns in the gulf is presented in Alenius et al. (1998), Soomere et al.
(2008), Leppäranta and Myrberg (2009). One of the aims of this chapter is to present
an insight about how the entire pool of different forcing and steering factors are in-
terrelated in a subtle way, giving rise to a certain order in the seemingly almost ran-
dom system of motions. We make an attempt to present a selection of results, based
both on classical measurements and numerical modelling, which at least to some
extent explain the complicated interactions behind the existing and numerically re-
vealed circulation patterns. This information is a critical constituent of the mod-
ern field experiments and an essential background for all advanced high-resolution
numerical modelling exercises. Its quality implicitly delineates whether or not the
efforts towards numerical replication of the local circulation, carried out nowadays
for a number of applications, incl. environmental management of the basin, lead to
reliable results or could be interpreted simply as an advanced computer game.
An overview of existing physical knowledge is thus presented from the viewpoint
of high-resolution numerical modelling of its dynamics. Several models used in this
basin are described in detail in Chaps. 4 and 9. We start with the description of the
bathymetry and overall stratification conditions, and then take a short glance at mix-
ing, meteorological forcing and ice conditions that are the main driving factors for
water movements in the Gulf of Finland. The focus is on the circulation dynamics
from mesoscale features up to basin-scale patterns. A common way of its approxi-
mate quantification is to introduce the basic scales of the motions of interest. This
material characterizes the basics of the physics of the gulf based on both the past
investigations and in terms of classical oceanographic analyses of current fields and
largely follows the three above-mentioned sources (Alenius et al. 1998; Soomere
et al. 2008; Leppäranta and Myrberg 2009).
This leads us to a topic of specific importance—the balance between wind-

induced motions, thermohaline circulation, the overall circulation pattern and wa-
ter exchange properties. The results gained from the modern observations of cur-
rents and measured time series of currents in the Gulf of Finland are discussed very
shortly, only in the context of their use for the verification and validation of basin-
scale models. We intentionally skip several aspects of the physical oceanography of
this water body (such as surface wave climatology, marine optics, most applications
of remote sensing) and only concentrate on processes that are customarily accounted
for or parameterized in contemporary circulation models, or may substantially affect
the accuracy of their outcome. The reader is referred to Wulff et al. (2001), Feistel
et al. (2008), Dera and Wozniak (2010) for further information and references about
these processes in the entire Baltic Sea.
The importance of such a focus on a selection of features becomes clear from
the following pair of coupled questions: (i) what new information have we learned
about the circulation in the gulf during the last decades via numerical simulations,
and (ii) how much of this information is reliable and to which extent can the details
and processes in the Gulf of Finland and its interaction with the rest of the Baltic
Sea be investigated using numerical modelling tools?
More generally, the ever increasing challenge for marine scientists is to distin-
guish which features highlighted by the model reflect the reality and which ones are
implicitly built-in artefacts of the numerical scheme and gaps in the forcing patterns.
There are definitely areas where numerical models cannot be replaced by real mea-
surements and where the results, even if not exact, are unrivalled. Today’s models
are definitely able to recognize to which extent the circulation in the gulf is a local
feature and how much of it is just a reflection of the open Baltic Sea conditions, or to
provide sensible estimates of the water exchange between the gulf and the Northern
Gotland Basin.
The modelling exercises have improved our understanding of the structure of the
mean circulation, its stability and the role of various driving forces in different time
scales. These issues serve as basic agents for transport and mixing. In addition, we
have learned that the current dynamics in the very thin upper layer of a few metres
may dramatically differ from that below: i.e., the cyclonic, relatively stable general
circulation in the lower layers is in some cases reversed into an anticyclonic gyre
in the uppermost water layer (Soomere et al. 2011a). Even if this feature is not per-
fectly quantified by the model, it is consistent with known theoretical solutions for
the circulation patterns in a stratified medium and the structure of the local forcing,
and is supported by some measurements (Suursaar 2010). For being fully convinced,
however, this feature should be verified in situ.
For many environmental problems it is important to know the residence time of
the water. This is a complicated topic which can be studied only by using modelling
tools because measurements can only give very rough estimates of its basic features.
A major development during the latter decade is that the new generation of high-
resolution numerical models allows us to study the mesoscale features of circulation
and, if not yet to resolve these scales in detail, at least to adequately replicate the
statistics of their impact on the circulation and transport.
Fig. 6.1 Scheme of the Gulf of Finland. Graphics by M. Viška
6.2 General Features of the Gulf of Finland Affecting

the Circulation Dynamics: The Forcing
6.2.1 Topography
The elongated, narrow and relatively shallow Gulf of Finland is the easternmost
basin in the Baltic Sea, surrounded by Finland, Russia and Estonia (Fig. 6.1). The
gulf lies between 59°11 N, 22°50 E and 60°46 N, 30°20 E (Alenius et al. 1998;
Leppäranta and Myrberg 2009). Even in the Baltic Sea scales, it is a relatively small
basin. Its length is about 400 km and the width varies from 48 km between Tallinn
and Porkkala to 135 km between Narva Bay and Kotka. Its volume and surface
area (1,103 km3 and 29,948 km2 , respectively) are about 5 % of the volume and
about 7.5 % of the surface of the whole Baltic Sea. The gulf nevertheless provides
relatively large impact to the Baltic Sea through its drainage area (420,990 km2 , i.e.,
about 20 % of the total drainage area of the Baltic Sea) and associated voluminous
river runoff.
The Gulf of Finland has no sill towards the Gotland Sea (Fig. 6.2). Topographi-
cally, the gulf is a direct continuation of the Northern Gotland Basin and just be-
comes shallower toward its eastern end. Its western boundary has been defined
historically (Witting 1910) as the line between Estonia and Finland: Põõsaspea–
Osmussaar–Hanko town. The deepest areas (80–100 m) are located in the western
and southern parts, the maximum depth (Paldiski Deep, the unofficial name of a
gap located just outside the town of Paldiski, Estonia) being 123 m (Leppäranta and
Myrberg 2009). The central part of the gulf is quite deep (over 60 m) until longi-
tude 28° E and has uninterrupted deep-water connection with the northern Gotland
Sea. The complicated shape of deeper areas (Andrejev et al. 2010) enhances the
challenge to adequately replicate the dynamics of water masses in this basin.
Various parts of the gulf have very different geometric and bathymetric prop-
erties. The Finnish nearshore is a typical example of a skären-type coast. Its ex-
Fig. 6.2 The bathymetry of the Gulf of Finland. Graphics by M. Viška
tremely stable granite formations are almost insensitive with respect to the wave
action (Granö and Roto 1989). As this area has a very limited amount of finer
material, the bathymetry is governed by the character of granite formations. The
nearshore is shallow (depth 20–40 m), has an extremely irregular coastline and
rugged bathymetry, and contains extensive archipelago areas.
In contrast, the north-eastern nearshore of the Northern Gotland Basin contains
extensive shallow areas with sedimentary seabed to the north and east of the island
of Hiiumaa. Although the southern (Estonian) coast of this gulf hosts several bays
deeply cut into the mainland, on smaller scales (of a few km) its appearance is quite
regular. This wave-dominated coast (Soomere and Healy 2011) largely consists of
locally almost straight sections with a rather steep but still comparatively regular
slope. A few islands mostly consist of till shaped by the last ice sheet (except for
Osmussaar that has a limestone core). They have been further moulded into a more
regular shape and provided with a gently sloping nearshore by the wave action. An
exception is the vicinity of Neugrund at the entrance to the gulf, the steep underwater
cliffs of which have been created by a massive hit by an asteroid.
A few major islands with granite core and steep slopes lie far offshore in the east-
ern part of the gulf: Gogland (also Suursaari or Suursaar) between Kotka and Narva,
and Tjuters (also Tytärsaari or Tütarsaar) and Seskar (also Seiskari or Seskär) further
east (Fig. 6.1). The transition zone between the area frequently impacted by deep
waters from the Northern Gotland Basin and the area where shallow water effects
predominate (between longitudes 28° E and 29° E) with gradually decreasing width
and depth is sometimes called the Seskar (Seskär) basin. It plays an essential role
in the transport of water and different substances (sediments and nutrients) between
the domain governed by the river discharge and the rest of the Gulf of Finland. Ow-
ing to a combination of massive discharge of fluvial sediment to the eastern gulf
and the deep enough water to prevent resuspension by surface waves, this intense
deposition area acts as a buffer preventing the spreading of various substances west-
wards. The easternmost part of the gulf between the island of Kotlin and the River
Fig. 6.3 The annual water budget of the Baltic Sea according to Leppäranta and Myrberg (2009).
The largest component here is the river runoff R (440 km3 /yr). Precipitation P (215 km3 /yr) ex-
ceeds slightly evaporation E (175 km3 /yr). Due to the large fresh water surplus, the outflow from
the Baltic (1660 km3 /yr) is larger than the corresponding inflow (1180 km3 /yr). No such budget
exists in the international literature for the Gulf of Finland
Neva estuary, sometimes called Neva Bay or Neva Bight, is an about 22 km long
and 14–15 km wide very shallow area, with a mean depth of only some 5 m (except
for the dredged waterway to Saint Petersburg).
6.2.2 Water Budget
The positive fresh water budget of the Gulf of Finland is an important issue not only
for the stratification of the gulf itself but also for the entire Baltic Sea (Fig. 6.3).
Most of the fresh water enters the easternmost parts of the gulf. The River Neva pro-
vides the largest single river inflow to the entire Baltic Sea. The relevant data since
1859 (Bergström and Carlsson 1994) show that its discharge is some 75.5 km3 /yr
(2400 m3 /s) on average but ranges from 42 km3 /yr (observed in 1940) to 115 km3 /yr
(in 1924). The Rivers Kymi and Narva, also located in the eastern gulf, contribute
9.5–12.5 km3 /yr (300–400 m3 /s) of fresh water while the discharge from the River
Luga is only 3 km3 /yr (Mikulski 1970). The share of other rivers is minor in the
overall fresh water budget of the gulf.
In toto, the long-term mean annual river runoff into the gulf is about 114 km3 /year.
Thus, this small gulf contributes up to about 25 % of the total fresh water input to
the Baltic Sea. This feature has substantial implications to the diversity of water
masses of this water body. It considerably enhances spatio-temporal variations in
the salinity and stratification. The resulting complexity of hydrographic fields and
circulation patterns makes it much more difficult to predict their details here than in
many other parts of the Baltic Sea.
The large fresh water input initiates certain specific water movements. It is of
particular importance in spring, when the rivers have their annual maximum runoff
and the winds and atmospheric pressure gradients are weak. At that time of the year
this ‘juvenile fresh water’ (Stigebrandt 2001) flows out of the Gulf of Finland in
the surface layer. This feature may substantially modify the transport patterns in the
surface layer during the spring season. As it only occurs in a quite thin upper layer,
ocean models of a modest vertical resolution do not always resolve this transport. It
can also be easily overlooked in oil and pollution propagation models. In summer
and autumn the fresh water input is smaller and the wind-induced circulation takes
over.
Evaporation E and precipitationP more or less balance each other in the annual
mean over the whole Baltic Sea area (Ehlin 1981; Leppäranta and Myrberg 2009).
However, the P –E budget strongly depends on the place and time. In the northern
regions of the Baltic Sea, incl. the Gulf of Finland, precipitation exceeds evapora-
tion in the annual average but both components have pronounced seasonal and in-
terannual variations (HELCOM 1986; Omstedt et al. 1997). The existing estimates
are not very accurate and definitely need further studies (Leppäranta and Myrberg
2009). The water budget of this basin furthermore depends on the water exchange
with the open Baltic, the water from which substantially affects the stratification and
circulation in the gulf.
6.2.3 Horizontal and Vertical Structure of Salinity
The stratification of the Gulf of Finland has several features that are characteristic
for large estuaries. The entire gulf can be treated as a transition zone from almost
fresh water in the Neva Bay and its vicinity to the typical brackish waters of the
Gotland Sea. As there is no sill between the gulf and the Gotland Sea, no specific
topographically-isolated water masses exist in the deeper part of the gulf. The char-
acteristic features are (i) a fresh water inflow at the landward end, (ii) a saltier wedge
of water penetrating into the estuary along the bottom from the seaward end com-
bined with (iii) strong mixing (in a part of the gulf it is strongly reduced due to the
halocline) and (iv) large gradients of salinity, temperature and density in the whole
water body (Alenius et al. 1998; Myrberg 1998; Soomere et al. 2008). The seasonal
and interannual variability of salinity and temperature in both horizontal and verti-
cal directions is pronounced due to a large natural variability in wind forcing, fresh
water input and energy budget. A specific feature of the gulf is that horizontal gra-
dients of salinity and temperature can occasionally be extremely large as a result of
local upwelling.
In the Gulf of Finland, like in the entire Baltic Sea but unlike most of the World
Ocean, salinity mostly determines the stratification of water masses. In other words,
it affects density much more than temperature. The distribution of salinity in the
Baltic Sea develops in a specific manner as described in Chaps. 2 and 5. The water
in the bottom layers of the Baltic Sea is formed from relatively saline waters of the
North Sea that at times enter the Baltic Sea and move further to inner parts of the
sea. The entering waters are denser than the ambient surface waters and thus sink
Fig. 6.4 Salinity

cross-section through the
Gulf of Finland and Gotland
Deep in summer 2010
(source: Finnish Environment
Institute, SYKE)
down to the level of neutral buoyancy. This process results in a permanent two-layer
structure of the deeper parts of the Baltic Sea: the upper layer and the bottom layer
(or lower layer), separated by a relatively sharp halocline. The halocline in the Baltic
Sea is usually at a depth of some 60–80 m and thus extends into the deepest parts
of the western and central Gulf of Finland. The depth and strength of the halocline
are controlled by the topography (that allows for a penetration of the relatively salty
water originating from the North Sea, albeit substantially mixed on its way, deep into
the gulf), wind-induced and convective mixing, the water budget and the advection
of water masses in the bottom layer.
The eastern Gulf of Finland has a different salinity dynamics. This difference
makes the modelling of the stratification and circulation in the gulf extremely com-
plex and demanding in terms of the required model resolution and the accuracy of
the description of mixing and advection. The classical concept of the shallow east-
ern part of the gulf is that the permanent halocline may be missing there and that the
salinity increases approximately linearly with depth. There is, however, a rich vari-
ety of different phenomena in this area. On average, the upper layer is homohaline,
while the bottom layer is continuously stratified. Another interesting feature is the
interplay of salinity and temperature stratification. The former is influenced by the
formation of the seasonal surface layer in summer. This warm, well-mixed layer is
largely isolated from the deeper water. The spring and summer runoff furthermore
lowers its salinity and enhances its stability (Fig. 6.4).
The strong halocline plays a crucial role in the overall stratification and mixing.
It prevents vertical mixing of the water body down to the bottom. Normally, the
deep waters below the halocline are to a large extent decoupled from the direct
atmospheric forcing. The renewal of the deep waters in the western part of the gulf
is therefore caused by the horizontal advection of deep water from the Gotland Sea.
Towards the east, the difference between surface and bottom salinity decreases and
the halocline gradually weakens until it vanishes due to the massive fresh water
surplus and strong mixing in shallower areas.
The influence of river runoff is seen in the vertical structure of salinity both in the
long-term and seasonal variations (Haapala and Alenius 1994) along the whole Gulf
of Finland. The salinity of the surface layer decreases from winter to midsummer
Fig. 6.5 Long-term seasonal variability of salinity (‰) at different depths (m, scale at right) near
the island of Utö in 1911–2012 (Riikka Hietala and Pekka Alenius, personal communication; data
courtesy of Finnish Meteorological Institute). Modified from Leppäranta and Myrberg (2009)
simultaneously with an increase in the salinity of the deep layer (Fig. 6.5). In early
spring the salinity isolines are evenly inclined. During the spring and summer the
isohalines render a more horizontal orientation in the surface layer while in the
bottom layer they shift towards the inner domains of the gulf. This suggests that the
observed seasonal salinity variations are caused by advection and that the dynamics
of the gulf plays an essential role in their formation.
The variations in the surface layer are apparently related to the melting of the
ice cover and the increased spring-time River Neva runoff. The water outflow in
the surface layer may generate an inflow into the Gulf of Finland in the deeper lay-
ers (Haapala and Alenius 1994). This driver of circulation will be discussed later.
In addition to this specific mechanism, the seasonal cycle of surface salinity is of-
ten explained by suppressed vertical convection in summer and by erosion of the
halocline in winter (Reissmann et al. 2009).
The salinity has also considerable horizontal variations all along the gulf except
in its easternmost narrow area and the Neva Bay. The resulting extreme character of
the background for the dynamics of the gulf—the baroclinic salinity field varying in
space and time—is a challenge for both experimental studies and numerical mod-
elling (Fig. 6.6). The salinity increases from east to west and from north to south.
The surface salinity increases from zero2 at the mouth of the River Neva up to 6–
6.5 ‰ in the west. In the bottom layer of the western gulf, where a halocline exists,
the salinity is 7–9 ‰ (occasionally even 10 ‰). In the bottom layer of the central
part it is 5–8 ‰ and in the east 0–5 ‰ (Fig. 6.6).
2 Although the salinity unit ‰ (per mill) is discouraged since 1978, it has been customary in a large
part of the oceanographic and popular literature to use this notion. We only use this unit in data
and estimates extracted from older sources. Note that since January 2010, the seawater standard
EOS-80 is obsolete. The new international standard TEOS-10 uses absolute salinity values in g/kg
(Millero et al. 2008) rather than psu. This choice has the advantage that volumes of sea water can
be properly converted to masses of salt contained, which is not the case for psu as Practical Salinity
is not a mass fraction by definition.
Fig. 6.6 Salinity distribution (‰) in the surface layer (upper panel) and bottom layer (lower panel)
in the Gulf of Finland based on a high-resolution dataset from June 1 to September 1 1996 (Myr-
berg et al. 2010b). The scale at right is linear from 0 to 10 ‰
6.2.4 Temperature
In the Gulf of Finland, as in the whole Baltic Sea, the water temperature follows
the two-layer structure determined by the salinity (Fig. 6.4). The most remarkable
feature is the strong seasonal cycle of the temperature in the upper layer, forced by
the radiation budget and air–sea interaction. The temperature in the lower layer is
decoupled from atmospheric forcing due to the halocline and largely dictated by the
advection from the North Sea. The latter is most pronounced in the southern Baltic
but is felt also in the gulf.
As expected, the seasonal evolution of temperature in the Gulf of Finland
(Fig. 6.7) is very different in the upper and lower layers. Due to the large seasonal
variations in the energy budget of the upper layer, the water temperature at sea sur-
face reaches its maximum in summer and minimum in winter, whereas in winter
at least a part of the gulf freezes. The seasonal variability of the temperature in the
lower layer is weak because of the decoupling from atmospheric forcing. The tem-
Fig. 6.7 Long-term mean annual course of temperature (°C) at different depths (m, scale at right)
near Utö at the mouth of the Gulf of Finland in 1911–2012 (Riikka Hietala and Pekka Alenius,
personal communication; data courtesy of Finnish Meteorological Institute). Modified from Lep-
päranta and Myrberg (2009)
peratures are typically 2–4 °C and most probably largely governed by the advection
of warm water from the Northern Gotland Basin where deep water is warmer than in
the gulf. Certain temperature flux via mixing through the halocline is also possible,
but these processes are still largely unknown.
In the Gulf of Finland an inverse thermocline might be formed in wintertime;
the temperature is at the freezing point at the surface and increases with depth to
the temperature of maximum density (2–3 °C). A dicothermal layer (see Chap. 2,
Sect. 2.2.2 for details) forms in summer at the bottom of the upper layer where the
water mass is heated from the top but the heat flux to colder water further down is
very limited due to the density jump at the halocline. The layered structure is clearly
evident in the mean temperature profiles (see Fig. 2.4 in Chap. 2). In spring, after the
melting of the ice, a thin upper layer is heated due to solar radiation (Fig. 6.8). The
surface waters quickly reach the temperature of maximum density Tm (1.5–2 °C).
In the western part of the gulf the wintertime surface temperature may remain
close to Tm . After the temperature exceeds Tm , the surface waters become lighter
than the water masses below, affecting dramatically the local mixing conditions.
Thus the convection stops and during the rest of the spring and summer the seasonal
thermocline splits the upper layer into two, separating the warm surface layer waters
from the remarkably colder waters below it. Its formation starts in the beginning of
May. In summertime it is located at a depth of 15–20 m and its deepening due to
autumn cooling begins in early September (Fig. 6.9).
As the sea reacts quite slowly to the heat input from the atmosphere because of its
thermal inertia, the sea surface temperature follows the air temperature with a certain
lag. During the summer the thickness of the mixed surface layer slowly increases
mainly due to the mechanical mixing caused by the wind forcing (Fig. 6.9). The
summer thermocline is strong, with a temperature drop up to about 10 °C across
a few meters. It prevents to a large extent the wind-induced mixing from affecting
the layer below. The resulting strongly weakened exchange of material and heat has
important effects on biogeochemical processes.
Usually, the thermocline can be easily determined from a single profile, but there
are also cases where the interpretation is complicated. The surface layer is not al-
Fig. 6.8 The temperature (°C) of the Baltic Sea water in different seasons according to Lentz
(1971) along the central axis Kattegat–Gulf of Finland, Kattegat–Eastern Gotland Basin in west–
east direction and along the central axis of the Gulf of Bothnia: (A) February, (B) May. In February
only the Kattegat–Gulf of Finland cross-section is available due to ice conditions. From Leppäranta
and Myrberg (2009)
ways homogeneous. Usually denoted as the ‘(well-)mixed’ surface layer, it is often

defined as a layer where the vertical change of temperature does not exceed some
prescribed value, say 0.1 °C. Still, it frequently possesses a transient echelon (mi-
cro)structure consisting of minor thermoclines.
The presence of a strong thermocline favours the fresh water from rivers to re-
main in the surface layer. This strengthens the density difference across the thermo-
cline. Thus the warming and fresh water content of the surface layer have a positive
feedback to each other. Typically, maximum sea surface temperature (usually 15–
20 °C, at most up to 23 °C) are observed in July–August. The horizontal gradients
can occasionally be large owing to local upwelling events.
Fig. 6.9 The same as for Fig. 6.8 but for (A) August, (B) November. From Leppäranta and Myr-
berg (2009)
In the eastern regions of the Gulf of Finland, to the east of Gogland (Fig. 6.1),
the thickness of the upper mixed layer usually does not exceed 20 m, and typically
it is 12–15 m. It decreases to the east together with the typical intensity and the
predominant length of waves, the major mixing agent in this practically non-tidal
area. In front of the Saint Petersburg flood barrier, the mixing depth, even after
strong storms, is as small as about 10 m. The maximum vertical density gradient
(sometimes called the jump layer core) lies about 1–3 m below the lower boundary
of the mixed layer. Since the water depth is comparable with the typical depth of
the mixed layer, strong storms may cause mixing of the entire water column and at
times yield large, vertically homogeneous patches.
6.2.5 Upwelling and Turbulence
Other key elements of the stratification-related dynamics in the Gulf of Finland are
mechanically forced vertical motions of water masses called upwelling and down-
welling. In both the gulf and the Baltic Sea upwelling events are predominantly
of coastal type and based on Ekman transport, resulting from the horizontal di-
vergence of wind-driven motions in the surface layer. In essence, upwelling is the
penetration of dense, cooler and usually nutrient-rich water towards the sea sur-
face. It not only affects dramatically the stratification but has an important effect
on redistributing nutrients and other substances in the vertical direction (see, e.g.,
Lips et al. 2009). Its complementary process—downwelling—takes warmer, usually
nutrient-depleted surface water to the lower layers but usually does not play such an
important role in sea dynamics as upwelling does.
In the Gulf of Finland upwelling is typically triggered by alongshore winds (Haa-
pala 1994). Owing to the complex coastline and the presence of many islands, wind
from virtually any direction may cause up- or downwelling near a certain coast with
accompanying vertical mixing and displacement of water masses. The predominant
westerly winds normally cause upwelling in the Finnish nearshore and downwelling
at the Estonian coast. Easterly winds lead to a reversed pattern of these phenomena.
At the Finnish coast of the gulf upwelling occurs as frequently as 15–30 % of
the time (Lehmann et al. 2012). A wind event with a duration of about 60 hours or
a wind impulse of the order of 4000–9000 kg/(m s) (Haapala 1994) is usually re-
quired for such an event to develop. During summer and autumn, when the sea sur-
face is warm, upwelling results in a local surface temperature drop of up to 10 °C,
between the upwelled water and the surrounding surface water, which is easily de-
tected by infrared satellite measurements (Fig. 6.10). The horizontal gradient may
reach 1 °C/km and the vertical velocity 3 × 10−5 m/s (Myrberg and Andrejev 2003).
The nutrient-rich waters gradually spread offshore via advection and horizontal dif-
fusion in the form of so-called upwelling filaments (Zhurbas et al. 2007), which
are often steered by local topographical features. The major primary effects are ob-
served in a 5–20 km broad coastal zone (sometimes from 5 km and/or up to 40 km)
in the entire Baltic Sea (Lehmann et al. 2002, 2012; Myrberg and Andrejev 2003;
Lehmann and Myrberg 2008; Myrberg et al. 2010b) (Fig. 6.10) and extend to about
100 km alongshore (Gidhagen 1987). Their lifetime spans from several days to sev-
eral weeks.
Fig. 6.10 Upwelling in the Gulf of Finland on September 3, 2002 seen from a satellite
(NOAA/AVHRR figure received by the Finnish Meteorological Institute and processed by the
Finnish Environment Institute SYKE, Leppäranta and Myrberg 2009). The blue colours reflect
cold, upwelled waters. The red colours represent warm surface waters. The scale is given in °C
The intensity of upwelling-driven vertical mixing also depends on the fresh water
flux and long-wave forcing (Laanearu 2003; Laanearu and Lips 2003). The direc-
tional structure of winds over the Gulf of Finland reveals that relatively strong and
long events of winds may blow alternatively from the east and west. A resulting
sequence of upwellings and downwellings may substantially modify the average
properties of the water masses across the gulf (Talpsepp 2008).
The existing (albeit fairly limited) pool of studies of turbulence-related mixing
processes in the Gulf of Finland have demonstrated that horizontal turbulence in
its coastal regions plays a significant role in the transport and spreading of various
substances (e.g., the contaminants discharged from rivers and the runoff of nutrients
from agricultural activities). The parameters of turbulence that govern the intensity
of the exchange between the coastal zone and the offshore are highly variable. Their
characteristic values are still necessary for assessment of the impact of local turbu-
lence and for modelling efforts, and should be determined for the region of interest
from field measurements.
A frequently used parameter in ecosystem modelling (e.g., when aiming at the
optimization of the disposition and arrangement of sewage discharges) is the eddy
diffusivity coefficient. Its mean values in the eastern part of the Gulf of Finland as
well as the rate of turbulent kinetic energy dissipation, the scale of turbulence (the
effective dimensions of eddies), the lifetime of a diffusing substance cloud and its
path length were estimated using current measurements at different distances from
the coast (Ivanov and Mikhailov 1972; Mikhailov 1974). Horizontal turbulence was
relatively restricted by the smaller dimensions of the basin, but the rate of energy
dissipation was greater in this part of the gulf. Rather different results were obtained
based on moored current measurements (Nikolaev and Luvsk 1975).
Alternatively, parameters of horizontal turbulence and the spreading rate of
closely packed drifters in selected parts of the Gulf of Finland were estimated using
the Lagrangian approach, by tracking floating indicators (drifters, buoys, dye spots).
Airborne tracking of floating drifters in the eastern gulf for different ‘phases’ (rise
and fall) of sea level oscillations showed that horizontal turbulence appears to be
less intense in coastal regions than in more open areas (Mikhailov 1981).
A series of experiments using drifters located in the uppermost 1 m layer and
drifting up to 8 days in summer and a few months in ice conditions have been per-
formed in 2007 in the middle of the gulf, the western Estonian archipelago and in
the eastern sector of the Northern Gotland Basin (Kõuts et al. 2010; Verjovkina et al.
2010). Extensive experiments in summer and autumn 2010 with lightweight drifters
in the uppermost 1.5 m thick layer in several domains of the gulf (Soomere et al.
2011b) showed that the spreading rate of closely packed drifters reveals two dis-
tinct regimes depending on the distance and thus may require the use of different
coefficients of eddy diffusivity for applications with different horizontal resolution
(Chap. 8).
Recently new data have also been gained about vertical turbulence in the Gulf
of Finland, about which only very little information was available in the past (Ale-
nius et al. 1998). Microstructure measurements were performed in three different
wind forcing regimes and three different background density stratification and cur-
rent velocity shear situations at the entrance of the gulf (Lilover and Stips 2011).
It was remarkable that all three standard parameterizations of the eddy diffusivity
(using Richardson number, Brunt–Väisälä frequency and k–ε turbulence closure)
gave biased results in comparison to the measurements. A new parameterization is
thus necessary which would be able to account for the internal wave kinetic energy
of the super-inertial frequency band (Lilover and Stips 2011).
6.3 Meteorological Forcing
6.3.1 Wind Forcing
Meteorological forcing, the principal factor causing water movements, is extremely

complicated in the Gulf of Finland and not fully understood yet. The sparse ob-
servational network and the very limited amount of open sea data make the esti-
mations of wind patterns very difficult. The gulf is characterized by a remarkable
wind anisotropy (Launiainen and Saarinen 1982; Soomere and Keevallik 2003; Kee-
vallik and Soomere 2010). Complicated patterns of atmospheric temperature (that
develop due to the very variable surface roughness) modify local wind fields and
correspondingly air–sea interaction in and around the gulf, giving rise to the spatio-
temporal variability of the momentum and heat exchange between the sea and the
atmosphere. The latitudes where the gulf is located experience a great variability in
the overall meteorological conditions and solar radiation throughout the year and
between different years, which in turn affects the circulation dynamics and strat-
ification. It is thus not unexpected that the modelling of atmospheric forcing and
air–sea interaction in the Gulf of Finland area is a major challenge.
The winds in the region surrounding the Gulf of Finland are governed by the
large-scale circulation patterns over the whole Baltic Sea. This domain belongs to
the zone of westerlies (where typically a strong zonal circulation type exists with
cyclones moving to the east) controlled by the NAO3 index. A positive value of the
NAO index (as for most of the winters in the recent past) favours the predominance
of westerly winds. During winters characterized by negative values of the NAO
index (like 2010, 2011 and 2012) the winds blow mostly from northerly and easterly
directions. In the summertime winds are weaker and the role of local features (like
the land-sea breeze) is larger.
The wind regime of the open part of the gulf reflects a superposition of the
most frequent south-western and north-north-western winds (predominant in the
entire Baltic Sea basin) and more local but also relatively frequent western and
eastern winds (east-north-eastern winds on the southern coast) blowing along the
gulf. On the northern coast, moderate (6–10 m/s) and strong winds (>10 m/s) come
mostly from south or south-west, but on the southern coast they blow more often
from south-west or west. South-eastern winds are infrequent and relatively weak
(Soomere and Keevallik 2003). The average wind speed has been reported to ex-
hibit opposite trends on the opposite coasts (Keevallik and Soomere 2004), proba-
bly because of long-term changes in the land surface roughness in the vicinity of
some measurement sites. It may also be a reflection of the considerable changes in
the overall air flow patterns (Keevallik and Rajasalu 2001; Keevallik and Soomere
2008) (Fig. 6.11).
Measurements from vessels and at lighthouses reveal that wind speeds on the
open sea are considerable greater than at or near coasts (Niros et al. 2002). The
probability of strong storms (wind speeds >20 m/s) was dramatically greater in
the 1990s than previously. Wind speeds exceeded 10 m/s with a probability of 24 %
(13.6 % for the period 1977–1982) and 20 m/s with a probability of 0.2 % (0.008 %).
The mean wind speed during the most severe, long-lasting storms in the Gulf of
Finland is by 2–3 m/s less than in the open part of the Baltic Sea. Southern or south-
western winds are the strongest: the 3-hour mean wind speed may reach 24–25 m/s
once in a century. Eastern winds are confined to a narrow direction span and may
reach 23 m/s (Soomere and Keevallik 2003). Further estimates of extreme wind
properties in the eastern part of the gulf are presented in Lopatukhin et al. (2006a,
2006b).
The local winds have a relatively large role in the summertime when the large-
scale winds are relatively weak. The frequent, relatively strong afternoon surface
winds observed along the northern coast of the gulf in typical summer conditions
may reflect the interplay of the basic flow, sea breeze and the geometry of the gulf
(Savijärvi et al. 2005). The meanders of the sea breeze, amplified by the unidi-
rectional basic flow, may become evident as relatively strong winds, at times lo-
cated just above the opposite coast—the Finnish coast—for the predominant south-
western winds.
3 The North Atlantic oscillation (NAO) is a climatic phenomenon in the northern Atlantic Ocean.
It is characterized by the NAO index, which is the difference of atmospheric pressure at sea level
between the Icelandic low and the Azores high.
Fig. 6.11 Long-term mean wind speed (m/s, direction shown by arrows) in the Baltic Sea area
in January (upper left panel), April (upper right), July (lower left), October (lower right). From
Leppäranta and Myrberg (2009); redrawn from Mietus (1998)
6.3.2 Ice Conditions
The Gulf of Finland is one of the three Baltic Sea basins—in addition to the Bay of
Bothnia and the Gulf of Riga—where significant amounts of ice form every winter.
On average, sea ice is present in this gulf for five months each winter, usually from
December to April. The average freezing date is 1 December in the Neva Bay, and
the last drift ice floes, observed off Vyborg, have typically melted by 1 May (SMHI
and FIMR 1982; Feistel et al. 2008). The range of the freezing and break-up dates
is 15 November–15 January and 15 April–15 May, respectively (Fig. 6.12). The ice
season is more severe towards the east and the north of the gulf. Normally, the gulf
becomes fully ice covered, but in mild winters only the part east of the Narva–Kotka
Fig. 6.12 Extension of ice

cover in the Gulf of Finland
in differently severe ice
winters. Open water is found
to the west of the ice edge
curve (Seinä and Palosuo
1993). Graphics by M. Viška
line freezes. The freezing mostly occurs during January and the ice melts in April,
giving an average length of three months for the extensive ice coverage. The ice
reaches its maximum extent in February or March. The maximum annual thickness
of coastal fast ice varies from 30 to 80 cm near Vyborg and in the Neva Bay.
The ice conditions are not evenly distributed in the different parts of the Gulf
of Finland. The heat inflow due to the coastal current from the Northern Gotland
Basin and the predominance of southern and south-western winds keep the Estonian
coastal area free of ice some time. The north–south asymmetry of the ice conditions
is enhanced by the coastal morphology, which supports a broad, landfast ice zone
along the coast of Finland but leaves almost no fast ice at the Estonian coast. The
width of the landfast ice zone in coastal areas, where the ice is stationary for most
of the ice season, depends on the bottom topography: islands and grounded sea ice
ridges serve as support points to stabilize the ice sheet. The edge of the landfast ice
in the gulf is in the neighbourhood of the 10 m isobath (Leppäranta 1981).
Sea ice has a remarkable influence on virtually all aspects of the physical, biolog-
ical and chemical oceanography of the Gulf of Finland. Ice formation and melting
influence the stratification. Particularly important is the freshening of the surface
layer in spring. Especially thick ice substantially reduces the transfer of momentum
from the wind to the water body. The circulation is forced only by boundary fluxes
at the ice edge (Leppäranta 2010) and possibly by the ice drift. Under ice cover,
circulation in the subsurface layer (Andrejev et al. 2004a) may thus extend over the
entire upper layer (Soomere and Quak 2007).
Even more dramatic is the non-linear dynamic behaviour of wind- and current-
driven drift ice outside the landfast ice zone, with a highly non-linear mechanical be-
haviour. In open ice fields the floes drift independently and, like any floating drifters
on the sea surface, with a velocity close to that of the surface current. The theoreti-
cal free-drift speed of ice fields is 2 % of the wind speed, and drift direction 30° to
the right of the wind direction (Leppäranta 2010). Wind is a purely external force,
but the dynamics of ice and water are coupled. The interfacial stress depends first
of all on the ice-water velocity difference. Sometimes the drift can be as large as
20–30 km/day. There is no general simple rule for compact drift ice. The mobility
of a compact ice cover has been evaluated from experimental data. The ice with a
thickness <10 cm offers little resistance to wind and can be easily pushed to the
eastern end of the Gulf of Finland. At the other extreme, a compact ice field may
stand stationary even under considerable forcing. In harsh winters the entire gulf
can be covered with landfast ice. An ice thickness of 50 cm is usually strong enough
to resist wind forcing, except when the wind is just pushing ice out of the gulf into
the Northern Gotland Basin.
Horizontal turbulence under the ice cover was studied in Smelov et al. (1985)
using radioactive indicators in the northern part of the Neva Bay and in the sea
area to the east of Kotlin. The measurements covered a fairly small horizontal scale
range (up to 100 m) and showed the dependence of the eddy diffusivity coefficient
on the scale of the turbulence and on ice conditions. The influence of ice led to an
amplification of the hydraulic resistance reducing the turbulent diffusivity. It also
induced additional disturbances in the water adjacent to the lower surface of the ice.
6.3.3 Energy Budget
The decoupling of the dynamics of the upper layer from the lower layer has impli-
cations also on the energy budget of the water masses. In the Gulf of Finland, like
in the entire Baltic Sea, the heat budget of the upper layer is driven mainly by solar
and atmospheric forcing. The heating and cooling of the largely decoupled lower
layer mostly occurs owing to advection of heat and water masses from the Gotland
Sea.
The heat content of the Gulf of Finland has an annual amplitude of 46 × 1018 J,
which corresponds to a mean temperature amplitude of 10.2 °C (Jurva 1937). The
main factor is the annual cycle in the solar radiation, which is close to zero in
mid-winter, reaches 200–300 W/m2 (on daily average) and peak values of about
600 W/m2 in summer. The monthly mean solar radiation at Jokioinen (60°49 N,
23°30 E) varies from 10 W/m2 in December–January to 250 W/m2 in June. The
radiation emitted by the sea surface usually exceeds the terrestrial (backward) radi-
ation from the sky), except for very rare situations. The air–sea surface temperature
difference may have either sign. When the two temperatures are equal, the sea sur-
face loses total radiation of about 50 W/m2 . The total (terrestrial + solar) radiation
budget turns positive from sometime in March–April and becomes negative for au-
tumn and winter from September–October.
The sensible heat flux is usually within ±50 W/m2 . The monthly values are typ-
ically a few W/m2 upwards in April, July and August, while a downward flux pre-
vails in May and June. The latent heat flux is mostly negative, with the monthly
mean values about 25 W/m2 or even more and generally directed upwards from
April to August. Therefore, the turbulent fluxes together usually do not exceed the
terrestrial radiation losses and can predominate only when the temperature or water
vapour pressure difference between air and sea is large and the wind speed high
(e.g., in late autumn). Heat transfer from precipitation takes place as sensible heat
exchange and as phase changes. If the precipitation were solid, the melting would
take heat of 30 W/m2 and cool down the surface layer.
The net heat flux through the sea surface is on the order of 100 W/m2 , being
positive in summer and negative in winter (see also Meier and Döscher 2002). This
value corresponds to heating or cooling at a rate of about 0.04 °C/day for the whole
upper layer of a thickness of ∼50 m or 0.2 °C/day for a ∼10 m thick surface layer.
The presence of ice in every winter considerably affects the air–sea fluxes in the
gulf as it radically reduces the terrestrial radiation and turbulent losses. Snowfall
also lowers the surface temperature and weakens the turbulent transfer.
Horizontal advection and diffusive heat transfer can produce an impact to the
upper layer comparable with the atmospheric heating. For typical values of the hori-
zontal net transport speed of 5 cm/s (Soomere et al. 2011a) and temperature gradient
of ∼0.01 °C/km, the advective change will be ∼0.04 °C/day (Soomere et al. 2008).
Assuming the horizontal diffusivity coefficient to be ∼106 m2 /s and temperature
variations of ∼0.01 °C/km, the diffusive smoothing rate will be ∼0.01 °C/day.
6.4 Circulation Dynamics
6.4.1 Scaling of the Equations of Motions and the Rossby Radius
The equations governing large-scale and mesoscale ocean dynamics have been in-
troduced and discussed in Chaps. 2–4. Traditionally, the x-axis is directed to the
east, the y-axis to the north and the z-axis upwards. The molecular viscosity is
generally neglected in the analysis of such motions and only the horizontal part
of the Coriolis acceleration is included into the equations. As the intensity of hor-
izontal currents (about 10 cm/s) exceeds the typical values of vertical velocities
(0.1 mm/s) by several orders of magnitude, the equation for the vertical velocity
w can be simplified using the hydrostatic approximation. The radical difference in
the magnitudes of the vertical and horizontal velocity U = (u, v) is traditionally
expressed by introducing different values for the representative horizontal AH and
vertical Av turbulent (eddy) viscosities with AH ∼ 105 m2 /s and Av ∼ 0.05 m2 /s.
The resulting equations of motions are traditionally written in the following form
(Cushman-Roisin and Beckers 2011):

∂U ∂U 1 ∂ ∂U
+ U · ∇H U + w + f k × U = − ∇ H p + AH ∇ H 2
U+ Av ,
∂t ∂z ρ ∂z ∂z
(6.1)
∂w ∂p
+ ∇H · U = 0, = −ρg, (6.2)
∂z ∂z
where ∇ H is the horizontal gradient operator, f = 2Ω sin φ is the Coriolis param-
eter at latitude φ, Ω = 0.7292 × 10−4 1/s is the Earth’s rotation rate, k is the unit
vector directed upwards, ρ is the density of sea water, p is pressure, g is the accel-
eration due to gravity, and AH and Av are the horizontal and vertical eddy viscosity
coefficients, respectively. Equations (6.1), (6.2) are purely dynamic and should be
amended by the equation of state for sea water and equations for the budget of heat
and salt as demonstrated in Chap. 4.
The magnitudes of different terms in these equations can be radically different

for different classes of motions as described in Chaps. 2–4. There is an essentially
never-ending discussion about which choice of scaling at best represents the mo-
tions in a particular basin. A sensible choice obviously depends on the guesses of
the scientist but is still to some extent predefined by the class of motions under con-
sideration. The task of calculating the statistics of transport of pollutants in surface
currents intrinsically requires accounting for the scales from the basin dimensions
down to, at least, typical parameters of mesoscale motions. Then a few parameters
such as the typical horizontal extension L = 50 km, vertical extension (basically
equivalent to the mean depth here) H = 25 m, the basin-scale inclination of the
sea surface β ∼ 1 mm/(1 km) and the horizontal pressure gradient in the surface
layer ρ −1 ∇p = −gβ ∼ 10−5 m/s2 are externally defined and mostly coincide with
similar values for the open Baltic Sea (Chap. 2).
The limited north–south extension of the Gulf of Finland favours using the con-
stant value of the Coriolis parameter f = 1.26 × 10−4 1/s at φ = 60° N and a local
plane projection, which rotates around the local vertical axis (the so-called f -plane
approximation). The beta-plane approximation, in which the Coriolis parameter
varies in the north–south direction, is used infrequently in applications involving the
dynamics of the gulf. The typical scales of horizontal velocity U = |U| = 10 cm/s
and (eddy turnover) time T = 5 days are more site-specific although they coincide
here with those for the entire Baltic Sea. Similarly, representative turbulent (eddy)
viscosities AH = 105 m2 /s and Av = 0.05 m2 /s apparently are the same in the Gulf
of Finland and for the entire Baltic Sea.
The corresponding characteristic magnitudes of the terms of Eq. (6.1) of horizon-
tal motions also coincide with their typical values for the whole Baltic Sea (in units
of 10−6 m/s2 , see Table 2.3 in Chap. 2): inertia U/T ∼ 0.2, advection U 2 /L ∼ 2,
Coriolis term f U and pressure gradient ρ −1 H p/L both ∼10, and internal fric-
tion terms AH U/L2 ∼ 10 and Av U/H 2 ∼ 4. As the characteristic value of the term
representing vertical velocity is also U 2 /L, all the advection terms have an equal
weight. As for the entire Baltic Sea, they are by almost one order of magnitude
smaller than the dominating terms (the Coriolis acceleration, pressure gradient and
vertical friction) for the motions with a chosen scale. Note that the friction term in-
cludes the transfer of wind stress to the sea surface and the damping of motions by
bottom friction. Differently from the deep ocean, the pressure gradient is not neces-
sarily large in the Gulf of Finland and in many cases the Coriolis acceleration and
vertical friction are approximately balanced.
Although in this greatly simplified scale analysis the role of advection and hor-
izontal friction terms is smaller than that of the Coriolis term, advection does play
an important role in intensive mesoscale dynamics (when U becomes large for rel-
atively small values of L). Horizontal friction becomes important near the coasts
where L decreases. The inertia term is significant in relatively fast processes with
a time scale T of some hours and in so-called inertial oscillations (see Sect. 2.3.4
in Chap. 2). The boundary conditions are rather simple for the Gulf of Finland. The
gulf is largely surrounded by mainland (which is a passive solid boundary) except
for point sources of fresh water in the river mouths and a dynamic inflow-outflow
system at the open boundary towards to the Northern Gotland Basin.
Fig. 6.13 The baroclinic (internal) Rossby radius of deformation (km) in the Gulf of Finland
(Alenius et al. 2003). Reproduced from Soomere et al. (2008)
The extensive variation in the stratification of the Gulf of Finland substantially

affects several scaling factors that govern the character of circulation patterns. The
key ‘measure’ of the large-scale and mesoscale dynamics of the sea is the internal
(baroclinic) Rossby radius of deformation R1 . In essence, it is the ratio of the phase
speed of internal waves to the Coriolis parameter (see Sect. 2.3.2 in Chap. 2 for a
more detailed discussion). Although it is not present explicitly in the above scale
analysis, it implicitly defines the typical scale of mesoscale features such as eddies,
fronts and local jets.
Owing to strong density gradients, R1 is very small, usually 2–4 km and exhibits
a large spatial and temporal variability in the Gulf of Finland (Alenius et al. 2003)
(Fig. 6.13). The highest values occur in summer in the deep parts of the gulf. In the
shallow coastal regions and in the eastern part of the Gulf of Finland, R1 is even
smaller, down to about 0.5 km (Nekrasov and Lebedeva 2002). Its small values
compared to those in the open Baltic (about 3–10 km, Fennel et al. 1991) require a
very high horizontal resolution of both measurement campaigns and eddy-resolving
numerical models in the gulf. To properly describe the mesoscale eddies, fronts and
jets, the grid size of circulation models should not exceed 1/2–1/3 of R1 (Drijfhout
1989; Lindow 1997).
6.4.2 In Search of the Mean Residual Circulation
An analysis of circulation dynamics in the Gulf of Finland should be started with the
question: What is the physical forcing that causes the currents? Their main forcing
factor in the gulf is definitely the wind stress. It is highly anisotropic as the main
wind direction is generally from the south-west (Mietus 1998). The pronounced
horizontal buoyancy (density) gradients caused by variations of temperature and
salinity play an important role in the overall circulation. The slope of the sea surface
that results from the voluminous water supply to the eastern part of the gulf also
contributes appreciably to the existing circulation pattern (Alenius et al. 1998). The
Fig. 6.14 Average mean

surface circulation in the Gulf
of Finland in (A) June,
(B) August, (C) October,
(D) June–October (Witting
1910). The numbers in the
figure show the persistency of
the current direction (%) at
different locations. From
Alenius et al. (1998)
above scale analysis confirms that the gulf is large enough to feel the effects of the
Earth’s rotation on the circulation (Hela 1946). Currents excited by the listed forc-
ing and steering factors are further modified by the hydrographic and topographic
steering.
The basic appearance of the general circulation pattern has already been known
for 100 years. There are still substantial gaps in our perception of the detailed struc-
ture of the circulation, especially of the mesoscale dynamics. The classical stud-
ies (Witting 1912; Palmén 1930) elucidated both the resultant cyclonic (counter-
clockwise in the northern hemisphere) surface current of the Gulf of Finland and
the remarkable horizontal variability of the entire field of currents. The residual
speed was estimated to be about 1–3 cm/s. The long-term persistency (defined as
the ratio of the vector and scalar mean speed) was between 6 % and 26 % (Palmén
1930), whereas it was much larger during different seasons (Fig. 6.14).
Although those early estimates gave a fair understanding of the existing circula-
tion, the view was based on sparse observations, to some extent oversimplified and
did not account for mesoscale features which were discovered only in the 1970s.
Even though there is a large continuous fresh water input from the rivers in the east-
ern regions of the gulf, the mean velocity of the resulting buoyancy-driven current
is almost by an order of magnitude smaller than the average speed near the Finnish
coast.
The relatively poor long-term persistency of the currents reflects the importance
of the variability in the wind forcing and also indicates that the mean circulation
is a statistical property rather than a constant feature. This is often forgotten and
the cyclonic pattern inappropriately treated as a permanent circulation type like the
local ‘Gulf Stream’. In fact, the instantaneous currents almost totally mask the long-
term mean circulation. Numerical simulations suggest that even an anticyclonic gyre
may exist in the eastern part of the gulf in selected years (Soomere et al. 2011a).
The persistency of the currents is smaller along the northern nearshore than along
the southern coast (Andrejev et al. 2004a), apparently owing to the impact of local
topographic and hydrographic features. Periodic processes like inertial oscillations
(see Sect. 2.3.4 in Chap. 2) and seiches of the gulf and of the combined Gulf of
Finland–Northern Gotland Basin system additionally complicate the current system.
Although many studies have been devoted to the problem of determining the
mean circulation in the Gulf of Finland, the question is still partially open. The early
analyses (Palmén 1930; Hela 1952) were based on observations from the lightships
of Helsinki and Tallinn. They represent the very surface layer and it is necessary
to eliminate the local drift currents in order to extract the residual (more or less
stationary) flow. The drift current was defined as the flow in a stationary wind and
current system (which very seldom exist). The dependence of the resulting residual
flow on the wind (Hela 1952) indicates that the interactions between different forc-
ing and steering factors are very complex and it is difficult to give an unambiguous
interpretation of the results. Another way to roughly estimate the residual flow is to
calculate the resultant of the currents observed during calm weather (Hela 1952).
The sum of the residual flow and an additional (Ekman) current deflected to the
right of the wind was denoted as the characteristic current in Hela (1952). It was
very slow, 0.6–2.0 cm/s, off Helsinki and somewhat faster, 1.7–4.9 cm/s, off Tallinn.
The best match between the total surface current vc and the wind speed Ww was
vc = 0.0137Ww . Thus, as a rule of thumb, the characteristic surface current has a
speed of about 1.4 % of the prevailing wind speed and is deflected by 19° to the right
of the wind direction. This deflection angle is naturally not valid near coasts. These
historical results are still of importance for understanding the circulation dynamics
of the Gulf of Finland and have a direct application for the forecast and hindcast of
pollution and contaminants in the surface layer during a large part of the year. Their
outcome is equivalent to that of the studies of net current-driven transport over time
scales larger than the synoptic scale (Soomere et al. 2011a).
The question of a residual flow has been further addressed using a linear regres-
sion between the easterly and northerly wind components and the easterly compo-
nent of the current at a depth of 5 m off Kotka in the northern Gulf of Finland based
on 25 days of observations (Sarkkula 1989, 1991; Pitkänen et al. 1993). The pro-
portionality coefficient of 4.8 cm/s (r 2 = 0.69) was interpreted as the speed of the
residual flow driven by the joint effect of horizontal salinity gradient and bottom
relief (Sarkisyan et al. 1975).
The obtained speed was somewhat larger than in the above estimates for the ice-
free period. Note that Palmén (1930) used 5 years of observations in which seasonal
differences tend to cancel each other while the more recent results mostly reflect a
single wind event. Another source of the difference stems from the short-term basin-
wide dynamics of the gulf. The predominant south-western winds pile up the water
at the eastern end of the gulf. If the wind stress vanishes, the westward barotropic
return flow towards an equilibrium stage may simply override the baroclinic effects
and circulation caused by the river water runoff. The linear regression tends to over-
estimate the residual flow because the wind stress is seldom zero for a long enough
time for the equilibrium stage to be reached. Despite several debatable issues, this
was the first time when the background current was actually identified.
Importantly, the measurements were carried out at a depth of 5 m, which is quite
substantial in the light of the existing vertical structure of water masses in the Gulf
of Finland. The difference between the classical results for the surface flow and
for the currents at a depth of 5 m may be interpreted as an important evidence of
the layered structure of velocity fields in the gulf. Simulations of Andrejev et al.
(2004a) suggest that the uppermost layer with a thickness of 2–3 m may be almost
completely decoupled from the subsurface layer in terms of instantaneous velocities,
persistency and the overall structure of the flow. Although it is very likely that sea
currents are highly variable over the first meters of depth at least for some seasons
(Suursaar 2010), the details of the forecast or hindcast layered structure of currents
still cannot be assumed as true until confirmed by massive in situ experiments. Rapid
development of profiling velocity meters will eventually fill this major gap in our
understanding very soon.
The long-term mean flow towards the west was indirectly determined from the
time difference between the annual minimum of the surface salinity near Utö and the
maximal river runoff (Launiainen 1982). The time lag of 2.5 months implies a mean
flow of a couple of cm/s (Mälkki and Tamsalu 1985). Even if this estimate does
not represent a true assessment of the circulation, it still gives a clue of the scale of
the motion and matches the typical east-west net transport speed in relatively coarse
numerical models (Soomere et al. 2011a).
The separation of the residual circulation from the observed current field is some-
what artificial (and virtually meaningless in contemporary high-resolution simula-
tions of circulation). Its importance mainly becomes evident in long-term estimates
of the drift of substances, but even there the interpretation is not trivial. Leppäranta
and Peltola (1986) compared cases with and without a permanent flow derived from
Palmén (1930) in a random transport model. The probability distribution of the
transport after a one-year period from a starting point at Helsinki had a probability
maximum in the west but the probability density was non-zero in the east. There-
fore, eastward net transport does occur in some cases even if the direct impact of
mesoscale features is filtered out. The actual patterns of currents are evidently much
more complex. The wind- and density-driven currents are coupled in a strongly non-
linear manner that becomes evident for example via the formation of eddies.
One of the key questions of the functioning of the Gulf of Finland is how the wa-
ter exchange occurs between the gulf and the Northern Gotland Basin, with no sill
in between. A proper answer to this question is necessary, for example, to estimate
how long the water remains in the gulf. Early estimates of the exchange volumes and
time scale (Witting 1912; Palmén 1930) were based on rather sparse field measure-
ments, from which short-term variability has been filtered out. They should thus be
treated as ‘educated guesses’ rather than accurate estimates. Straightforward budget
estimates using the classical Knudsen’s formula (Knudsen 1900) resulted in annual
water in- and outflows of 480 and 600 km3 /yr, respectively (Witting 1912). The cor-
responding mean current speed ∼1 cm/s is clearly underestimated compared with
more recent observations (Sarkkula 1991; Mikhailov and Tshernyshova 1997). De-
spite a significant discrepancy, it is remarkable that the excess of water (mainly input
of river water to the Gulf of Finland), around 115 km3 /yr (Bergström and Carlsson
1994) was recovered correctly.
6.4.3 Complexity of Observed Motion Patterns
Detailed investigations of the spatial and temporal variability of the currents during
the last decades have considerably improved our understanding of the circulation in
the Gulf of Finland (Alenius et al. 1998). Although there is a widely spread opinion
that tidal flows and especially seiches (Jönsson et al. 2008) are relatively weak in
the entire gulf, remarkable seiche-driven currents with periods of 31, 24, 19.5, 16
and 11 hours have been identified in the gulf from in situ measurements near the
island of Naissaar (Lilover et al. 2011). The longest seiche periods correspond, al-
beit not perfectly, to the well-known Baltic Sea–Gulf of Finland oscillations, with a
theoretically estimated period in the range of 27.5–31 hours (Neumann 1941; Wüb-
ber and Krauss 1979). The correlation between winds and corresponding currents
varied between 0.69 and 0.9 as in the earlier estimates (Hela 1952; Sarkkula 1991).
The current was rotated 35° to the right from the wind (Lilover et al. 2011), that is,
slightly more than in the classical estimates (Alenius et al. 1998; Elken et al. 2011).
The average speed of tidal flows was close to 10 cm/s on certain occasions (Lilover
2012).
Measurements using moored current meters (Talpsepp 1986; Talpsepp et al.
1994) revealed that current speeds were generally higher near the Estonian coast
than on the Finnish side. Strong currents (about 20–30 cm/s) in the Estonian
nearshore can sometimes even be called coastal jets. The long-term mean current
speed off Helsinki was only about 10 cm/s. The current variability was reported to
be conspicuous for its 18–20 day periods (Talpsepp et al. 1994). Unidirectional cur-
rents prevailed over several days and then changed direction (Laakkonen et al. 1981;
Talpsepp et al. 1994).
Mean surface current speeds were low in summer (3–5 cm/s) and in autumn
(4–6 cm/s) off the Hanko peninsula (in the Tvärminne area) (Laakkonen et al. 1981;
Haapala et al. 1990) and near Kotka (Alenius 1986). The mean current direction was
westwards at the northern coast, supporting the cyclonic circulation idea. However,
strong westerly winds turned the currents to the east near the Finnish coast even
for several weeks at a time. Somewhat counter-intuitively, the largest instantaneous
flow speeds were recorded in summer, apparently because the energy of wind-driven
motions is concentrated in the relatively thin upper mixed layer above the thermo-
cline. Eddy-like short-lived structures that affect the circulation in the near-bottom
layer were observed in a number of studies (Mälkki and Talpsepp 1988).
Measurements carried out by the Finnish Institute of Marine Research in summer
and autumn 1994–1995 (unpublished data) at five stations across the entrance region
of the Gulf of Finland showed that the flow direction varied across the gulf as ex-
pected for a cyclonic circulation. The typical surface-layer velocity was 10–20 cm/s.
Simultaneous Acoustic Doppler Current Profiler (ADCP) data (Stipa 2004) revealed
more clearly the regions of oppositely flowing currents separated by about 10 km.
Because the flow remained quite steady for several days, the ADCP data across the
gulf can be considered as representative of that situation.
On the Finnish side the mean velocity was 5–9 cm/s westwards during the sum-
mer and the persistency of the current was high (near to or above 50 %). In the
Fig. 6.15 The

semi-permanent salinity (‰)
front at the entrance of the
Gulf of Finland according to
Kononen et al. (1996). From
Leppäranta and Myrberg
(2009)
middle of the gulf the current direction was highly variable. Near the Estonian coast
the current mainly flowed to the north-east. In the autumn there was also a long pe-
riod of inflow along the Finnish coast. This might be interpreted as a reflection of a
semi-persistent anticyclonic circulation but it could equally well be a fingerprint of
transport of water masses pushed into the gulf by stronger south-western winds (that
are associated with the export of saltier water at larger depths, Elken et al. 2003).
There have been very few in situ observations of mesoscale eddies in the Gulf
of Finland. One example, with a diameter of 15–20 km (about 4 times the internal
Rossby radius) was formed at the entrance of the gulf during a rapid splitting of
the eastward downwelling jet into an offshore cyclonic and an onshore anticyclonic
branch (Pavelson 2005). The centre of the eddy was within the seasonal thermocline
at 30–35 m depth. The maximum velocities in its core reached 0.35 m/s. Its vertical
structure was typical of that of a geostrophically balanced eddy (Pavelson 2005).
The structure of local currents in deep semi-enclosed bays at the southern coast
of the gulf frequently mimics the multi-layered flow of the entire gulf. For example,
the general anticyclonic flow in the upper layer and the cyclonic flow in the lower
layer, to some extent steered by the local topography, were established in Muuga
Bay (Raudsepp 1998). A periodic variability of the velocity of the current between 5
and 25 cm/s was interpreted in terms of travelling coastally trapped waves (Talpsepp
2006). Although only a quarter of the eastward travelling wave period of ∼40 days
was covered by the measurements, the wave supposedly reversed the cross-shore
flow direction (Raudsepp 1998).
The diversity of water masses in the Gulf of Finland and the richness in fronts be-
tween them follow from its estuarine character and are enhanced by the voluminous
river runoff, the vigorous modulation of currents by bathymetry, and the frequently
occurring upwelling events. Fronts can be found all over the entire water body, but
those occurring in the uppermost layer reveal important coupling with biogeochem-
ical processes (Pavelson 2005) and can be easily tracked and quantified by remote
sensing.
Probably the best known example is the quasi-permanent salinity front at the
entrance of the gulf (Fig. 6.15). It is typically oriented in the south-western–north-
eastern direction and is positioned approximately parallel to the isobaths (Kononen
et al. 1996; Pavelson et al. 1997; Laanemets et al. 1997; Pavelson 2005). It is cre-
ated by the interplay of waters of different salinity and/or temperature flowing in
and out the gulf. The more saline water of the Northern Gotland Basin enters the
Fig. 6.16 Surface currents

observed in the eastern Gulf
of Finland (to the east of
29° E), averaged over periods
from two weeks to six months
and illustrated via the major
and minor axes of the
scattering ellipses and the
mean vector (Belyshev and
Preobrazhensky 1988). From
Alenius et al. (1998)
gulf along the Estonian coast, and fresher water flows out along the opposite coast.
With easterly winds, the denser (more saline) water mass moves offshore. The front
becomes sharp and is strongly inclined to the sea surface. When winds are from
the west, the less dense water of the gulf forms a surface sublayer, giving rise to a
secondary pycnocline approximately in the middle of the upper layer. The changes
in the potential energy and stratification conditions in the upper layer are coupled
with differential advection induced by along-front wind stress and wind-generated
vertical mixing (Pavelson et al. 1997).
The presented material demonstrates that the system of currents in the Gulf of
Finland hosts motions with a large range of spatial and temporal scales, from basin-
wide circulation down to local synoptic vortices. In particular, the broad band of
mesoscale processes encompasses phenomena with dimensions from ∼1 km to tens
of kilometres. This diversity of processes and scales, partially caused by small val-
ues of the Rossby radius in some regions of the gulf (Nekrasov and Lebedeva 2002;
Alenius et al. 2003) requires considerable efforts and extremely high-resolution
models, down to 0.25 nautical miles (Andrejev et al. 2010) to numerically resolve
their dynamics.
The eastern end of the Gulf of Finland is dynamically similar to an estuarine
domain. Its current system is largely decoupled from the circulation in the rest of
the gulf and reveals extremely high variability, with pulsating components usually
exceeding the mean (residual) values. This is clearly seen in the scattering ellipses
of the observed currents (Fig. 6.16). Though the mean flow is basically to the west
with a velocity reaching 10 cm/s, the direction is highly variable. It is unlikely that
any persistent ‘streams’ exist in this part of the gulf, except possibly in the Neva Bay
and in the vicinity of the Saint Petersburg flood barrier where the influence of the
River Neva is most clearly manifested. Still the movements more resemble a system
of alternating jets as the major axis of the scattering ellipses is 3–5 times as long as
the minor one.
6.4.4 Advanced Numerical Modelling of Mesoscale Dynamics
Although the historical studies of the Gulf of Finland dynamics by Witting and
Palmén (Witting 1910, 1912; Palmén 1930) correctly captured several of its funda-
Fig. 6.17 Mean circulation in the subsurface layer of the Gulf of Finland (2.5–7.5 m). Arrows
show the flow direction and the colour bar denotes current velocity (cm/s) (Andrejev et al. 2004a)
mental properties, many fascinating details have been added to this picture through
contemporary high-resolution modelling tools. First of all they have revealed a man-
ifold of mesoscale features as expected for flows with a small internal Rossby radius
(Andrejev et al. 2004a). Both the mean and instantaneous circulation patterns con-
tain numerous relatively persistent eddies with a typical size exceeding the internal
Rossby radius and probably steered to some extent by the complex bathymetry. In
the eastern part of the central gulf vortices are relatively small.
The cyclonic mean circulation (Fig. 6.17) is generally discernible in the Gulf
of Finland but the resulting patterns and the persistency of the currents deviate to
some extent from the classical analyses (Andrejev et al. 2004a, 2004b). The mod-
elled circulation patterns reveal certain nontrivial and temporally and spatially vary-
ing features. The Neva Bay and the easternmost narrow part of the Gulf of Fin-
land host strong and persistent, mostly west to north-west directed but still strongly
meandering currents evidently supported by the voluminous runoff from the River
Neva. These currents are accounted for in the commercial ship routing software (see
Chap. 11).
The uppermost 2.5 m thick layer is characterized mainly by an Ekman-type drift.
A quite persistent (up to 50 %) inflow near the southern coast is visible at all depths
matching the data in Stipa (2004). Its typical velocities are 1–4 cm/s, whereas the
most intense flow (7–10 cm/s) occurs near the surface (Fig. 6.18). A compensat-
ing outflow exists in the rest of the gulf. It is highly persistent (up to 80 %) in the
subsurface layer (2.5–7.5 m) and below it slightly north of the axis of the gulf but
weakens considerably near the bottom. Even if the model resolution was not ex-
tremely high, the results suggest that the uppermost layer and the layer just below it
(2.5–7.5 m) may have clearly different dynamics. This feature may have substantial
implications for the reconstruction of trajectories of pollutants and floating objects
at the sea surface and definitely calls for the use of very high vertical resolution for
the ocean models underlying the oil spill and pollution transport models.
Fig. 6.18 Simulated

five-year mean of inflow
(positive) and outflow
(negative) speed (cm/s) at the
entrance of the Gulf
(Hanko–Osmussaar line).
Finland is located on the left,
Estonia on the right
(Andrejev et al. 2004a)
Fig. 6.19 Drifter in the Gulf

of Finland. Photo by Maria
Gästgifvars
There have been very few drifter experiments to verify the modelling studies in
the Gulf of Finland prior to those described in Chap. 8. Two relatively short (5–32
hours) experiments were performed in May 2003 using Current Spy surface buoys
(Fig. 6.19) with a drogue extending to a maximum depth of 0.7 m (Gästgifvars
et al. 2006). In moderate wind conditions the buoys moved with a velocity of about
2 % of the wind velocity and with a deviation angle of 0–10° to the right from the
local wind direction. The results match well the earlier observations, where wind-
driven surface currents had 1–3 % of the wind speed (Hela 1952) albeit the veering
angle was much smaller. The behaviour of drifters was quite different in weak wind
conditions when the buoys drifted to some 60° to the left of the wind.
An important implication is that the vertical structure of currents may vary a lot
and that the dynamics of the lower layers may at times override the local wind-
induced drift. Another lesson is that the role of the dynamics of deeper layers for
the drift of various substances in the uppermost layer may be much larger than
expected from standard modelling efforts. Therefore, operational drift models based
solely on the local wind information may need improvement in order to reproduce
accurately the actual hydrodynamic flow field and the corresponding drift in the
Gulf of Finland.
6.4.5 The Intriguing Entrance to the Gulf of Finland
The multitude of processes at the entrance to the Gulf of Finland certainly makes
their modelling extremely difficult. Internal wave activity, a process scarcely re-
flected in recent studies (Kurkina et al. 2011; Lilover and Stips 2011), is apparently
high in this area. The production of eddies (Pavelson 2005), topographically con-
trolled currents and fronts is frequent; diapycnal mixing is therefore intense in the
entire entrance region. The anisotropic wave regime in the Baltic, with frequent
occurrence of very high waves in the north-eastern part of the Northern Gotland
Basin (Schmager et al. 2008), apparently also contributes to the forming of specific
features of the vertical structure of water at the entrance to the gulf.
The previous dynamic consideration of the water cycle in the Gulf of Finland
(Alenius et al. 1998) mirrored the sporadic one-way salt water inflow through the
Danish Straits (e.g., Lass and Matthäus 1996; Elken and Matthäus 2008) into the
deeper layers of the Baltic Sea (Chap. 2). Instead, a two-way exchange and even
an almost total export of saline bottom water from the gulf, accompanied by drastic
variations in the halocline’s position, or by its almost complete disappearance, may
take place at the entrance of the gulf (Elken et al. 2003).
Whilst the classical estuarine dynamics is governed mostly by the balance be-
tween the fresh water flow and the impact of open sea water masses, the strongly
anisotropic wind forcing (Soomere and Keevallik 2003) plays an important role in
the water exchange and in the formation of the vertical structure of the water masses
here. Whereas the standard estuarine circulation is supported by most of the wind
directions, long-lasting, strong south-western winds push a large amount of fresher
surface water back into the gulf. The excess volume of water increases the hydro-
static pressure in the gulf and may lead to a gradual export of the salt wedge in the
bottom layer of the gulf (Elken et al. 2003) if the wind speed exceeds a mean value
of 4–5.5 m/s.
A major consequence of this kind of reversal of the estuarine transport is a weak-
ening of the stratification of water masses at the entrance of the gulf, accompa-
nied by an intensification of vertical mixing (Elken et al. 2006). Its practical con-
sequences for the functioning of the deep-water ecosystem are not fully understood
yet; for example, it may result in unexpectedly effective supply of oxygen to the
deep layers of the Northern Gotland Basin and thus play a great role in the function-
ing of the entire Baltic Sea.
Perhaps even more importantly, this finding has led to a substantial revision of the
traditional concept of mostly decoupled lower layer dynamics and reveals that both
the surface and near-bottom layers respond rather actively to wind forcing. While
the layers are mostly decoupled on local scales as discussed above, the coupling
here occurs on the scale of the entire basin. The basin-scale barotropic flows are
frequently transformed into topographically-driven baroclinic mesoscale motions

characterized by large isopycnal displacements (more than 20 m within a distance
of 10–20 km) and high intra-halocline current speeds (>20 cm/s).
Similarly to the observations, also in models the Finnish side is primarily domi-
nated by offshore outflow down to the depth of 40–50 m and with a mean velocity
of 8 cm/s (Meier 1999; Andrejev et al. 2004a). This outflow can be interpreted as a
persistent buoyancy current stabilized and amplified by certain subtle mechanisms.
Generally, baroclinic instability leads to disintegration of the flow but a relatively
gently sloping bottom at the Finnish coast may cause an internal baroclinic adjust-
ment of the flow to a quasi-stable state (Stipa 2004).
Alternatively, the outflow can be interpreted as an analogue to large-scale nearly
zonal ocean currents where the impact of the beta-effect is mimicked by an equiv-
alent impact of the sloping bottom (Cushman-Roisin and Beckers 2011). Such
baroclinic currents are generally more pronounced than barotropic ones (Soomere
1995). Somewhat counter-intuitively, a comparatively intense transport (speed about
8 cm/s; apparently an Ekman-type drift) goes into the Gulf of Finland on the Finnish
side in the uppermost model layer (0–2.5 m) (Andrejev et al. 2004a). This feature
once more confirms that the surface layer and the layer below may have clearly
different dynamics.
Lehmann and Hinrichsen (2000) were the first to numerically analyse the water
exchange between the Gulf of Finland and the Northern Gotland Basin. The net
outflow was the same, around 130 km3 /yr, as in Witting (1910) but the volumes
of in- and outflow were much higher. The large deviations are caused by the com-
plexity of water dynamics at the entrance of the gulf, which makes some of the
volumes strongly dependent on the temporal resolution of the processes (Andrejev
et al. 2004a). Leaving aside the small-scale dynamics, which is at present accu-
rately reflected neither by measurements nor in numerical simulations, processes at
two basic scales contribute to the water budget.
The estimates of the net in- and outflow in Witting (1910) reflect the amount of
water entering the interior of the gulf, remaining there for a long time, or ultimately
being transported out of it. Relatively persistent or quasi-periodic, mesoscale fea-
tures (local jets, synoptic eddies, inertial oscillations) induce comparatively short-
term transport of water across the entrance line. The water is usually not trans-
ported far from its original location and does not affect the interior of the gulf. These
motions were not accounted for by Witting (1910). When the water exchange was
summed every 30 minutes (accounting for the full dynamics of the mesoscale cir-
culation), the average in- and outflows were 3154 and 3273 km3 /yr for 1987–1992
(Andrejev et al. 2004a). A similar estimate from the average velocities for the 5-year
period (only accounting for the quasi-stationary circulation pattern) gave 1417 and
1532 km3 /yr, respectively. The net outflow (119 or 115 km3 /yr) was close to the
total river runoff to the Gulf of Finland.
Fig. 6.20 Water age in the Gulf of Finland (days) at depths 2.5–7.5 m (Andrejev et al. 2004b)
6.4.6 Water Age
The amounts of in- and outflowing water provide rather limited information on the
water renewal rates and water quality in the Gulf of Finland. The concept of water
age is a helpful tool to study these matters. A widely used definition of the age of a
water particle is the time elapsed since the particle left the sea surface (Deleersnijder
et al. 2001). This definition is relevant in the gulf, where the salt wedge is usually
quasi-stationary and water is supplied either as near-surface currents or through pre-
cipitation, or by river discharge. According to this definition, the oldest water (about
8.3 years)4 is evidently located at the bottom of the gulf (Meier 2005).
Another definition of water age accounts for both horizontal and vertical dis-
placements (Bolin and Rodhe 1973) and characterizes the residence time of water
particles in a selected area while a renewal index (Andrejev et al. 2004b) portrays
the amount of original water in the area at a given time. The local maxima of these
two quantities do not necessarily coincide, however; simultaneous analysis of these
variables provides much better information about the water quality. An estimated
average age of Gulf of Finland water in terms of water exchange with the North
Sea in the entire Baltic Sea water renewal cycle is about 20 years (Meier 2007).
The water age in the gulf with respect to water exchange with the Northern Got-
land Basin is at most only two years. Such small values confirm the very intense
interaction between the gulf and the rest of the Baltic Sea. The water age shows
a pronounced spatial and temporal variability (Andrejev et al. 2004b). Its smallest
values are found in the inflow regions and at river mouths (Fig. 6.20), the highest
ones in the outflow regions. The overall distribution of the water age mimics the
cyclonic mean circulation.
4 The numerical values for the water age substantially depend on the choice of the ‘outer’ basin,
which was the North Sea in Meier (2005) and the Northern Gotland Basin in Andrejev et al.
(2004b).

The mesoscale and basin-scale dynamics of the Gulf of Finland appear very com-
plex, not only due to the pronounced variations in forcing functions and energy
budget but also owing to rapidly changing hydrography and complex bathymetry.
As in the entire Baltic Sea, the four fundamental mechanisms to induce currents in
the gulf are: (i) wind stress, (ii) surface pressure gradient, (iii) thermohaline horizon-
tal gradient of density and (iv) tidal forces. Voluminous river runoffs can produce
considerable local changes in the sea level height and drive significant currents. The
resulting currents are steered by the Coriolis acceleration, topography and friction.
Due to the small depth and size of the gulf, bottom friction exerts far stronger impact
on the currents compared to the deep ocean and even many shelf seas.
The general circulation pattern is typical for a stratified system and mimics the
similar pattern in the entire Baltic Sea. Inflowing waters into a basin are placed at
the depth where the ambient water has an equal density: The fresher water goes into
the upper layer and saltier water masses into a certain lower layer. Owing to strong
stratification the fresher waters leave the gulf near the surface whereas the inflow of
more saline water takes place mostly at larger depths.
The motions in the gulf represent many different time scales. It is customary to
associate the motions averaged over longer time intervals (from several months to
years) with a baroclinic, virtually wind-independent circulation. Its presence is intu-
itively expected due to the positive fresh water budget and resulting large horizontal
gradient of salinity but its identification from measurements and simulations is still
a challenge (Elken et al. 2011). In short time-scales (1–10 days) the currents are
caused by the wind stress. Due to the large variability of the winds, the resulting
long-term wind-driven mean circulation is weak, and transient currents are by one
order of magnitude larger than the average ones.
Drift currents produce in coastal areas upwelling and downwelling features that
may be further affected by Kelvin-type waves. The water is laterally mixed by
mesoscale eddies and deep-water circulation. In the time scale from 1 hour to 1
day, there are several periodic dynamical processes. The most important are inertial
oscillations (about 14 hours) partially overlapping with tidal flows (12–13 or 24–26
hours) and seiches (less than 40 hours). The near-bottom layer below the halocline
is governed by advection and mixing and is virtually decoupled from direct wind
forcing.
A major challenge in the modelling of the vertical structure of currents and espe-
cially the properties of transport driven by surface currents, performed in the follow-
ing chapters, is that the circulation may be substantially different in the very surface
layer and in the one below it. The uppermost layer with a thickness of about 2 m is
typically characterized by Ekman-type flow. The resulting average circulation pat-
tern, at least in certain model simulations (Soomere et al. 2011a) is not necessarily
cyclonic. The layer below, at depths starting from about 3 m is apparently much
more strongly involved in the dynamics of the entire water column and expresses
classical features such as a quasi-permanent cyclonic circulation superposed with
mesoscale eddies. Under certain wind conditions the uppermost layer is eventually
‘entrained’ into the motions in the subsurface layer (Gästgifvars et al. 2006).
An adequate representation of such a complicated multi-layered circulation

needs refined and tailored schemes not only for mesoscale motions but also for
the description of vertical turbulence in the Gulf of Finland. The methods for its pa-
rameterization used in contemporary circulation models of the Baltic Sea are taken
from general ocean models and do not necessarily fit the specific conditions in the
gulf. Several studies have suggested that the improper replication of the vertical tur-
bulence might be a major reason for the mismatch of the model results and measure-
ments in the Baltic (Myrberg et al. 2010a). An improvement of this shortcoming has
a large potential for a more reliable replication of both the dynamical and statistical
features of current-driven transport of various pollutants and contaminants.
Finally, we would like to emphasize that the processes affecting the long-term
mean surface circulation and its modelled or observed outcome in the Gulf of Fin-
land are essentially non-linear and exhibit a number of sophisticated feedbacks. For
example, the wind-independent baroclinic mean circulation and the mean wind-
driven circulation interact with each other in a highly non-linear manner and it is
virtually impossible to specify which one is more important. Moreover, the answer
to this question depends on the case studied and on the time scale under investiga-
tion. The circulation studies set extremely high demands for the modelling work and
measurement campaigns not only because of the small internal Rossby radius in this
basin but, perhaps more importantly, due to the above-mentioned complex dynam-
ics that is not directly evident from simple scaling exercises. The role of the model
resolution becomes clearly evident in the following chapters that are dedicated to
detailed studies of circulation and associated transport and drift properties.
Acknowledgements This study was performed in the framework of the BalticWay project, which
was supported by the Estonian Science Foundation, the Academy of Finland and by funding from
the European Commission’s Seventh Framework Programme (FP7 2007–2013) under grant agree-
ment No. 217246 made with the joint Baltic Sea research and development programme BONUS.
The research was partially supported by the Estonian Science Foundation (grant No. 9125), tar-
geted financing by the Estonian Ministry of Education and Research (grant SF0140007s11), and by
the European Regional Development Fund via support to the Centre of Excellence for Non-linear
Studies CENS. The authors are particularly grateful to Dr. J. Laanemets and to an anonymous
reviewer for highlighting several gaps and misinterpretations in the original version of the chapter.
References
Alenius P (1986) Kotkan merihiekka 1985—virtaustutkimus. Merentutkimuslaitos, Sisäinen ra-
portti 5/1986. [Kotka sea sand 1985—current measurements. Finnish Institute of Marine Re-
search, Internal report 5/1986] 28:1–12 (in Finnish)
Alenius P, Myrberg K, Nekrasov A (1998) Physical oceanography of the Gulf of Finland: a review.
Boreal Environ Res 3:97–125
9:1–16
Belyshev AP, Preobrazhensky LY (1988) Structure of currents in the Neva Bight and eastern Gulf
of Finland based on field measurements. Proc State Hydrol Inst 321:4–16 (in Russian)
Bolin B, Rodhe H (1973) A note on the concepts of age distribution and transit term in natural
reservoirs. Tellus 25:58–62
numerical aspects. Academic Press, San Diego, 828 pp
Deleersnijder E, Campin J-M, Delhez EJM (2001) The concept of age in marine modelling. I. The-
ory and preliminary model results. J Mar Syst 28:229–267
Dera J, Wozniak B (2010) Solar radiation in the Baltic Sea. Oceanologia 52:533–582
Ehlin U (1981) Hydrology of the Baltic Sea. In: Voipio A (ed) The Baltic Sea. Elsevier, Amster-
dam, pp 123–134
Elken J, Matthäus W (2008) Physical system description. In: The BACC author team, assessment
of climate change for the Baltic Sea basin. Springer, Berlin, pp 379–386
Elken J, Mälkki P, Alenius P, Stipa T (2006) Large halocline variations in the northern Baltic
Proper and associated meso- and basin-scale processes. Oceanologia 48(S):91–117
Elken J, Nõmm M, Lagemaa P (2011) Circulation patterns in the Gulf of Finland derived from the
EOF analysis of model results. Boreal Environ Res 16(Suppl A):84–102
Feistel G, Nausch G, Wasmund N (2008) State and evolution of the Baltic Sea, 1952–2005. Wiley,
Hoboken, 703 pp
Res 11:23–36
Gidhagen L (1987) Coastal upwelling in the Baltic Sea—satellite and in situ measurements of
sea-surface temperatures indicating coastal upwelling. Estuar Coast Shelf Sci 24:449–462
Granö O, Roto M (1989) The duration of shore exposure along the emerging Finnish coast. J Coast
Res 5:49–55
Haapala J (1994) Upwelling and its influence on nutrient concentration in the coastal area of the
Hanko Peninsula, entrance of the Gulf of Finland. Estuar Coast Shelf Sci 38:507–521
Haapala J, Alenius P (1994) Temperature and salinity statistics for the northern Baltic Sea 1961–
1990. Finn Mar Res 262:51–121
Haapala J, Launiainen J, Keynäs K, Pokki J (1990) Hankoniemen edustan virtaustutkimus. Loppu-
raportti. Helsingin Yliopisto, geofysiikan laitos, 59 pp (in Finnish)
Hela I (1946) Coriolis-voiman vaikutuksesta Suomenlahden hydrografisiin oloihin. Terra 58:52–59
(in Finnish)
Hela I (1952) Drift currents and permanent flow. Commentationes physico-mathematicae, vol XVI.
Societas scientarium Fennica, Helsinki, 14, 27 pp
HELCOM (1986) Water balance of the Baltic Sea. Baltic Marine Environment Protection
Commission—Helsinki Commission, Helsinki, Finland. Baltic Sea environment proceedings,
vol 16, 174 pp
Ivanov VN, Mikhailov YD (1972) Estimation of velocity of turbulent energy dissipation and hor-
izontal turbulent diffusion coefficient in the Baltic Sea. Proc State Hydrol Inst 6:1–12 (in Rus-
sian)
Jönsson B, Döös K, Nycander J, Lundberg P (2008) Standing waves in the Gulf of Finland and
their relationship to the basin-wide Baltic seiches. J Geophys Res—Oceans 113:C03004
Jurva R (1937) Laskelmia meriemme lämpövarastosta. Suomi Merellä 12:162–191 (in Finnish)
Keevallik S, Rajasalu R (2001) Winds on the 500 hPa isobaric level over Estonia (1953–1998).
Phys Chem Earth 26:425–429
Keevallik S, Soomere T (2004) Trends in wind speed over the Gulf of Finland 1961–2000. In:
Isemer H-J (ed) Fourth study conference on BALTEX, Gudhjem, Bornholm, Denmark, 24–28
May 2004. International BALTEX Secretariat, Publication No 29, pp 129–130
Keevallik S, Soomere T (2008) Shifts in early spring wind regime in North-East Europe (1955–
2007). Clim Past 4:147–152
Kononen K, Kuparinen J, Mäkelä K, Laanemets J, Pavelson J, Nõmmann S (1996) Initiation of
cyanobacterial blooms in a frontal region at the entrance to the Gulf of Finland, Baltic Sea.
Limnol Oceanogr 41:98–112
Kõuts T, Verjovkina S, Lagemaa P, Raudsepp U (2010) Use of lightweight on-line GPS drifters
for surface current and ice drift observations. In: 2010 IEEE/OES US/EU Baltic international
symposium, Riga, Latvia, August 25–27, 2010. IEEE Press, New York
Knudsen M (1900) Ein Hydrographischer Lehrsatz [A hydrographical theorem]. Ann Marit Mete-
orol 28:316–320 (in German)
Kurkina O, Pelinovsky E, Talipova T, Soomere T (2011) Mapping the internal wave field in
the Baltic Sea in the context of sediment transport in shallow water. J Coast Res 64(Special
Issue):2042–2047
Laakkonen A, Mälkki P, Niemi Å (1981) Studies of the sinking, degradation and sedimentation of
organic matter off Hanko Peninsula, entrance of the Gulf of Finland, in 1979. MERI 9:3–42
Laanearu J (2003) On coastal waves and related upwellings in the Narva Bay. In: Brebbia CA, Al-
morza D, Lopez-Aguayo F (eds) Coastal engineering VI, computer modelling and experimental
measurements of seas and coastal regions. WIT Press, Southampton, pp 53–61
Laanearu J, Lips U (2003) Observed thermohaline fields and low-frequency currents in the Narva
Bay. Proc Est Acad Sci, Eng 9:99–106
Laanemets J, Kononen K, Pavelson J (1997) Nutrient intrusions at the entrance area to the Gulf of
Finland. Boreal Environ Res 2:337–344
Lass HU, Matthäus W (1996) On temporal wind variations forcing salt water inflows into the Baltic
Sea. Tellus A 48:663–671
Launiainen J (1982) Variation of salinity at Finnish fixed hydrographic stations in the Gulf of
Finland and river runoff of the Baltic Sea. In: Gulf of Finland seminar, Leningrad, August 16–
20, 1982, 12 pp
Launiainen J, Saarinen J (1982) Examples of comparison of wind and air–sea interaction charac-
teristics on the open sea and in the coastal areas of the Gulf of Finland. Geophysica 19:33–46
25:333–357
Lehmann A, Myrberg K, Höflich K (2012) A statistical approach to coastal upwelling in the Baltic
Sea based on the analysis of satellite data for 1990–2009. Oceanologia 54:369–393
Lentz W (1971) Monatskarte der Temperatur der Ostsee dargestellt für verschiedene Tiefenho-
risonte. Dtsch Hydrogr Z. Ergänz.heft Reihe B 11:1–148 (in German)
Leppäranta M (1981) On the structure and mechanics of pack ice in the Bothnian Bay. Finn Mar
Res 248:3–86
378 pp
Leppäranta M, Peltola J (1986) Satunnaiskulkumallin soveltyvuudesta Itämerelle [On the applica-

bility of a random walk model to the Baltic Sea]. Finnish Institute of Marine Research, Internal
Report 4, Helsinki, 31 pp (in Finnish)
Lilover M-J (2012) Tidal currents as estimated from ADCP measurements in “practically non-
tidal” Baltic Sea. In: Proceedings of the IEEE/OES Baltic 2012 international symposium
“Ocean: past, present and future. Climate change research, ocean observation & advanced tech-
nologies for regional sustainability”, Klaipėda, May 8–11, 2012. IEEE Press, New York, 4 pp
Lilover MJ, Stips AK (2011) An alternative parameterization of eddy diffusivity in the Gulf of
Finland based on the kinetic energy of high frequency internal wave band. Boreal Environ Res
16(Suppl A):103–116
Lilover M-J, Pavelson J, Kõuts T (2011) Wind forced currents over the shallow Naissaar Bank in
the Gulf of Finland. Boreal Environ Res 16(Suppl A):164–174
Lindow H (1997) Experimentelle Simulationen windangeregter dynamischer Muster in hochau-
flösenden numerischen Modellen. Meereswissenschaftliche Berichte No 22, Institut für Ost-
Lips I, Lips U, Liblik T (2009) Consequences of coastal upwelling events on physical and chemical
patterns in the central Gulf of Finland. Cont Shelf Res 29:1836–1847
Lopatukhin LI, Bukhanovsky AV, Ivanov SV, Tshernyshova ES (eds) (2006a) Handbook of wind
and wave regimes in the Baltic Sea, North Sea, Black Sea, Azov Sea and the Mediterranean.
Russian Shipping Registry, Saint Petersburg, 450 pp (in Russian)
Lopatukhin LI, Mironov ME, Pomeranets KS, Trapeznikov ES, Tshernysheva ES (2006b) Esti-
mates of extreme wind and wave conditions in the eastern part of the Gulf of Finland. Proc
VNIIG 245:145–155 (in Russian)
Mälkki P, Talpsepp L (1988) On the joint Soviet-Finnish experiment in the Gulf of Finland in May
1987: the results of hydrophysical measurements. In: Proceedings of the 16th conference of
Baltic oceanographers, Kiel, pp 687–695
Mälkki P, Tamsalu R (1985) Physical features of the Baltic Sea. Finn Mar Res 252, 110 pp
Meier HEM (1999) First results of multi-year simulations using a 3D Baltic Sea model. SMHI
Reports Oceanography, No 27, Norrköping, Sweden, 48 pp
Meier HEM, Döscher R (2002) Simulated water and heat cycles of the Baltic Sea using a 3D
coupled atmosphere–ice–ocean model. Boreal Environ Res 7:327–334
Mietus M (co-ordinator) (1998) The climate of the Baltic Sea Basin, Marine meteorology and
related oceanographic activities. Report No 41, World Meteorological Organisation, Geneva,
64 pp
Mikhailov YD (1974) Statistical characteristics of turbulent field of current velocities and their
estimation in the coastal zone of the Baltic Sea. Proc State Oceanogr Inst 122:90–107 (in Rus-
sian)
Mikhailov YD (1981) Horizontal turbulence in the eastern Gulf of Finland. Collect Works Leningr
Hydrometeorol Obs 12:29–33 (in Russian)
Mikhailov AE, Tshernyshova ES (1997) General water circulation. In: Davidan IN, Savchuk OP
(eds) “Baltica” project. Issue 5. Part 2. Ecosystem models. assessment of the modern state of
the Gulf of Finland. Gidrometeoizdat, Saint Petersburg, pp 245–260 (in Russian)
Mikulski Z (1970) Inflow of river water to the Baltic Sea in 1961–1970. Nord Hydrol 4:216–227
Millero FJ, Feistel R, Wright DG, McDougall TJ (2008) The composition of standard seawater and
the definition of the reference-composition salinity scale. Deep-Sea Res I 55:50–72
Myrberg K (1998) Analysing and modelling the physical processes of the Gulf of Finland in the
Baltic Sea. Monographs of the Boreal Environment Research, vol 10
Myrberg K, Andrejev O, Lehmann A (2010a) Dynamic features of successive upwelling events in
the Baltic Sea—a numerical case study. Oceanologia 52:77–99
Söderkvist J, Sokolov A, Kuosa H, Anderson TR, Lehmann A, Skogen MD (2010b) Validation
BFU Res Bull 4(5):89–93
Neumann G (1941) Eigenschwingungen der Ostsee. Arch Deutchen Seewarte Mar Obs 61:1–59
(in German)
Nikolaev S, Luvsk T (1975) Spatial variability of characteristics of large-scale turbulence and
mixing in the Baltic Sea. Izv, Fiz Atmos Okean 11:86–869 (in Russian)
Niros A, Vihma T, Launiainen J (2002) Marine meteorological conditions and air–sea exchange
processes over the northern Baltic Sea in 1990s. Geophysica 38:59–87
Omstedt A, Meuller L, Nyberg L (1997) Interannual, seasonal and regional variations of precipita-
tion and evaporation in the Baltic Sea. Ambio 26(8):484–492
Omstedt A, Axell LB (2003) Modeling the variations of salinity and temperature in the large Gulfs
of the Baltic Sea. Cont Shelf Res 23:265–294
Commentationes physico-mathematicae, vol 12. Societas Scientarium Fennica, Helsinki (in
German)
Pavelson J (2005) Mesoscale physical processes and related impact on the summer nutrient fields
and phytoplankton blooms in the western Gulf of Finland. PhD Thesis, Tallinn University of
Technology, Estonia
Pavelson J, Laanemets J, Kononen K, Nõmmann S (1997) Quasi-permanent density front at the
entrance to the Gulf of Finland: response to wind forcing. Cont Shelf Res 17:253–265
Pitkänen H, Tamminen T, Kangas P, Huttula T, Kivi K, Kuosa H, Sarkkula J, Eloheimo K, Kaup-
pila P, Skakalsky B (1993) Late summer trophic conditions in the north-eastern Gulf of Finland
and the river Neva estuary, Baltic Sea. Estuar Coast Shelf Sci 37:453–474
Raudsepp U (1998) Current dynamics of estuarine circulation in the lateral boundary layer. Estuar
Coast Shelf Sci 47:715–730
Reissmann J, Burchard H, Feistel R, Hagen E, Lass HU, Mohrholz V, Nausch G, Umlauf L, Wiec-
zorek G (2009) State-of-the-art review on vertical mixing in the Baltic Sea and consequences
for eutrophication. Prog Oceanogr 82:47–80
Sarkkula J (1989) Measuring and modeling flow and water quality in Finland. Vituki monogra-
phies, vol 49. Water Resources Research Centre, Budapest, 39 pp
Sarkkula J (1991) Measuring and modeling water currents and quality as a part of decision making
for water pollution control. PhD thesis, Tartu University, Estonia
Sarkisyan AS, Staśkiewicz A, Kowalik Z (1975) Diagnostic calculations of summer circulation in
the Baltic Sea. Okeanologiya 15:1002–1009 (in Russian)
Savijärvi H, Niemela S, Tisler P (2005) Coastal winds and low-level jets: simulations for sea gulfs.
Q J R Meteorol Soc B 131:625–637
Schmager G, Fröhle P, Schräder D, Weisse R, Müller-Navarra S (2008) Sea state, tides. In: Feis-
tel R, Nausch G, Wasmund N (eds) State and evolution of the Baltic Sea 1952–2005. Wiley,
Hoboken, pp 143–198
Seinä A, Palosuo E (1993) The classification of the maximum annual extent of ice cover in the
Baltic Sea 1720–1992. MERI 20:5–20
Smelov NF, Fomichev BA, Chernov VM (1985) Turbulence of waters in the Neva Bight below the
ice. Collect Works Leningr Hydrometeorol Obs 2(15) (in Russian)
SMHI and FIMR (1982) An ice atlas for the Baltic Sea, Kattegat, Skagerrak and Lake Vänern.
Norrköping, Sjöfartsverket
Soomere T (1995) Generation of zonal flow and meridional anisotropy in two-layer weak
geostrophic turbulence. Phys Rev Lett 75:2440–2443
Soomere T, Healy T (2011) On the dynamics of ‘almost equilibrium’ beaches in semi-sheltered
bays along the southern coast of the Gulf of Finland. In: Harff J, Björck S, Hoth P (eds) The
Baltic Sea basin. Central and eastern European development studies, vol 5. Springer, Heidel-
berg, pp 255–279
Soomere T, Keevallik S (2003) Directional and extreme wind properties in the Gulf of Finland.
Proc Est Acad Sci, Eng 9:73–90
Soomere T, Myrberg K, Leppäranta M, Nekrasov A (2008) The progress in knowledge of physical
oceanography of the Gulf of Finland: a review for 1997–2007. Oceanologia 50:287–362
Soomere T, Delpeche N, Viikmäe B, Quak E, Meier HEM, Döös K (2011a) Patterns of current-
A):49–63
Soomere T, Viidebaum M, Kalda J (2011b) On dispersion properties of surface motions in the Gulf
of Finland. Proc Est Acad Sci 60:269–279
Stigebrandt A (2001) Physical oceanography of the Baltic Sea. In: Wulff F, Rahm L, Larsson P
(eds) A system analysis of the Baltic Sea. Ecological studies, vol 148. Springer, Berlin, pp 19–
74
Stipa T (2004) Baroclinic adjustment in the Finnish coastal current. Tellus A 56:79–87
Suursaar Ü (2010) Waves, currents and sea level variations along the Letipea–Sillamäe coastal
section of the southern Gulf of Finland. Oceanologia 52:391–416
Talpsepp L (1986) On the jet current in the Gulf of Finland. In: Proceedings of the 15th conference
of the Baltic oceanographers, Copenhagen, Denmark, pp 613–626
Talpsepp L (2006) Periodic variability of currents induced by topographically trapped waves in the
coastal zone in the Gulf of Finland. Oceanologia 48(S):75–90
Talpsepp L (2008) On the influence of the sequence of coastal upwellings and downwellings on
surface water salinity in the Gulf of Finland. Est J Eng 14:29–41
Talpsepp L, Nõges T, Raid T, Kõuts T (1994) Hydrophysical and hydrobiological processes in the
Gulf of Finland in summer 1987: characterization and relationship. Cont Shelf Res 14:749–763
Verjovkina S, Raudsepp U, Kõuts T, Vahter K (2010) Validation of Seatrack Web using surface
drifters in the Gulf of Finland and Baltic Proper. In: 2010 IEEE/OES US/EU Baltic international
symposium, Riga, Latvia, August 25–27, 2010. IEEE Press, New York, 7 pp
Zhurbas VI, Oh S, Park T (2007) Formation and decay of a longshore baroclinic jet associated with
transient coastal upwelling and downwelling: a numerical study with applications to the Baltic
Sea. J Geophys Res—Oceans 111:C04014
Witting R (1910) Rannikkomeret. Suomen Kartasto. Karttalehdet Nr 6b, 7, 8 ja 9. Finnish Institute
of Marine Research, Helsinki (in Finnish)
Meerbusens und der nördlichen Ostsee nach den Untersuchungen bis Ende 1910, Finnländische
hydrographisch-biologische Untersuchungen No 7, 82 pp (in German)
Wübber C, Krauss W (1979) The two-dimensional seiches of the Baltic Sea. Oceanol Acta 2:435–
446
Wulff F, Rahm L, Larsson P (eds) (2001) A systems analysis of the Baltic Sea. Ecological studies,
vol 148. Springer, Berlin, 455 pp
Part II
Lagrangian Dynamics and Inverse
Problems
Chapter 7
TRACMASS—A Lagrangian Trajectory Model
Kristofer Döös, Joakim Kjellsson, and Bror Jönsson
Abstract A detailed description of the Lagrangian trajectory model TRACMASS

is presented. The theory behind the original scheme for steady state velocities is
derived for rectangular and curvilinear grids with different vertical coordinates for
the oceanic and atmospheric circulation models. Two different ways to integrate
the trajectories in time in TRACMASS are presented. These different time schemes
are compared by simulating inertial oscillations, which show that both schemes are
sufficiently accurate not to deviate from the analytical solution.
The TRACMASS are exact solutions to differential equations and can hence be
integrated both forward and backward with unique solutions. Two low-order trajec-
tory subgrid parameterizations, which are available in TRACMASS, are explained.
They both enable an increase of the Lagrangian dispersion, but are, however, too
simple to simulate some of the Lagrangian properties that are desirable. The mass
conservation properties of TRACMASS are shown to make it possible to follow the
water or air masses both forward and backward in time, which also opens up for
all sorts of calculations of water/air mass exchanges as well as Lagrangian stream
functions.
7.1 Introduction
The specification of a flow field can be made in an Eulerian or a Lagrangian frame of

reference. The Eulerian method is when the fluid flow is observed from a point fixed
in space, while the Lagrangian method is instead working from the perspective of the
K. Döös (B) · J. Kjellsson

Department of Meteorology, Bolin Centre for Climate Research, Stockholm University, Svante
Arrhenius väg 16C, 106 91 Stockholm, Sweden
e-mail: doos@misu.su.se
J. Kjellsson
e-mail: joakim@misu.su.se
B. Jönsson
Department of Geosciences, Princeton University, Guyot Hall, Princeton, NJ 08544, USA
e-mail: bjonsson@princeton.edu

DOI 10.1007/978-3-319-00440-2_7,
226 K. Döös et al.
flow. This can be illustrated by a cyclist, who passes an immobile traffic jam. In this
case the static car driver sees the moving cyclist from an Eulerian perspective, while
the moving cyclist observes the static traffic jam from a Lagrangian perspective. The
zigzagging path of the cyclist between the cars constitutes a Lagrangian trajectory.
Most analytical and numerical models in fluid dynamics are made in the Eulerian
framework, since it is then straightforward to describe the motion as a function of
position and time. This is why in nearly all ocean general circulation models the
equations of motion are discretized with finite differences on a fixed grid so that the
motion of the water and its tracers such as salinity and temperature are described
from the Eulerian perspective with different values in each grid box, even if the
vertical discretization often has a time dependent component related to the motion
of the fluid. Lagrangian trajectories are, however, still possible to calculate from the
model simulated Eulerian velocity fields on the model grid.
The present chapter will present the TRACMASS Lagrangian trajectory model,
which uses the Eulerian velocity fields, which have been simulated by ocean or
atmosphere general circulation models (GCM). The trajectories are calculated off-
line, i.e., after the GCM has been integrated and the velocity fields have been stored.
This makes it possible to calculate many more trajectories than would be possible
on-line, i.e., simultaneously with the GCM run. TRACMASS has been applied to
many different general circulation models, both for the ocean and for the atmo-
sphere.
The original feature of the method is that it solves the trajectory path through
each grid cell with an analytical solution of a differential equation which depends
on the velocities on the walls of the grid box. The scheme was originally devel-
oped in Döös (1995), Blanke and Raynaud (1997) for stationary velocity fields and
hereafter further developed in de Vries and Döös (2001) for time-dependent fields
by solving a linear interpolation of the velocity field both in time and in space over
each grid box. This is in contrast to the time schemes such as simple Euler forward or
more advanced fourth order Runge–Kutta methods (Butcher 2008; Fabbroni 2009)
where the trajectories are integrated forward in time with as short time steps as pos-
sible.
A consequence of solving the trajectory paths analytically over a certain time is
that the solutions are unique and can be integrated forward in time and then back-
ward in time and arriving exactly at the same position, which is not possible with
the other trajectory methods. This makes it possible to trace origins of water or air
masses as long as the subgrid parameterization is not activated.
The TRACMASS code has been further developed over the years and used in
many studies of the global ocean (Döös and Coward 1997; Drijfhout et al. 2003;
Döös et al. 2008) and regional ones for the Mediterranean and Baltic Seas (Döös
et al. 2004; Jönsson et al. 2004; Engqvist et al. 2006; Soomere et al. 2011) as well
as the large scale atmospheric circulation (Kjellsson and Döös 2012).
The code was originally written in Fortran 77 for the FRAM ocean model at
the Institute of Oceanographic Sciences, Deacon Laboratory (IOSDL) in Wormley,
UK in the early 1990s. The name TRACMASS comes from the EU project with
the same name, where it served together with the similar trajectory code Ariane by
7 TRACMASS—A Lagrangian Trajectory Model 227
Blanke and Raynaud (1997). The present code is written in Fortran 95 and can be
driven by velocity fields from most GCMs based on finite differences. The TRAC-
MASS code is continuously upgraded and adapted. The code can be downloaded
from http://tracmass.org/. The user must be familiar, in order to be able to use the
TRACMASS code, with (1) the equations of motion for the ocean-atmosphere cir-
culation (described in Chaps. 2, 4 and 6), (2) the finite differences of these equations
(Chap. 3), (3) the TRACMASS theory (the introduction to which is presented in this
chapter), (4) Unix and (5) Fortran.
The Lagrangian trajectory approach has many similarities with the Eulerian
tracer approach but at the same time many differences. The two approaches are
often confused due to their similarities. They are both advected passively by the
velocity fields of the GCM, which makes it possible to trace water/air masses or
substances such as pollutants as they are carried with the ocean currents or winds.
The tracer equation generally needs to be integrated ‘on-line’ with the GCM while
the Lagrangian trajectories can be both ‘on-line’ and ‘off-line’. The ‘off-line’ cal-
culation of Lagrangian trajectories is by far the most rapid way since one only
needs to read the already simulated velocity fields in order to calculate the trajecto-
ries.
The tracer equation includes explicitly a diffusion term, which represents a pa-
rameterization of the unresolved subgrid scales. There is also a numerical reason
to include this since GCMs generally need some diffusion and viscosity to remain
numerically stable in order to dissipate energy or to eliminate numerical noise due
to the truncation errors in the numerical schemes. The passive tracers also have
a numerical diffusion due to the finite difference approximation error, which by
itself often would be enough as diffusion. The tracer approach is therefore often
too diffusive but has been improved with better numerical advection schemes dur-
ing the last decade. The Lagrangian trajectories are passively advected with the
currents or winds and the subgrid parameterization is included in the sense that
the GCM has been integrated with viscosity and diffusion. An extra diffusion can,
however, if desired, be added to the trajectories. Another advantage of the trajec-
tories is that it is possible to follow particles from their release points to the end
both forward and backwards, which is impossible with passive tracers that can-
not be integrated backward in time due to the numerical and parameterized diffu-
sion.
The present chapter will describe the basic theory for the TRACMASS trajectory
calculations and is organized as follows. In Sect. 7.2 we present the basic equations
for a rectangular grid, which is then extended in Sect. 7.3 to the more general case
with non-rectangular grids and for atmospheric GCMs in Sect. 7.4. The TRAC-
MASS analytical time dependent scheme based on de Vries and Döös (2001) is
presented in Sect. 7.5 followed by the presentation of two simple sub-grid param-
eterizations in Sect. 7.6 and how the mass conservation in TRACMASS enables
analysis of the water/air mass transports in Sect. 7.7. In Sect. 7.8, we summarize
and discuss the TRACMASS approach and its possible improvements in the future.
Fig. 7.1 Left: B-grid, Right: C-grid
Fig. 7.2 Illustration of a

trajectory [x(t), y(t)] through
one grid box. The model
velocities are defined at the
walls of the grid box
7.2 Trajectory Solution for Rectangular Grids

This section is here only for pedagogical reasons, since it is only valid for rect-
angular Cartesian grids. The TRACMASS code is written in a more general way in
order to enable TRACMASS to work with curvilinear grids, which are used by most
GCMs, and will be presented in the next section.
Most finite difference GCMs use B- or C-grids (Mesinger and Arakawa 1976)
as shown in Fig. 7.1, where i, j, k denote the discretized longitude, latitude and
model level, respectively. The zonal velocity ui,j,k and meridional velocity vi,j,k
are located differently in these two grids, while the vertical velocity wi,j,k is located
in the middle at the top of the box in both grids (Figs. 7.2, 7.3a). Both these types
of grids can be used in TRACMASS. The velocities in TRACMASS are set on
a C-grid, which makes it straightforward when using a C-grid model. Although
B-grid velocities just need to be projected on the C-grid by making a meridional
average uC i,j,k = 0.5(ui,j,k + ui,j −1,k ) of two zonal velocities and a zonal average
B B
C
vi,j,k = 0.5(vi,j,k
B + vi−1,j,k
B ) of two meridional velocities for each grid box.
In a finite difference model there is no information of scales below the grid size.
The tracers are regarded as homogeneous within each grid box and the velocities
Fig. 7.3 Vertical trajectory discretization in model grids
are only defined on the grid side walls. It is, however, possible to define the velocity
inside a grid box by interpolating linearly between the discretized velocity values of
the opposite walls. For the zonal x-direction one obtains
x − xi−1
u(x) = ui−1,j,k + (ui,j,k − ui−1,j,k ). (7.1)
x
Local zonal velocity and position are related by u = dx/dt. The approximation in
Eq. (7.1) can now be written in terms of the following differential equation:
dx
+ βx + δ = 0, (7.2)
dt
with β ≡ (ui−1,j,k − ui,j,k )/x and δ ≡ −ui−1,j,k − βxi−1 . Using the initial con-
dition x(t0 ) = x0 , the zonal displacement of the trajectory inside the considered grid
box can be solved analytically and is given by

δ −β(t−t0 ) δ
x(t) = x0 + e − . (7.3)
β β
The time t1 when the trajectory reaches a zonal wall can be determined explicitly:
1 x1 + δ/β
t1 = t0 − log , (7.4)
β x0 + δ/β
where x1 = x(t1 ) is given by either xi−1 or xi . For a trajectory reaching the wall
x = xi , for instance, the velocity ui must necessarily be positive, so in order for
Eq. (7.4) to have a solution, the velocity ui−1 must then be positive also. If this is
not the case, then the trajectory either reaches the other wall at xi−1 or the signs of
the transports are such that there is a zero zonal transport somewhere inside the grid
box that is reached exponentially slow. For the meridional and vertical directions,
similar calculations of t1 are performed determining the meridional and vertical dis-
placements of the trajectory, respectively, inside the considered grid box. The small-
est transit time t1 − t0 and the corresponding x1 denote at which wall of the grid box
Fig. 7.4 The Ocean Conveyor Belt with velocities simulated by the OCCAM model. The red
trajectories are part of the shallow warmer part of the Conveyor Belt with transports toward the
North Atlantic. The blue trajectories represent the flow of the dense and cold North Atlantic Deep
Water from the North Atlantic into the Indo-Pacific
the trajectory will exit and move into the adjacent one. The exact displacements in
the other two directions are then computed using the smallest t1 in the correspond-
ing Eq. (7.3). The resulting trajectory through the grid box is illustrated by Figs. 7.2
and 7.3a.
Note that a consequence of solving the trajectories analytically with Eq. (7.3) is
that the solution is unique. The trajectory can hence be integrated forward in time
and then backward in time and arriving back exactly in the same point where it
started.
7.3 Scheme for Volume or Mass Transports

and Non-rectangular Grids
The disadvantage with the scheme presented in the previous section is that it requires
rectangular grid cells and GCMs generally use some sort of spherical or curvilin-
ear grids as in the case of the Ocean Circulation and Climate Advanced Model
(OCCAM) model presented in Fig. 7.4, where two spherical grids have been used
for the world ocean. The longitudinal (xi,j ) and the latitudinal (yi,j ) grid lengths
will hence be a function of their horizontal positions i, j on a curvilinear grid. The
depth level thickness zk will similarly vary but with layer level k.
Trajectories can, however, be calculated for the curvilinear grids by replacing the
velocities by volume transports. The transport Ui,j,k through the eastern wall of the
i, j, k grid box is given by
Ui,j,k = ui,j,k yi,j zk . (7.5)

The distance is non-dimensionalized by using r = x/x, and the linear interpola-

tion of the velocity (Eq. (7.1)) is replaced by
U (r) = Ui−1,j,k + (r − ri−1 )(Ui,j,k − Ui−1,j,k ). (7.6)
The local transport and position are now related by U = dr/ds, where the scaled
time variable is s ≡ t/(xi,j yi,j zk ), the denominator being the volume of the
particular grid box. The differential equation (7.2) is replaced by
dr
+ βr + δ = 0, (7.7)
ds
with β ≡ Ui−1,j,k − Ui,j,k and δ ≡ −Ui−1,j,k − βri−1 . Using the initial condition
r(s0 ) = r0 , the zonal displacement of the trajectory is now given by

δ −β(s−s0 ) δ
r(s) = r0 + e − . (7.8)
β β
The scaled time s1 becomes

1 r1 + δ/β
s1 = s0 − log , (7.9)
β r0 + δ/β
where r1 = r(s1 ) is given by either ri−1 or ri . With the use of Eq. (7.5), the loga-
rithmic factor can be expressed as log[U (r1 )/U (r0 )].
For a trajectory reaching the wall r = ri , for instance, the transport U (r1 ) must
necessarily be positive, so in order for Eq. (7.9) to have a solution, the transport
U (r0 ) must then be positive also. If this is not the case, then the trajectory either
reaches the other wall at ri−1 or the signs of the transports are such that there is
a zero zonal transport somewhere inside the grid box that is reached exponentially
slow. The calculations of s1 are performed determining the zonal, meridional and
vertical displacements of the trajectory, respectively, inside the considered grid box.
The smallest transit time s1 − s0 and the corresponding r1 denote at which wall
of the grid box the trajectory will exit and move into the adjacent one. The exact
displacements in the other two directions are then computed using the smallest s1 in
the corresponding Eq. (7.8).
The scheme is mass conserving since it deals with the transport across the grid
walls just as in the GCM and the transport is only linearly interpolated between two
opposite walls in a grid box.
The trajectories will hence never cross a grid wall.
The solutions for the meridional and vertical directions are calculated similarly
as the zonal one but using the meridional and vertical transport, respectively, defined
as
Vi,j,k = vi,j,k xzk , (7.10)

Wi,j,k = wi,j,k xy. (7.11)
The scheme is also mass conserving in the sense that the vertical transport is
directly calculated from the continuity equation in the same way as in the ocean
GCM, which is due to the incompressibility in the ocean
∂u ∂v ∂w
+ + =0 (7.12)
∂x ∂y ∂z
that is discretized with finite differences on a C-grid into
ui,j,k − ui−1,j,k vi,j,k − vi,j −1,k wi,j,k − wi,j,k−1
+ + = 0. (7.13)
xi,j yi,j zk
Equation (7.13) simply reflects the condition that the sum of all the volume fluxes
in or out of the grid box is zero. The vertical volume transport through the top of the
grid box is obtained from Eqs. (7.11) and (7.13),
Wi,j,k = Wi,j,k−1 − (Ui,j,k − Ui−1,j,k + Vi,j,k − Vi,j −1,k ), (7.14)
which can be computed by integration from the bottom and upwards with the bottom
boundary condition Wi,j,0 = 0. Since the trajectory solutions are exact and the con-
tinuity equation is respected the TRACMASS trajectories will therefore never hit
any solid boundary such as the coast or the sea floor. This feature should be taken
into account when the TRACMASS model is used for calculations of the transport
of tracers or pollution to the coast. As described in Chap. 9, the virtual coastline
should be set to a certain distance from the model coastline.
The depth level thickness z in the above derivations depends only on the depth
level k. TRACMASS can, however, be integrated, with other GCM vertical coordi-
nates that may depend on something more than just the depth level. Options of ver-
tical coordinates for TRACMASS hence exist for (1) depth level models, (2) sigma-
coordinate models, where the thickness depends on the total depth, which varies in
each horizontal grid point, (3) z-star coordinates, where the layer thickness depends
on sea surface elevation, (4) isopycnal models, where z is the density layer thick-
ness, which was implemented in TRACMASS by Marsh and Megann (2002) and
(5) hybrid vertical coordinates for atmospheric GCMs, which will be presented in
the next section. See Chap. 3 for a discussion of some properties of such models.
7.4 Scheme for Atmospheric Hybrid Vertical Coordinates

The atmospheric version of TRACMASS uses conservation of mass instead of vol-
ume. Most atmospheric GCMs today use terrain-following vertical coordinates. Fol-
lowing Simmons and Burridge (1981) the atmosphere is divided into NLEV layers,
which are defined by the pressures at the interfaces between them and these pres-
sures are given by pk+1/2 = Ak+1/2 + Bk+1/2 pS for k = 0, 1, . . . , NLEV , with k = 0
at the top of the atmosphere and k = NLEV at the Earth’s surface. The Ak+1/2 and
Bk+1/2 are constants, whose values effectively define the vertical coordinate and pS
is the surface pressure. The dependent variables, which are the zonal wind u, the
meridional wind v, the temperature T and the specific humidity q are defined in the
middle of the layers, where the pressure is defined by pk = 12 (pk−1/2 + pk+1/2 ), for
k = 1, 2, . . . , NLEV . The vertical coordinate is η = η(p, pS ) and has the boundary
value η(0, pS ) = 0 at the top of the atmosphere and η(pS , pS ) = 1 at the Earth’s
surface.
For the ocean, in the previous sections, we used volume transport because of the
incompressibility approximation. In the atmosphere we need instead to use mass
transport so Eq. (7.5) is now replaced by the zonal and meridional mass transports
in the model layers:
ypk xj pk
Ui,j,k = ui,j,k , Vi,j,k = vi,j,k , (7.15)
g g
where pk = Ak + Bk pSi,j,k , Ak = Ak+1/2 − Ak−1/2 and Bk = Bk+1/2 −
Bk−1/2 . Note that with hybrid coordinates, the pressure at model layer interfaces
pi,j,k varies in both space and time as surface pressure varies.
The mass transport between model layer interfaces, here denoted W as a vertical
flux, can be calculated using the continuity equation from Simmons and Burridge
(1981) as done in Kjellsson and Döös (2012):

∂pk
k
∂p
η̇ =− − ∇ · (um , vm )pm . (7.16)
∂η k ∂t
m=1
This gives the vertical velocity due to the variations in the pressure at the interface
and the divergence above. This quantity can be translated into mass flux by multi-
plying by xj y:

∂p
Wi,j,k = xy η̇
∂η k
k
=− Ui,j,m − Ui−1,j,m + Vi,j,m − Vi,j −1,m
m=1

psn − psn−1
+ xyBm , (7.17)
t
where we use ∂pk /∂t = ∂(Bk ps )/∂t.
The mass conservation of a grid box is illustrated in Fig. 7.3b. The integration
over the model levels is done from the top down, with the assumption Wi,j,0 = 0.
This may lead to Wi,j,NLEV = 0, if the fields are not perfectly mass-conserving,
which is the case for reanalysis data (see Berrisford et al. 2011 for a study on ERA-
Interim) as used by Kjellsson and Döös (2012). In such a case, the vertical flux at
the surface must be explicitly set to zero.
The trajectory differential equation (7.7) and its solutions (7.8)–(7.9) remain the
same but now fed with mass transports on atmospheric terrain-following vertical
Fig. 7.5 Example of mid-tropospheric atmospheric trajectories. The wind velocities are from the
ERA-Interim reanalysis from the ECMWF
coordinates and the scaled time is now s ≡ gt/(xi,j yi,j pk ). The atmospheric
TRACMASS code has been used to study the atmospheric Hadley and Ferrel cells
as well as the inter-hemispheric air mass exchange (Kjellsson and Döös 2012). Fig-
ure 7.5 shows an example of atmospheric TRACMASS trajectories calculated with
winds from the ERA-Interim reanalysis from the European Centre for Medium-
Range Weather Forecasts (ECMWF).
7.5 Time Integration

The trajectory schemes in the previous sections with the differential Eqs. (7.2) and
(7.7) are only valid for stationary velocity fields. We will now present two possi-
ble ways to incorporate the temporal variability of the velocity and surface eleva-
tion fields in the TRACMASS trajectory calculations. One (called time-stepping)
method is based on previous sections and one is more advanced, where the differ-
ential equation is extended in time and solved analytically in both space and time.
Note that nearly all GCMs today have some sort of free surface, which will make
the level thickness z also dependent of time and will hence affect the mass trans-
port across the grid walls. It is therefore necessary to have both the velocity and the
surface elevation fields in order to compute the trajectories with TRACMASS.
7.5.1 Time-Stepping Method
The time-stepping method consists of assuming that the velocity and surface ele-
vation fields are in steady state during a limited time interval. The fields are then
Fig. 7.6 Schematic illustration of how the velocity fields u(t) can be updated in time, with new
GCM data at regular intervals tG in green and linearly interpolated velocity points in red with
the time step ti . The number of intermediate time steps between two GCM velocities is in this
example IS = tG /ti = 4
updated successively as new fields are available. If this is made ‘on-line’, i.e., in
the same time as the GCM is integrated, then this time interval will simply be the
same as the time step the GCM is integrated with, which is typically of the order of
minutes in a global GCM. If instead the trajectories are calculated ‘off-line’ it will
be at least as often as the fields have been stored by the GCM.
A linear time interpolation of the velocity fields between two GCM velocity fields
enables a simple way to have shorter time steps by which the fields are updated in
time. The time interval between two GCM velocity fields is tG and the shorter
time interval at which the fields are interpolated is ti as illustrated by Fig. 7.6. The
number of intermediate time steps is hence the ratio IS = tG /ti .
7.5.2 Analytical Time Integration
In the present section, we will present a time dependent scheme, which was intro-
duced in TRACMASS by de Vries and Döös (2001) that is solved analytically in
time over tG between two GCM time steps.
Given a set of velocities Vn for each model point, where n represents a dis-
cretized time, a bi-linear interpolation of transport in space as well as in time leads
to
F (r, s) = Fi−1,n−1 + (r − ri−1 )(Fi,n−1 − Fi−1,n−1 )

s − sn−1
+ Fi−1,n − Fi−1,n−1
s

+ (r − ri−1 )(Fi,n − Fi−1,n − Fi,n−1 + Fi−1,n−1 ) , (7.18)
which is the general expression for the three directions where i signifies either a
longitudinal, meridional, or vertical direction. The transport is F = (U, V , W ) and
as before r = (x/x, y/y, z/z), s ≡ t/(xyz), where the denominator is
the volume of the particular grid box and s is the scaled time step between two
data sets:
s = sn − sn−1 = (tn − tn−1 )/(xyz) = tG /(xyz), (7.19)

where tG is the time step between two data sets in true time dimension (seconds).
Connecting the local transport to the time derivative of the position with F =
dr/ds, we get the differential equation
dr
+ αrs + βr + γ s + δ = 0, (7.20)
ds
where the coefficients are defined by
1
α≡− (Fi,n − Fi−1,n − Fi,n−1 + Fi−1,n−1 ), (7.21)
s
β ≡ Fi−1,n−1 − Fi,n−1 − αsn−1 , (7.22)
1
γ ≡− (Fi−1,n − Fi−1,n−1 ) − αri−1 , (7.23)
s
δ ≡ Fi−1,n−1 + ri−1 (Fi,n−1 − Fi−1,n−1 ) − γ sn−1 . (7.24)
Differently shaped analytical solutions exist for the three cases: α > 0, α < 0 and
α = 0, which together cover all possible values of α. Note that inside the grid box,
the acceleration d 2 r/ds 2 = −αr − γ consists of a constant and a linear r-dependent
term proportional to α. For α > 0, the latter term implies a varying deceleration
across the grid box. √
If α > 0, we define the time-like variable ξ = (β + αs)/ 2α and get

γ ξ 2 −ξ 2 γ βγ − αδ 2
D(ξ ) − eξ0 −ξ D(ξ0 )
2 2
r(s) = r0 + e0 − + (7.25)
α α α α
using Dawson’s integral
ξ
D(ξ ) ≡ e−ξ
2 2
ex dx (7.26)
0
√ r(s0 ) = r0 . If α < 0, ξ becomes imaginary. By defining

and the initial condition
ζ ≡ iξ = (β + αs)/ −2α, Eq. (7.25) can be re-expressed as

γ ζ 2 −ζ 2 γ βγ − αδ π ζ2
r(s) = r0 + e 0 − − e erf(ζ ) − erf(ζ0 ) , (7.27)
α α α −2α
√ ζ −x 2
where the error function is erf(ζ ) = (2/ )π 0 e dx. The case α = 0 will occur
occasionally in practice. The corresponding solution of Eq. (7.20) is

δ −β(s−s0 ) δ γ
r(s) = r0 + e − + 2 1 − βs + (βs0 − 1)e−β(s−s0 ) . (7.28)
β β β
A major difference compared with the time-stepping method (solution of Eq.
(7.8)) is that now the transit times s1 − s0 cannot in general be obtained explicitly.
Using the solutions (7.25)–(7.28), the relevant root s1 of
r(s1 ) − r1 = 0 (7.29)
has to be computed numerically for each direction. In the following subsection, we

describe how the roots s1 and the corresponding exiting wall r1 can be determined.
The displacement of the trajectory inside the considered grid box then proceeds as
discussed previously for stationary velocity fields.
We will now determine the roots s1 of Eq. (7.29) and the corresponding r1 needed
to compute trajectories inside a grid box. In the following, sn−1 s0 < sn and the
relevant roots s1 are to obey s0 < s1 sn . We also focus on the cases α > 0 and
α < 0, since the considerations below can easily be adapted for α = 0. For numerical
purposes, we use
βγ − αδ Fi,n Fi−1,n−1 − Fi,n−1 Fi−1,n

= , (7.30)
α Fi,n − Fi−1,n − Fi,n−1 + Fi−1,n−1
γ Fi−1,n − Fi−1,n−1
= − ri−1 , (7.31)
α Fi,n − Fi−1,n − Fi,n−1 + Fi−1,n−1
Fi−1,n−1 − Fi,n−1 + α(s − sn−1 )
ξ= √ , (7.32)
2α
Fi−1,n−1 − Fi,n−1 + α(s − sn−1 )
ζ= √ . (7.33)
−2α
The coefficient in (7.30) appearing in (7.25) and (7.27) is exactly zero when either
the ri−1 or ri wall represents land, the transport Fi or Fi−1 being zero for all n,
respectively. In these instances, the opposite wall fixes r1 , and the root s1 > s0 can
then be computed analytically. If there is no solution, we take s1 = sn . When all
three transit times equal sn , the trajectory will not move into an adjacent grid box
but will remain inside the original one. Its new position is subsequently computed,
and the next time interval is considered.
If (βγ − αδ)/α = 0, the computation of the roots of Eq. (7.29) can only be
done numerically. This is also true for locating the extrema of the solutions (7.25)
and (7.27). Alternatively, one can consider F (r, s) = 0 using Eq. (7.18) to analyse
where possible extrema are located. It follows that in the (s–r)-plane, extrema lie
on a hyperbola of the form r = (as + b)/(c + ds). Of course, only the parts de-
fined by sn−1 ≤ s ≤ sn and ri−1 ≤ r ≤ ri are relevant. Depending on which parts
of the hyperbola, if any, lie in this ‘box’ and on the initial condition r(s0 ) = r0 , the
trajectory r(s) exhibits none, one, or at most two extrema. In the latter case, the
trajectory will not cross either the wall at ri−1 or the one at ri (see Fig. 7.7 for an
example). Hence, those trajectories r(s) determining the transit time s1 − s0 will
have at most one extremum, that is, there is at most one change of sign in the local
transport F .
An efficient way to proceed then is as follows. First, consider the wall at ri . For a
trajectory to reach this wall, the local transport must be nonnegative, which depends
on the signs of the transport Fi−1,n and Fi,n . Four distinct configurations may arise:
1. F (ri , s) > 0 for sn−1 < s < sn .
2. Sign of F (ri , s) changes from positive to negative at s = s̃ < sn .
Fig. 7.7 Example of

trajectory r(s) exhibiting two
extrema (zero-transport
points) inside the relevant r–s
‘box.’ Regions with positive
and negative transports are
shown. Extrema for
trajectories with differing
initial conditions must lie on
the hyperbola (dotted curves)
3. Sign of F (ri , s) changes from negative to positive at s = ŝ < sn .

4. F (ri , s) < 0 for sn−1 < s < sn .
For case 1, we evaluate r(sn ) using the appropriate analytical solution. If r(sn ) ≥ ri ,
the trajectory has crossed the grid-box wall for s1 ≤ sn . If the initial transport
F (r0 , s0 ) < 0, the trajectory may have crossed the opposite wall at an earlier time.
The latter is only possible if case 3 applies for the wall at ri−1 and ŝ > 0, in which
case one checks whether r(ŝ) ≤ ri−1 . If this is not so, then there is a solution to
r(s1 ) − r1 = 0 for r1 = ri and s0 < s1 ≤ sn . Subsequently, this root can be simply
calculated numerically using a root-solving algorithm. But if r(sn ) < ri or, if appli-
cable, r(ŝ) ≤ ri−1 , we continue with considering the other wall. The arguments for
the wall at ri−1 are similar to those relating to r. If case 2 applies and s0 < s̃, we
follow the considerations given for case 1 using s̃ instead of sn . If there is a root for
r1 = ri , then s0 < s1 ≤ ŝ. For case 3, we follow the considerations given for case 1.
If there is a root for r1 = ri , then ŝ < s1 ≤ sn . For case 4, no solution of Eq. (7.29) is
possible for r1 = ri . We must then turn attention to the other wall instead. All these
considerations are applied to each direction.
7.5.3 Evaluation of the Two Time Integration Methods
The two possible time schemes by which TRACMASS can be integrated in time,
which have been presented above, will here be evaluated by testing them on inertial
oscillations. Exact analytical solutions of the trajectories for inertial oscillations can
be found as well as the corresponding velocity fields. The experiment was origi-
nally set up by Fabbroni (2009) to test four different trajectory algorithms. One of
these algorithms was Ariane (Blanke and Raynaud 1997), which is based on the
same equations as the version of TRACMASS that uses the time-stepping method.
The three other trajectory algorithms were based on Euler forward and Runge–Kutta
schemes. The trajectories, simulated by Ariane, deviated clearly from the analytical
solution and the other trajectory schemes. It was thus concluded that Ariane was not
as accurate as the other schemes.
In the present study we will repeat one of the Fabbroni (2009) tests for the two
TRACMASS schemes and evaluate them by comparing them with the exact ana-
lytical inertial oscillation solution. The test consists of using the analytical solution
of damped inertial oscillations, which are carried away with a mean geostrophic
current so that the equations of motion are
∂u
− f v = −γ u,
∂t
(7.34)
∂v
+ f u = −γ v + f ug ,
∂t
which describe particle circles with a drift to the east due to a geostrophic velocity
ug and with a decreasing oscillation radius depending on the linear friction coeffi-
cient γ . The solutions for the velocities are
u = ug e−γg t + (u0 − ug )e−γ t cos f t,

(7.35)
v = −(u0 − ug )e−γ t sin f t
and for the particle trajectories

ug (u0 − ug )f γ γ
x = x0 + 1 − e−γg t + + e −γ t
sin f t − cos f t ,
γg f2 +γ2 f f
(7.36)
(u0 − ug )f −γ t γ
y = y0 − 1−e cos f t + sin f t .
f2 +γ2 f
We used the same coefficients as Fabbroni (2009) with u0 = 0.3 m/s, ug = 0.04 m/s
and a damping time of td = 1/γ = 2.89 days and tg = 1/γg = 28.9 days. The lat-
itude was set to be 45◦ N. The velocities are read into TRACMASS every hour
(tG = 1 hour) to mimic a GCM that stores the data once an hour. This in contrast
to Fabbroni (2009) who read in the velocities as often as every 3 minutes, which is
unrealistically high to be run off-line with.
TRACMASS was then integrated forward in time using the time-stepping
method with intermediate time steps so the velocities were updated with linear in-
terpolation between two such mimicked ‘GCM’ velocities. The results are shown
in Fig. 7.8. Only the red curve that corresponds to the case with no intermediate
time steps deviates clearly from the true analytical solution. The blue curve calcu-
lated with 10 intermediate time steps has a slight difference. The green (for which
1000 intermediate time steps have been used) lies almost exactly under the purple
curve (that reflects the analytical time integration scheme). The small differences
between the results from the truly analytical solution and the analytical time inte-
gration scheme and time-stepping scheme with 1000 intermediate time steps are
likely due to that the velocities are only read into TRACMASS every hour on the
model grid and not continuously in both time and space since it is suppose to mimic
the reading of GCM fields, which are stored only every hour.
We do not know why we obtain clearly different and better results using TRAC-
MASS here compared to what (Fabbroni 2009) got with Ariane, since both codes,
Fig. 7.8 Comparison of solutions for inertial oscillations. The black curve describe the pure an-
alytical solution, the red, blue and green curves reflect the results from the time-stepping method
using 0, 10 and 1000 intermediate steps between two ‘GCM’ velocities. The purple curve depicts
the results obtained using the analytical time integration method
we believe, should be based on the same method. A model bug on some level in
the Fabbroni (2009) experiment is one possible explanation unless Ariane is not as
similar to TRACMASS as we have supposed.
7.6 Subgrid Turbulence Parameterizations

The trajectory solutions in the previous sections only include the implicit large scale
diffusion due to along-trajectory changes of temperature and salinity/humidity, and
by the GCM’s parameterization of turbulent mixing in the momentum equations.
These trajectories do not, however, explicitly represent subgrid scale turbulence.
There are two ways to incorporate a representation of subgrid-scale turbulence in
TRACMASS. One where an additional random velocity is added called the ‘turbu-
lence parameterization’ and one that adds a random displacement to the trajectory
position, which is named ‘diffusion’. These two subgrid turbulence parameteriza-
tions will be presented here.
7.6.1 Turbulence Parameterization
This scheme, which was introduced by Döös and Engqvist (2007), adds a fluctuation
u , v to the GCM-simulated velocity fields U , V . These fluctuations are expected to
somehow model the deviations of the trajectories from the exact ones owing to the
impact of subgrid turbulence, which is illustrated by Fig. 7.9. These are the instan-
taneous GCM velocities U, V , which are updated with the GCM output time step
and from which the trajectories are calculated when no subgrid parameterization is
added.
The turbulent velocities u , v are added to each horizontal grid-cell wall for each
trajectory calculation and changed at every trajectory time step t. The trajectories
Fig. 7.9 Schematic

illustration of the changed
particle position by the
subgrid turbulence
parameterization due to the
added random velocities u , v
are hence calculated with the TRACMASS code as it is, but with a velocity field,
u = U +u , that is somewhat shaken, resulting in a stirring of the trajectory particles.
The amplitude of the random turbulent velocity is proportional to the velocity
of the circulation model velocity U so that u = RU . Here R is a random number √
uniformly distributed between −a and a, with standard deviation equal to 3a.
This amplitude was set to the constant a = 1 in Döös and Engqvist (2007), but has
here been tuned to obtain a relative dispersion similar to that of the surface drifters.
The amplitude was furthermore adapted in Döös et al. (2011) so that the trajectory
time step t in the TRACMASS code did not affect the results. This was obtained
by setting a = κ/(t)1/3 . The best fit for an amplitude of the relative dispersion
similar to that of the surface drifters was obtained for κ = 160. Using this √ scheme
in practice we add a random noise with a standard deviation on the order of 3aσu ,
where σu is the Lagrangian standard deviation of the unperturbed velocity field.
The effect of this superimposed subgrid turbulence is clearly visible in Fig. 7.10,
where a particle cluster is traced with and without this subgrid parameterization.
The turbulence smoothes the trajectory positions and spreads them more evenly.
The stirred particles in Fig. 7.10b fill visibly regions where no particles were present
without subgrid turbulence in Fig. 7.10a.
7.6.2 Diffusion
This scheme adds a random displacement to the trajectory position in order to in-
corporate a subgrid parameterization of the non-resolved scales as illustrated by
Fig. 7.11. The scheme was introduced in TRACMASS in Levine (2005) and tested
in a relative dispersion study (Döös et al. 2011).
The horizontal advection-diffusion equation is
∂P ∂P ∂P
+U +V = ∇ · (AH ∇P ), (7.37)
∂t ∂x ∂y
where AH is the horizontal eddy viscosity coefficient. Equation (7.37) is equivalent
(see, e.g., Rupolo 2007) to the zeroth-order Markov process:
Fig. 7.10 A cluster of particles released and followed as trajectories until they exit the model
domain. The colour scale indicates time in hours from the release, with trajectories’ positions
plotted every hour. The black line is the mean position of the trajectory cluster as it evolves in time.
(a) without and (b) with subgrid turbulence parameterization. From Döös and Engqvist (2007)
Fig. 7.11 The added displacement due to diffusion. Left panel shows in blue the original trajectory
and in light blue the changed one due to the added displacement. Right panel zooms in on the added
random displacement, where R is defined by Eq. (7.42)
dxi dw
= Ui + 2AH . (7.38)
dt dt
√ impulse is represented by the increment dη = (xd + yd ) . It

Here the stochastic 2 2 1/2
equals to dη = 2AH dw, where w is a Wiener process with a zero mean and a
second order moment dw · dw
= 2dt. The corresponding Gaussian distribution is
2
1 xd + yd2
P (xd , yd , t) = √ exp − . (7.39)
2πAH t 2AH t
Figure 7.11 illustrates the displacements added to the original position of the
particle after each time step of length t. The added random walk for the particles
is given by

xd = −AH t log(1 − q1 ) cos 2πq2 , (7.40)

yd = −AH t log(1 − q1 ) sin 2πq2 , (7.41)

zd = −Av t log(1 − q3 ) cos 2πq4 . (7.42)
Here AH and Av are the horizontal and vertical eddy viscosity coefficients and qn
are random numbers between 0 and 1. The added displacement in the horizontal and
vertical planes will hence be respectively

rH = xd2 + yd2 = −tAH log(1 − q1 ),
(7.43)
rV = −tAv log(1 − q3 ),
with horizontal and vertical standard deviations that are respectively

RH = tAH and RV = tAv . (7.44)
This implies that about 70 % of the particles will be within this distance from
their original positions and that the new velocity field will be characterized by an
extra standard horizontal deviation on the order of (AH /dt)1/2 , where dt is the
Lagrangian integration time step.
It is important to distinguish between this subgrid parameterization of the hori-
zontal and vertical mixing of the Lagrangian trajectories and that of the GCM itself.
The velocity fields are generally simulated by the GCM with some sort of Laplacian
diffusion. The mixing is hence included in a trajectory as it progresses and changes
its tracer properties by contact with its surroundings (Koch-Larrouy et al. 2008).
On the one hand one could therefore argue that adding a component to this velocity
field would be redundant since the mixing has already been included in the GCM.
These trajectories in themselves do not, however, explicitly represent subgrid-scale
turbulent motion since they are passively advected by the model-simulated currents
with no subgrid scales apart from the linear interpolations of the velocities between
the grid points. On the other hand, Lagrangian trajectories are the equivalent of in-
tegrating Eq. (7.37) with no effects of velocity scales under the grid scale, which
clearly must exist in the real ocean. Furthermore when Eq. (7.37) is discretized and
integrated in an OGCM for the tracers it will also include the numerical diffusion,
which is not the case for our trajectories since they are exact analytical solutions to
the velocity fields in TRACMASS. It is however important to note that we can only
evaluate or validate the OGCM itself when we do not add any subgrid parameteri-
zation to the model trajectories.
7.6.3 Subgrid Parameterization Questions
Döös et al. (2011) compared the relative dispersion of 5854 pairs of surface drifters
with that of simulated TRACMASS trajectories. The coefficients were tuned in or-
der to match the magnitude of the relative dispersion of the surface drifters after
32 days. The ‘diffusion’ parameterization, which adds a stochastic term to the tra-
jectory in accordance with Eq. (7.38), attains realistic relative dispersion rates for
AH = 2500 m2 /s. By calibrating the amplitude of the extra horizontal turbulent
velocities u , v (cf. Appendix B of Döös et al. 2011), also the turbulence param-
eterization reaches realistic values. The absolute dispersion is not much affected
when the diffusion parameterization is added, but gives far too high values for the
‘turbulence’ subgrid parameterization. The modelled trajectories with added diffu-
sion/turbulence also manifest values of the residual velocities which are similar to
real data, but with decidedly smaller values of the Lagrangian correlation time. In
other words, realistic particle separation rates are obtained using a large diffusivity
value, but at the cost of totally changing correlation properties and energy partition-
ing in the frequency domain.
A more realistic representation of the unresolved scales would require a higher
order subgrid parameterization. Griffa (1996) showed that a random walk does
not describe the turbulent dispersion behaviour of ocean tracers and that a better
quantitative agreement can be reached using an Ornstein–Uhlenbeck process. This
work has been refined by Pasquero et al. (2001) who observed that the Ornstein–
Uhlenbeck model assumes Gaussian velocity distributions, while the ocean displays
exponential-like tails associated with the mesoscale dynamics (Bracco et al. 2000a).
Those tails are common to 2D turbulent flows (Bracco et al. 2000b) and to La-
grangian trajectories in an oceanic eddy-resolving model (Bracco et al. 2003). Based
on these similarities Pasquero et al. (2001) built a family of two-process stochastic
models that provided a better parameterization of turbulent dispersion in rotating
barotropic flows.
Berloff and McWilliams (2002) and Berloff et al. (2002) also explored in detail
the issue of (horizontal) stochastic parameterizations for oceanic flows, suggesting
an alternative model to the one of Pasquero et al. (2001). It is therefore to be ex-
pected that the zeroth-order Markov process used in the present study will not pro-
vide a good representation of the surface drifters. The relative dispersion rates can
hence only be tuned to match the total value at a particular moment. The shape of
the power spectrum of the modelled velocity without parameterizations is therefore
more realistic in its shape even if too weak.
7.7 Mass Transport and Lagrangian Stream Functions
The mass conservation of the TRACMASS schemes makes it possible to calculate

mass transports between different sections in the model domain. A particular water
or air mass can be isolated and followed as a set of trajectories between specific
Fig. 7.12 The Lagrangian

stream function discretization
on a grid box seen from
above, with the grid lengths
x and y. An example of
one trajectory passing
through so that the transport
through the walls is
y
Ti,j,k,n = Ti,j
x
−1,k,n = Tn and
y
Ti−1,j,k,n = Ti,j,k,n
x =0
Fig. 7.13 Schematic

illustration of how the
transport of two trajectories is
counted on each grid cell
wall. The orange dots
correspond to meridional
transport and the red dots to
vertical transport, which are
then summed in order to
compute the Lagrangian
stream functions
initial and final sections. Each trajectory, indexed by n, is associated with a volume
transport Tn given by the velocity, initial area, and number of trajectories released
(Fig. 7.12). During transit from the initial to the final section the volume transport
remains unchanged; the transport/velocity field is thus non-divergent, permitting
representation in terms of stream functions. The volume transport linked to each
trajectory is inversely proportional to the number of trajectories released, viz the
Lagrangian resolution (which should be sufficiently high to ensure that the stream
function does not change when the number of trajectories is further increased).
A non-divergent 3-D volume-transport field is obtained by recording every in-
stance of a trajectory passing a grid-box wall (Fig. 7.13). Every trajectory entering
a grid box also exits, and hence this field exactly satisfies
y y
x
Ti,j,k,n − Ti−1,j,k,n
x
+ Ti,j,k,n − Ti,j −1,k,n + Ti,j,k,n
z
− Ti,j,k−1,n
z
= 0, (7.45)
x y z
where Ti,j,k,n , Ti,j,k,n and Ti,j,k,n , are the trajectory-derived volume transports in
the zonal (i), meridional (j ), and vertical (k) directions, respectively.
A Lagrangian stream function can be calculated by summing over trajectories
representing a desired path (Blanke et al. 1999). By integrating vertically over the
transport and over the trajectories one obtains the Lagrangian barotropic stream
Fig. 7.14 Lagrangian zonal overturning stream function in the Gulf of Finland decomposed with
water particle trajectories starting in the east at the exit of the River Neva (left panel) or at the
longitude 23◦ E from the Northern Gotland Basin (right panel). The green arrows show where
the particles have been released and the black thin arrows the direction of the flow. Contours of
500 m3 /s
LB :
function Ψi,j
y
LB
Ψi,j − Ψi−1,j
LB
= Ti,j,k,n LB
or Ψi,j − Ψi,j
LB
−1 = −
x
Ti,j,k,n . (7.46)
k n k n
By instead integrating zonally one obtains the Lagrangian meridional overturning

LM :
stream function Ψj,k
y
LM
Ψj,k − Ψj,k−1
LM
=− Ti,j,k,n or LM
Ψj,k − ΨjLM
−1,k =
z
Ti,j,k,n . (7.47)
i n i n
Finally by integrating meridionally one obtains the Lagrangian zonal overturning

LZ :
stream function Ψi,k
y
LZ
Ψi,k − Ψi,k−1
LZ
=− Ti,j,k,n LZ
or Ψi,k − Ψi−1,k
LZ
= z
Ti,i,k,n . (7.48)
i n i n
An example of a zonal Lagrangian stream function is shown in Fig. 7.14.

The indices i, j, k do not have to be the horizontal or vertical discretization of
the model grid. They can also be replaced by, e.g., temperature, salinity, density,
specific humidity, geopotential height or pressure.
7.8 Conclusion and Discussion

In this chapter we have presented the theory behind the trajectory model TRAC-
MASS by summarizing many articles, which have introduced new options and im-
proved TRACMASS. There are, however, still things that would be desirable to im-
prove or add. The TRACMASS subgrid parameterizations, which were introduced
in Levine (2005), Döös and Engqvist (2007), Döös et al. (2011) will need to be
ameliorated with a higher order Markov model (see, e.g., Rupolo 2007).
It would also be desirable to evaluate the precision of the different TRACMASS
schemes in more detail and compare them with other trajectory schemes such as
the Runge–Kutta scheme. Fabbroni (2009) compared Ariane (Blanke and Raynaud
1997), which is based on the same equations as the time step version of TRAC-
MASS with other trajectory schemes. The trajectories, simulated by Ariane, devi-
ated clearly from the analytical solution and the other trajectory schemes in her
study and Ariane was concluded not to be as accurate as the other schemes. In the
present study we repeated the Fabbroni (2009) test of inertial oscillations, with exact
analytical solutions. We found in contrast to her test that the TRACMASS scheme
gave nearly exactly the same results as the analytical solution. The TRACMASS
time-stepping method, which is comparable to the Ariane method, requires, how-
ever, that one uses sufficiently intermediate velocity time steps between the GCM
time steps. The TRACMASS time-stepping method, when using 1000 intermediate
time steps, gave almost exactly the same precision as the TRACMASS the method
of analytical time integration. From these tests, we would like to argue that the
TRACMASS schemes give at least as accurate trajectories as any other scheme and
it is hard to argue that it would be of any use to have even more accurate schemes for
geophysical fluid applications given all the missing physics and scales in a GCM.
A more detailed and quantitive study would, however, be necessary to measure this.
One of the major advantages of TRACMASS is that it is mass conserving and
now can calculate all sorts of mass transports between different sections in the
ocean or the atmosphere as well as Lagrangian stream functions for chosen wa-
ter/air masses. This can be particularly useful when performing analysis of, e.g.,
the inter-ocean exchange of water masses or the large scale atmospheric circulation.
TRACMASS has also turned out to be very useful in completely different applica-
tions such as studies of genetic connectivity, dispersion of radionuclides or identifi-
cation of transport patterns in the surface layer as in the present book. The number
of possible TRACMASS applications will certainly continue to grow in the future
as long as GCMs are based on finite differences.
Acknowledgements The authors wish to thank Tarmo Soomere and Ewald Quak for construc-
tive comments. This work was originally motivated by the BONUS+ project BalticWay that was
supported by the funding from the European Community’s Seventh Framework Programme (FP7
2007–2013) under grant agreement No. 217246 made with the joint Baltic Sea research and devel-
opment programme BONUS+ and by the Swedish Research Council for Environment, Agricultural
Sciences and Spatial Planning (Formas, Ref. No. 2008–1900).
References
Berloff P, McWilliams J (2002) Material transport in oceanic gyres. Part II: Hierarchy of stochastic
models. J Phys Oceanogr 32:797–830
Berloff P, McWilliams J, Bracco A (2002) Material transport in oceanic gyres. Part I: Phenomenol-
ogy. J Phys Oceanogr 32:764–796
Berrisford P, Kållberg P, Kobayashi S, Dee D, Uppala S, Simmons AJ, Poli P, Sato H (2011)
Atmospheric conservation properties in ERA-Interim. Q J R Meteorol Soc 137:1381–1399
Blanke B, Raynaud S (1997) Kinematics of the Pacific Equatorial Undercurrent: a Eulerian and
Lagrangian approach from GCM results. J Phys Oceanogr 27:1038–1053
Blanke B, Arhan M, Madec G, Roche S (1999) Warm water paths in the equatorial Atlantic as
diagnosed with a general circulation model. J Phys Oceanogr 29:2753–2768
Bracco A, LaCasce J, Provenzale A (2000a) Velocity probability density functions for oceanic
floats. J Phys Oceanogr 30:461–474
Bracco A, LaCasce J, Pasquero C, Provenzale A (2000b) The velocity distribution in barotropic
turbulence. Phys Fluids 12:2478–2488
Bracco A, Chassignet EP, Garraffo Z, Provenzale A (2003) Lagrangian velocity distributions in
a high resolution numerical simulation of the North Atlantic. J Atmos Ocean Technol 8:1212–
1220
Butcher JC (2008) Numerical methods for ordinary differential equations, 2nd edn. Wiley, New
York
de Vries P, Döös K (2001) Calculating Lagrangian trajectories using time-dependent velocity
fields. J Atmos Ocean Technol 18:1092–1101
Döös K (1995) Inter-ocean exchange of water masses. J Geophys Res—Oceans 100:13,499–
13,514
Döös K, Coward AC (1997) The Southern Ocean as the major upwelling zone of the North Atlantic
Deep Water, pp 3–17. WOCE Newsletter, No 27, July 1997
Döös K, Engqvist A (2007) Assessment of water exchange between a discharge region and the
open sea—a comparison of different methodological concepts. Estuar Coast Shelf Sci 74:585–
597
Döös K, Nycander J, Coward AC (2008) Lagrangian decomposition of the Deacon Cell. J Geophys
Res—Oceans 113:C07028
Döös K, Rupolo V, Brodeau L (2011) Dispersion of surface drifters and model-simulated trajecto-
ries. Ocean Model 39:301–310
Drijfhout S, de Vries P, Döös K, Coward A (2003) Impact of eddy-induced transport of the
Lagrangian structure of the upper branch of the thermohaline circulation. J Phys Oceanogr
33:2141–2155
Engqvist A, Döös K, Andrejev O (2006) Modeling water exchange and contaminant transport
through a Baltic coastal region. Ambio 35:435–447
Fabbroni N (2009) Numerical simulations of passive tracers dispersion in the sea. Alma Mater
Studiorum—Universita di Bologna, PhD Thesis, 164 pp
Griffa A (1996) Applications of stochastic particle models to oceanographic problems. In: Adler
RJ, Müller P, Rozovoskii BL (eds) Stochastic modelling in physical oceanography. Birkhäuser,
Basel, pp 114–140
Jönsson B, Lundberg P, Döös K (2004) Baltic sub-basin turnover times examined using the Rossby
Centre Ocean model. Ambio 33:2257–2260
Kjellsson J, Döös K (2012) Lagrangian decomposition of the Hadley and Ferrel Cells. Geophys
Res Lett 39:L15807
Koch-Larrouy A, Madec G, Blanke B, Molcard R (2008) Water mass transformation along the
Indonesian throughflow in an OGCM. Ocean Dyn 58:289–309
Levine RC (2005) Changes in shelf waters due to air-sea fluxes and their influence on the Arctic
Ocean circulation as simulated in the OCCAM global ocean model. University of Southampton,
Faculty of Engineering Science and Mathematics, School of Ocean and Earth Science, PhD
Thesis, 225 pp
Marsh R, Megann AP (2002) Tracing water masses with particle trajectories in an isopycnic-
coordinate model of the global ocean. Ocean Model 4:27–53
Pasquero C, Provenzale A, Babiano A (2001) Parameterization of dispersion in two-dimensional
turbulence. J Fluid Mech 439:279–303
Rupolo V (2007) Observing turbulence regimes and Lagrangian dispersal properties in the ocean.
In: Griffa A, Kirwan AD, Mariano AJ, Özgökmen TM, Rossby T (eds) Lagrangian analysis and
prediction of coastal and ocean dynamics (LAPCOD). Cambridge University Press, Cambridge,
pp 423–479
Simmons AJ, Burridge DM (1981) An energy and angular-momentum conserving vertical finite-
difference scheme and hybrid vertical coordinates. Mon Weather Rev 109:758–766
Soomere T, Delpeche N, Viikmäe B, Quak E, Meier HEM, Döös K (2011) Patterns of
16(Suppl A):49–63
Chapter 8
Evaluation and Tuning of Model Trajectories
and Spreading Rates in the Baltic Sea Using
Surface Drifter Observations
Joakim Kjellsson, Kristofer Döös, and Tarmo Soomere
Abstract Results from experiments with surface drifters in the Baltic Sea in 2010–
2011 are presented and discussed. In a first experiment, 12 SVP-B (Surface Velocity
Program, with Barometer) drifters with a drogue at 12–18 m depth were deployed
in the Baltic Sea. In a second experiment, shallow drifters extending to a depth
of 1.5 m were deployed in the Gulf of Finland. Results from the SVP-B drifter
experiment are compared to results from a regional ocean model and a trajectory
code. Differences between the observed SVP-B drifters and simulated drifters are
found for absolute dispersion (i.e., squared displacement from initial position) and
relative dispersion (i.e., squared distance between two initially paired drifters). The
former is somewhat underestimated since the simulated currents are neither as fast
nor as variable as those observed. The latter is underestimated both due to the above-
mentioned reasons and due to the resolution of the ocean model.
For the shallower drifters, spreading in the upper 1–2 m of the Gulf of Finland
is investigated. The spreading rate is about 200 m/day for separations <0.5 km,
500 m/day for separations below 1 km and in the range of 0.5–3 km/day for sep-
arations in the range of 1–4 km. The spreading rate does not follow Richardson’s
law. The initial spreading, up to a distance of about d = 100–150 m, is governed
by the power law d ∼ t 0.27 whereas for larger separations the distance increases as
d ∼ t 2.5 .
J. Kjellsson (B) · K. Döös

Department of Meteorology, Bolin Centre for Climate Research, Stockholm University, Svante
Arrhenius väg 16C, 106 91 Stockholm, Sweden
e-mail: joakim@misu.su.se
K. Döös
e-mail: doos@misu.su.se
T. Soomere
T. Soomere

DOI 10.1007/978-3-319-00440-2_8,
252 J. Kjellsson et al.
8.1 Background
Many studies of the ocean rely on Lagrangian trajectories. Knowledge of the ori-
gin and destination of a water particle, as well as the spreading of several initially
closely located particles, is necessary for a number of purposes, for example esti-
mating the fate of oil spills (Soomere et al. 2010) or living organisms (Corell et al.
2012), as well as for planning rescue operations or finding lost items. Trajectories
can be either observed using drifters or floats, or simulated using a computer model
of the ocean and a trajectory algorithm. Model-simulated trajectories may be used
to track entire water masses (Döös 1995; Blanke and Raynaud 1997; Döös et al.
2004), or to map transport and dispersion in the ocean (Pizzigalli et al. 2007). As
these studies become more frequent, the need for evaluating the modelled results
against observations is continuously growing.
In the World Ocean, several studies have used surface drifters or floats to vali-
date model-simulated trajectories. These studies have covered, e.g., the North At-
lantic (Garraffo et al. 2001; McClean et al. 2002; Lumpkin et al. 2002), the Pacific
(Garfield et al. 2001), and the global ocean (Döös et al. 2011). Although they have
employed different models, and different sets of Lagrangian observations, a com-
mon conclusion is that the squared displacement from the initial position (the abso-
lute dispersion) shows a fair agreement between models and observations. Discrep-
ancies between them are often found when studying the squared distance between
two initially paired drifters (relative dispersion), and/or the variability of the currents
(eddy kinetic energy). Relative dispersion and/or eddy kinetic energy is found too
low1 in models of resolution 1/4◦ –1◦ , thus the simulated drifters do not separate as
much as the observed ones (Lumpkin et al. 2002; McClean et al. 2002; Döös et al.
2011).
Discrepancies between models and observations can partly be attributed to the
coarse model resolution, which does not take turbulence on small scales into ac-
count, and implies a need for parameterizing subgrid-scale motions (Döös et al.
2011; Griffa et al. 2004). For example, an extensive analysis of the performance of
six circulation models was performed for the Gulf of Finland (Myrberg et al. 2010).
Errors in the wind forcing can also explain some of the differences in some models,
e.g., Keevallik and Soomere (2010) highlighted systematic bias between modelled
and measured wind directions and air flow properties in the central part of the Gulf
of Finland.
To be able to implement realistic parameterizations of subgrid turbulence it is
crucial to understand the circulation on small scales. For the Baltic Sea, there have
been several studies using model-simulated trajectories (Döös et al. 2004; Soomere
et al. 2010; Corell et al. 2012), but very little observational Lagrangian data. To
the authors’ knowledge, there has been only one experiment using Surface Velocity
Program (SVP) drifters similar to the ones used in this study (Håkansson and Rahm
1 The notion ‘too low’ is used here to denote the situation when a modelled quantity is systemati-
cally smaller than its values estimated from measurements. Similarly, the notion ‘too high’ or ‘too
large’ denotes the case when a modelled quantity systematically exceeds its measured values.
8 Trajectories and Spreading of Observed and Simulated Drifters in the Baltic Sea 253
1993; Launiainen et al. 1993). These drifter data sets have not been used to evaluate
the accuracy of any ocean circulation model. This lack of observations can, at least
partly, be explained by the small horizontal extent over which the depth of the Baltic
Sea exceeds the depth of the SVP drifter drogue (which sits between 12 and 18 m
depth), and by the heavy ship traffic. The risk of a surface drifter getting caught up
in too shallow waters or colliding with a ship is likely higher in the Baltic Sea than
in the open ocean.
For the Baltic Sea, only a few short-term drifter experiments have been per-
formed in the uppermost 1–2 m thick layer. For instance, shallow drifters have been
used to validate the output of the High Resolution Operational Model of the Baltic
Sea (HIROMB) and to study ice drift in the Gulf of Finland (Kõuts et al. 2010).
Also, experiments targeting at the validation of the Seatrack Web oil spill model
were performed with the same drifters in the middle of the Gulf of Finland, western
Estonian archipelago and in the eastern sector of the Northern Gotland Basin (Ver-
jovkina et al. 2010). Their duration ranged from 8 hours up to 7 days. The longest
distance covered by a drifter during a single experiment was 52 nautical miles.
This chapter presents an overview of results of experiments with SVP surface2
drifters and shallow (∼1.5 m deep) drifters (called GPS/GSM drifters below) de-
ployed in the Baltic Sea in 2010–2011. The advection of SVP drifters has also been
simulated using a regional ocean model and a trajectory code. The drifter obser-
vations and model trajectories were compared using several statistical measures.
Results from the comparison may then be applied to tune the algorithms used to
simulate the trajectories to obtain a better fit to observations. The realism of the
original and tuned model trajectories is discussed, as well as the implications for
Lagrangian modelling in the Baltic Sea. We also present results of a series of exper-
iments with the GPS drifters in the Gulf of Finland that were designed to quantify
the spreading of floats’ pairs in the uppermost 1–1.5 m thick layer. This behaviour
is discussed in the context of predictions from the turbulence theory.
8.2 Surface Drifters in the Baltic Sea

Twelve SVP-B (Surface Velocity Program, with barometer port) drifters (Lumpkin
and Pazos 2007) were deployed in the Baltic Sea in 2010–2011 (Kjellsson and Döös
2012). Their drift was tracked over the time interval of 14 July 2010–19 November
2011. The drifters were manufactured by Marlin–Yug Ltd. in Sevastopol, Ukraine,
and are approved by the Naval Oceanographic Office (NAVOCEANO) to comply
with the World Ocean Circulation Experiment (WOCE) type. SVP-B drifters are
equipped with a surface buoy containing GPS-sensors for measuring the position,
sea surface temperature and atmospheric pressure, and a system to transmit the data
to the Argos or Iridium satellites. Attached to the surface buoy is a holey sock an-
chored between 12 and 18 m depth. This configuration allows for the drifters to
2 They actually follow subsurface currents at depths 12–18 m.

Fig. 8.1 (a) The 12 SVP-B surface drifters used in this study mapped on top of the bathymetry
(Kjellsson and Döös 2012). Dots indicate the start of a segment of the drifter’s trajectory after every
256 hours. Surface drifters are given a letter corresponding to the deployment event. A and B are
the two pairs deployed in 2010, C and D the two triplets in 2011, and E is the pair deployed in
2011 in the Gulf of Finland. (b) Simulated segments of SVP drifters’ trajectories. 36 simulated
segments were released for each SVP drifter segment except those during the winter (December—
January–February) season. Depth <18 m is indicated by cyan shading
mostly follow subsurface currents at these depths. A schematic picture of the SVP
drifters can be found in Lumpkin and Pazos (2007). Data—including the state of
the drifter—are transmitted once an hour. The position data retrieved from the Ar-
gos and Iridium satellites allows for the identification and elimination of any (occa-
sional) erroneous GPS positions. Our data set did not contain any erroneous posi-
tions.
Two pairs of SVP drifters were deployed in the Baltic Sea in July and August
2010 (Fig. 8.1a and Table 8.1). The deployment point was chosen to be near the
major fairway where an oil spill could potentially occur. A triplet of drifters was
then deployed at the same point in June 2011. As four of these seven SVP drifters
got lost (apparently owing to collisions with ships) within a relatively short time
period, another triplet was deployed at some distance to the northwest of the pre-
vious location in August 2011. Two of the drifters from the two pairs stranded and
were then re-deployed as a third pair in the Gulf of Finland in November 2011. All
deployments, except the last pair, were made from the ferry Silja Festival on her
regular crossing from Stockholm to Riga.
Drifters of a different kind were deployed in the Gulf of Finland to study pair
separations in the uppermost layer (Soomere et al. 2011). These drifters were not of
the WOCE standard. They were fixed within the uppermost ∼1.5 m thick layer. The
active component (a high sensitivity (−159 dB) GPS/GSM device CT-24, Sanav
Corp., Taiwan) of these lightweight floating buoys reported its position 4 times an
hour. The device was mounted on the top of a 2 m long and 50 mm in diameter
Table 8.1 The first transmitted positions (∼ release positions) of all 12 SVP surface drifters, and
their lifetime
Drifter Initial Lon/Lat Start date Lifetime
#1 20.6984◦ E/58.3559◦ N 14 July 2010 96 days

#2 20.6988◦ E/58.3562◦ N 14 July 2010 102 days
#3 20.6967◦ E/58.3568◦ N 17 August 2010 317 days

#4 20.6971◦ E/58.3567◦ N 17 August 2010 11 days
#5 20.7002◦ E/58.3539◦ N 9 June 2011 22 days

#6 20.6976◦ E/58.3532◦ N 9 June 2011 63 days
#7 20.6987◦ E/58.3539◦ N 9 June 2011 25 days
#8 19.8129◦ E/58.8452◦ N 10 August 2011 101 days

#9 19.8128◦ E/58.8447◦ N 10 August 2011 100 days
# 10 19.8096◦ E/58.8462◦ N 10 August 2011 101 days
# 11 24.7421◦ E/59.6804◦ N 7 November 2011 12 days

# 12 24.7415◦ E/59.6796◦ N 7 November 2011 12 days
plastic pipe, about 2/3 of which was submerged and about 1/3 (60 cm) was above
the water surface. A picture of one of the drifters can be found in Soomere et al.
(2011). Three deployments were made with altogether 8 drifters. Each time drifters
were deployed at a distance of about 50–150 m from each other and let to drift from
a few days to a few weeks. Two deployments took place about 8 km west of the
island of Naissaar and one in Muuga Bay.
8.3 Simulated Drifters
Simulated drifters were computed using the Lagrangian trajectory code TRACMASS
(see Chap. 7 and Döös 1995; Blanke and Raynaud 1997) and velocity fields from the
Rossby Centre regional Ocean climate (RCO) model (Meier et al. 1999, 2003). The
RCO model (Chap. 4) is a coupled ice–ocean circulation model, and a regionalized
version of the global Ocean Circulation and Climate Advanced Model (OCCAM)
(Webb et al. 1997). Here, the model grid covers the Baltic Sea with an open bound-
ary at Kattegat. The resolution is 1/15◦ × 1/30◦ , i.e., ∼2 nautical miles, with 41
vertical levels, where each layer in the upper 40 m is 3 m thick. RCO uses a k–ε
mixing scheme (Meier et al. 2003) to simulate effects of subgrid turbulence, which
applies to both velocity components (u, v) in the momentum equations and tracers
(temperature, salinity, etc.).
The ocean model RCO and the atmospheric model RCA (Kjellström et al.
2005) are both components of the regional coupled atmosphere–ocean model RCAO
(Döscher et al. 2002) that is used for simulating the future climate over Scandinavia.
In our data set, the RCO model has been run in hindcast mode using atmospheric
forcing from the ERA-40 reanalysis (Uppala et al. 2005) and observed river runoff.
The ERA-40 fields have been downscaled to the RCO model grid by using them
as input for the regional atmospheric model RCA. Winds have subsequently been
corrected to include some gustiness (Höglund et al. 2009). Several studies have val-
idated the RCO model output data against observations of temperature, sea surface
height, and salinity. Meier (2002) used data from four separate stations in the Baltic
Sea, and found that the model data agreed reasonably well. See Chap. 4 for a detailed
overview of many features of this model. In the present study, the output data of the
RCO model, including temperature, salinity, and three-dimensional (3D) velocity,
were available every 6 hours for June 1961–May 2005.
The TRACMASS trajectories are computed off-line, that is, after the fields from
the RCO model have been integrated and stored. This allows for faster and less
memory consuming computations. A thorough discussion of the pro’s and con’s of
the on-line and off-line methods of trajectory calculations is presented in Chap. 7.
The simulated trajectories were not fully Lagrangian: to mimic the motion of the
drifters, only horizontal components of the velocities were used to calculate the
advection of the particles. The velocity of advection was evaluated as a weighted
average of the modelled currents for depths between 12 and 18 m. The simulated
drifters were locked at the depth of 15 m. To simulate (the impact of) drifter strand-
ing, any model drifter that at some time instant reached a depth shallower than 18 m
was considered as stranded and the relevant data was removed from the statistics.
The TRACMASS code includes tunable parameterizations of subgrid-scale tur-
bulence and diffusion to imitate subgrid-scale motions (Döös and Engqvist 2007;
Döös et al. 2011). The turbulence scheme adds a random perturbation to the veloc-
ity fields, while the diffusion scheme adds a random perturbation to the position. In
the zonal direction, the impact of turbulence is included as
unew = uorig + uturb , (8.1)

1
uturb = κ (q − 0.5)uorig , (8.2)
(tmin )1/3
where q is a random number between 0 and 1. The same value of q is used at both
the eastern and western grid box walls. The quantity tmin is the time until the
trajectory has moved through the grid box or until the velocity fields are updated
(every hour) and may therefore be individual for each trajectory (see Chap. 7 for
details). There is thus no uniquely defined time step of the ‘upgrade’ of the velocity
or position in the parameterization of turbulence or diffusion. The random increment
added by the turbulence scheme is proportional to the mass flux through the grid
box, the time step, a random number, and a parameter κ. The equations are similar
in the meridional direction.
The value of the parameter κ was set by simulating trajectories with no turbu-
lence parameterization and comparing the results to observed SVP drifters. An em-
pirical value can then be estimated by simulating trajectories using different κ and
comparing to observed drifter trajectories.
The scheme is a low-order one in the sense that it does not take into account
properties such as Lagrangian time scales or Lagrangian velocity autocorrelation.
Schemes like this are often called ‘Markov 0’ processes (Rupolo 2007). Several
more advanced processes of ‘Markov 1’ or ‘Markov 2’ type have been tested in
other studies but never used with the TRACMASS code, and thus are not used in
this study. The possible advantages of using Markov 1 or Markov 2 models will be
discussed in Sect. 8.9.
8.4 Lagrangian Statistics

Absolute dispersion is a measure of the square of travelled distance from the origin,
i.e., the trajectory length, as a function of time. An average of this quantity over M
trajectories is here evaluated by integrating the velocity
M 2
1

DA (tn ) ≡
2
ui,m (tn ) dt , (8.3)
M
m=1
where tn is the time (discrete steps), dt is the time step (1 hour in experiments with
the SVP drifters), m is the trajectory number, and i indicates which velocity compo-
nent is used. The turbulent absolute dispersion, DA 2 (t ) is found by integrating the
n
turbulent velocity u = u − u, where u is a time average of the trajectory, in a similar
fashion. The mean displacement is defined as the displacement from the origin as a
function of time

M 2
DD (tn ) ≡
1 x (t ) − x (0)2 , (8.4)
i,m n i,m
M
m=1 i=1
where t = 0 is associated with the beginning of a trajectory segment.

Relative dispersion is often defined as the square of the distance from the mean
position at a certain time. However, here it is defined as half of the pair separation,
i.e., the square of half of the distance between two drifters at a given time step, which
is equal to the squared distance from the mean position of the two drifters. With the
same notations as for the absolute dispersion, the average relative dispersion over P
pairs is defined as
2
1 di,p (tn ) 2
P
DR (tn ) ≡
2
, di,p (tn ) = xi,q (tn ) − xi,r (tn ), (8.5)
P 2
p=1 i=1
where di,p is the pair separation and p is the pair consisting of drifters r and q. The
square of the separation ensures positive values.
The Lagrangian velocity is obtained by using a non-centred finite difference
scheme
dxi,m (tn ) xi,m (tn ) − xi,m (tn−1 )
ui,m (tn ) ≡ ≈ , (8.6)
dt tn − tn−1
with the indices having the same meaning as above. Similarly, the acceleration was
calculated using a similar finite difference scheme applied to the particular velocity
component:
dui,m (tn ) ui,m (tn ) − ui,m (tn−1 )
ai,m (tn ) ≡ ≈ . (8.7)
dt tn − tn−1
Note that velocity is not defined at the first time instant (or position), and accelera-
tion is not defined at the first and second time instant.
The Lagrangian velocity autocorrelation describes the correlation of the velocity
at one time with that of previous times:
σ 2 (τ ) σq2
R(τ ) = ≈ R q = , (8.8)
σ 2 (τ = 0) σ02
where σ 2 (τ ) and σ 2 (τ = 0) are the Lagrangian velocity autocovariances for time
lag τ and no lag (τ = 0), respectively. Here q is the discrete time step and Rq is the
autocorrelation at time step q. The quantity σ 2 (τ ) is defined as
T
2 −q−1
N
1 1
σ (τ ) = lim
2
u (t + τ ) · u (t) dt ≈ σq2 ≡ u i,n u i,n+q ,
T →∞ T 0 N −q −1
i=1 n=1
(8.9)
where u i,n = ui,n − ui and ui is a time average of the segment. Note that the total
velocity autocovariance is the sum of its zonal and meridional components: σ 2 =
2 + σ2 .
σi=1 i=2
Using the autocorrelation R(τ ), the Lagrangian integral time scale TL is defined
as
t1
TL = R(τ ) dτ. (8.10)
0
This is a measure of the memory of a trajectory, that is, the time lag during which the
Lagrangian velocity is correlated. When computing this integral, the upper bound,
t1 , is the point where R(τ ) = 0 occurs for the first time. This truncation is perhaps
the most commonly used one, due to the often noisy character of the autocorrelation
function R(τ ) for large τ . Lumpkin et al. (2002) compared this choice with several
other approximations, and found that all approaches produced essentially the same
results. Thus it is natural to assume that the approximation used here is a robust one.
8.5 Results
8.5.1 The Surface Drifters
The 12 drifter trajectories were of different length because of variable drifter life-
time. The mean drifter lifetime was ∼80 days (Table 8.1). For this reason, each
Fig. 8.2 Pair separation

(distance between drifters) for
each individual drifter pair
drifter trajectory was split up into shorter segments of fixed length (called SVP seg-
ments or simply segments below). To make the Fast Fourier Transformation (FFT)
calculations more convenient, the segments’ length was taken as a power of two,
28 = 256 hours. It was important that the segments (i) were long enough for con-
secutive segments of each trajectory to be uncorrelated (this property will be dis-
cussed later) and (ii) yield a reasonably long time series of single-particle statistics.
They also needed to be shorter than the shortest SVP drifter trajectory (∼11 days).
This resulted in 85 segments altogether. From our viewpoint of statistical properties
of dispersion these segments can be interpreted as trajectories of 85 independent
drifters.
Tracks of the SVP drifters are shown in Fig. 8.1a. The drifter deployments
yielded 9 pairs (Fig. 8.2). It should be noted that the pair separation (related to rela-
tive dispersion by Eq. (8.5)) grew at very different rates and that the pairs had very
different lifetimes. For instance, the pair ‘E1 & E2’ dispersed rapidly after 5 days,
while the pair ‘D1 & D3’ stayed essentially paired for more than 20 days.
Velocities at each time step, except for the last one, were calculated for each SVP
drifter from Eq. (8.6). The drifter velocities show strong currents near Poland, west
of Estonia and west of Gotland (Fig. 8.3a). The Lagrangian time scales computed
for each segment from Eq. (8.10) are shown in Fig. 8.3b. In this case, the turbulent
velocities were calculated by subtracting the time-averaged velocity of the segment,
not the full trajectory. A comparison of Figs. 8.3a and 8.3b shows that regions of
high velocity tend to have higher frequencies of variability, and thus shorter La-
grangian velocity time scales. Note that the relevant time scales are ∼1–2 days,
similar to those calculated from SVP drifters in other parts of the World Ocean
(Rupolo 2007), and much shorter than the segment length.
The SVP drifters also collected data of sea surface temperature (Fig. 8.4) and
atmospheric pressure. During summertime temperatures reached ∼24 °C, while de-
scending to near 0 °C in winter. The only SVP drifter surviving the winter always
measured temperatures above 0 °C. By using available ice charts it was verified that
the drifter avoided ice-covered waters.
Fig. 8.3 (a) SVP drifter trajectories coloured by their total Lagrangian velocities. The colour range
is from 0 to 0.4 m/s. (b) SVP drifter trajectories coloured by the Lagrangian integral time scales.
Calculations of Lagrangian time scales yielded one value per drifter segment (256 hours). Colours
range from 0 to 2 days
Fig. 8.4 The sea surface

temperature measured by
each SVP drifter. Horizontal
time axis is days from the first
deployment, 14 July 2010, up
to December 2011 when the
data collection was stopped.
The SSTs were constantly
above 0 °C, suggesting
ice-free conditions
8.5.2 Model Evaluation
The years 2010–2011 were not available in our RCO data set and thus a direct com-
parison of measured and modelled statistical features of drift was not possible. Also,
the first and last years (1961 and 2005) in the RCO data set were incomplete and
could therefore not be used. Statistics extracted from the SVP drifter segments was
therefore compared to statistics derived from the motion of simulated drifters in
1962–2004. This comparison is not exactly straightforward because the SVP drifter
data in 2010–2011 were collected under ice-free conditions but large parts of the
Baltic Proper3 and the Gulf of Finland have been frozen during some winters in
1962–2004. The presence of ice definitely impacts both the modelled and real veloc-
ity fields but in our comparison the related effects only become evident for simulated
drifters (assuming that the RCO model simulated the sea ice correctly). Including
simulated winter drifters in some ice-free years while excluding those from years
of a frozen Baltic Sea would give extra weight to some years in the statistics and
would thereby bias the comparison. For this reason, all segments recorded by the
SVP drifters during the period December 2010–February 2011 were removed from
the data, leaving 76 drifter segments for the comparisons with the modelled data.
The starting longitude and latitude as well as hour, day, and month for each of the
76 drifter segments were used as starting points for simulated drifters in each full
model year (1962–2004). Thus, 36 simulated drifters were released around each
starting point in each model year, resulting in 36 × 76 = 2736 segments for each
model year. Four of the 36 simulated drifters around each starting point originated
in the same grid box as the surface drifter segment, while the others were spread
horizontally in the eight adjacent grid boxes.
As each grid box was about two nautical miles wide, the described process con-
verted each starting point of a SVP drifter segment into a ‘cloud’ (with a radius of
about 3 nautical miles or ∼5.5 km) of model drifters. Doing so made it possible to
take into account the natural subgrid-scale variability of currents around the starting
point of each drifter. The large number of simulated drifters also resulted in clearer
statistics. In an additional experiment, the number of simulated drifters was dou-
bled, resulting in no significant difference in the statistical parameters. It was thus
concluded that the set of 36 simulated drifters per each SVP drifter formed a suffi-
cient pool of samples to calculate the necessary statistics. No turbulence or diffusion
parameterization was used at this point. In order to better replicate the conditions of
the motion of SVP drifters, the advection of the simulated ones was calculated using
the horizontal velocities at 12–18 m depth, with no vertical velocity. Moreover, only
simulated drifters that stayed for 256 hours in waters deeper than 18 m were used in
the comparison. Simulated drifter segments from the year 1962 (Fig. 8.1b) covered
similar sea regions as the SVP drifter segments but their separation rate was smaller.
The positions of simulated drifters were stored every hour to have the same tem-
poral resolution as for the SVP drifters. However, the velocity fields from the RCO
model were only available every 6 hours. Hence, variations on time scales shorter
than 12 hours were not resolved by the model, and time scales slightly longer were
poorly resolved. The average velocity power spectrum of all SVP drifter segments
(Fig. 8.5, left panel), shows a peak near the frequency of 2 cycles per day, corre-
sponding to a period of ∼14 hours. This peak obviously reflects the presence of
inertial oscillations (see Chap. 2, Sect. 2.3.4 for details).
While this peak was very pronounced for the SVP drifters, it was also visible
for the pool of all simulated drifters in each model year but in this case appeared
as a double peak. This distortion is most likely an effect of the inertial oscillations
3 We use here the notion Baltic Proper to denote the Eastern, Northern and Western Gotland Basin,
Bornholm Basin and Gdańsk Bay (Chap. 2, Fig. 2.1).

Fig. 8.5 Power spectra of the Lagrangian velocity for SVP drifters (thick black line) and simulated
drifters in each year 1962–2004 (thin coloured lines) for the unfiltered data (left) and with a 14-hour
running mean applied to both SVP and simulated drifter positions (right)
being poorly resolved, leading to aliasing (errors in the frequency). It is very likely
that similar distortions occurred in the statistics of trajectories simulated using the
RCO velocity fields in Chaps. 9 and 10. In order to remove the distorted part of
statistics, a 14-hour running mean was applied to the drifter positions of both the
SVP and modelled drifters before further processing of the data. After filtering out
the inertial oscillations the power spectra for the SVP and simulated drifters were
quite similar (Fig. 8.5, right panel).
The mean displacement was calculated for both SVP drifter segments and sim-
ulated drifters using Eq. (8.4). The procedure yielded one time series for all SVP
drifter segments and one for each model year (1962–2004) (Fig. 8.6c). The total
and turbulent components of absolute dispersion (Eq. (8.3)) were also calculated
(Figs. 8.6a, b). The absolute dispersion increased with time during the entire dura-
tion of the segments, while the mean displacement started to level off at the end of
the segments (∼10 days). This indicates that the horizontal extent over which the
mean depth is >18 m in the Baltic Sea is rather small, and that drifters can only drift
for ∼40 km until they become influenced by this.
The discrepancies between observed and simulated drifters in Fig. 8.6 indicate
that the simulated velocities are lower than the observed ones. Indeed, the distri-
bution of absolute velocities for all drifter segments, and for all simulated drifters
in each model year (Fig. 8.7) showed the latter to be narrower (thus exhibiting less
variability) and centred over lower values than the former. Note that there are far
more simulated drifters in a given model year than the 76 observed SVP drifter
segments. For this reason all the distributions are normalized to have the integral
equal to 1. The results thus suggest that the horizontal velocities in the RCO model
generally are lower and less variable than those of the SVP drifters.
Every trajectory is characterized by two Lagrangian velocity autocorrelation
functions, one for the zonal and another for the meridional velocity. The total ve-
locity autocorrelation is the average of the two. It was calculated and averaged over
all SVP drifter segments and also over all simulated drifters (Eqs. (8.8)–(8.9)) in
2 on a log–log scale; (b) abso-

Fig. 8.6 (a) Absolute dispersion by integrating the total velocity DA
2
lute dispersion by integrating the turbulent velocity DA ; (c) mean displacement. Solid black lines
indicate the ensemble mean for SVP drifter segments, solid coloured lines are ensemble means
for simulated segments for each year 1962–2004. Dashed black lines show the 10th and 90th per-
centiles for SVP drifter segments
Fig. 8.7 Normalized

distributions of the total
Lagrangian velocities
(Eq. (8.6)) for SVP drifter
segments (black line) and
simulated drifters in all model
years (colour lines). The
distributions of simulated
velocities are narrower and
displaced towards lower
values than the distribution of
SVP drifter velocities
each year (Fig. 8.8). Note that the velocity autocorrelation is normalized autoco-
variance (Eq. (8.9)) and that the autocovariance is calculated from u and v defined
as u = u − u and v = v − v (Rupolo 2007).
Fig. 8.8 Ensemble-mean Lagrangian velocity autocorrelation R(τ ) for SVP drifter segments
(thick black line) and for model trajectories in each model year (colour lines). Dashed lines show
the 10th and 90th percentiles of the SVP drifter segments. The horizontal coordinate is the time
lag τ ranging up to 8 days. Note that all trajectories, observed and modelled, were filtered using
a 14-hour running mean. The smallest τ for which R(τ ) = 0 is the upper limit of the integral in
Eq. (8.10)
As noted above, a good agreement between statistical characteristics of simulated

and SVP drifter segments was found only after the inertial oscillations had been
filtered out with a 14-hour running mean (Fig. 8.8) since the inertial oscillations
are poorly resolved in the RCO model output (Fig. 8.5). The Lagrangian integral
time scale TL was calculated by integrating the autocorrelations using Eq. (8.10),
where the lowest τ for which R(τ ) = 0 was used as upper limit of the integral (see
Lumpkin et al. 2002 for comments on this, and other, methods). This was done for
both zonal and meridional velocity autocorrelations. The total Lagrangian integral
time scale is defined as the average of the two time scales. Hence, there was one
time scale for each drifter segment.
The distributions of TL for all the SVP drifter segments and simulated drifters in
each model year were both centred over time scales ∼1 day (Fig. 8.9). The distri-
butions were normalized as there were more simulated drifters than observed ones.
There was good agreement between the drifter data and corresponding model re-
sults, largely attributable to the agreement in velocity autocorrelations. This can
be interpreted as indicating that the variabilities in the observed and simulated La-
grangian velocities have similar time scales.
The relative dispersion cannot be calculated from the SVP drifter segments
since the drifters need to be paired initially. For this reason, only the 12 com-
plete SVP drifter trajectories and the corresponding simulated drifters originating
from the starting points of these SVP trajectories were used. Both were filtered
with the 14-hour running mean as mentioned before. As each of the SVP drifter
pairs was modelled by 2 × 36 = 72 simulated drifters, and each of the triplets
using 3 × 36 = 108 simulated drifters, this yielded 9 pairs of SVP drifters, and
Fig. 8.9 Normalized

distribution of the total
Lagrangian integral time
scales calculated using the
autocorrelations in Fig. 8.8.
Shown are SVP drifter
segments (thick black line)
and simulated drifters
(thinner, coloured lines) in
each model year
>5000 simulated pairs for each model year. The SVP drifters were initially sepa-
rated by4 O(100) m, but the simulated drifters had an initial separation ranging from
O(100) m to O(1000) m as they were spread more widely around the SVP drifter
starting point. Initially, only modelled drifter pairs with initial separation of <1 km
(equivalently, DR2 (0) < 0.25 km) were used for the comparison with the SVP drifter
pairs.
The relative dispersion was evaluated as an ensemble average over all SVP drifter
pairs and also over all simulated drifter pairs for each model year. This quantity was
found to be much lower for the simulated pairs of trajectories than for the SVP
drifter pairs (Fig. 8.10). Part of this difference may be due to the fact that the RCO
model has a resolution of ∼5.5 km. Therefore, processes that lead to separation
on scales below this value are generally not resolved, and no parameterization of
turbulence or diffusion was used. Studies with an OGCM (Poje et al. 2010) have
shown that although DR2 increases with t, the values are generally lower at coarser
resolutions, which is consistent with Fig. 8.10. However, in this study, the growth
rates of DR2 (t) are different for SVP drifters and model drifters, which was not
the case when running the OGCM and merely varying the resolution. Hence, the
observed difference is due to more than just the coarse resolution of the RCO model.
To check whether or not the moderate resolution of the RCO model is the domi-
nant reason for a too low separation of the modelled pairs, the motions of simulated
pairs with an initial separation in the range of 4–12 km (approx. 1 to 3 grid boxes)
were also compared to the behaviour of pairs of SVP drifters (Fig. 8.11). The simu-
lated relative dispersion should then be compared to the relative dispersion of SVP
drifters after their distance has reached 4 km. This process takes about 9 days. Fig-
ure 8.11 shows the relative dispersion as a function of pair separation. Even when
the simulated pairs were separated by one grid box, the relative dispersion of mod-
4 The ‘Big O’ notation is commonly used to describe the limiting behaviour of a function f by com-
paring it to the behaviour of a simpler function g (Chap. 3). Here we use it to roughly characterize
the distance between drifters.
Fig. 8.10 (a) Mean pair separation. (b) Mean relative dispersion. The black solid line shows the
mean of all SVP drifter pairs, and the coloured lines show the mean of all simulated pairs in each
model year 1962–2004. Dashed lines show the 10th and 90th percentile of the SVP drifter pairs.
The initial separation of pairs of both drifters and model trajectories is O(100) m, that is, much
smaller than the model grid box width
Fig. 8.11 As Fig. 8.10, but an average is calculated of all model years 1962–2004 (red line). The
blue line shows the same average of model years, but with drifter pairs initially separated by >4 km
elled pairs did not agree with that of the SVP drifter pairs. This finding indicates
that the discrepancies are not only due to the limited resolution of the RCO model.
8.6 Tuning the Trajectories

As previously noted, the simulated velocities of drifters were generally lower and
less variable than the observed ones (Fig. 8.7). From Fig. 8.12 the mean displace-
ment DD can be estimated as ∼37 km after 256 hours for the SVP drifter segments.
The corresponding average of modelled drifters over all model years is ∼30 km af-
ter 256 hours. This indicates that the modelled mean displacement is generally only
4/5 of the observed one.
As a first approximation to resolve this discrepancy, all model velocities were
multiplied by 1.25 (since (30/37) × 1.25 ≈ 1) and the simulations of model tra-
jectories repeated. This resulted in the mean displacement and absolute dispersion

2 . (b) Turbulent absolute dispersion, D 2 . (c) Mean
Fig. 8.12 (a) Total absolute dispersion, DA A
displacement of SVP drifter segments (black), and for all simulated trajectories in all model years
1962–2004 (red). Also shown are the results when including parameterized subgrid turbulence
(blue), and when multiplying the modelled velocity fields by 1.25 (yellow)
of the simulated drifters shifted to values closer to those calculated from the drifter
segments (Fig. 8.12). This suggests that the simulated velocities were too low in the
original RCO simulation. Interestingly, this increase in the velocity magnitude by
25 % did not, however, yield more variability in the motion of modelled drifters.
In another simulation, we employed the turbulence scheme for TRACMASS as
introduced in Döös and Engqvist (2007), Döös et al. (2011). This scheme added
extra velocity to the simulated drifters, but not necessarily in the direction of the
modelled advection. All data was filtered with a 14-hour running mean as before.
The magnitude of the subgrid turbulence, controlled by the parameter κ in Eq. (8.2),
was tuned to get the values of mean displacement close to that of the SVP drifter
data. A fair fit was achieved for κ = 200 (Fig. 8.12). This estimate is close to the
value of κ used in Döös et al. (2011) for the open ocean conditions. For turbulent
absolute dispersion DA 2 using subgrid turbulence led to better match of the statistics
of modelled and SVP drifters than simply multiplying the velocities by 1.25.
For both above-mentioned methods, the distributions of Lagrangian integral time
scales, TL , are shown in Fig. 8.13. Note that TL is calculated using the deviations
from the time-averaged velocities, u and v defined as u = u − u and v = v − v.
The random motions introduced by the subgrid turbulence shortened the Lagrangian
integral time scales even though the data was filtered by a 14-hour running mean.
Fig. 8.13 Normalized distribution of the Lagrangian integral time scales for all SVP drifter seg-
ments (thick black line) and all simulated trajectories in each model year (thin colour line). Left
panel shows for model simulations with added subgrid turbulence of κ = 200, and the right shows
model simulations where the horizontal velocities at each point and time step are multiplied by
1.25
This did not occur when the velocities were simply multiplied by 1.25. The reason
is that the latter operation does not introduce any new motion and merely amplifies
the advection that is already present, while a ‘Markov 0’ model has no memory
since the turbulence is completely random. As mentioned above, there exist more
advanced turbulence parameterizations in other Lagrangian models that account for
velocity autocorrelation (e.g., ‘Markov 1’ and ‘Markov 2’ type). Such schemes are
not available yet in TRACMASS.
The relative dispersion calculated from the motion of trajectories simulated us-
ing the two above-described variations of the evaluation of velocities is shown in
Fig. 8.14. Multiplying the simulated velocities by a constant factor did not increase
the relative dispersion significantly compared to the data presented in Fig. 8.11.
Adding subgrid turbulence, however, resulted in an increase of both relative disper-
sion and pair separation. The resulting values of these quantities were closer to those
extracted from the SVP drifter data. For initially close drifters (DR2 (0) < 0.25 km),
adding the parameterization of subgrid turbulence resulted in a good agreement at
times >10 days (Fig. 8.14). Note, however, that the magnitude of the impact of sub-
grid processes used in this calculation was chosen to give a good fit for absolute
dispersion and mean displacement, and thus some differences in relative dispersion
are not unexpected.
8.7 Spreading Rates in the Uppermost Layer of the Gulf

of Finland
The separation of drifters can be also approximated by a power function or an expo-
nential law of the time t elapsed since the release of the particles. The spreading rate
Fig. 8.14 Mean pair separation (left) and relative dispersion (right) for SVP drifter pairs (thick
black line), and for all simulated drifters in all model years (colour lines). As in Fig. 8.11, two sets
of trajectories are shown for each simulation; those with initial pair separation <1 km (red, green,
orange) and those with >4 km (blue, cyan, purple). The simulations shown are the original (red,
blue), one with added subgrid turbulence (orange, purple) and one where all modelled velocities
were increased by 25 % (green, cyan). Multiplying the velocity fields by 1.25 does not alter the
relative dispersion, so the green and red as well as blue and cyan lines overlap
is then the rate of change of the separation distance with time (and thus related to rel-
ative dispersion, see Eq. (8.5)). The classical notion for the average distance between
two particles in a turbulent velocity field, Richardson’s law (Richardson 1926), ap-
plies in fully developed 3D turbulent flows where the average difference between
velocity fluctuations u(r, t) follows the (Kolmogorov’s) power law (Falkovich et al.
2001):

u(x, t) − u(x + x, t) = A|x|a . (8.11)
Here angle brackets denote averaging over the coordinate x = (x, y, z) and/or over
the ensemble of flows, A is a constant and the exponent a = 1/3 is specific to the
fully developed 3D turbulent flow. In such an environment, the average distance d
between a pair of particles scales as
1
d ∝ t b, where b = . (8.12)
1−a
In the case of Kolmogorov’s law a = 1/3, the corresponding exponent is b = 3/2.
The character of spreading is markedly different in two-dimensional (2D) flows
where, at scales smaller than the energy input scale, the velocity spectrum is domi-
nated by the enstrophy cascade and a = 1. In this situation the exponent b → ∞ in
Eq. (8.12) and an exponential growth of the distance with time (Lin’s law) occurs
(Lin 1972; Falkovich et al. 2001; LaCasce 2008). Thus, for an ideal 2D turbulence
with a single energy input scale λ, Lin’s law is expected to be valid for scales below
λ whereas Richardson’s law is related to large-scale circulation (Salazar and Collins
2009). Both these flow regimes have been observed using observed and simulated
drifters in the open ocean (Ollitrault et al. 2005) and in the Baltic Sea for different
scales (Döös and Engqvist 2007).
The spreading rate of pairs of drifters in the uppermost layer of the Gulf of Fin-
land was studied using surface drifters (called GPS/GSM drifters) following mo-
tions at a depth down to about 1.5 m (Soomere et al. 2011). The drifters were there-
fore constrained to 2D flow. However, they may be influenced by the 3D turbulence
of the ocean. Unlike for the SVP drifters, the inertial oscillations were not filtered
out. As the filter did little change to the relative dispersion for the SVP drifters, it
should not markedly influence the results from the GPS/GSM drifters either. Dif-
ferently from a number of similar studies (Ollitrault et al. 2005; Döös and Engqvist
2007; Lumpkin and Elipot 2010) the experiments performed in the western and
central part of the Gulf of Finland in August–October 2010 were concentrated on
relatively small initial separations of the drifters (∼100 m). The deployments re-
sulted in 7 pairs of drifter trajectories. As the signal from the drifters was lost at
times, it was not possible to adequately calculate the detailed statistics of the drift
but the recorded data still allowed quantification of the temporal evolution of the
pair separations.
The observed trajectories reflected a variety of phenomena characteristic of the
currents in the Gulf of Finland (Fig. 8.15): relatively small mesoscale eddies with a
diameter of about 5 km to the north of Naissaar, inertial oscillations in the open part
of the gulf, and relatively rapid and almost straight drift sections (cf. Kõuts et al.
2010; Verjovkina et al. 2010). While most of the trajectories were relatively short
(shorter than 50 km), one drifter covered more than 150 km during about two weeks
and left the Gulf of Finland to the Gotland Sea.
Contrary to the previously mentioned theoretical expectation, Richardson’s law
has been found to describe spreading properties for small distances fairly well while
Lin’s law has proven a better fit for larger distances (Döös and Engqvist 2007).
A probable reason for this counter-intuitive observation is that the character of
spreading is particularly complicated for 2D flows occurring on the surface of 3D
flows, which is often the case in strongly stratified environments. Velocity fields in
such flows may be highly compressible5 and may exhibit a considerable decrease in
the exponent b in Eq. (8.12) compared to the pure 2D case (Bec et al. 2004; Kalda
2007). For a review of relevant laboratory experiments see Cressman et al. (2004).
For realistic geophysical flows one might expect quite a large variation in the range
of 1.5 ≤ b < ∞ of this exponent. Note that the Gulf of Finland is rather narrow and
the separation process of drifters eventually becomes affected by the boundaries.
The gulf is O(105 ) m wide, but the drifters were not deployed in the middle of the
it. There may therefore be some influence of the southern coast on the drift, e.g., the
drifters could not move as far south as in the other directions.
The typical spreading rate was almost constant for all the pairs within the first
10–15 hours, or until the drifters were separated by about 150 m, and increased con-
siderably afterwards (Fig. 8.16). Although the estimates for initial distances below
5 The (flow) compressibility (equivalently, the compressibility of the associated velocity field) is
defined as the relative weight of the potential component in the decomposition of the net velocity
field into solenoidal and potential components. See a more detailed discussion of this quantity in
the context of the Baltic Sea in Giudici et al. (2012), Kalda et al. (2013).
Fig. 8.15 Trajectories of drifters deployed on 12.08.2010 (upper panel) and on 26.08.2010 (bot-
tom panel) in the Gulf of Finland. The deployment site is indicated by an empty circle. Thin straight
sections of the trajectories represent intervals when the GSM signal was not available (Soomere
et al. 2011)
100 m should be interpreted as indicative because of possible uncertainties of GPS-

measured locations, such a behaviour suggests the presence of different regimes of
spreading (either ballistic or Richardson’s law) for drifter pairs up to separations of
about 150 m.
For drifters initially separated by less than 1 km the separation rate varied from
about 100 m/day to 700 m/day. For even larger distances between the drifters
(>600 m) it revealed somewhat different behaviour for different pairs. The dis-
tance persistently increased for several pairs but revealed quasi-regular oscillations
for some other pairs. This phenomenon is common for the Gulf of Finland (cf. Ver-
Fig. 8.16 Time series of the distance between pairs (colour lines) and the average distance (black
thick line) in linear coordinates. Circles show the beginning and end of sensible measurements of
the pairs’ locations. The beginning time is chosen so that the initial separation of each pair matches
the average distance of pairs deployed with initially smaller separation. The insert shows the pairs
separation during the first 20 hours (Soomere et al. 2011)
jovkina et al. 2010) and apparently caused by relatively small mesoscale eddies with
a diameter as small as about 400 m.
The spreading rate owing to the impact of random walks on small scales (on the
order of O(100) m) (Lumpkin and Elipot 2010) can be estimated from the initial
sections of the trajectories of the pairs (i.e., from the parts that revealed no extensive
quasi-periodic variations in the distance due to coherent mesoscale structures) of the
drifter motion in Fig. 8.16. This rate is about 200–300 m/day, that is, about twice as
large as hypothesized in Andrejev et al. (2010). This should not be confused with
the random walk of drifters that are further apart (d much larger than the baroclinic
Rossby radius), which also experience random walk when the motion of initially
paired drifters has become uncorrelated.
8.8 Power Law Representation of the Spreading Rate
Data from the drifters in the uppermost layer suggest that the structure of small-scale
turbulence in the study area may contain motions of substantially different charac-
ter at different scales. The substantial decrease in the average spreading rate for
distances of 1.6–3.2 km during a certain time interval (Fig. 8.16) is apparently due
to a substantial impact from mesoscale eddies with a diameter matching the local
baroclinic Rossby radius that, ideally, should be resolved by a regional circulation
model.
Fig. 8.17 Time series of the distance between pairs of drifters (colour lines) and of the average
distance of pairs (black thick line) in linear–power law 2/3 coordinates Soomere et al. (2011)
It is interesting to analyse whether the dynamics of the study site is mostly gov-
erned by the 3D (local) turbulence or by a 2D (large-scale) motion system. The dis-
tance between paired drifters increased approximately linearly in the linear–power
law 2/3 coordinates (corresponding to the theoretical spreading rate for the 3D tur-
bulence) up to values of about 400 m or about 25 hours, after which the separation
rate increased for most pairs. Remarkably, two pairs (2 & 3 in Fig. 8.17) revealed a
linear increase in their distance in this framework after 2–2.5 days, which suggests
that they were involved into mostly 3D turbulent motions. Again, these drifters were
not filtered like the SVP. At the end of this part of motion, the distance between the
drifters was more than 4 km, a scale which is usually resolved by contemporary re-
gional circulation models of the Baltic Sea. The drifters in question were deployed
on 12 August 2010 in relatively calm weather conditions and thus were only weakly,
if at all, impacted by the wind.
The above findings suggest that there probably exists no single proper fit of the
exponent b in the power law d ∼ t b in Eq. (8.12). This assumption is confirmed
by the analysis in log–log coordinates (Fig. 8.18). For relatively small separations
(<70 m in the initial phase of the drift, up to 8 hours) the exponent b was in the
range 0.23–0.3, with a mean value of 0.27. Therefore, the separation rate is gov-
erned by a ballistic law rather than Richardson’s law in this situation. As none of
these laws dominated, certain specific mechanisms, such as shear dispersion (parti-
cle separation due to spatial variations in the velocity field) or specific surface-layer
dispersion (induced by the gradient of the energy dissipation rate in the turbulent
surface layer, Skvortsov et al. 2010) may govern the initial particle separation rate.
Modelling of separations on small scales (Orre et al. 2006) has revealed spreading
rates slightly lower than expected, possibly due to the lateral boundaries affecting
the drifters. It is possible that this, to some extent, also is the case in this study.
Fig. 8.18 Time series of the distance between pairs of drifters (colour lines) and of the average
distance (black thick line) in log–log coordinates. Bold dashed lines correspond to the power laws
with b = 0.27 (time interval 1–10.5 hours) and b = 2.5 (time interval 8–105 hours). Thin dashed
lines correspond to Richardson’s law with b = 1.5 (Soomere et al. 2011)
Starting from a separation of about 100 m (or a drift time of 10 hours), the separa-
tion rates were different. Two pairs in Fig. 8.18 displayed coherent motions and thus
separated poorly, reflected by b = 1.3 and b = 0.88 being the best fits for these pairs.
All other pairs revealed a surprising agreement in spreading rates. The exponent b
for them varied from 2.12 to 2.72, with the average value of b ≈ 2.5. Including or
excluding the two pairs mentioned before did not change this average significantly.
The resulting value is of a reasonable magnitude compared to the infinite exponent
characterizing 2D flows, yet clearly larger than the classical value of b = 1.5 that
corresponds to Richardson’s law and is characteristic to the 3D turbulent motions.
Therefore, in the study area the dynamics was predominantly governed by 3D flows
but the contribution of a 2D motion system was still substantial.
8.9 Discussion and Conclusions
Results from deployments of surface and subsurface drifters in the Baltic Sea during
the summers of 2010 and 2011 have been presented. Two types of drifters were used.
The SVP (sub)surface drifters have a 18 m drogue depth and represent motions of
water masses between 12–18 m depth (Fig. 8.1a). The other (GPS/GSM) drifters
were designed to represent only the currents in the uppermost 1.5 m thick layer
(Fig. 8.15).
The average lifetime of a SVP drifter was 80 days (Table 8.1). These drifters were
used to map some geographical aspects of subsurface currents in the Baltic Proper
(Fig. 8.3a), and to obtain values of mean displacement and absolute and relative
dispersion (Figs. 8.6 and 8.11). Statistical parameters of drift characterizing a sub-
stantial part of the Baltic Proper were derived from the SVP drifter data by means
of splitting the time series into basically uncorrelated segments of ∼11 days each.
The resulting statistics was compared to a similar one extracted from the motion of
drifters simulated by a trajectory model starting at the same position and time as the
SVP drifter segments (Fig. 8.1b).
In order to remove inertial oscillations, which were well observed in the drifter
data but not very well in the model results, all drifter segments and the simulated
drifters were filtered using a 14-hour running mean. As the ocean model data were
not available for 2010 and 2011, the drifter segments had to be compared to sim-
ulated drifters in the years available, 1962–2004. As such, the motion of a drifter
segment could not be directly compared to any specific simulated drifter. However,
some indications followed from a statistical comparison.
The mean displacement and absolute dispersion were found to be significantly
lower for the simulated than for the observed drifters (Fig. 8.6). This property was
attributed to the model velocities being lower and less variable, as shown by com-
paring the probability distributions of Lagrangian velocities from drifters to that of
the model trajectories (Fig. 8.7).
Near-surface currents are, to some extent, wind-driven on time scales comparable
to the duration of the drifter segments (Leppäranta and Myrberg 2009). The quality
of simulated near-surface currents thus partly depends on the quality of the wind
forcing, mixed-layer depth, and parameterization. Meier (2002) compared tempera-
ture and salinity profiles modelled using the RCO model to observations and found
good agreement in mixed-layer depth. The wind forcing (ERA-40 winds, dynam-
ically downscaled by the RCA model) was corrected using a parameterization of
wind gusts by Höglund et al. (2009) as the wind speeds were found not to be variable
enough. The correction yielded somewhat more realistic frequency distributions of
the wind speeds. However, this does not imply that the wind at a specific point or
time became more realistic. In particular, the root-mean-square errors may very well
have increased with this correction. Furthermore, the study of Höglund et al. (2009)
was limited to the Swedish coastal regions, as no observations over open water were
available. Thus, there is no information about the quality of the wind forcing over
open water, although it is likely to share some of the problems of the coastal winds.
It is thus conceivable that errors in the subsurface currents, to a large part, are due
to errors in the wind field from the RCA model output.
In principle, the discrepancy may reflect the possibility that the years 2010–2011
could have been ‘extreme’ in terms of absolute dispersion. However, Fig. 8.6, sug-
gests that such ‘extreme’ years are quite uncommon, and the likelihood that they
would occur over a two-year period is thus even smaller. The number of SVP drifters
is also relatively small, and the data may be somewhat biased as two of the drifter
pairs stayed together for nearly 20 days, and thus both drifters in the pairs sampled
the same dynamical region. However, this only affects a few segments out of the
total 76 used.
SVP drifters are seldom fully submerged into the water and may thus be affected
by the winds. Langmuir currents, Stokes drift, and waves may also effect the drifters
in a manner not represented in the RCO model. However, due to the drogue anchored
at 12 to 18 m depth, the SVP drifters in the World Ocean have been shown to follow
the currents at the drogue depth with only small deviations (Niiler et al. 1995).
A difference between the open ocean and the Baltic Sea is that the latter has a
much shallower thermocline, which in some cases is situated at the drogue depth.
This could result in a shear on the drogue, in which case the effect on the drifter
is uncertain. For this reason, it would be interesting to equip future drifters with a
temperature sensor at both the top and bottom of the drogue to identify these events.
Multiplying the velocities of the RCO model output by a factor 1.25 or includ-
ing parameterized subgrid turbulence resulted in a clearly better agreement between
several statistical characteristics of observed and simulated drifters such as distribu-
tion of velocities, absolute dispersion and mean displacement. Although both meth-
ods yielded similar results for mean displacement, the results were very different in
terms of some other metrics. The Lagrangian integral time scale was severely short-
ened after adding subgrid turbulence. The resulting time scales were shorter than
those of the recorded drifter segments.
The random motions introduced through the turbulence parameterization (Döös
et al. 2011) do not take the original velocity into account, thus changing both the
velocity and properties of the trajectory somewhat. The transport speed can thus be
improved with this parameterization, but at the cost of changes in, e.g., the transport
direction. If one needs to tune the motion of simulated drifters for single-particle
statistics, multiplying the simulated velocities by a constant factor would then be
a reasonable choice: it increases the speed but does not alter the properties of the
trajectory. This choice is however poor if the relative dispersion needs to be adjusted
since small, subgrid scale changes in the direction are needed to separate initially
closely located simulated drifters.
Furthermore, even when the simulated drifters were separated by at least one
grid box, the further separation rates were lower than the average rate for the
SVP drifters. Therefore, subgrid parameterization is needed also on larger scales
(DR2 > 4 km). In this study, simulated drifters were tuned to resemble the mean dis-
placement and absolute dispersion of observed SVP drifters. It would be possible
to instead tune them to yield good fit to observed relative dispersion, depending on
whether transport or spreading is the most crucial metric for the study. It is possi-
ble that using a higher-order turbulence scheme such as ‘Markov 1’ or ‘Markov 2’
(Rupolo 2007), thereby taking, e.g., velocity autocorrelation into account, would
give better results by combining the benefits of both methods used in this study.
The higher-order Markov models have been found to better describe the motions
of water particles than the ‘Markov 0’ model (Griffa 1996). Such schemes may be
implemented in TRACMASS in the future.
The comparison between simulated drifters and observed SVP-B drifters indi-
cates that the RCO model simulates correctly neither the mean flow nor the turbulent
flow. Using values roughly estimated from Figs. 8.6 and 8.10, some implications
for Lagrangian modelling without tuning can be identified. If the simulated drifters
have 4/5 of the mean displacement of the SVP drifter segments, this would mean
that if simulated particles, on average, travel 100 km in 10 days, a drifter, or a real
water particle, would, on average travel ∼125 km in those 10 days. By the same
argument, if simulated particles are estimated to reach the coast in 10 days, real wa-
ter particles would make the same journey in <8 days. Furthermore, water particles
contained within a <1 km radius initially would spread over an area of 12 km radius
in 25 days, while their model counterparts would disperse to cover <10 km radius.
Such conclusions would have impacts when estimating the fate of oil spills or other
pollutants. However, we wish to stress that we have compared observations from
2010–2011 with model data from 1962–2004, and that the drifter data is limited.
To make the comparison of modelled and observed properties fair, and to yield
more confidence in the magnitudes of the tuning needed would require model data
for the years 2010–2011. This could, however, take a few years as the RCO model
is being decommissioned in favour of a new regional ocean model, based on the
Nucleus of European Modelling (NEMO) ocean circulation model (Madec 2009).
The wind forcing will most likely also need updating, as the ERA-40 data set is
to be replaced by ERA-Interim and eventually by ERA-75, while the RCA model
used for downscaling is also likely to be decommissioned. If SVP-B drifter data is
continued to be gathered for the Baltic Sea, this data could be used to validate and
perhaps tune the next generation of Baltic Sea models.
The presented results from the drifters in the uppermost layer of the Gulf of
Finland indicate substantial differences in the dynamics of the vertically integrated
relatively thick subsurface layer and the uppermost layer with a thickness of 1–
1.5 m (which is where, e.g., oil spills or smaller lost items are transported). Although
the average spreading rate generally increased with time or the distance between
drifters, the well-known Richardson’s law did not satisfactorily explain the transport
in the uppermost layer of the Gulf of Finland. For the separation d of drifter pairs
on short time scales (less than 8 hours) or separation distances (from the first tens
of meters up to about 100–150 m), a power law d ∼ t 0.27 described the spreading
much better. Starting from this threshold, the distance then increased, in average,
according to a power law d ∼ t 2.5 . The spreading rate was about 200 m/day for
separations below 0.5 km, 500 m/day for separations below 1 km and in the range
of 0.5–3 km/day for separations in the range of 1–4 km.
As a considerable part of the drifters following the uppermost 1.5 m thick layer
was above the water surface, their drift was impacted by wind properties to some
extent. For example, a wind speed of 5 m/s may, technically, yield a contribution of
about 10 cm/s to the drift speed (Soomere et al. 2011). Although this value is on the
order of the current speed, it does not significantly effect the separation of drifters,
as the wind patterns over sea surface are much more homogeneous compared to
similar winds over the mainland. Therefore, it is natural to expect that the impact of
wind on closely located drifters mostly resulted in their concurrent downwind drift.
This would imply that winds can influence absolute dispersion while their impact
on relative dispersion would be negligible.
The results from the drifters in the uppermost layer suggest that a realistic param-
eterization of subgrid-scale processes in the Gulf of Finland strongly depends on the
resolution of the ocean model. It is well known that models with spatial resolution
coarser than 2 km cannot resolve mesoscale dynamics in this region (see Chap. 6
for more details). Their parameterization of subgrid processes should correspond to

a typical spreading rate of about 2 km/day. The same rate is reasonable for models
with a resolution of about 1–2 km while the models with a resolution of ∼1 km
might use the rate of about 700 m/day. Parameterizations leading to spreading rates
of 300–500 m/day may be recommended for extremely high-resolution models with
a grid step of ∼0.5 km. As the drifters in the uppermost layer have experienced a
certain impact of the local wind and waves on their drift the presented rates may to
some extent overestimate the actual spreading rates but the order of magnitude for
the spreading effects extracted from the experiments evidently is realistic.
The parameters characterizing the dynamics of spreading of objects in the upper-
most layer are of utmost importance for the technique developed in this book. Its
key idea is to use the Lagrangian dynamics of currents to develop methods for the
reduction of environmental risks. Its key component is statistical analysis of large
sets of Lagrangian trajectories of simulated drifters or water particles. The results
are evidently highly sensitive with respect to the parameterization of subgrid-scale
processes that may randomly redirect drifters to largely different sea areas compared
to the modelled fields of currents (Döös 1995; de Vries and Döös 2001; Griffa et al.
2004; Andrejev et al. 2010).
The problem is even more complicated in strongly stratified sea areas such as
the Gulf of Finland where the drift is frequently steered by multi-layered dynamics
(Andrejev et al. 2004; Gästgifvars et al. 2006) and where it is not clear beforehand
which theoretical framework (predomination of 2D or 3D motion systems) should
be used in the analysis.
Similar problems intrinsically arise in the attempts of modelling pathways of dif-
ferent water masses (Meier 2007) and especially in simulations, both in forecast and
hindcast modes, of pollution transport by regional ocean models such as HIROMB
or Seatrack Web (Funkquist 2001; Gästgifvars et al. 2006; Verjovkina et al. 2010).
This chapter has contributed by presenting results from recently deployed surface
drifters of different types as well as drifters simulated with an ocean model and a
trajectory code, and comparing these to each other and the theoretical expectations
and discussing the implications for Lagrangian modelling in the Baltic Sea.
Acknowledgements This study was performed in the framework of the BalticWay project, which
was jointly supported by the funding from the by the Swedish Research Council for Environ-
ment, Agricultural Sciences and Spatial Planning (Formas, Ref. No. 2008–1900), Estonian Sci-
ence Foundation and the European Commission’s Seventh Framework Programme (FP7 2007–
2013) under grant agreement No. 217246 made with the joint Baltic Sea research and development
programme BONUS. The research was partially supported by the Estonian Science Foundation
(grant No. 9125), targeted financing by the Estonian Ministry of Education and Research (grant
SF0140007s11), and by the European Regional Development Fund via support to the Centre of
Excellence for Non-linear Studies CENS. The authors wish to thank Tallink Silja shipping com-
pany for allowing us to deploy the SVP drifters from the Stockholm–Riga line and in particular we
wish to thank Captain Lembit Uustulnd and his crew on M/S Silja Festival for permission and help
in deployment of the drifters. We also acknowledge and appreciate the help given by Prof. Peter
Lundberg and Dr. Anders Engqvist in deployment. The experiments with the surface drifters in
the Gulf of Finland were performed very professionally by Mr Mikk Viidebaum. His cooperation
towards deployment and rescue of SVP drifters is also gratefully acknowledged. Finally, we also
express our gratitude to Markus Meier and Anders Höglund at the Swedish Meteorological and
Hydrological Institute for providing the RCO data as well as help and fruitful discussions about it.
All trajectory simulations have been made using supercomputers maintained by NSC at Linköping
University, Sweden.
References
9:1–16
Bec J, Gawedzki K, Horvai P (2004) Multifractal clustering in compressible flows. Phys Rev Lett
92:224501
Blanke B, Raynaud S (1997) Kinematics of the Pacific Equatorial Undercurrent: an Eulerian and
Lagangian approach from GCM results. J Phys Oceanogr 27:1038–1053
Corell H, Moksnes PO, Engqvist A, Döös K, Jonsson P (2012) Larval depth distribution criti-
cally affects dispersal distance and optimum size for marine protected areas. Mar Ecol Prog Ser
467:29–46
Cressman J, Davoudi J, Goldburg W, Schumacher J (2004) Eulerian and Lagrangian studies in
surface flow turbulence. New J Phys 6:53
13,514
Döös K, Rupolo V, Brodeau L (2011) Dispersion of surface drifters and model-simulated trajecto-
ries. Ocean Model 39:301–310
597
Döscher R, Willén U, Jones C, Rutgersson A, Meier HEM, Hansson U, Graham LP (2002) The de-
velopment of the regional coupled ocean–atmosphere model RCAO. Boreal Environ Res 7:183–
192
Falkovich G, Gawedzki K, Vergassola M (2001) Particles and fields in fluid turbulence. Rev Mod
Phys 73:913–975
Funkquist F (2001) HIROMB, an operational eddy-resolving model for the Baltic Sea. Bull Mar
Inst Gdańsk 28:7–16
Garfield N, Maltrud M, Collins C, Rago T, Paquette R (2001) Lagrangian flow in the California
Undercurrent, an observation and model comparison. J Mar Syst 29:201–220
Garraffo ZD, Mariano AJ, Griffa A, Veneziani C, Chassignet EP (2001) Lagrangian data in a high-
resolution numerical simulation of the North Atlantic I. Comparison with in situ drifter data.
J Mar Syst 29:157–176
Giudici A, Kalda J, Soomere T (2012) On the compressibility of surface currents in the Gulf of
Finland, the Baltic Sea. In: Proceedings of the IEEE/OES Baltic 2012 international symposium
“Ocean: past, present and future. Climate change research, ocean observation & advanced tech-
nologies for regional sustainability”, Klaipėda, Lithuania, May 8–11, 2012. IEEE Press, New
York, 8 pp
Griffa A (1996) Applications of stochastic particle models to oceanographic problems. In: Adler
RJ, Müller P, Rozovskii BL (eds) Stochastic modelling in physical oceanography. Birkhäuser
Boston, Cambridge, pp 113–140
Håkansson B, Rahm L (1993) Swedish Lagrangian current experiments. In: Gulf of Bothnia year
1991—physical transport experiments, vol 15, pp 41–55. SMHI, RO
Sweden
Kalda J (2007) Sticky particles in compressible flows: aggregation and Richardson’s law. Phys Rev
Lett 98:064501
Kalda J, Soomere T, Giudici A (2013) On the finite-time compressibility of the surface currents in
the Gulf of Finland, the Baltic Sea. J Mar Syst. doi:10.1016/j.jmarsys.2012.08.010
Kjellsson J, Döös K (2012) Surface drifters and model trajectories in the Baltic Sea. Boreal Environ
Res 17:447–459
Kjellström E, Bärring L, Gollvik S, Hansson U, Jones C, Samuelsson P, Rummukainen M, Uller-
stig A, Willén U, Wyser K (2005) A 140-year simulation of European climate with the new
version of the Rossby Centre Regional Atmospheric Climate Model (RCA3). Reports meteo-
rology and climatology, vol 108
Launiainen J, Stipa T, Grönwall H, Vihma T (1993) Finnish Lagrangian current experiments. In:
Gulf of Bothnia year 1991—physical transport experiments, vol 15, pp 55–67. SMHI, RO
LaCasce J (2008) Statistics from Lagrangian observations. Prog Oceanogr 77:1–29
Leppäranta M, Myrberg K (2009) Physical oceanography of the Baltic Sea. Springer, Berlin,
378 pp
Lin J (1972) Relative dispersion in the enstrophy-cascading inertial range of homogeneous two-
dimensional turbulence. J Atmos Sci 29:394–396
Lumpkin R, Elipot S (2010) Surface drifter pair spreading in the North Atlantic. J Geophys Res—
Oceans 115:C12017
Lumpkin J, Pazos M (2007) Measuring surface currents with Surface Velocity Program drifters: the
instrument, its data, and some recent results. In: Griffa A, Kirwan A, Mariano A, Özgökmen T,
Rossby T (eds) Lagrangian analysis and prediction of coastal and ocean dynamics. Cambridge
University Press, New York, pp 39–68
Lumpkin R, Treguier AM, Speer K (2002) Lagrangian eddy scales in the Northern Atlantic Ocean.
J Phys Oceanogr 32:2425–2440
Madec G (2009) NEMO ocean engine. Note du Pole de modélisation. Institut Pierre-Simon
Laplace (IPSL), France, No 27, ISSN No 1288-1619, 217 pp
McClean JL, Poulain PM, Pelton J, Maltrud ME (2002) Eulerian and Lagrangian statistics from
surface drifters and a high-resolution POP simulation in the North Atlantic. J Phys Oceanogr
32:2472–2491
Meier HEM, Döscher R, Coward AC, Nycander J, Döös K (1999) RCO—Rossby Centre regional
Ocean climate model: model description (version 1.0) and first results from the hindcast period
1992/93. SMHI reports oceanography, vol 26
Meier HEM (2002) Regional ocean climate simulations with a 3D ice-ocean model for the Baltic
Sea. Part 1: Model experiments and results for temperature and salinity. Clim Dyn 19:237–253
Sea: application to salt inflow. J Geophys Res—Oceans 108:C3273
Niiler PP, Sybrandy AS, Kenong B, Poulain PM, Bitterman D (1995) Measurements of the water-
following capability of holey-sock and tristar drifters. Deep-Sea Res 42:1951–1964
Ollitrault M, Gabillet C, Colin de Verdiere A (2005) Open ocean regimes of relative dispersion.
J Fluid Mech 533:381–407
Orre S, Gjevik B, LaCasce JH (2006) Characterizing chaotic dispersion in a coastal tidal model.
Pizzigalli C, Rupolo V, Lombardi E, Blanke B (2007) Seasonal probability dispersion maps in
the Mediterranean Sea obtained from the Mediterranean forecasting system Eulerian velocity
fields. J Geophys Res—Oceans 112:C05012
Poje AC, Haza AC, Özgökmen TM, Magaldi MG, Garraffo ZD (2010) Resolution dependent rel-
ative dispersion statistics in a hierarchy of ocean models. Ocean Model 31:36–50
Richardson LF (1926) Atmospheric diffusion shown on a distance-neighbour graph. Proc R Soc A
110:709–737
Rupolo V (2007) Observing turbulence regimes and Lagrangian dispersal properties in the ocean.
In: Griffa A, Kirwan A, Mariano A, Özgökmen T, Rossby T (eds) Lagrangian analysis and
prediction of coastal and ocean dynamics. Cambridge University Press, New York, pp 231–275
Salazar JPLC, Collins LR (2009) Two-particle dispersion in isotropic turbulent flows. Annu Rev
Fluid Mech 41:405–432
Skvortsov A, Jamriska M, DuBois TC (2010) Scaling laws of passive tracer dispersion in the
turbulent surface layer. Phys Rev E 82:056304
Soomere T, Viidebaum M, Kalda J (2011) On dispersion properties of surface motions in the Gulf
Webb DJ, Coward AC, de Cuevas BA, Gwilliam GS (1997) A multiprocessor ocean circulation
Chapter 9
Statistics of Lagrangian Transport Reveals
Hidden Features of Velocity Fields
Tarmo Soomere
Abstract The potential of the systematic use of Lagrangian trajectories to identify

normally hidden properties and semi-persistent patterns of the dynamics of surface
currents is analysed and tested for the Gulf of Finland and the south-western Baltic
Sea. These patterns are highlighted using statistical analysis of a large number of
Lagrangian trajectories constructed by using horizontal velocity fields in the upper-
most layer generated by three-dimensional circulation models. The evaluation of
environmental risks and the identification of patterns of rapid Lagrangian transport
rely on a specific discretization of the direct problem of the propagation of passive
tracers. This discretization contains several time scales and other parameters, the
optimum choice of which is analysed based on examples from the Gulf of Finland.
It is shown how the dependence of the overall ratio of net and bulk transport can be
used for highlighting the nature of the surface dynamics. Semi-persistent patterns of
net transport are identified for the Gulf of Finland using the Rossby Centre Ocean
Model (RCO) velocity fields and the offline trajectory model TRACMASS for the
time period 1987–1991. Rapid transport pathways are mostly aligned along certain
coastal segments but pathways across the gulf are evident in transitional months
(March–May and August–October). The results are applicable for estimates of the
transport of neutrally buoyant substances in the uppermost layer of the sea.
9.1 Introduction
International ship transport has dramatically increased in the Baltic Sea (Fig. 9.1)
over the last two decades. At present it accounts for up to 15 % of the world’s
cargo transportation. The largest threat to the vulnerable boreal environment of this
region is oil transportation that has increased by more than a factor of two in 2000–
T. Soomere (B)
T. Soomere

DOI 10.1007/978-3-319-00440-2_9,
284 T. Soomere
Fig. 9.1 Major fairways

crossing the Baltic Sea in
2011. Graphics by Maija
Viška
2006 (HELCOM 2009) and continues to increase. Sustainable management of this

traffic flow is a major challenge in this region, which is designated as a particularly
sensitive sea area by the International Maritime Organization (Kachel 2008).
The shipping conditions are harsh in the entire Baltic Sea during a large part
of the year. Frequent stormy winds, a short period of daylight and cold weather in
autumn and winter put the seamen under great pressure. Almost every winter the
presence of heavy ice in large sea domains, especially in semi-enclosed subbasins
such as the Gulf of Riga, the Gulf of Finland and the Sea of Bothnia, drastically com-
plicates the navigation. Therefore the entire Baltic Sea area needs novel approaches
concerning the management of the associated risk.
The technology described in this book and presented in more detail in Chaps. 10
and 11 aims at a systematic optimization of the potential influence of various off-
shore activities. These include not only ship traffic but also oil platforms, pipelines,
mining, dumping, etc. The goal is to develop a method that would be able to iden-
tify and highlight areas (of reduced risk), where a potential accident would have
minimum impact on vulnerable or high-cost marine and coastal regions. The focus
here is on the reduction of the impact of adverse substances that are carried by sea
currents.
The system of the Baltic Sea surface currents is extremely complicated and
hardly predictable. There is, though, a certain order in this system. This order be-
comes evident in the form of regularly occurring flow patterns in offshore domains
9 Statistics of Lagrangian Transport Reveals Hidden Features of Velocity Fields 285
that persist during certain time intervals. Such patterns have been recently identified
for different areas of the Baltic Sea both numerically (Lehmann et al. 2002; Meier
2007; Lu et al. 2012) and experimentally (Osiński and Piechura 2009). Their typi-
cal lifetime ranges from a few days to weeks up to a few months, that is, from the
synoptic scale (in the Gulf of Finland and in the south-western Baltic Sea) up to
seasonal scale.
Such more or less regularly occurring patterns of currents may definitely pro-
vide a joint effect similar to that of jet currents. They apparently relocate not only
water masses but also transport key components of the ecosystem such as nutri-
ents, oxygen or fish larvae between certain sea areas, and thus are extremely im-
portant for the functioning of the entire ecosystem. On top of that, they inherently
cause relatively fast current-driven transport of adverse impacts and thus obviously
have a high potential for the systematic transport of pollution between specific do-
mains, and also from the offshore to vulnerable coastal and protected marine areas
(Delpeche-Ellmann and Soomere 2013).
At present, man is not able to steer or influence this dynamics but it is in our
power to highlight such patterns and to account for their presence. The least we can
do is to avoid the release of adverse impacts in those regions from where such cur-
rents may rapidly transport water masses to high-value domains. The smart use of
such patterns is a feasible way towards a reduction of anthropogenic impact to vul-
nerable areas. This can be achieved, for example, by routing ships along a current
that would rapidly carry a potential oil spill out of a vulnerable domain (Soomere
and Quak 2007). Alternatively, and more interestingly, human activities (such as
marine traffic) might be placed in specific regions (areas of reduced risk), from
which the transport of pollution to vulnerable or high-cost areas is unlikely. Avoid-
ing ship traffic (and associated pollution) in certain offshore domains and time in-
tervals might considerably reduce the consequences of potential accidents .
The presence of such (normally concealed) patterns can be used as an indicator
of the applicability of the method developed in this book. Such patterns eventually
shape the pollution propagation as well as the drift of various items (such as vessels
without propulsion, rescue boats or lost containers). Their presence is a strong argu-
ment in favour of the existence of areas of reduced risk. Strictly speaking, achieving
a substantial decrease in industry-induced remote environmental risks is not neces-
sarily directly connected with the presence of such patterns. However, their absence
is a signal that there is little hope to benefit from favourable features of current-
driven transport.
The idea of using favourable patterns of transport of adverse impacts for environ-
mental management, ideally, should be applied to the full three-dimensional (3D)
motions in the entire water column and be combined with the wind- and wave-driven
transport. As explained in more detail in Chap. 1, in this book we focus, however,
on the current-driven transport of adverse impacts that are locked in the uppermost
layer of the sea. Moreover, in this chapter we leave out the impact of wind and
waves. Doing so is partially justified by the fact that the properties of wind and
wave fields and the associated transport are relatively well (albeit not perfectly)
known (ASCE (American Society of Civil Engineers) 1996; Reed et al. 1999; Ard-
huin et al. 2009). Compared to this knowledge, the prediction of currents and related
286 T. Soomere
transport is much more challenging even for surface currents (Vandenbulcke et al.
2009).
The presented results are thus directly applicable for persistent neutrally buoyant
substances that are dissolved in strongly stratified environments under calm condi-
tions when the contaminants (e.g., dissolved radioactive substances) largely remain
in the uppermost layer and are mostly carried by surface currents (e.g., Periáñez
2004). This setup is only conditionally valid for oil pollution, lost containers, ves-
sels without propulsion, debris, litter, etc., as it does not consider other metocean
drivers, chemical processes, buoyancy effects, weathering of oil, and so forth.
The use of the simulated field of currents to calculate the drift of dissolved sub-
stances and objects located within the uppermost layer is most applicable in the
Gulf of Finland during a large part of the spring and early summer when the wind is
fairly weak (below 5 m/s) (Mietus 1998) and also during the presence of ice cover
(up to six months in some years) (Jevrejeva and Leppäranta 2002; Sooäär and Jaa-
gus 2007) when the field of currents is almost totally disconnected from the direct
atmospheric influence. The setup in use is therefore of clear value to improve the
understanding about the potential role of a practical use of the intrinsic dynamics
of currents to preventively reduce the costs of accidents at sea (Soomere and Quak
2007).
This chapter demonstrates the potential of the analysis of Lagrangian trajectories
of water particles to identify usually concealed semi-persistent patterns and asso-
ciated Lagrangian transport driven by surface currents in areas hosting extremely
complicated systems of currents. The focus is on two areas of the Baltic Sea—the
Gulf of Finland and the south-western (SW) region of the Baltic Sea. The basic tool
is the statistical analysis of properties of a large number of Lagrangian trajectories
constructed from precomputed Eulerian velocity fields. A similar technique (but in
an Eulerian framework) has been applied in Chap. 4. The same technique will be
used in Chaps. 10 and 11 to evaluate the environmental risks, to identify the areas
of reduced risk and to find optimum fairways for the test areas. The technique relies
on a specific discretization of the direct problem of propagation of passive tracers.
This discretization contains several time scales and other parameters, the optimum
choice of which is analysed based on examples from the two test areas. Finally, the
developed technique is applied to identify several normally concealed features of
the dynamics of the Gulf of Finland.
9.2 Hydrodynamics of the Test Regions
The two test regions, the Gulf of Finland and SW Baltic Sea, not only host extremely
dense ship traffic but also play in a certain sense a key role in the functioning of
the entire Baltic Sea. The major marine highway approaches the Baltic Sea via the
Kattegat and Kiel Channel, branches into two flows through the Danish Straits and
merges back into a powerful flux in the Arkona Basin (Fig. 9.2). Its major branch
Fig. 9.2 Major fairways in

the south-western Baltic Sea,
Danish Straits, Kattegat and
Skagerrak. Graphics by Maija
Viška
stretches across the Baltic Proper1 to the north-east, through the Gulf of Finland and
finally reaches Saint Petersburg, the major population and industrial centre in this
area, and a number of new harbours in its vicinity.
9.2.1 South-Western Baltic Sea
The Kattegat, the Danish Straits, the Belt Sea and the Arkona Basin in the SW
Baltic are the gateway through which all the ships travelling between the North Sea
and the Baltic Sea (including those using the Kiel Channel) have to pass (Fig. 9.2).
A detailed description of the dynamics of water masses and the relevant modelling
efforts in this region is given in Chap. 5. All the listed domains are mostly shal-
low (see Fig. 5.20 in Chap. 5) and have a moderate bottom slope. The excess of
brackish water flows out of the Baltic Sea through this transit area and through the
Danish Straits. The outflow is sporadically replaced by large-scale saline water in-
flows (Matthäus and Lass 1995), forced by specific atmospheric conditions. During
inflows this area serves as a vital channel for water exchange between the Baltic Sea
1 We use here Baltic Proper to denote the open part of the Baltic Sea, consisting of the Northern
Gotland Basin, Eastern Gotland Basin, Bornholm Basin and Gdańsk Bay; see Fig. 2.1 in Chap. 2
(Leppäranta and Myrberg 2009).
288 T. Soomere
and the Kattegat that supplies the deeper parts of the Baltic Sea with oxygen-rich
waters. The dynamics of water masses in this area is described in more detail in
Sects. 5.2.2, 5.2.5 and 5.3.3.
The interplay of brackish Baltic Sea waters with saline North Sea water masses
results in strong salinity gradients and small values of the baroclinic (internal)
Rossby radius2 R1 in this area. This measure usually varies from 1–4 km in the Belt
Sea (Fennel et al. 1991) and is about 5 km in the Arkona Basin and the Pomera-
nian Bay (Osiński et al. 2010). Therefore most of the medium-resolution circulation
models with a grid step around 2 nautical miles (nm) are barely eddy-permitting
and a finer resolution is needed to properly resolve the dynamics in this domain.
Consequently, the application of the technology of preventive coastal protection for
this region (Lu et al. 2012) presents an intricate challenge for modellers.
9.2.2 The Gulf of Finland
The situation is equally intricate in the Gulf of Finland. The width of its naviga-
ble part is at places below 50 km and extensive domains in its eastern part are too
shallow for large ships. The concentration of traffic is exceptionally high in several
narrow passages that can host the largest ships sailing to the eastern end of the gulf
(see Fig. 10.1 in Chap. 10). The major long-haul fairway from the Northern Gotland
Basin (see Fig. 2.1 in Chap. 2 or Leppäranta and Myrberg 2009 for the contempo-
rary partition of the Baltic Sea) to the eastern region of the gulf crosses an intense
fast ferry link between Helsinki and Tallinn. In the recent past more than 50 gulf
crossings took place daily along this route during the high season (Parnell et al.
2008). The traffic flow has decreased to some extent since then but still hosts 30–
40 gulf crossings a day (Kurennoy et al. 2011). The listed circumstances increase
the odds for a release of various adverse impacts (oil or chemical pollution) or haz-
ardous objects into the sea owing either to an accident, to technical problems or to
human mistakes or misbehaviour. The associated impact may be dangerous to both
the environment and to other vessels.
A thorough overview of the physical oceanography of the Gulf of Finland is
presented in Chap. 6. This easternmost prolongation of the Baltic Sea has a total
length of about 400 km, maximum width of 125 km and a mean depth of only 37 m
(Soomere et al. 2008). Similarly to the gateway to the Baltic Sea, this basin also has
complicated geometry and bathymetry, strong gradients, small baroclinic Rossby
radius and extremely complex internal dynamics. The overall nature of its currents is
well known. A traditional view is that the mean circulation (integrated over the entire
water column) is cyclonic with an average speed of a few cm/s (Alenius et al. 1998).
This gyre not necessarily becomes evident on the sea surface where the predominant
2 See Chaps. 2–6 for a definition of this measure and for the discussion of its role in the dynamics.
SW winds (Mietus 1998) usually drive Ekman surface transport3 to the south-east
(SE).
A combination of this transport with the presence of numerous synoptic4 (also
called mesoscale) eddies with a typical radius matching the baroclinic Rossby radius
leads to almost perfectly random surface flow in terms of (directional) persistency
over longer time intervals (Lehmann et al. 2002; Andrejev et al. 2004a, 2004b). The
measure of persistence used in Andrejev et al. (2004a) was originally defined as
the ratio of the average water flow speed and the mean water speed (Witting 1912;
Palmén 1930):
√
ū2 + v̄ 2
Rp = × 100 %, (9.1)
|u|rms
where ū and v̄ are the average horizontal velocity components and |u|rms is the
root-mean-square speed over a certain set of measurements or a time interval. This
measure is also called the ratio of vector and scalar mean speeds (Andrejev et al.
2004a) and basically expresses the persistence of the direction of currents. Low val-
ues of persistence, however, do not exclude the existence of semi-persistent transport
pathways, the search for which is the central goal of this chapter.
9.2.3 Aperiodic Flow Patterns and Subsurface Currents
The properties of current-induced transport can be relatively easily predicted for

(quasi-)stationary jet currents. The situation is different in areas where such cur-
rents are absent and where mesoscale motions govern the dynamics of water masses.
Usually no systematic transport patterns exist in such cases except for a slight pre-
dominance of long-term average flow along parallels in the ocean of constant depth
or along the f/ h isolines (f is the local value of the Coriolis parameter character-
izing ambient vorticity and h is the water depth) in areas hosting large-scale bottom
gradients (Cushman-Roisin and Beckers 2011).
The presence of coasts and medium-scale bottom features usually induces a cer-
tain order into the field of synoptic motions. Coastal jets and the overall circulation
in semi-enclosed basins (cyclonic on the northern hemisphere) are common exam-
ples. Certain repeating flow patterns that may result in a predominant direction of
the transport are also common along major bathymetric features. For instance, reg-
ular separation of coastal currents at headlands may provide a continuous, robust
3 The nature and properties of Ekman transport are presented in more detail in Chap. 2, Sect. 2.3.5.
4 The notion of synoptic motions is used in meteorology to denote weather systems ranging in size
from several hundred kilometers to several thousand kilometres. The relevant (synoptic) scale is
understood as the scale of migratory high and low pressure systems and is characterized by the
so-called baroclinic (internal) Rossby radius. After the discovery of offshore synoptic vortices in
the ocean in the 1970s, this notion has been extended to describe the typical scales of such vortices
(frequently called mesoscale eddies) and associated phenomena in different water bodies. See, for
example, Cushman-Roisin and Beckers (2011) or Chaps. 2 and 6.
290 T. Soomere
Fig. 9.3 Five-year (1990–1994) average surface currents in the SW Baltic Sea, Danish Straits and
Kattegat, averaged for the whole period (upper left panel), during outflow (upper right panel) and
inflow situations (lower panel) (Lu et al. 2012)
feature that drives the cross-shelf transport from the coast to the deep ocean (Barth
et al. 2000). Similarly, the frequent presence of trapped motions or fronts at bathy-
metric features may result in certain robust features of surface transport (Bisagni
et al. 1996). Alternatively, voluminous fresh water discharge from large rivers often
drives a highly coherent flow extending to substantial distances (Morey et al. 2003).
These features are examples of natural semi-persistent patterns of transport.
Such features naturally exist in the SW Baltic Sea not only in the Danish Straits
but also in certain offshore domains (Lu et al. 2012). They are hardly distinguishable
in the average of Eulerian velocities over longer periods because of the frequent
change in the flow direction in the Straits (Fig. 9.3). In such long-term averages of
surface flow only certain gyres become evident in the Belt Sea and the Kattegat.
The situation is largely different when the averaging is performed over certain sit-
uations corresponding to naturally occurring specific forcing conditions. For the SW
Baltic the forcing patterns corresponding to the inflow and the outflow conditions5
5 The inflow and outflow conditions are defined in Lu et al. (2012) as certain patterns of atmospheric
forcing and are not directly related with the direction of flow in the Danish Straits.
Fig. 9.4 Five-year persistency Rp (%) of modelled surface currents in the Gulf of Finland in
1987–1991 (Andrejev et al. 2004a)
are highly different (e.g., Lu et al. 2012). If the Eulerian velocities are averaged over
only one type of situation, the presence of highly ordered flow systems is evident
not only in the Straits but also in the southern part of the Kattegat, across the entire
Belt Sea, over the Darss Sill and up to the island of Rügen and the Pomeranian Bay
(Fig. 9.3).
There is less evidence about such patterns in the offshore domains of the Gulf of
Finland. Similarly to the Danish Straits, the stratification is strong and the surface
drift is frequently steered by multi-layered dynamics in this gulf (Gästgifvars et al.
2006). Owing to a small baroclinic Rossby radius (normally 2–4 km, in the eastern
part of the gulf down to 0.5 km) (Alenius et al. 1998; Soomere et al. 2008), the
dimensions of mesoscale features may be as small as a few km. The driving forces
are not only highly variable both seasonally and annually but also exceedingly in-
termittent within shorter time intervals (Myrberg et al. 2010). This combination of
different factors favours an extreme complexity of motions and results in a generally
very low persistence of surface currents in almost the entire basin (Fig. 9.4). Only
the easternmost narrow region of the Gulf of Finland adjacent to shallow Neva Bay
hosts a well-defined current driven by the runoff from the River Neva and certain
segments of coastal currents (e.g., a branch of inflow near the southern coast) have
a relatively high persistency (up to Rp = 50 %).
The presence of such a highly complicated system of currents makes the predic-
tion of even surface currents extremely complex and sometimes hardly possible even
when using the most advanced models and computational facilities. Given such a
highly variable system of currents, it is not unexpected that a substantial oil spill may
cross the entire Gulf of Finland within only one typical autumn day (Anonymous
2002) and become an acute danger to large sections of the nearshore on the next day.
The dynamics of the gulf, however, is not completely unpredictable and contains
several ordered elements. An amazing property of the Gulf of Finland is the pres-
ence of a very persistent motion (Rp up to 80 %) that exists in the subsurface layer
(2.5–7.5 m) slightly north of the gulf axis (∼30 km off the Finnish coast) in circu-
lation simulations of Andrejev et al. (2004a) (Fig. 9.5). This flow to the east covers
the entire gulf except for a short segment between Lahemaa and Helsinki. Physical
mechanisms supporting this flow are discussed in Soomere et al. (2008).
292 T. Soomere
Fig. 9.5 Five-year persistency Rp (%) of modelled subsurface currents (depths 2.5–7.5 m) in the
Gulf of Finland in 1987–1991 (Andrejev et al. 2004a)
As the thickness of the subsurface layer is two times as large as that of the upper-
most layer in the model used by Andrejev et al. (2004a), the subsurface flow appar-
ently dominates the surface dynamics during certain seasons. This predominance
is likely when the wind forcing is relatively weak, for example under ice cover or
during the calm season. This phenomenon may implicitly protect the northern coast
of the Gulf of Finland from the impact of adverse substances released to the south
of this flow: it may carry a part of the adverse impact out of the Gulf of Finland.
Therefore, rerouting ship traffic into the region of the most intense and persistent
subsurface flow could decrease the probability of coastal pollution because a part of
the pollution will be carried out of the gulf before it reaches the nearshore (Soomere
and Quak 2007).
Note that calculations in Anonymous (2002) only indicate the region possibly
reached by the oil spill but do not provide the associated probability. This proba-
bility can be substantially different if the current-driven surface transport is highly
anisotropic such as the subsurface flow in Fig. 9.5. The presence of such a flow may
increase the average odds for pollution released into certain sea areas to reach the
coast and may decrease such odds for other sea areas.
9.3 Tools for the Search for the Patterns of Lagrangian

Transport
9.3.1 Semi-persistent Patterns
The search for semi-persistent patterns of surface currents and especially the asso-
ciated pathways of current-driven transport is a highly interesting challenge as their
location and lifetime are not known beforehand. It is not unexpected that the long-
term persistency Rp of surface currents of the Gulf of Finland over longer time
intervals (several times exceeding the typical turnover time of mesoscale eddies)
is very small as averaging over longer time intervals (five years in Andrejev et al.
2004a, 2004b) inevitably smoothes out transient patterns with a relatively short life-
time. A careful investigation of drift patterns at various time scales thus serves as
a prime test whether it would be possible to choose specific areas for potential ac-
cidents (or to impose specific travel routes) that minimize the environmental risk.
A feasible way to tackle this issue is to rely on the analysis of a large pool of numeri-
cally simulated transport patterns for shorter time intervals of varying length. Doing
so eventually allows identifying the presence and basic features of semi-persistent
current patterns that may later be used for practical purposes.
For the search of such patterns, we develop a technique that will be used in
Chaps. 10 and 11 for approximately solving the inverse problem of current-driven
transport. The starting point is a 3D circulation model that provides surface ve-
locity fields for a longer time interval at fixed points. These velocities reflect the
so-called Eulerian specification of the flow field for each fixed location in space as
time passes. In marine conditions one has to use fixed moorings (e.g., at offshore
structures) in order to properly measure the Eulerian velocities. As it is much easier
to model the flow in the Eulerian sense, this framework is mostly used in circulation
modelling. An average of such fields (Fig. 9.3) gives some idea about how the water
particles might move.
A large part of the complexity of the reproduction of the actual drift of water or
pollution particles, or current-driven transport, stems from its Lagrangian6 nature.
In this specification of the motion the observer follows an individual water particle
as it moves through space and time. The Lagrangian velocity and transport can be
directly measured using drifting buoys as described in Chap. 8. This transport is
usually reconstructed by first calculating the Eulerian velocities for a certain grid
and then constructing Lagrangian trajectories (pathways) for single water particles
(Fig. 9.6). Finally, the patterns of transport are evaluated and highlighted by consid-
ering large ensembles of Lagrangian pathways of various lengths. The aim of this
analysis is the identification and visualization of several properties of surface that
cannot be extracted directly from the current fields.
The number of studies concerning Lagrangian propagation and trajectories of
various substances in the marine environment is rapidly increasing. They frequently
address natural constituents such as propagation pathways of different water masses
(Meier 2007), suspended matter (Gräwe and Wolff 2010), fish eggs and larvae (Mar-
iani et al. 2010) or turtle hatchlings (Monzon-Argullo et al. 2010). Another major
class of studies of Lagrangian propagations relates to different adverse impacts such
as oil (Korotenko et al. 2004), microorganisms (Korajkic et al. 2009) or marine litter
(Yoon et al. 2010).
The technique of Lagrangian transport is commonly used to determine which
environment would be most seriously damaged so that it may receive priority pro-
tection (see Abascal et al. 2010 and references therein). The majority of the relevant
research efforts address the direct problem of current-induced propagation of pas-
sive tracers, with the aim to make clear where an object, particle or substance will
be transported if the initial fields of velocities, forcing and boundary conditions (or,
6 A deeper treatment of the Eulerian and Lagrangian specifications of the flow is presented in
Chap. 7. For further information we recommend classical sources such as Batchelor (1973), Lamb
(1994).
294 T. Soomere
Fig. 9.6 The complexity of trajectories of water particles calculated using the TRACMASS code
(Chap. 7) and the Rossby Centre Ocean (RCO) model data in the Baltic Sea entering the sea
through the Øresund (red), Great Belt (green) and from River Neva through the Gulf of Finland
(blue). Visualization by K. Döös (Soomere et al. 2011a)
alternatively, the properties of the Eulerian flow field) are known (e.g., Korotenko
et al. 2010). The studies cover a wide range of applications, from a verification
of the classical circulation models beyond that offered in an Eulerian assessment
(Ohlmann and Mitarai 2010) up to intricate statistical models of oil spill propaga-
tion based on a large number of propagation scenarios (Abascal et al. 2010), eval-
uation systems based on simulating the transport of the underlying contaminant or
toxic algae (Chrastansky and Callies 2009; Havens et al. 2010) or forecasting search
areas using ensemble modelling (Melsom et al. 2012). Following the widely used
approach in atmospheric sciences (e.g., Lin et al. 2004; Witham and Manning 2007),
the analysis of Lagrangian trajectories is now increasingly used for the restoration of
the link between the source and impact areas of pollution in the marine environment
(Chrastansky et al. 2009, among others).
9.3.2 Circulation Models
A detailed study of reasonable parameters and temporal and spatial scales of the cal-
culations of Lagrangian trajectories, suitable for highlighting semi-persistent current
patterns in the Gulf of Finland, is presented in Viikmäe et al. (2010), Soomere et al.
(2011a). They used the surface-layer velocity fields from 3D current velocity data
simulated for 1987–1991. This time period matches the one used in earlier circula-
tion simulations (Andrejev et al. 2004a, 2004b) and was also used in a subsequent
series of studies into properties of probability distributions for coastal hits (events
when a tracked particle reaches the coast) and particle age (also residence time, un-
derstood here as the time elapsed from the release until the first coastal hit) in the
Gulf of Finland (Soomere et al. 2010, 2011a) described in Chaps. 10 11. These years
were quite typical in terms of wave intensity (Broman et al. 2006) and thus also in
terms of energy supply to water masses.
The velocity data used in these studies were calculated by the Rossby Centre
Ocean Model (RCO) for the entire Baltic Sea. As this model and the corresponding
forcing are described in detail in Chap. 4, only the basic information is provided
here for convenience. The RCO model is a Bryan–Cox–Semtner primitive equation
circulation model following (Webb et al. 1997) with a free surface (Killworth et al.
1991) and open boundary conditions (Stevens 1991) in the northern Kattegat. It is
coupled to a Hibler-type sea ice model (Hibler 1979) with elastic-viscous-plastic
rheology (Hunke and Dukowicz 1997). Subgrid-scale mixing is parameterized us-
ing a turbulence closure scheme of the k–ε type with flux boundary conditions to
include the effect of an enhanced layer of turbulence due to breaking surface gravity
waves (Meier 2001). A flux-corrected, monotonicity-preserving transport scheme
following (Gerdes et al. 1991) is embedded. No explicit horizontal diffusion is ap-
plied.
The horizontal resolution of the model grid is 2 × 2 nm and the model uses 41
vertical levels in z-coordinates. The thickness of the vertical layers varies between
3 m close to the surface and 12 m in 250 m depth. The model run, data from which is
used in Viikmäe et al. (2010), Soomere et al. (2011a) is forced with 10 m wind, 2 m
air temperature, 2 m specific humidity, precipitation, total cloudiness and sea level
pressure fields from a regionalization of the ERA-40 reanalysis over Europe using
a regional atmosphere model with a horizontal resolution of 25 km during 1961–
2007 (Höglund et al. 2009; Samuelsson et al. 2011). It also accounts for river inflow
and water exchange through the Danish Straits. As the atmospheric model tends to
underestimate wind speed extremes, the wind is adjusted using simulated gustiness
to improve the wind statistics (Höglund et al. 2009). Standard bulk formulae are
used to calculate the air–sea fluxes over open water and over sea ice. For further
details of the model set-up and an extensive validation of the model output the reader
is referred to Meier (2001, 2007), Meier et al. (2003).
Given the very small values of the baroclinic Rossby radius in the Gulf of Finland
(typically 2–4 km, Alenius et al. 2003), the RCO model apparently resolves a certain
part of the mesoscale dynamics in this gulf but an exact representation of the location
and properties of single eddies cannot be expected. In other words, models with this
resolution are barely eddy-permitting in this basin (cf. Andrejev et al. 2010, 2011).
The barotropic time step was 15 s and the baroclinic time step 150 s. In order to
keep the velocity data set within a reasonable limit, the model output was saved
once in 6 hours. Therefore, the simulations resolved inertial oscillations in the gulf
but owing to the coarse temporal resolution of saved data (about 1/2 of the period of
inertial oscillations at the latitude of the Gulf of Finland) these oscillations are not
resolved in the drift of particles (Chap. 8).
296 T. Soomere
9.3.3 Reconstruction of Trajectories of Water Particles
Calculation of a Lagrangian trajectory of a particle, carried passively by sea cur-

rents, is equivalent to solving the Cauchy problem for a simple system of differential
equations
dX dY dZ
= u, = v, = w,
dt dt dt (9.2)
Xt=t0 = x0 , Yt=t0 = y0 , Zt=t0 = z0 ,
where (X, Y, Z) is the instantaneous location of the particle, (x, y, z) are spatial
coordinates, (u, v, w) are the velocity components, t is time and (x0 , y0 , z0 ) is the
location of the particle at the starting time instant t0 (Chap. 7). The results of such
calculations obviously depend on the quality and spatial and temporal resolution of
the underlying velocity fields. It is therefore crucial to use the best available circu-
lation models and forcing and boundary conditions in order to obtain an adequate
representation of the statistics of the modelled Lagrangian trajectories.
As mentioned above, in this and the following chapter we only use the horizontal
velocities (u, v) in the uppermost model layer to model the Lagrangian transport of
selected water particles. The vertical velocity w and the equation for Z(x, y, t) are
ignored. Thus the resulting trajectories are not truly Lagrangian (except for areas
where w ≥ 0 in this layer) and essentially represent the transport of particles or
substances that are locked within the uppermost layer.
Although both wind drag and wave-driven impact on the trajectories of drifting
objects could be substantial (Rohrs et al. 2012), we focus on the problem of statis-
tics of purely current-driven transport. In other words, the impact of wind drag on
the floating objects (lost containers, debris, litter, oil spill, algae, etc.) is ignored
as well as wave-driven Stokes drift and other wave-induced effects (Breivik et al.
2011). A partial reason for this simplification is that the impact of wind and waves
is very site- and object-specific (Breivik et al. 2012; Rohrs et al. 2012). The realistic
behaviour and fate of oil spill under the joint impact of currents, wind and waves is
studied in Chap. 11.
There is extensive evidence that the drift of oil spills (and eventually other sub-
stances that form a thin film on the sea surface) usually differs considerably from
the trajectories of other objects, incl. the devices specifically designed to follow
the oil spill (Fingas 2011). This situation calls for us to take a step back and make
more efforts towards a better understanding of current-driven transport, which still
seems to be the largest source of uncertainties in the prediction of surface-layer drift
(Vandenbulcke et al. 2009).
The use of the output of circulation models for solving Eqs. (9.2) is justified
for many occasions in several domains of the World Ocean. This setup is generally
applicable for studies of contaminants or dissolved radioactive substances in a thin
surface layer of lighter water overlying denser water masses. Hydrometeorological
conditions supporting such a manner of transport occur, for example, regularly and
during relatively long time periods in the Gulf of Finland. In the spring season the
wind is normally very weak there (Mietus 1998) and the uppermost well-mixed
Fig. 9.7 A selection of 60-day long trajectories of water particles calculated using the
TRACMASS code and the RCO model data in the Gulf of Finland for November–January 1987.
Red circles show the starting points of the trajectories (Soomere et al. 2011a)
layer is very thin (Chap. 6). This situation often persists for months, during which
the drift of virtually all items in the uppermost layer is mostly governed by currents.
Owing to the voluminous river discharge into the eastern Gulf of Finland a similar
layer of fresher water exists under ice. Note, however, that the bottom of the ice
cover is normally not smooth (Leppäranta 2013) and floating items tend to stick to
the ice and drift with the speed of the ice.
The TRACMASS model (Chap. 7) (Döös 1995; de Vries and Döös 2001) com-
putes the trajectories off-line, i.e., after the circulation models have been integrated
and the velocity fields have been stored. This model uses a linear approximation
of the simulated current velocities at the walls of grid cells to solve the first two
equations (9.2) exactly for each particle in this cell. Doing so allows explicitly cal-
culating the time instant at which the particle would enter another cell. The position
of each particle is updated either at the end of each time step of the velocity data
(six hours for the RCO model output) or when it crosses any wall between the cells,
whatever event occurs first. In order to keep the resulting data set reasonable, the
coordinates of the trajectory points were saved once in six hours. Doing so some-
times caused side-effects such as trajectories seemingly crossing some peninsula or
islands (Fig. 9.7) but apparently did not substantially affect the resulting statistics
of transport over several days.
The resulting trajectories evidently depend to some extent on the available tem-
poral resolution of both the circulation data and the trajectory calculation scheme.
In extreme cases, the resulting differences may lead to rapid divergence of initially
close trajectories. The experience with the TRACMASS code, however, reveals that
the spreading of the calculated trajectories is normally much smaller than the spread-
ing of real drifters owing to the effect of sub-grid turbulence, and initially close
trajectories have an overly tendency to stay close (Chap. 8). This feature implic-
itly suggests that the potential impact of the choice of the temporal resolution in
trajectory reconstruction is minor in terms of the statistics of a large number of
trajectories. These questions are addressed in Chaps. 7 and 8.
298 T. Soomere
Fig. 9.8 Sketch of splitting

the simulation period into
time windows
9.3.4 Splitting the Simulation Periods
Lagrangian trajectories are the key test elements used below to highlight certain hid-
den transport properties. The procedure employed for the construction of a required
set of trajectories was the same in all the implementations of the developed technol-
ogy. The entire time period of interest [t0 , t0 + tD ] with duration tD was divided into
time windows of fixed length tW (Fig. 9.8). The initial locations of a certain number
of water particles (interpreted as carrying the adverse impact) were specified. Their
motion paths (interpreted as trajectories of current-driven propagation of the adverse
impact) were first simulated over the interval [t0 , t0 + tW ]. In some simulations (e.g.,
Soomere et al. 2010, 2011a; Lu et al. 2012) all the resulting trajectories were saved
for further analysis, whereas other authors (e.g., Andrejev et al. 2011) only saved
certain distributions calculated based on these trajectories.
In order to increase the number of independent trajectories, simulations for the
same initial positions of particles were restarted at another time instant [t0 , t0 + tS ].
The trajectories were again calculated over a time window with duration tW . While
in the simulations of Soomere et al. (2010, 2011a) the time windows largely overlap,
Andrejev et al. (2010, 2011) used tW = tS and sequential time windows. The process
was repeated (tD − tW )/tS times (Fig. 9.8). The long-term properties of Lagrangian
transport were found via averaging of the results over a proper selection of time
windows. For example, for a yearly simulation with a time window of tW = 20
days and with a lag tS = 10 days, the averaging was performed over 34 ensembles
of trajectories, the last examples of which started on 07 December and ended at
midnight of 26 December.
9.3.5 Simulating Statistically Independent Trajectories
The suitability of the resulting sets of trajectories to reflect the important features of
surface transport or to be used for the developed technology depends on the choice
of several parameters and options. Some of them (e.g., the spatial resolution of the
circulation model, the trajectory calculation scheme, the temporal resolution of the
velocity data, and the methods of modelling the spreading of trajectories) are usu-
ally prescribed externally. Among these parameters the horizontal resolution of the
circulation model has probably the largest impact (Andrejev et al. 2011). For nar-
row sea areas with complicated internal dynamics such as the Gulf of Finland the
medium-resolution RCO model does not reproduce many local bathymetric features
(cf. Andrejev et al. 2010). The use of a proper representation of the nearshore for
the TRACMASS model (three grid cells, see below) means that the offshore of the
Gulf of Finland is only represented by about 10 grid points in its narrowest part.
The temporal resolution of saved velocity data (6 hours for the RCO model data)
evidently distorts the impact of inertial oscillations and only partially accounts for
the mesoscale motions.
The control over other parameters is still in the hands of the operator of the
technology. The resulting discrete problem contains three intrinsic time scales:
• Duration of single trajectories (time window tW );
• Selection of water particles in time (time lag tS between subsequent simulations);
• Time interval [t0 , t0 + tD ] covered by the calculations.
Additionally, several other parameters may affect the appearance and reliability
of the results:
• Number of trajectories used for statistical estimates;
• Numerical representation of the vulnerable areas;
• Target function (e.g., whether or not the climatologically valid solution is tar-
geted, see Chap. 11).
Several items in the list can be specified by a suitable choice of time scales (e.g.,
for which semi-persistent patterns may be important in the particular basin, or the
typical time for coastal hits). The use of very short propagation times (trajectory
lengths less than a few hours) is normally imprudent as no adverse impact can reach
remote areas during such a short time. The propagation time should be long enough7
to allow for a significant number of particles to reach the vulnerable area(s). Also,
the hidden properties of Lagrangian transport can only be revealed by trajectories
that are long enough to highlight net dislodgements of surface water.
A rough estimate of the reasonable time scales could be extracted from a com-
parison of the appearance of the Eulerian and Lagrangian transport in various en-
vironments. For a rectilinear flow (a generalisation of a stationary jet current in the
sea) the Eulerian and Lagrangian velocities and transport are equivalent and a proper
time scale should match the lifetime of the jet currents.
These velocities and the relevant transport rates are usually very different in envi-
ronments containing numerous synoptic eddies. In the core of a stationary eddy the
Eulerian speed is constant and the Eulerian transport increases linearly with time.
The water particles ideally exert circular motions and periodically return to their
initial locations. The direction of the motion of a water particle (equivalently, its
Lagrangian velocity) varies periodically in time but the relevant speed is constant.
Its displacement (Lagrangian transport) periodically varies from zero to a certain
maximum value. Consequently in eddy-dominated (and even in eddy-containing)
7 It is, however, equally unwise to trace the trajectories during very long time intervals as short-
living patterns will be smoothed out and the properties of the tracked items or substances eventually
will change.
300 T. Soomere
flows it is generally necessary to average over time intervals longer than the typi-
cal turnover time of eddies in order to get a flavour of the properties of Lagrangian
transport. This scale is about five days in the Gulf of Finland and in the SW Baltic
Sea, and somewhat longer, some 10–15 days, in the Baltic Proper.
9.3.6 Subgrid Processes and Spreading of Trajectories
The impact of subgrid-scale processes inevitably leads to certain deviations of the

simulated trajectories from the measured paths of water particles or tracers. For
example, in the Ligurian Sea, the simulated and measured trajectories are usually
highly correlated only within 1–1.5 days (Vandenbulcke et al. 2009). The situation
is even more complicated for the Gulf of Finland and in the Northern Gotland Basin,
where the High Resolution Oceanographic Model of the Baltic Sea (HIROMB) cir-
culation model (Kõuts et al. 2010) and Seatrack Web (Verjovkina et al. 2010) (basi-
cally also relying on the HIROMB engine) demonstrated a reasonable match during
about 10 hours in autumn 2007. A more satisfactory reproduction of pathways of
surface drifters was obtained in the Gulf of Finland (Gästgifvars et al. 2006) us-
ing the OAAS model (see Chap. 10 for its description). Trajectories simulated for
longer than about one day will therefore usually not match the motions of real water
particles in the test areas.
Numerical models will probably never be able to replicate the smallest details
of ocean turbulence. Although a certain increase in the accuracy of simulations of
circulation and particular trajectories is viable, e.g., through an improvement of the
spatial resolution or via a decrease in the time step of the circulation model or by
means of an increase in the complexity of the tracking scheme (Gräwe et al. 2012),
exact matches of longer simulated trajectories with real motions of tracers are very
unlikely. The loss of this exact matching is not critical as a large number of different
pathways may lead to equivalent displacements. These displacements and associated
transport patterns can still be extracted using statistical analysis of a large number
of trajectories. For such an analysis to be reliable, it is necessary to adjust the pa-
rameters and options listed above so that these features will be adequately captured
and/or highlighted.
A proper representation of statistical features of transport requires the use of
a sufficient number of single trajectories. The number of simultaneously calculated
independent trajectories is implicitly limited by the number of grid cells in the circu-
lation model. The overview in Chap. 7 demonstrates that the version of the TRAC-
MASS trajectory model used in Soomere et al. (2010, 2011a) does not reproduce
the spreading of real water particles or drifters. This does not mean that the TRAC-
MASS model is wrong. It correctly uses all the data from the circulation model but
cannot guess into which direction and how rapidly the real drifter would move under
the impact of factors that are not resolved by the circulation model.
It is natural to assume that the ignoring of subgrid processes does not consid-
erably alter the statistics of net displacement of the tracked water particles if the
number of test particles and their initial separation are large enough. The limited
spreading of TRACMASS trajectories means that initially closely packed particles
will move together and the resulting trajectories will be strongly correlated. There-
fore, it makes no sense to pack many test particles next to each other at a certain
time instant. This feature is accounted for in all described experiments that use
TRACMASS. The largest number of particles in one grid cell was 5 in Soomere
et al. (2011a), whereas Lu et al. (2012) seeded one particle into each fourth cell.
The trajectory models that contain an internal spreading scheme (that simulates
the impact of subgrid-scale motions) can easily make reasonable use of 10 trajecto-
ries per each grid point (Andrejev et al. 2010, 2011). Such schemes usually insert
a random disturbance to each trajectory after certain time intervals. This procedure
corrects certain statistical features of large pools of trajectories but obviously does
not improve the match of the resulting trajectories with the motion of real water par-
ticles. The best way to increase the number of trajectories is therefore to improve the
spatial resolution of the circulation model or to increase the time interval covered
by the circulation modelling.
9.3.7 Overlapping Simulations
Another feasible way of increasing the number of uncorrelated trajectories is to

restart the calculations with a given set of water particles already before the end of
the previous run. Restarting the calculations after very short time intervals should
be avoided as doing so would lead to equivalent trajectories and thus serially corre-
lated results, and would not improve the final statistics. The surface current speed
in the test areas is typically well below 0.5 m/s and about 0.2–0.3 m/s in the Gulf of
Finland. Within one time step of the RCO velocity data (6 hours) a particle would
normally move by 4–6 km, that is, at least into a neighbouring grid cell. It is rea-
sonable to assume that trajectories that are initially separated by a distance of about
the grid size of the circulation model (equivalently, start at least 6 hours after the
previous one in the case of the 2 nm resolution RCO model), can be treated as inde-
pendent ones.
The use of a longer time lag between subsequent calculations leads to a decrease
in the chances for having equivalent (or strongly correlated) trajectories. As the re-
sulting set of trajectories is smaller, doing so may still adversely affect the reliability
of the conclusions (Viikmäe et al. 2010). An analysis of the pools of trajectories
shows that the results are relatively insensitive with respect to variations in the time
lag tS from 1 to 10 days for the RCO data for 1987–1991 (Viikmäe et al. 2010).
This suggests that the accuracy and reliability of the results first of all depends on
the number of trajectories involved and less on how densely they are distributed in
time. This conjecture becomes important in the optimization of long-term calcula-
tions based on high-resolution simulations (Andrejev et al. 2010).
The presented estimates and their more exact specification in particular experi-
ments only concern the reasonable values of different parameters necessary for the
302 T. Soomere
identification of semi-persistent patterns8 but do not provide any information about

their existence in reality. In general, for each study area, circulation and trajectory
model, it is essential to evaluate the necessary spatial and temporal scales in order
to reach representative results.
Several parameters may have different optimum values depending on the partic-
ular problem addressed. For example, a time interval covering at least three years
is necessary to reach an appreciable approximation to the climatologically valid op-
timum fairway in the Gulf of Finland (Andrejev et al. 2011). For certain sea areas
(such as the vicinity of the Danish Straits) a climatological solution might be a poor
compromise and the optimum should be determined separately for dynamically dif-
ferent but irregularly occurring flow regimes such as the inflow and outflow condi-
tions (Lu et al. 2012). The solutions may also exhibit substantial seasonal variations
Soomere et al. (2011a).
9.4 Simulations of Environmental Risks
9.4.1 The Coast as the Vulnerable Region
It is straightforward to specify the boundaries of the vulnerable regions that are lo-
cated offshore where their boundaries do not modify the currents and current-driven
transport (Delpeche-Ellmann and Soomere 2013). The treatment of such boundaries
usually needs some care when the vulnerable area is at the coast or in the nearshore.
The circulation models usually prescribe that the velocity component normal to the
rigid boundaries (bottom and coast) vanishes in the boundary cells (Chaps. 3 and 4).
This feature, although physically correct, constrains the simulated flow to be largely
alongshore in the nearshore and in many cases suppresses the motions towards the
coast at a distance of a few grid cells.
The rate of suppression of the cross-shore motions depends on the details of
the circulation model. The relevant features are usually not explicitly described and
should be requested from the developers of the models. The problem becomes ev-
ident in modelling the beaching of oil spills. The exact current-driven trajectory of
an oil spill (or any other item) follows the velocity data and, in principle, cannot
reach the land even if it has been brought into a grid cell adjacent to land by, e.g., a
strong downwelling event. In particular, a coastal hit of a trajectory calculated using
TRACMASS without spreading is impossible (Chap. 7). A coastal hit may happen
if, e.g., a particular trajectory is constructed using a too long time step and low-order
difference scheme.
In the real ocean wind drag and wave-driven effects substantially contribute to
the beaching of different floating items. The process of beaching is usually solved
8 Here we have in mind patterns of currents or transport that are directly related neither to single
synoptic eddies, particular events of coastal (or otherwise topographically controlled) jet currents
nor to the mean circulation in the particular basin.
Fig. 9.9 Starting points of trajectories in simulations of coastal hits. Dark grid points indicate the
3 grid cell wide nearshore area. The entrance to the gulf is set along 59◦ N, 21◦ 48 E. Graphics by
B. Viikmäe
indirectly, for example, by including the local wave- and wind-induced transport
into the Seatrack Web (Ambjörn 2007) or by defining the control line for coastal
hits at a certain distance from the geographical coast (Broström et al. 2011). The
use of artificial spreading in trajectory simulations would also do the job (Andrejev
et al. 2011).
The described ‘repulsion’ from the coast of modelled trajectories may lead to
inadequate estimates of the typical drift time to vulnerable nearshore areas and may
also distort the spatial distribution of the coastal hits. An example of an estimate
of the magnitude of this effect, used for the decision-making about a reasonable
location of the model boundary of the vulnerable nearshore, is demonstrated in Vi-
ikmäe et al. (2010). They performed simulations using the 2-mile RCO model and
the TRACMASS code for particles seeded once a day into centres of 93 cells along a
line roughly representing the axis of the Gulf of Finland (Fig. 9.9) for the year 1987.
This year was a usual one: there were no exceptional storms and the annual mean
wind speed was just a few percent lower than the average for 1987–1991 (Andrejev
et al. 2011).
The suitable location of the virtual coast was simulated by means of three zones
with a typical width of 1, 2 and 3 grid cells from the coast (called alert zones 1–3).
A hit to each alert zone was counted when a trajectory reached the seaward edge of
the zone for the first time. As the gulf had an open boundary, a part of the particles
were carried out of the gulf. The presence of each particle in each alert zone (or its
drift out of the gulf) was accounted for only once.
The monthly average number of hits to the alert zones and the fraction of particles
leaving the gulf revealed substantial seasonal variability (Fig. 9.10). The probability
to enter alert zone 1 (that is, into the grid cells directly adjacent to the land mask)
was only 4.86 %. This probability is larger, about 11 % on average, for alert zone 2.
In other words, the behaviour of 90–95 % of the trajectories did not contribute to
the estimates of environmental risks to these zones (Viikmäe et al. 2010). As a con-
sequence, the statistics of coastal hits would be based on quite a small number of
trajectories and would have quite large uncertainty.
304 T. Soomere
Fig. 9.10 Percentage of hits to the nearshore for 15-day long trajectories for different starting in-
stants in 1987 with a time lag of tS = 1 day. The uppermost dashed line shows the total percentage
of particles that have either hit zone 3 or drifted out of the gulf (Viikmäe et al. 2010). A similar set
of tests for the SW Baltic Sea is presented in Lu et al. (2012)
9.4.2 Time to Reach the Coast and the Hitting Rate Reflect
the Surface Dynamics
The proper definition of the nearshore is to some extent connected with the adequate
choice of the length of trajectories. This length generally cannot be extracted from
the modelled or measured Eulerian velocities as they do not necessarily reflect ad-
equately the net displacement speed of surface particles for eddy-containing flows.
Although calculations in Anonymous (2002) suggest that oil pollution may cross the
Gulf of Finland within 1–1.5 days on typical autumn days, Fig. 9.10 reveals that the
drift time from the central part of the gulf into the nearshore is generally quite long.
In many cases the trajectories first enter the nearshore area after about 10 days of
propagation (Viikmäe et al. 2010) and it takes about 15 days for half of the particles
released at the centreline of the gulf to reach the coast.
The typical time over which the particles reach the vulnerable area (nearshore)
in such flows can be extracted from similar experiments. A further analysis of the
3–15-day long trajectories of the above series of simulations of clusters of 93 par-
ticles released along the centreline of the gulf (Fig. 9.9) revealed a substantial sea-
sonal variability of this time (Viikmäe et al. 2010). The monthly average fraction
of particles hitting alert zone 3 increased rapidly (and almost linearly) when tW in-
creased from 3 to 15 days (Fig. 9.11). Both the hitting count and the increase rate
were very low for the calmest months, namely March and May. The largest num-
ber of coastal hits, about 30–40 % of the particles, occurred in the windiest months
(October and November). The percentage of particles hitting the coast showed a
considerably lower increase rate after about half of the particles had either already
hit the nearshore or left the gulf (after about a 10-day long drift). Although this rate
considerably decreased during the second week, it still did not stabilize by the 15th
day.
The low probability of entering zones 1 and 2 (Fig. 9.10) does not mean that the
particles continuously stay far offshore. The annual average probability of entering
alert zone 3 is 19 %, whereas it is as large as about 50 % during the windy months.
Fig. 9.11 Monthly mean percentage for coastal hits (black) and leaving the gulf (white) for dif-
ferent lengths of the time window for alert zone 3 in 1987. For each month, columns from left to
right show the percentage for tW = 3, 5, 7, 10, 13 and 15 days (Viikmäe et al. 2010)
The joint probability for a particle to either enter alert zone 3 or to leave the gulf
varies from about 5 % in spring to close to 100 % during the windiest period, with
an annual average of about 25 %. Therefore statistics calculated with the use of alert
zone 3 as a model of the nearshore is generally representative for the velocity data
in use. The use of alert zones 1 and 2 may only be justified during the windy season
or if the total number of trajectories would be considerably increased. The relevant
values apparently will be larger for particles distributed randomly over the Gulf of
Finland but even then the statistics for the calm seasons would be poor.
The number of particles drifting out of the gulf increases almost linearly with the
increase in tW . The absence of an evident correlation with the number of nearshore
hits suggests that the ‘open sea’ and ‘nearshore’ dynamics in the Gulf of Finland are
relatively well separated even when the nearshore is 11 km wide and covers up to
40 % of the width of the gulf. These features are consistent with the dynamics of the
Gulf of Finland. The depth-averaged circulation in this basin is mostly cyclonic with
a typical velocity of a few cm/s (Alenius et al. 1998; Leppäranta and Myrberg 2009).
This slow gyre serves as a background for a number of mesoscale eddies that usu-
ally drift to the west (Andrejev et al. 2004a, 2004b) and in this way carry entrained
surface particles gradually towards the Northern Gotland Basin. Such a structure
becomes evident in all eddy-permitting and eddy-resolving simulations (e.g., An-
drejev et al. 2004a, 2004b). As is typical for estuarine circulation of subbasins with
an excess of fresh water, outflow from the Gulf of Finland largely occurs in the
surface and subsurface layers (Andrejev et al. 2004b), which results in the drift of
selected particles out of the gulf.
The trajectories of particles that leave the Gulf of Finland (typically about 10 %
of the released ones for tW = 15 days; Figs. 9.10, 9.11) are not accounted for in the
calculations of the flow properties. As their number exceeds the number of trajecto-
ries hitting the coast during selected months, this outflow may lower the accuracy of
the estimates of coastal hits. This number suggests that the surface water exchange
between the Northern Gotland Basin and the Gulf of Finland may be much more
intense than the overall water exchange in the entire water column (Andrejev et al.
2004b). If about 10 % of the surface water leaves the gulf within two weeks, it might
take only about half a year for the total removal of the surface water from the gulf.
306 T. Soomere
Earlier simulations indicate that much of the water is apparently transported back
and forth at the entrance to the gulf (Andrejev et al. 2004a, 2004b) and that the net
exchange forms a relatively small fraction of the total exchange. The almost constant
number of particles leaving the gulf in a month (Fig. 9.11) suggests that the outflow
of surface water is much more regular than the processes in the entire water column.
Only in August and September (which are usually much windier than May, June or
July, Mietus 1998) it is less intense.
This counter-intuitive behaviour probably reflects the different proportion of the
wind-induced Ekman drift (see Chap. 2, Sect. 2.3.5 for details) in different sea-
sons. For example, strong SW winds (that are common in autumn) generate surface
Ekman drift to the east or SE, that is, towards the gulf interior. The surface dynam-
ics seems to be relatively weakly correlated with the dynamics of the underlying
water masses during windy months and sometimes may even form an anticyclonic
gyre (Soomere et al. 2011a). In calm seasons and under ice cover, however, the un-
derlying dynamics evidently will play a much larger role in the surface dynamics
(Gästgifvars et al. 2006).
The number of particles hitting the coast exhibits substantial temporal variabil-
ity (Fig. 9.11). The lesson is that applications of the presented method require a
careful choice of the governing parameters for each particular sea area and circu-
lation and trajectory model. A reliable statistics of coastal hits requires a sensible
amount of selected particles (carrying the adverse impact) to reach a properly de-
fined nearshore. On the one hand, the time window should provide enough time for
a reasonable number of coastal hits to occur. For example, values of tW below 10
days are probably inappropriate for the analysis of coastal hits in the Gulf of Finland
during most months. In single years and the calmest months only a small fraction of
even 15–20-day long trajectories would enter the nearshore. On the other hand, the
average percentage of coastal hits should be well below 100 % to properly resolve
spatial details of the ‘ability’ of different sea points to provide coastal pollution.
As a rule of thumb, tW should not be much longer than the typical time at which
the largest number of hits occurs (Viikmäe et al. 2010); otherwise the rest of the
computing time of trajectories will not be justified. Still, the use of relatively long
time windows generally better resolves the spatial structure of maps of the time it
takes to reach the coast for particles released in different offshore domains. The
length of the time window used in experiments for the Gulf of Finland varies from
10 days (Soomere et al. 2010, 2011a; Andrejev et al. 2011; Lu et al. 2012) to 20
days (Soomere et al. 2011c).
As noted above, the calculation scheme does not account for the impact of
subgrid-scale turbulence on the shape of the trajectory. Although it is not likely
that the relevant effects could substantially modify the statistics of trajectories, it is
still interesting to roughly estimate the magnitude of related effects under the very
unlikely assumption that they always tend to deviate the simulated path from its
actual appearance. Its impact apparently is small in terms of statistics of isotropic
flow patterns and/or more or less circular basins but may considerably affect the
probability of coastal hits in elongated basins such as the Gulf of Finland.
First of all, the typical spreading of initially closely located particles over the
time window should remain below the width of the narrowest part of the gulf. If
this condition is violated, the uncertainty in the positioning of the particle caused
by subgrid-scale turbulence would be about the same size as the extension of the
open sea area and the related statistics of coastal hits could become meaningless.
Numerical simulations of Andrejev et al. (2010) assume that the typical spreading
rate is about 2 mm/s. Drifter experiments in the Baltic Proper (Kjellsson and Döös
2012) (see also Chap. 8) and in the Gulf of Finland (Soomere et al. 2011b) suggest
that this rate apparently is somewhat larger.
Therefore, the subgrid turbulence may separate the particles on average by 15 km
within about three weeks. This suggests that for time windows longer than about 20
days the final position of the particle would be basically random. As the majority
of coastal hits occurs much earlier and the actual rate of deviation of the modelled
path from the theoretically valid one is on average much smaller, the resulting dis-
tributions of the probabilities of coastal hits, the time it takes for the pollution to
reach the coast and net and bulk transport patterns evidently reflect well the reality
as simulated by the circulation model.
Summing up, the basic time scales for a reasonable calculation of the Lagrangian
trajectories, for current-induced semi-persistent features of transport of surface wa-
ter and for the adequate location of the model boundary of the vulnerable areas
should be analysed separately for each basin. For the RCO model and the non-
spreading version of the TRACMASS trajectory model it is appropriate to use a
nearshore area about 3 grid cells wide as a proper representation of the coastal zone.
A sensible length of time windows in calculations of coastal hits in the Gulf of Fin-
land is at least 10–15 days. About 10 % of the released particles drift out of the gulf
and about one-third of the particles released in the central part of the gulf enter the
model nearshore during this time.9
9.4.3 Temporal Scales for Transport Patterns in the Gulf

of Finland
We demonstrate several of the above-discussed issues and the importance of the

proper choice of the length of the time window for the example of areas of rapid net
and bulk Lagrangian transport in the Gulf of Finland. The net transport is defined as
the distance between the start and end positions of a trajectory (Fig. 9.12) and the
bulk transport as the length of the entire trajectory. Patterns of intense net transport
may become evident in many cases of current fields that have low overall long-term
persistency Rp . A typical example is a system of coastal currents where the flow
direction alternates in weekly scales. They may rapidly and systematically relocate
patches of various substances between different domains even when the long-term
current system is almost perfectly random. Their location and magnitude can be
identified numerically using the above discretization of calculations into a sequence
of numerical simulations of Lagrangian trajectories of water particles.
9 The long-term average of this rate becomes explicitly evident below as the long-term average
probability of coastal hits for the entire basin, see Chap. 10.
308 T. Soomere
Fig. 9.12 Sketch of the net

and bulk transport occurring
along a Lagrangian trajectory
In contrast to the problem of coastal hits, relatively short time windows might
be used to highlight pathways of rapid transport that may persist only for a few
days in some areas. The areas of rapid net transport for a single, relatively short
time window obviously coincide with areas of large Eulerian current speeds. The
use of very short time windows therefore will result in smoothed patterns of the
average Eulerian velocities. The areas of fast flow will generally be different for
different time windows as the local jets and mesoscale eddies emerge, relocate and
decay over time. The use of a too long window would result in a variation of the
mean circulation pattern. A reasonable solution is the use of a variable-length time
window for the search of such patterns similar to the use of wavelet analysis (Ad-
dison 2002) for the identification of coherent transient wave patterns in complex
wave fields. An analysis of results obtained using a selection of properly chosen
time windows has the largest potential to highlight regions of systematic and in-
tense water transport, for example areas where jets alter their direction over certain
time scales.
As a rule of thumb, the relevant time window should roughly match the typical
synoptic eddy turnover time. Numerical simulations and a few available observa-
tions of single eddies in the Gulf of Finland (Soomere et al. 2008) suggest that
their typical core diameter is 10–20 km and turnover time is about 4–5 days. The
overall ability of a set of trajectories to highlight rapid pathways of net transport
can be roughly estimated by comparison of the average speed of net transport for
this set with the long-term average current speed (Viikmäe et al. 2010). The differ-
ence apparently is the largest for short time windows when the net transport speed
matches the instantaneous current speed. At a sensible upper limit for tW the net
transport speed becomes close to the long-term average current speed. For even
longer time windows the semi-persistent flow patterns will probably be averaged
out.
This difference was estimated in Viikmäe et al. (2010) based on the surface ve-
locity data from the RCO model for 1987–1991 with tW extending from 2 to 15
days and a time lag of tS = 1 day. One particle was released into each of 3131 grid
cells in the Gulf of Finland (Fig. 9.9). Their evolution was tracked with the non-
spreading version of the TRACMASS model. The average speed of net transport
decreased from about 5 cm/s to 3.4 cm/s when tW increased from 2 to 10 days. It
reached values close to the long-term average speed (about 2.5 cm/s) for tW ≥ 15
days (Fig. 9.13). Therefore, the range of suitable lengths for time windows for this
purpose is 5–15 days in the Gulf of Finland.
Fig. 9.13 Dependence of the average net transport speed on the length of the time window in the
Gulf of Finland for 1987–1991 (Viikmäe et al. 2010)
The developed analysis can be applied to detect certain internal characteristics

of the flow. Such estimates implicitly characterize the ability of the underlying cir-
culation model to represent the basic structure of currents in the area in question.
An example of the use of the dependence of the ratio of net and bulk transport on
the length of the time window to estimate the properties of mesoscale eddies is de-
scribed in Soomere et al. (2011a). This ratio is close to 1 for very short time windows
when the trajectories are approximately straight. For long time windows it remains
close to 1 for jet currents, approaches a certain limiting value for a gradually trans-
lating eddy field and vanishes for stationary eddies. For eddy-dominated systems it
is expected to decrease rapidly with the increase in tW until tW exceeds the turnover
time of the most energetic eddies, after which the attenuation rate should become
smaller.
The bending point at about 5 days in the relevant graph for the Gulf of Finland
(Fig. 9.14) evidently indicates the typical turnover time of synoptic eddies. Together
with a rough estimate of the maximum speed in the eddy cores it leads to quite a
reasonable estimate of their typical radius (the distance from the rotation centre to
the area hosting largest velocities) of about 5–6 km. This value about by factor of
two exceeds the estimates of the baroclinic Rossby radius for this water body (2–
5 km, Alenius et al. 2003). This feature probably indicates that the RCO model fails
to properly replicate the population of smaller mesoscale eddies in this basin.
9.4.4 Patterns of Net and Bulk Lagrangian Transport
An exercise towards the identification of semi-persistent patterns of Lagrangian

transport in the surface layer of the Gulf of Finland using the described discretiza-
tion, RCO velocity fields for the period of 1987–1991 and the non-spreading version
of TRACMASS for trajectory calculations is presented in Soomere et al. (2011a).
The focus was on making clear how ‘ordered’ the surface transport could be. The
kinematic properties of the current field (such as the annual average longitudinal
and latitudinal flow speed of 0.02–0.07 m/s and the average speed for the five-year
310 T. Soomere
Fig. 9.14 The dependence of the ratio of net to bulk transport on the length of the time window for
1992 over the Gulf of Finland (3131 grid points, one tracer released into each grid centre; time lag
6 hours). Dashed lines show linear trends for tW in the range of 1–5 and 6–30 days, respectively
(Soomere et al. 2011a)
period of 0.033 m/s) as well as strong seasonal variations of these quantities were
characteristic to the Gulf of Finland (Alenius et al. 1998). The drift was mostly to
the east but no predominant direction of the meridional transport was identified. The
surface flow was on average usually to the south in windy seasons, whereas in calm
seasons the motion was directed to the north.
The results indicated that the flow in the surface layer of the Gulf of Finland was
frequently decoupled from the dynamics of the underlying water masses. It not nec-
essarily follows the cyclonic circulationand may develop an apparent anticyclonic
gyre in the central and western area of the gulf. This pattern is not unique in water
bodies of similar size. For example, an anticyclonic gyre apparently dominates in
the southern basin of Lake Michigan (Beletsky et al. 2006) when the upper mixed
layer is very thin (which is frequently the case in the Gulf of Finland). A simi-
lar anticyclonic gyre has been noted in the southern Kattegat (Lu et al. 2012). The
decoupling may reflect the impact of predominant moderate and strong SW winds
(Soomere and Keevallik 2001, 2003). They create Ekman transport to the east or
SE, that is, opposite to the cyclonic circulation in the northern part of the gulf. An
additional contribution to the southwards-directed Ekman transport (not resolved in
the forcing fields of the RCO model) may stem from a temporary turn of SW winds
to the right (more to the west) in the gulf interior (Savijärvi et al. 2005; Keevallik
and Soomere 2010).
A comparison of the net Lagrangian transport over certain time windows tW
with the average Eulerian velocity allows identifying the areas that frequently host
strong flow of appreciable duration even if the flow direction varies over longer
time. As discussed above, by varying the length of the time window and applying
the averaging over selected time periods it is possible to identify patterns persisting
over different time intervals.
Test calculations in Soomere et al. (2011a) tracked particles seeded after each 6
hours at the centre of each grid cell within tW = 4 days. This time scale accounts for
the internal circulation of a part of mesoscale eddies in the Gulf of Finland and is
Fig. 9.15 Average net transport speed (cm/s) for 1987–1991 (left) and for 1987 (right) (Soomere
et al. 2011a)
Fig. 9.16 Average net transport speed (colour scale, cm/s) during the calm season 1988 (upper
left), windy period of 1988/1989 (upper right), spring transitional season (March–May) of 1988
(lower left) and autumn (August–October) of 1989 (lower right) (Soomere et al. 2011a)
about a half of the typical time of the hitting the coast for tracers released in the sur-
face layer of this water (Soomere et al. 2010). The pointwise average speeds of net
transport were calculated over the entire pool of the trajectories and over selected
time intervals or seasons. The elongated areas of intense net transport (Fig. 9.15)
evidently indicate pathways of fast displacements of surface water. The fastest net
flow over 1987–1991 occurred as a wide inflow band along the northern coast of
the Gulf of Finland. The only persistent transport to the west exists along the north-
eastern (NE) coast of Estonia. The match of the average Eulerian current and the La-
grangian net transport is only partial. For example, intense net transport to the west
along the NW coast of Estonia is not visible in the average current field (Soomere
et al. 2011a). The patterns for the average net transport are very similar for all single
years although the patterns of Eulerian currents may vary.
The net transport patterns for the windy (October–March) and calm (May–
August) periods were quite different. This may reflect the large role of subsurface
currents in low wind conditions (Gästgifvars et al. 2006). During calm seasons in-
tense net transport usually occurred along the southern and especially SE coast of
the gulf (Fig. 9.16). During windy seasons a similar band near the NW coast is the
most prominent. In the transitional months for spring (March–May) both the above
features are present. Relatively fast net transport is occasionally observed in the cen-
tral area of the gulf during late summer and early autumn (Fig. 9.16). Interestingly,
fast net transport across the gulf is identified during transitional seasons. The trans-
312 T. Soomere
Fig. 9.17 Ratio of net and bulk transport (colour scale) for 1987–1991 (left) and for 1991 (right)
(Soomere et al. 2011a)
port direction varies: in 1987 it was to the north whilst in 1988 it was to the south,
with a typical speed up to 0.05 m/s, that is, about half of that for the meridional
transport.
Several rapid pathways of net transport revealed substantial excursions of the
flow. The ratio of net and bulk transport additionally showed how straight the mo-
tions of the tracers are. This measure of the variability in the current direction al-
lowed in addition to distinguish the areas hosting unidirectional currents from eddy-
dominated regions. Its reasonable calculation required time windows much longer
than the typical turnover time of eddies.
The relevant distributions found using tW = 15 days in 1987–1991 (Soomere
et al. 2011a) highlighted coastal currents in the SE Gulf of Finland and to a lesser
extent the runoff of the River Neva (Fig. 9.17). This ratio had limited interannual
variability but exhibited substantial spatial variations during different seasons. Inter-
estingly, the regions of high values in calm seasons (Soomere et al. 2011a, Fig. 12)
matched the location of the area of overall high persistency of motions in the sub-
surface layer (Andrejev et al. 2004a).
This chapter presented a selection of outcomes from test simulations developed to

understand the role of the basic time scales, parameters and options of the discretiza-
tion of the inverse problem of environmental management. The obtained knowledge
and established sensible parameters and options serve as a cornerstone of the solu-
tions described in the next chapter. The key message is a demonstration of the ability
of the systematic use of Lagrangian trajectories to reveal usually concealed features
of the surface flow. Examples using this technique indicate the potential richness
of semi-persistent structures in water bodies with complicated dynamics such as
the Gulf of Finland. These patterns are normally hidden in the classical Eulerian
representation. Some of them evidently cannot be extracted from a straightforward
analysis of the Eulerian velocity fields at all.
The presence of relatively intense mostly meridional transport pathways in some
seasons and sections of the Gulf of Finland is highly intriguing. Although they do
not occur every year, their existence is of major importance in the context of the
identification of areas of high risk in terms of pollution of selected coastal seg-

ments. Their presence may substantially impact the necessary length of trajectory
calculations in order to obtain a sufficient amount of coastal hits. This length obvi-
ously depends on both the distance from the release site of an adverse impact to the
vulnerable area and on the cross-shore component of the net Lagrangian transport.
Different transport patterns and pathways contribute to an early evaluation of the
potential benefit of the entire approach via a demonstration how strong the con-
cealed features of current-driven transport are. Such features may become evident,
for example, in the field of the average net transport rate in different directions
and the ratio of average net and bulk transport (equivalently, the ratio of the final
displacement and the length of the trajectories). The viable options for their identi-
fication are an important ingredient of a corresponding decision support system.
The associated implicit possibility for roughly estimating whether or how prop-
erly the underlying circulation model reproduces the mesoscale dynamics can be
used in various contexts. It is natural that some of the results of such a check are
intuitively obvious or match the known picture of the dynamics of the surface layer
in the given sea area. The performed analysis, however, is a step forward towards a
deeper understanding of the role of different dynamic and transport phenomena in
the context of environmental management of vulnerable sea and coastal areas.
The established extensive (mostly seasonal) variability of many revealed patterns
shows on the one hand that unambiguous conclusions about their spatial distribu-
tion, persistency, relative strength or time scales of their formation are not always
possible. On the other hand it suggests that a proper way forward is to develop dif-
ferent solutions for different seasons (or more irregularly occurring flow regimes
such as the inflow or outflow conditions in the SW Baltic Sea) (Lu et al. 2012).
Several highlighted aspects may serve as starting points of future research. This
first concerns sea areas with very small baroclinic Rossy radius and short turnover
time of synoptic eddies where several properties of the Lagrangian transport may
have time scales on the order of a few weeks. This time scale considerably exceeds
the synoptic time scale (which is about a week in the Gulf of Finland) but is sub-
stantially shorter than the length of typical seasonal variations (2–4 months). Such
a separation of the synoptic and seasonal time scales encourages the search for phe-
nomena that persist over ‘intermediate’ time scales of a few weeks. This range is
the most promising for the detection of yet unknown features in the dynamics of
the Gulf of Finland. A similar separation is hardly possible in the open ocean where
the synoptic time scale is about a month and the lifetime of many synoptic features
overlaps with the duration of seasonal variations.
In conclusion, the presented results demonstrate the feasibility of the approach
of Lagrangian trajectories for the identification of semi-persistent transport patterns
in the surface layer. The next step consists in merging the detected patterns with the
probability analysis of vulnerable regions being hit by adverse impacts stemming
from different sea areas. The techniques for the calculation of Lagrangian trajecto-
ries and the developed technology can be generalized in a straightforward manner
to the full 3D analysis of the propagation of adverse impacts.
314 T. Soomere
Acknowledgements The presented results were obtained in the framework of the BalticWay
project, which was supported jointly by the funding from the Estonian Science Foundation and the
European Commission’s Seventh Framework Programme (FP7 2007–2013) under grant agreement
No. 217246 made with the joint Baltic Sea research and development programme BONUS. The
follow-up research was partially supported by the Estonian Science Foundation (grant No. 9125),
targeted financing by the Estonian Ministry of Education and Research (grant SF0140007s11), and
by the European Regional Development Fund via support to the Centre of Excellence for Non-
linear Studies CENS.
References
Abascal AJ, Castanedo S, Medina R, Liste M (2010) Analysis of the reliability of a statistical oil
spill response model. Mar Pollut Bull 60:2099–2110
Addison PS (2002) The illustrated wavelet transform handbook. Introductory theory and applica-
tions in science, engineering, medicine and finance. Taylor & Francis, London
Ambjörn C (2007) Seatrack Web, forecasts of oil spills, a new version. Environ Res Eng Manag
3(41):60–66
9:1–16
334
Anonymous (2002) An updated assessment of the risk for oil spills in the Baltic Sea area. Pre-
sented as a Status report on risk analyses for use in response to oil pollution in the Baltic Sea by
Dr S Ovsienko, Fifth meeting of the Sea-based Pollution Group HELCOM SEA, Turku, Fin-
land, 13–17 May 2002. Helsinki Commission, Helsinki, 77 pp. http://www.helcom.fi/stc/files/
shipping/RiskforOilSpillsReport2002.pdf
Oceanogr 39:2820–2838
ASCE (American Society of Civil Engineers), ASCE Committee on modeling oil spills, Water
Resources Engineering Division (1996) State-of-the-art review of modeling transport and fate
of oil spills. J Hydraul Eng 122:594–609
Barth JA, Pierce SD, Smith RL (2000) A separating coastal upwelling jet at Cape Blanco, Oregon
and its connection to the California Current System. Deep-Sea Res II 47:783–810
Batchelor GK (1973) An introduction to fluid dynamics. Cambridge University Press, Cambridge
Beletsky D, Schwab D, McCormick M (2006) Modeling the 1998–2003 summer circulation and
thermal structure in Lake Michigan. J Geophys Res—Oceans 111:C10010
Bisagni JJ, Beardsley RC, Ruhsam CM, Manning JP, Williams WJ (1996) Historical and recent
evidence of Scotian Shelf Water on southern Georges Bank. Deep-Sea Res II 43:1439–1471
Breivik Ø, Allen AA, Maisondieu C, Roth JC (2011) Wind-induced drift of objects at sea: the
leeway field method. Appl Ocean Res 33:100–109
Breivik Ø, Allen AA, Maisondieu C, Roth JC, Forest B (2012) The leeway of shipping containers
at different immersion levels. Ocean Dyn 62:741–752
Broman B, Hammarklint T, Rannat K, Soomere T, Valdmann A (2006) Trends and extremes of
wave fields in the North-eastern part of the Baltic Proper. Oceanologia 48(S):165–184
Broström G, Carrasco A, Hole LR, Dick S, Janssen F, Mattsson J, Berger S (2011) Usefulness of
high resolution coastal models for operational oil spill forecast: the Full City accident. Ocean
Sci 7:805–820
Chrastansky A, Callies U (2009) Model-based long-term reconstruction of weather-driven varia-
tions in chronic oil pollution along the German North Sea coast. Mar Pollut Bull 58:967–975
Chrastansky A, Callies U, Fleet DM (2009) Estimation of the impact of prevailing weather condi-
tions on the occurrence of oil-contaminated dead birds on the German North Sea coast. Environ
Pollut 157:194–198
numerical aspects. Elsevier/Academic Press, Amsterdam/San Diego, 828 pp
Delpeche-Ellmann NC, Soomere T (2013) Investigating the Marine Protected Areas most at risk
of current-driven pollution in the Gulf of Finland, the Baltic Sea, using a Lagrangian transport
model. Mar Pollut Bull 67:121–129
13,514
Res 11:23–36
Fingas MF (2011) Buoys and devices for oil spill tracking. In: Proceedings of the 34th AMOP
technical seminar on environmental contamination and response, environment, Canada, Ottawa,
ON, pp 213–228
Gerdes R, Köberle C, Willebrand J (1991) The influence of numerical advection schemes on the
results of ocean general circulation models. Clim Dyn 5:211–226
Gräwe U, Wolff J-O (2010) Suspended particulate matter dynamics in a particle framework. Envi-
ron Fluid Mech 10:21–39
Gräwe U, Deleersnijder E, Shah SHAM, Heemink AW (2012) Why the Euler scheme in particle
tracking is not enough: the shallow-sea pycnocline test case. Ocean Dyn 62:501–514
Havens H, Luther ME, Meyers SD, Heil CA (2010) Lagrangian particle tracking of a toxic di-
noflagellate bloom within the Tampa Bay estuary. Mar Pollut Bull 60:2233–2241
HELCOM (2009) Ensuring safe shipping in the Baltic. In: Stankiewicz M, Vlasov N (eds) Helsinki
commission, Helsinki, 18 pp
Hibler WD III (1979) A dynamic thermodynamic sea ice model. J Phys Oceanogr 9:817–846
Sweden, 29 pp
Hunke EC, Dukowicz JK (1997) An elastic-viscous-plastic model for sea ice dynamics. J Phys
Oceanogr 27:1849–1867
Jevrejeva S, Leppäranta M (2002) Ice conditions along the Estonian coast in a statistical view.
Nord Hydrol 33:241–262
Kachel MJ (2008) Particularly sensitive sea areas. Hamburg studies on Maritime Affairs, vol 13.
Springer, Berlin, 376 pp
Killworth P, Stainforth D, Webb D, Paterson S (1991) The development of a free-surface Bryan–
Cox–Semtner ocean model. J Phys Oceanogr 21:1333–1348
316 T. Soomere
Kjellsson J, Döös K (2012) Surface drifters and model trajectories in the Baltic Sea. Boreal Environ
Res 17:447–459
Korajkic A, Badgley BD, Brownell MJ, Harwood VJ (2009) Application of microbial source track-
ing methods in a Gulf of Mexico field setting. J Appl Microbiol 107:1518–1527
Korotenko KA, Mamedov RM, Kontar AE, Korotenko LA (2004) Particle tracking method in the
approach for prediction of oil slick transport in the sea: modelling oil pollution resulting from
river input. J Mar Syst 48:159–170
Korotenko KA, Bowman MJ, Dietrich DE (2010) High-resolution numerical model for predicting
the transport and dispersal of oil spilled in the Black Sea. Terr Atmos Ocean Sci 21:123–136
symposium, Riga, Latvia, August 25–27, 2010. IEEE Press, New York
Kurennoy D, Parnell KE, Soomere T (2011) Fast-ferry generated waves in South-West Tallinn Bay.
J Coast Res 64(Special Issue):165–169
Lamb H (1994) Hydrodynamics, 6th edn. Cambridge University Press, Cambridge
Leppäranta M (2013) Land-ice interaction in the Baltic Sea. Est J Earth Sci 62:2–14
378 pp
Lin CH, Wu YL, Chang KH, Lai CH (2004) A method for locating influential pollution sources
and estimating their contributions. Environ Model Assess 9:129–136
Mariani P, MacKenzie BR, Iudicone D, Bozec A (2010) Modelling retention and dispersion mech-
anisms of bluefin tuna eggs and larvae in the northwest Mediterranean Sea. Prog Oceanogr
86:45–58
Matthäus W, Lass HU (1995) The recent salt inflow into the Baltic Sea. J Phys Oceanogr 25:280–
286
Meier HEM (2001) On the parameterization of mixing in three-dimensional Baltic Sea models. J
Geophys Res—Oceans 106:30,997–31,016
Melsom A, Counillon F, LaCasce JH, Bertino L (2012) Forecasting search areas using ensemble
ocean circulation modelling. Ocean Dyn 62:1245–1257
Mietus M (co-ordinator) (1998) The climate of the Baltic Sea Basin, Marine meteorology and
related oceanographic activities. Report No 41, World Meteorological Organisation, Geneva,
64 pp
Monzon-Argullo C, Lopez-Jurado LF, Rico C, Marco A, Lopez P, Hays GC, Lee PLM (2010)
Evidence from genetic and Lagrangian drifter data for transatlantic transport of small juvenile
green turtles. J Biogeogr 37:1752–1766
Morey SL, Martin PJ, O’Brien JJ, Wallcraft AA, Zavala-Hidalgo J (2003) Export pathways for river
discharged fresh water in the northern Gulf of Mexico. J Geophys Res—Oceans 108:C3303
Ohlmann JC, Mitarai S (2010) Lagrangian assessment of simulated surface current dispersion in
the coastal ocean. Geophys Res Lett 37:L17602
Osiński R, Piechura J (2009) Latest findings about circulation of upper layer in the Baltic Proper.
In: BSSC 2009 abstract book, Tallinn, August 17–21 2009, p 103

Commentationes physico-mathematicae, vol 12. Societas Scientarium Fennica, Helsinki (in
German)
Parnell KE, Delpeche N, Didenkulova I, Dolphin T, Erm A, Kask A, Kelpšaitė L, Kurennoy D,
Quak E, Räämet A, Soomere T, Terentjeva A, Torsvik T, Zaitseva-Pärnaste I (2008) Far-field
vessel wakes in Tallinn Bay. Est J Eng 14:273–302
Rohrs J, Christensen KH, Hole LR, Broström G, Drivdal M, Sundby S (2012) Observation-based
evaluation of surface wave effects on currents and trajectory forecasts. Ocean Dyn 62:1519–
1533
Samuelsson P, Jones CG, Willén U, Ullerstig A, Gollvik S, Hansson U, Jansson C, Kjellström NE,
Wyser K (2011) The Rossby Centre Regional Climate model RCA3: model description and
performance. Tellus A 63:4–23
Savijärvi H, Niemela S, Tisler P (2005) Coastal winds and low-level jets: simulations for sea gulfs.
Q J R Meteorol Soc B 131:625–637
Sooäär J, Jaagus J (2007) Long-term variability and changes in the sea ice regime in the Baltic Sea
near the Estonian coast. Proc Est Acad Sci, Eng 13:189–200
Soomere T, Keevallik S (2001) Anisotropy of moderate and strong winds in the Baltic Proper. Proc
Est Acad Sci, Eng 7:35–49
Soomere T, Delpeche N, Viikmäe B, Quak E, Meier HEM, Döös K (2011a) Patterns of
16(Suppl A):49–63
Soomere T, Viidebaum M, Kalda J (2011b) On dispersion properties of surface motions in the Gulf
Soomere T, Berezovski M, Quak E, Viikmäe B (2011c) Modeling environmentally friendly fair-
ways using Lagrangian trajectories: a case study for the Gulf of Finland, the Baltic Sea. Ocean
Dyn 61:1669–1680
Stevens DP (1991) The open boundary condition in the United Kingdom fine-resolution Antarctic
Model. J Phys Oceanogr 21:1494–1499
Viikmäe B, Soomere T, Viidebaum M, Berezovski A (2010) Temporal scales for transport patterns
in the Gulf of Finland. Est J Eng 16:211–227
318 T. Soomere
Witham C, Manning A (2007) Impacts of Russian biomass burning on UK air quality. Atmos
Environ 41:8075–8090
Meerbusens und der nördlichen Ostsee nach den Untersuchungen bis Ende 1910, Finnländische
hydrographisch-biologische Untersuchungen No 7, 82 pp (in German)
Yoon J-H, Kawano S, Igawa S (2010) Modeling of marine litter drift and beaching in the Japan
Sea. Mar Pollut Bull 60:448–463
Chapter 10
Applications of the Inverse Problem of Pollution
Propagation
Tarmo Soomere
Abstract The quantification of the potential of different offshore domains to serve

as a remote source of danger to the vulnerable areas through pollution propagation is
developed into a technique for the environmental management of open sea regions.
The technique focuses on the identification of areas (of reduced risk) that provide the
lowest level of environmental concerns for offshore activities. An approximate so-
lution to the relevant inverse problem is constructed by means of statistical analysis
of a large number of Lagrangian trajectories of pollution particles. The method con-
tains an eddy-resolving circulation model, a scheme for the tracking of Lagrangian
trajectories, a technique for the calculation of quantities characterizing the poten-
tial of different sea areas to supply adverse impacts, and routines to construct the
optimum fairway. The distributions of the probability of current-driven transport of
pollution to vulnerable domains and its propagation time are used for the optimiza-
tion of the location of potentially dangerous activities. This technique is applied for
the optimization of marine fairways to minimize the risk to high-value areas in two
regions of the Baltic Sea. The environmental advantage is expressed in terms of the
probability of pollution transport to the nearshore and the associated time (particle
age). In the Gulf of Finland the use of the optimum fairway would decrease the
probability of coastal pollution by 40 % or increase the average time it takes for the
pollution to reach the coast from 5.3 to about 9 days.
10.1 Introduction
Modelling efforts addressing the transport of oil spills (or any other adverse im-
pacts such as chemical pollution, or items on the sea surface such as lost containers,
rescue boats, ships without propulsion, etc.) mostly focus on the direct problem of
T. Soomere (B)
e-mail: tarmo.soomere@cs.ioc.ee
T. Soomere

DOI 10.1007/978-3-319-00440-2_10,
320 T. Soomere
pollution propagation in the marine environment. The forecast is usually provided

for the given location of the source, the amount and the properties of the spilled
oil, and modelled metocean conditions (e.g., French et al. 1997; Reed et al. 1999;
French-McCay 2004; Ambjörn 2008). The results definitely assist in taking coun-
termeasures after an oil spill has occurred. The systematic use of the Lagrangian
approach makes it possible to estimate the probability of a region to be polluted, the
time it takes for the oil to reach a specific site, and the areas that could be a threat for
a given shoreline (Abascal et al. 2010). Sometimes backwards tracking of the drift
allows the origin of the spill to be determined and to catch the guilty party (Ambjörn
2008).
10.1.1 The Hidden Potential of Currents
The technique described in this chapter goes one step further. The goal is to estab-
lish for which release sites an oil spill will cause the least damage under the same
hydrometeorological conditions and to develop examples of engineering solutions
for optimizing some important activities accordingly.
Since the drift of oil is also affected by direct wind drag and wave-driven impact,
its adequate description generally presumes that these factors are considered. The
transport induced by wind drag and wave motion is mostly ‘downwind’ or ‘down-
wave’, respectively, and the odds for an area to be affected are roughly inversely
proportional to its distance from the pollution release site in those directions. The
regions associated with the lowest risk are thus as far ‘upwind/upwave’ from the
vulnerable location as possible. As the impact of wind and waves can be relatively
easily added to the model, the presentation here is limited to the quantification of
the current-driven transport. An example of the combined analysis of current-, wind-
and wave-driven transport is presented in Chap. 11.
The current-driven transport has the largest unused potential for the reduction
of coastal pollution. A pattern of currents is an integral reaction of water masses
to a variety of forcing factors. It is usually highly complicated even if the wind is
stationary. The currents are often directed against wave-induced transport or wind
drag (e.g., Andrejev et al. 2004a; Gästgifvars et al. 2006). The forecast of current-
induced transport of drifters is only reliable within about 0.5 days (Vandenbulcke
et al. 2009). Even small errors in the initial location of drifters can drastically change
the calculated trajectories (Griffa et al. 2004).
A necessary prerequisite for undertaking a systematic search for an optimum
release site is the presence of nontrivial, anisotropic internal dynamics of surface
currents: otherwise different release sites can simply be graded according to their
distance from the vulnerable spots. The anisotropy of transport patterns intrinsi-
cally characterizes jet-like currents but has also been shown to hold for sea areas
with seemingly highly random dynamics such as the Baltic Sea (Meier 2007; Lu
et al. 2012). Although trajectories of water particles may be extremely compli-
cated (see Fig. 9.6 in Chap. 9), they are never completely random. For example,
10 Applications of the Inverse Problem of Pollution Propagation 321
the net current-driven transport frequently has semi-persistent patterns on the sur-
face (Soomere et al. 2011d; Lu et al. 2012) or in the subsurface layer (Andrejev
et al. 2004a, 2004b). These usually concealed patterns make the probability of trans-
port of various substances between different sea areas highly variable. The resulting
current-driven transport may be at times directed across relatively narrow regions
(Soomere et al. 2011d) and/or may lead to unexpectedly high probability of trans-
port of adverse impacts between certain regions (Lekien et al. 2003). The challenge
addressed here is how to quantify the impact of this transport in some commonly
usable and clearly understandable categories.
10.1.2 The Value of Different Sea Areas
This challenge is only meaningful if some value is assigned to certain regions. The
difference in the value of various offshore and nearshore domains was recognized
many centuries ago (Bowden-Kerby 2001). It is legally formulated in the recent past
through establishing marine protected areas and developing the concept of partic-
ularly sensitive sea areas (Kachel 2008). Giving some regions a certain value nat-
urally generates a spatio-temporal distribution of costs of consequences. Namely,
similar accidents but occurring at different locations usually impact the valuable
areas differently.
This distribution obviously depends on the demarcation of the valuable areas and
on the assigned ‘price.’ For some cases (e.g., the collision with a whale, Stokstad
2009) the cost is evident immediately at the site and the use of its spatial distribution
(e.g., the frequency of observations of the whales) for environmental management is
straightforward. Still on many occasions a substantial part of the costs becomes ev-
ident later through some hidden or hardly predictable mechanism of remote impact.
Some of these situations (e.g., the propagation of an ash cloud after a volcano erup-
tion) are out of the control of society. Many others (incl. ship accidents), however,
allow at least limited human control.
A generic example of a dangerous event with substantial remote impact, used in
this book to illustrate the developed technique, is an oil spill. It is obviously unde-
sirable everywhere on the sea but the scale of the associated devastation drastically
increases when it drifts to some valuable area. It is not possible to control the wind,
waves or currents that govern the fate of an oil spill but it is viable to restrict, at least
to some extent, the location of the potential sources of oil spills.
If vulnerable spots are selected based on environmental criteria a smart use of
the factors that drive the oil propagation may provide a natural way to mitigate the
environmental damage. This chapter presents a technique for applying knowledge
of natural processes for this purpose and provides several examples about how to
use it for decision-making. An important step is the identification of areas of re-
duced risk, so that for accidents occurring there, the costs through current-driven
remote impact are lower than for adjacent areas. In many cases directing activities
to these areas involves marginal additional costs. A corresponding policy may serve
322 T. Soomere
as a qualitatively new way of protecting vulnerable areas (Soomere and Quak 2007;
Soomere et al. 2010).
There exist an enormous variety of ecological values and indicators of environ-
mental well-being (Rice 2003). Using oil pollution as an example of a dangerous
event, it is natural to treat the pollution of the nearshore (called a coastal hit below),
wherever it happens, as a highly undesired and costly consequence. This setting
underlines the importance of coastal domains in the marine ecosystem according
to the common opinion that shallow-water and nearshore areas frequently have the
largest ecological and environmental value (Gray 1997; Kokkonen et al. 2010). Like
any other assignment of values, it obviously generates an associated distribution of
costs of otherwise similar accidents occurring at different sites.
10.1.3 The Inverse Problem
The practical use of this idea generally requires solving an inverse problem of the
identification of the most favourable location for the oil release. Its straightforward
solution presumes inverse tracking of pollution propagation, similar to the problem
of establishing the origin of illegal oil spills. The methods for effectively solving in-
verse problems are considerably less developed than those used for direct problems.
Although many trajectory-tracking (such as TRACMASS described in Chap. 7) and
oil drift models (e.g., Ambjörn 2007) are formally invertible, such problems are fre-
quently mathematically ill-posed. A straightforward solution to these is not possible
and no universal method exists for their analysis.
The determination of the offshore domains that systematically offer the least dan-
ger to the environment in terms of the transport of various adverse impacts to vul-
nerable regions owing to different hydrometeorological factors is an extension of
the problem of the analysis of current-induced Lagrangian transport addressed in
the previous chapter. It is natural to use the same technique—statistical analysis of
a large pool of examples of current-driven pollution propagation—for solving it.
The variety of different environmental values naturally generates a similar vari-
ety of meaningful ways to characterize the ‘damage potential’ of the offshore areas.
They may be quantified, for example, according to the probability of pollution re-
leased in these areas to reach vulnerable regions, or in terms of time it takes for this
impact to reach the coast after an accident has happened. For any measure in use,
the goal is to identify the areas that offer the least ‘cost’ of accidents. As discussed
in Chap. 1, doing so is one of the two possible ways to reduce environmental risks
associated with this particular type of damage. For example, in terms of probabili-
ties those domains are preferable, from which the propagation of adverse impacts to
high-value areas is most unlikely. The resulting quantification makes it possible to
find optimum locations of oil platforms, or where to tow a leaking vessel, among a
set of technically feasible solutions.
A highly interesting application is the specification of optimum fairways for
semi-enclosed seas. Such sailing lines offer a reduced level of remote impact of ship
traffic on the environment in a fairly general framework that accounts not only for
the local costs of accidents but also reveals the costs that become evident later, far
from the accident site through the propagation of various adverse impacts released
either occasionally or as a consequence of an accident. This concept is a general-
ization of the idea of applying emerging knowledge about semi-persistent current
patterns to minimize the consequences of potential accidents (e.g., by routing ship
traffic through specific areas, Soomere and Quak 2007). A decrease in the aggregate
environmental risk (see Eq. (1.1) in Chap. 1) is achieved preventively, before the ac-
cident happens. Similar applications can also be developed to specify how far from
the open ocean coast an economically feasible and environmentally friendly sailing
line should be located (Soomere et al. 2011c).
10.2 The Use of Lagrangian Trajectories of Current-Driven

Transport in the Surface Layer
The problem of identifying the least dangerous offshore areas can be simplified by
considering an oil spill as a cluster of persistent oil particles of neutral buoyancy
that are passively carried with surface currents. Then an approximate quantification
of the level of danger posed by oil spills can be obtained by the statistical analysis
of a large number of Lagrangian trajectories of single pollution particles released
to the sea at different time instants and locations (Abascal et al. 2010; Soomere
et al. 2010). An approximation of the areas of reduced risk is thus sought through
the extraction of information from a large pool of particular solutions of the direct
problem of propagation of oil particles.
The previous chapter demonstrated the potential of this approach to unveil the
presence of very rich, usually concealed internal structures of semi-persistent trans-
port patterns in certain sea areas (Soomere et al. 2011d). This chapter focuses on
a subset of such patterns that may bring oil pollution close to the coast. The major
technical problems are the same: (i) how to extract useful information from the vast
amount of numerically simulated data, and (ii) how to realize an implementation for
the shipping or offshore industry based on the extracted information.
For simplicity, the potential of the technique is demonstrated here for the particu-
lar case when the trajectories of particles representing adverse impacts are locked in
the uppermost layer and thus reflect the behaviour of the lighter fractions of the oil
pollution. Although we only consider a few examples of simplified environmental
criteriaenvironmental criterion, the approach can be easily modified to account for
virtually any realistic set of measures or to be used in combination with practically
any set of values of different areas uniquely defined in time and space. The resulting
engineering solutions (e.g., an optimum fairway) of course depend on the particular
choice of the measures and values.
324 T. Soomere
10.2.1 Quantification of Environmental Risks
A commonly used measure is the probability of the valuable areas being hit by pol-
lution. It is straightforward to associate with each offshore point the probability of
oil propagation from this site to a certain vulnerable area. Its sensible use implic-
itly requires the setting of a (propagation) time scale over which this probability is
counted (Andrejev et al. 2010; Viikmäe et al. 2010). Doing so naturally leads to a
complementary quantity that can equally well characterize the potential of pollu-
tion released at a particular sea point to pose danger to the nearshore: the average
time it takes for the pollution to reach the valuable area. This measure is similar to
commonly used water age (Deleersnijder et al. 2001) and is called particle age in
the following discussion. It naturally characterizes the cost of consequences from
another viewpoint: the longer time pollution remains in the open sea, the larger
fraction of it will be weathered or may be removed before it hits a vulnerable spot.
These measures are of course not equivalent. The values of particle age intrin-
sically contain information about whether the particle had reached the vulnerable
area. A hit to a valuable area obviously has occurred if the age of a particle is less
than the propagation time. Therefore it is straightforward to evaluate the probability
of hits to the vulnerable areas once the values of the age of all particles are known.
An inversion of this operation is not possible because the values of probability only
contain information about whether or not a hit has occurred.
The quantification is performed in this chapter in the context of the propagation
of oil pollution from different offshore areas to the nearshore. It is obviously applica-
ble for any adverse impact that is passively carried by surface currents. Its possibil-
ities are demonstrated for relatively simple particular cases of constant-value coast.
These examples are rich enough in content to highlight not only the complexity of
the problem but also the variety of solutions depending on a particular set-up and in-
trinsic uncertainties of the technology. Accounting for the more complicated internal
structure of the high-value areas is also straightforward. For example, it is possible
to identify certain locations from where the oil is less likely to be transported to se-
lected vulnerable spots of any kind (e.g., marine protected areas, Delpeche-Ellmann
and Soomere 2013), or from where the transport will take a longer time (Viikmäe
and Soomere 2013).
10.2.2 Preventive Optimization of Dangerous Activities
Once a map of the probabilities of hitting a specific area (or particle age) has been
constructed, the optimum locations for potentially dangerous activities (areas of re-
duced risk) are the minima for the probabilities or the maxima for the particle age.
The obvious practical outcome is the possibility to plan (or redirect) dangerous ac-
tivities into such areas. These maps are, however, of much larger practical impor-
tance. They may serve as the basis for various engineering solutions, decision sup-
port systems and preventive methods for the environmental management of shipping
and offshore activities by means of optimization of the location of potential release

of adverse impacts.
An important application is their use for optimizing fairways in domains (such as
the North Sea or Baltic Sea and especially the narrow and shallow Gulf of Finland)
that host extremely heavy ship traffic in the immediate neighbourhood of valuable
areas or sectors of the coast. In this context, the formulated problem is equivalent
to the task of minimizing the environmental damage for a moving source of danger.
It can be naturally associated with the classical problem of smart ship routing once
the distribution of the underlying quantities is known. For example, an approximate
solution for the environmentally safest sailing line is to route the ships along the
minima of these probabilities (Andrejev et al. 2011).
Alternatively, one can benefit from a choice of the sailing line for a (chemical)
tanker that provides a systematic increase in the time it takes before the adverse im-
pact reaches a vulnerable area (Engqvist and Andrejev 2003; Engqvist et al. 2006).
The use of the maxima for the particle age (Andrejev et al. 2011) for this purpose
is equivalent to buying extra time to combat the leak while the spill travels to the
coast.
For elongated sea areas (such as the Gulf of Finland), it might also rational to
search for the equiprobability line, such that for each of its points it is equally prob-
able that the spill reaches one coastline or the opposite one during a certain time
interval (Soomere et al. 2010). A further challenge (addressed only partially below)
is to minimize the cost of damage so that the additional economic costs would re-
main manageable.
The goal of this technique is not to produce yet another operational model to
assist rescue and oil combating teams after an accident has happened. Instead the
goal is to identify beforehand the regions where it is statistically safer to travel. The
target is thus the preventive reduction of environmental risks: an optimization of
the location of potentially dangerous activities (e.g., ship traffic) so that the conse-
quences of the resulting unfortunate event (here in terms of the impact of substances
potentially released into the sea and transported by currents) upon high-cost areas
would be minimal once it indeed occurred (cf. Eide et al. 2007).
A precondition for the use of such a way of thinking is the heterogeneity of the
relevant physical or ecological fields (such as the resulting maps of the probabilities
of coastal hit or the particle age). A recent solution of this kind is the relocation of
the fairway entering Boston Harbour (Massachusetts, USA) to minimize the prob-
ability of collisions of ship traffic with globally endangered right whales (Stokstad
2009). This action was based on the existence of an area, which was only infre-
quently visited by whales and thus offered a clear decrease in the probability for a
collision.
10.3 Components of the Technique and Properties of Test Areas

The technology contains four basic steps (Soomere et al. 2011a). The basis is a high-
resolution three-dimensional (3D) ocean circulation model. As a prerequisite, the
326 T. Soomere
model is supposed to adequately resolve the current-driven transport in the area in

question and to provide (numerically simulated Eulerian) velocities at certain grid
points and with a certain temporal resolution over a long time period. Secondly,
the generated velocity data are used to compute Lagrangian trajectories of a large
number of water particles. These particles are interpreted as passive tracers repre-
senting potentially adverse substances of neutral buoyancy (e.g., single oil particles)
released at different locations and at different time instants.
Thirdly, the trajectories are analysed with respect to a given cost function and/or
problem setup using standard statistical methods. This analysis leads to a spatial
quantification of the offshore areas. Note that the resulting distributions character-
ize the points of release of adverse impacts rather than the vulnerable areas. Finally,
the spatial distributions of these quantities (optionally together with additional con-
straints such as the location of the ports) are used for decision-making, for example,
for the identification of the optimum location of the fairways, by choosing them as
close as possible to the local or global minima of the distributions.
Each of the steps requires a number of parameters, implicit time scales and op-
tions that may potentially affect the resulting 2D distributions (maps) and the further
decision-making process. The existing implementations have applied the steps con-
secutively. This made it possible to improve the particular methods used within steps
separately (Andrejev et al. 2010; Viikmäe et al. 2010). Some relevant issues are dis-
cussed in Chap. 9 and in Viikmäe et al. (2010). The dependence of the results on
the choice of the options was evident in the early analysis (Soomere et al. 2010)
where the equiprobability lines were calculated using two slightly different sets of
trajectories. The optimum fairways also depend substantially on the resolution of
the ocean model (Andrejev et al. 2010). The strong seasonal variation of net and
bulk transport patterns (Chap. 9; Soomere et al. 2011d) suggests that seasonally op-
timum fairways may be radically different for windy and calm seasons in the Gulf
of Finland as also discussed in Chap. 11. Lu et al. (2012) showed that extensive
variations in the optimum fairway correspond to the inflow and outflow situations
in the south-western (SW) Baltic Sea.
10.3.1 The Baltic Sea Under Pressure
There are several reasons why this technique has been first implemented for the
Baltic Sea (Fig. 10.1). This water body is under strong pressure from shipping and
other offshore activities. The concentration of ship traffic (including tankers of vari-
ous kinds) is exceptionally high (HELCOM 2009). More than 70 large ports handle
more than 1 million tonnes of cargo per year. The number of ship voyages (excluding
ferry traffic) is estimated at 150,000 per year (Gollasch and Leppäkoski 2007), and
it is assumed that it will increase considerably in the future. Since 1980 the Baltic
Sea has experienced on average one major shipping accident per year resulting in
an oil spill larger than 100 tonnes (WWF 2010).
Fig. 10.1 Baltic Sea and the

test regions in the Gulf of
Finland and the SW Baltic
Sea and the Danish Straits.
Graphics by M. Viška
The sea has been used for dumping massive quantities of waste, explosives and
extremely dangerous chemical warfare agents (Glasby 1997). Currently it hosts sev-
eral large-scale controversial constructions (Lidskog and Elander 2012) and a few
oil platforms (WWF 2010). The number of offshore installations may soon increase
considerably through ambitious plans for the erection of wind parks that might fill
large sections of the offshore with sizeable structures and cables. The resulting high
probability of the release of various undesirable substances into the sea as a result
of an accident, a technical problem, human error or deliberate action is alarming
because the consequences of a major pollution event could be devastating for this
particularly sensitive sea area (Kachel 2008).
The Baltic Sea is one of the most studied domains of the ocean. Many data
sets extend back more than two centuries (Leppäranta and Myrberg 2009). Reli-
able and thoroughly validated atmospheric and ocean models exist for this region
(Meier et al. 2003; Myrberg et al. 2010b). Its mesoscale dynamics has a complexity
comparable with that of the open ocean. The sea is known to have numerous semi-
persistent patterns of both surface and subsurface currents (Lehmann et al. 2002;
Andrejev et al. 2004a, 2004b; Meier 2007; Osiński and Piechura 2009; Soomere
et al. 2010).
The technology described here has been applied at a moderate resolution for the
entire Baltic Sea and at a higher resolution for two test areas, the SW Baltic Sea
and the Gulf of Finland. An other version based on Eulerian tracking of pollution
propagation (Höglund and Meier 2012) described in Chap. 4 has also been applied
328 T. Soomere
to the entire Baltic Sea. The test areas not only play a key role in the functioning of
the sea (as described in Chaps. 2 and 6) but also exhibit different challenges for the
technique.
10.3.2 The Western and Eastern Gateways to the Baltic Sea
The Danish Straits, the Belt Sea and the Arkona Basin form the gateway through
which all the ships travelling between the North Sea and the Baltic Sea have to pass
(see Figs. 9.1 and 9.2 in Chap. 9). This mostly shallow domain with moderately
sloping bottom serves as a transit area for the outflow of excess water through the
Danish Straits. More importantly, it is the gate through which the deeper parts of
the Baltic Sea are sporadically supplied under specific atmospheric conditions with
saltier, oxygen-rich North Sea waters (Matthäus and Lass 1995).
The complicated geometry of the coasts, the presence of narrow straits and the
small internal (baroclinic) Rossby radius R1 (about 5 km in the Arkona Basin and
the Pomeranian Bay, Osiński et al. 2010, and as small as 1–4 km in the shallow
Belt Sea, Fennel et al. 1991) make numerical modelling of water masses extremely
challenging. In order to properly resolve mesoscale dynamics, it is necessary to use
a horizontal grid size not larger than about 1/2 of the baroclinic Rossby radius (Dri-
jfhout 1989; Lindow 1997). Most medium-resolution Baltic Sea circulation models
have a grid step around 2 nautical miles (nm) and are barely eddy-permitting here.
Furthermore, fine vertical resolution is needed here to resolve the straits’ topography
and sharp stratification (Lu et al. 2012).
The Gulf of Finland (see Chap. 6) is an elongated estuary (length over 400 km,
width 48–135 km) in the north-eastern (NE) Baltic Sea with a mean depth of 37 m
only (Alenius et al. 1998). As it is quite narrow in some places and the water is too
shallow for marine transportation in others, there are several narrow passages where
the concentration of traffic is exceptionally high. Particularly intense passenger traf-
fic crosses the narrow section of the gulf between Tallinn and Helsinki (Parnell et al.
2008; Kurennoy et al. 2011).
The gulf is widely open to the Northern Gotland Basin. As there is no sill at
its entrance, continuous water exchange occurs between the two water bodies and
the dynamics of the Baltic Proper1 strongly affects currents in the western part of
the gulf. This region evidently plays a key role in the functioning of the northern
Baltic Sea. Under specific wind conditions the typical estuarine circulation may be
reversed, the salt water wedge may be exported from the gulf and the vertical strat-
ification may almost completely vanish (Elken et al. 2003). This process may lead
to enhanced vertical mixing of water masses. The eastern end of the gulf receives
voluminous fresh water inflow from the River Neva. This gives rise to a strong
east–west gradient in salinity and sea-level and leads to a small baroclinic Rossby
1 We shall use the notion Baltic Proper to denote the Eastern, Northern and Western Gotland Basin
(Table 2.1 in Chap. 2), Bornholm Basin and Gdańsk Bay.

radius (usually about 2–4 km) (Alenius et al. 2003) and enhanced buoyancy-driven
currents.
These features favour intricate spatio-temporal patterns in salinity and density,
and support a complicated interplay of the basin-scale cyclonic circulation, the ex-
change of water masses with the Baltic Proper, and the system of mesoscale (synop-
tic) eddies (Andrejev et al. 2004a, 2004b). In addition, these patterns become even
more complicated due to substantial seasonal variations in the wind field and the
incoming solar radiation, and to the possible presence of an ice cover (Myrberg and
Andrejev 2003; Andrejev et al. 2004a, 2004b; Myrberg et al. 2010b). Consequently,
the requirements for circulation modelling for the Gulf of Finland are as high as
for the SW Baltic Sea. The common circulation models with a grid step of 2–3 nm
(Myrberg et al. 2010b) apparently overlook a part of its dynamics, especially the
contribution by synoptic eddies to circulation and mixing.
10.4 Circulation and Trajectory Models
The quality of the circulation models is crucial for the success of the entire technol-
ogy. Three different ocean models have been used to simulate the 3D dynamics of
the Baltic Sea and in the test areas for the applications of the technique developed
in 2009–2012 in the framework of the BONUS BalticWay cooperation.
10.4.1 The Rossby Centre Ocean Model RCO
A set of velocity fields was calculated by the Swedish Meteorological and Hydrolog-
ical Institute using the Rossby Centre Ocean circulation model (RCO) for the entire
Baltic Sea (Chap. 4). The model details, set-up and extensive validation efforts are
described in a number of publications (Meier 2001, 2007; Meier et al. 2003, among
others). The bathymetric information is based on the data set of Seifert et al. (2001).
The RCO model is a further development of the Ocean Circulation and Climate Ad-
vanced Model (OCCAM) primitive-equation free-surface circulation model (based
on the Bryan–Cox–Semtner model, Webb et al. 1997) in z-coordinates. It covers the
entire Baltic Sea with a horizontal resolution of 2 nm and vertical resolution from
3 m in the uppermost layer down to 12 m in the deeper areas (for a total of 41 lev-
els). This horizontal resolution is usually sufficient for eddy-resolving runs in the
Baltic Proper (Lehmann 1995) but is barely eddy-permitting in the Gulf of Finland
(cf. Albretsen and Røed 2010).
The model contains several parameterizations with a special importance for the
Baltic Sea, such as a two-equation turbulence closure scheme of the k–ε type to pa-
rameterize subgrid-scale mixing, open boundary conditions in the northern Kattegat
(in the sea area to the west of southern Sweden in Fig. 10.1), a sea-ice model and a
parameterization of the effect of breaking surface gravity waves. The time-splitting
330 T. Soomere
is used with the time step of 150 s for the baroclinic and 15 s for the barotropic prop-
agation. The output is stored once in six hours to keep the size of output information
at a reasonable level.
The RCO model was forced with meteorological data from a regionalization of
the ERA-40 reanalysis over Europe using a regional atmosphere model covering the
Baltic Sea with a horizontal resolution of 25 km during 1961–2007 (Samuelsson
et al. 2011). This data set contains wind at the 10 m level (adjusted using simulated
gustiness, Höglund et al. 2009), air temperature, relative and specific humidity at
the 2 m level, total precipitation and total cloudiness with a temporal resolution
of 3 hours, sea level atmospheric pressure, snow depth, actual albedo, short- and
long-wave radiation and evaporation.
10.4.2 The OAAS Model
The numerical model OAAS, constructed for use in basins with complicated bathy-
metry and hydrography (Andrejev and Sokolov 1989, 1990) and named after the
authors (Oleg Andrejev ja Alexander Sokolov), was used to simulate the currents
in the Gulf of Finland to the east of 23◦ 27 E. This free-surface, 3D baroclinic z-
coordinate circulation model employs standard simplifications such as the Boussi-
nesq and hydrostatic approximation, and the no-slip condition for the entire seabed.
It is based on the primitive equations of horizontal momentum balance, the conti-
nuity equation and equations for the transport of heat and salt in Cartesian coordi-
nates.
Some of the model features (such as the equation of state) have been tuned for
the Baltic Sea conditions (Millero and Kremling 1976). The use of the govern-
ing equations in the flux form allows automatic maintaining of a number of inte-
gral constraints (Blumberg and Mellor 1987). The finite-difference method uses the
Arakawa C-grid (Mesinger and Arakawa 1976) and the method of splitting the time
step (Liu and Leendertse 1978). An overview of the model equations and the meth-
ods for their solving is presented in Andrejev and Sokolov (1989, 1990), Sokolov
et al. (1997), Andrejev et al. (2004a, 2004b, 2010).
As the winters during the period covered in simulations (1987–1991) were rather
mild and the Gulf of Finland was mostly free of ice, a simple parameterization
was used for ice phenomena. For water temperatures below freezing point, the wind
stress was decreased by a factor of 10 in order to mimic the presence of ice. At 0 ◦ C,
the vertical heat flux was stopped as long as cooling conditions prevailed. The loss
of heat during ice melting was approximated by decreasing the upward heat flux in
the early spring by a factor of four until the water temperature reaches +1 ◦ C.
The model resolution, originally restricted to 1 nm in order to match the available
bathymetric information (Seifert et al. 2001), has been increased to 0.25 nm for the
use of the technique in question (Andrejev et al. 2010). The model was applied in
three resolutions (2 nm, 1 nm and 0.5 nm) with otherwise identical set-up. The ver-
tical resolution was 1 m in the entire water column (except for the uppermost layer
that had a thickness of 2 m). The model has shown excellent performance in differ-
ent applications, from basin-scale estimates of the upwelling features in the entire
Baltic Sea (Myrberg and Andrejev 2003; Myrberg et al. 2010a), mean circulation
and water age (Andrejev et al. 2004a, 2004b) and various hydrophysical features of
the Gulf of Finland (Myrberg et al. 2010b) down to the small-scale reproduction of
surface buoy drift (Gästgifvars et al. 2006) and test simulations using a resolution
of up to 0.25 nm (Viikmäe et al. 2010).
The OAAS model setup, boundary conditions and forcing data are discussed in
Andrejev et al. (2010, 2011). The model was forced with the same meteorological
data as the RCO model. River discharge was approximated in the OAAS model
using monthly mean values for 1970–1990 (Bergström and Carlsson 1994). As a
first approximation, the salinity of river water was set to zero and its temperature
equal to the ambient sea water temperature at the river mouth. This approximation
(equivalent to ignoring both salinity and heat flux from the rivers) is sensible in the
Baltic Sea conditions, where the river water salinity is almost zero and the difference
in river and sea water temperature in shallow river mouth areas is moderate.
The modelling in the Gulf of Finland started from the resting water masses and
with the sea level in equilibrium with atmospheric pressure. The initial fields (wa-
ter temperature and salinity) and the boundary information (the 3D structure of
the salinity and temperature and sea level information) at the entrance to the gulf
(Fig. 10.1) were extracted (and optionally interpolated) from the output of the RCO
model at 6-hour resolution. To smooth the potential impact of the difference in the
resolution between the models, the lateral diffusivity coefficient was increased in
the OAAS model towards the boundary following a sine function in a sponge layer
of a width of 16 nm. The modelled fields were plausible from the very beginning of
calculations. The spin-up of the surface layer dynamics took ca 1–2 weeks.
10.4.3 The DMI/BSHcmod Model
Another circulation model, DMI/BSHcmod, was applied in the studies of the dy-
namics in the southern Baltic Sea. This is also a 3D primitive-equation, hydrostatic,
free-surface ocean model. It was developed originally by the German Federal Mar-
itime and Hydrographic Agency (BSH) (Kleine 1994; Dick et al. 2001) and fur-
ther developed by the Danish Meteorological Institute (DMI). Its version developed
for the Baltic Sea conditions is now known as the HIROMB-BOOS community
model (Funkquist 2001) and inter alia serves as the underlying circulation model of
Seatrack Web, the official web-based HELCOM tool for simulation of the propaga-
tion and fate of oil pollution in the Baltic Sea (Ambjörn 2007, 2008).
In order to accurately resolve the water exchange between the Baltic Sea and
North Sea through the Danish Straits, three nesting levels were applied by Lu et al.
(2012) (see also Chap. 5). A 2D model with a horizontal resolution of 6 nm covering
a large part of the north-eastern Atlantic provided boundary conditions for a local
3D Baltic Sea–North Sea model, with a horizontal resolution of 3 nm. A finer 3D
332 T. Soomere
model with a horizontal resolution of 0.5 nm and vertical resolution of 1 m covering

the transition area from Skagen to Bornholm (Fig. 10.1) was two-way nested into
the above Baltic Sea–North Sea model. A detailed description of this model and its
forcing and boundary conditions is presented by Lu et al. (2012). Some interesting
results from this model are also described in Chap. 5.
10.4.4 Trajectory Simulations Using TRACMASS
The test elements used to evaluate the risk of coastal hit by pollution released at a
particular sea point and then transported by surface currents to the nearshore were
numerically simulated Lagrangian trajectories of water particles passively carried
by the currents. The trajectories were locked in the uppermost layer (with a depth
of 3 m in the RCO model, 2 m in the OAAS model and 1 m in the DMI/BSHcmod
model) and exerted only horizontal advection. They were thus not truly Lagrangian
and basically represented the motion of items persistently located in the surface
layer and completely following its motion. Such items and substances are generally
also affected by wind drag or wave impact. The impact of these drivers is addressed
in Chap. 11. Here we only consider the current-driven advection of such items.
‘Off-line’ simulation of the pathways of selected particles (in which the circula-
tion modelling was separated from the trajectory calculation, see Chap. 7) was used
to build the trajectories from the output of the RCO model for the Gulf of Finland
(Soomere et al. 2011c, 2011d) and the northern Baltic Proper (Viikmäe et al. 2011),
and from the data of the DMI/BSHcmod model for the SW Baltic Sea (Lu et al.
2012). The trajectories were calculated with the TRACMASS code (Blanke and
Raynard 1997; Döös 1995; de Vries and Döös 2001) using precomputed 3D veloc-
ity fields to reconstruct the motion of the particles. See Chap. 7 for the underlying
theory and several examples of the use of this model.
The results depend to some extent on the temporal resolution of both the circula-
tion data and the accuracy of the trajectory calculation. The resulting inaccuracies of
the latter may lead to a large divergence of a few simulated trajectories from the real
ones (Gräwe and Wolff 2010). Given the large number of trajectories, it is natural
to assume that the potential impact of the time step of the circulation data and the
trajectory reconstruction scheme on the resulting statistics is minor.
In the marine environment the particles are not only transported with the instan-
taneous simulated velocity in each grid cell but are also affected by local motions
such as diffusive processes, small-scale turbulence, local vertical motions of water
particles, etc. (so-called subgrid-scale turbulence). The latter is only partially ac-
counted for in the circulation models and, even theoretically, cannot be adequately
represented in these trajectory models that exclusively rely on precomputed velocity
fields. It is only possible to replicate certain statistical features of trajectories in such
models.
The version of the TRACMASS code used in the described calculations did not
account for the effects of subgrid turbulence. In such cases the initially close trajec-
tories have an overly tendency to stay close while in the real atmospheric and marine
conditions the aggregated impact of small-scale features of the motions generally

tends to spread closely packed particles (Richardson 1926; Ollitrault et al. 2005).
It is well known that the spreading of the resulting trajectories in studies based on
this version of TRACMASS has been usually much smaller than the spreading of
real drifters (Jönsson et al. 2004; Engqvist et al. 2006; Döös and Engqvist 2007;
Döös et al. 2008). An implicit consequence from the lack of spreading is that the
nearshore had to be defined as a 3 grid cells wide zone near the coast when the
TRACMASS code was used for the Gulf of Finland (Viikmäe et al. 2010).
10.4.5 Trajectory Simulations in the OAAS Model
The estimates for the spreading of real drifters in the Gulf of Finland (Soomere
et al. 2011e, see also Chap. 8) show that it takes, on average, about 4 days for
two initially closely located particles to drift into different grid cells of the 2 nm
RCO model. This suggests that ignoring sub-grid scale processes may modify a
part of the trajectories but their majority will remain at a distance of less than one
grid step from their actual position. It is natural to assume that further spreading
of the trajectories will mostly occur due to the gradients in the simulated velocity
fields. It is also reasonable to suppose that for the particular horizontal resolution and
length of trajectories, ignoring the subgrid spreading does not significantly affect the
resulting 2D fields and might even suppress the noise in these fields compared to
artificial reproduction of the spreading. Doing so is, however, a major simplification
that might not be entirely justified (Andrejev et al. 2011).
The trajectories were simulated ‘on-line’ (simultaneously with the integration of
the model, see Chap. 7) in runs of the OAAS model (Andrejev et al. 2010, 2011;
Soomere et al. 2011a, 2011b). The scheme accounted for the effect of subgrid-scale
motions. Note again that the exact motions of the particles cannot be restored and the
major benefit from such efforts is that the statistics of spreading (e.g., the absolute
dispersion or the net distance covered by a particle over some time) is represented
more adequately.
The impact of subgrid turbulence is usually parameterized by adding certain ar-
tificial perturbations to the velocity components of each particle (see Chaps. 4 and
7 for more discussion). The displacement of the particles may be calculated, for
example, using the following equations:
d x̃ d ỹ
= uc (x̃, ỹ) + u , = vc (x̃, ỹ) + v , (10.1)
dt dt
where (x̃, ỹ) is the instantaneous location of the particle, (uc , vc ) are the numeri-
cally simulated velocity components from the circulation model and (u , v ) are the
artificial local velocity perturbations (Andrejev et al. 2010).
The trajectory module of the OAAS model reflects a specific feature of currents
in the Gulf of Finland. Jet-like patterns occur seldom here and synoptic eddies cover
a large part of the gulf (Andrejev et al. 2004a; Zhurbas et al. 2008). Consequently,
334 T. Soomere
a substantial part of the motions with periods from 2 to 36 h are strongly circularly
polarized (Lilover et al. 2011). This feature was accounted for by perturbing the x-
component of velocity based on the magnitude of the y-component and vice versa
(Andrejev et al. 2010). They chose u = crv, v = cru, where the random variable r
was uniformly distributed within the interval [−0.5, 0.5] and c was a tunable coef-
ficient reflecting the intensity of spreading.
Setting c = 0.5 in Andrejev et al. (2010) resulted in the average spreading rate
of 10-day long trajectories in December 1990–November 1991 of about 176 m/day
(∼2 mm/s) for the 1 nm model and 10 particles per each grid cell (with a typical ini-
tial distance of ∼500 m between particles). This value matches the estimates from
drift of surface buoys in 2011 for the separation rate of particles initially located
at a distance of 50–150 m from each other (Soomere et al. 2011e). The resulting
statistics apparently somewhat underestimates the actual spreading rate. The intro-
duced spreading was, though, reasonable to ensure a high enough rate of hits to
the nearshore (that was defined as the pool of wet points directly adjacent to the
land points): about 2/3 of all released particles entered this zone during 10 days of
propagation (Andrejev et al. 2011).
Each method has its advantages and disadvantages (Chap. 7). The on-line method
allows the use of the integration time step of the circulation model for solving the
trajectory equations and leads to more exact formal reproduction of single trajec-
tories. This is, however, no real advantage as subgrid-scale effects deform the real
trajectories in a manner that can be only reproduced statistically. The set of on-
line simulations is severely limited as re-running of a high-resolution circulation
model is usually very time-consuming. The accuracy of the trajectories from the
off-line method suffers from the coarse temporal resolution of the saved velocity
data (usually once in a few hours). This shortcoming is partially balanced by a
much larger flexibility: off-line trajectory calculations are much faster, it is easy
to re-parameterize the impact of subgrid motions, the already calculated trajectories
can be used for differently set problems, etc.
10.5 Quantification of the Risk of Coastal Pollution

The applications described in this book have assumed that a hit of any part of long
sections of the nearshore is equally undesired. This is equivalent to introducing a
constant cost function: reaching any section of the nearshore by a particle is ‘bad’
(optionally with a sign) and staying in the open sea (or leaving the test area towards
the North Sea or the Baltic Proper) is mostly considered ‘neutral.’
10.5.1 Launching the Particles
The basic requirements concerning the number of trajectories in order to create re-
liable statistics of coastal hits have been discussed in Chap. 9 and in Viikmäe et al.
Fig. 10.2 Major fairways

crossing the Gulf of Finland.
High-resolution simulation
using the OAAS model was
performed to the east of
longitude 23◦ 27 E. The blue
line shows the grid points of
the RCO model used for this
gulf. Graphics by M. Viška
(2010), Lu et al. (2012). The substantial seasonal variation in the circulation pat-
terns in the test areas requires performing the simulations over at least a few years
(Andrejev et al. 2011). In existing applications the simulations have been performed
for a five-year test period (May 1987–December 1991 for the Gulf of Finland and
1990–1995 for the SW Baltic Sea). Its length matches the renewal time of Gulf of
Finland waters (Andrejev et al. 2004b) and allows for separate consideration of the
inflow and outflow patterns in the SW Baltic Sea.
The simulations of trajectories over the entire period in question (typically
5 years in the existing applications) were divided into sequential (optionally par-
tially overlapping) time windows (see Chap. 9). Their length tw was 10–20 days for
the Gulf of Finland and up to 60 days in the Baltic Proper (Viikmäe et al. 2011).
At the beginning of each time window, a few (1–10) water particles were selected,
usually in all wet points of the circulation model, and their motion was tracked over
the time window. In on-line runs the statistics for the time window were calculated
immediately and in the off-line runs the results were saved for further analysis.
A subsequent simulation for the same configuration of particles was launched
with a certain time lag as described in Chap. 9. The process was repeated over the
entire time period of interest. The typical number of time windows ranged from
170 (Andrejev et al. 2010, 2011) up to >10,000 in test simulations with a time lag
of 6 hours (Viikmäe et al. 2010). The number of grid cells varied from 2270 in
the 2 nm version of the OAAS model (or about 3000 in the RCO model) for the
Gulf of Finland up to 31,838 in the 0.5 nm OAAS model in the Gulf of Finland
(Fig. 10.2) (Andrejev et al. 2010). The number of selected particles was reduced in
Lu et al. (2012) by choosing only every fourth cell of the ocean model for seeding of
particles. The usual number of trajectories per grid cell was on the order of 1,000–
10,000.
10.5.2 Indicators of Coastal Hit and Drift Time
The quantification procedure of the offshore domains used standard statistical anal-
mk and a mk associated with each released particle. Here,
ysis of auxiliary variables pij ij
336 T. Soomere
(i, j ) indicate the grid cell in Cartesian coordinates, m is the sequential number of
particles in this cell and k is the number of time window. For example, setting the
variable pijmk = 0 initially and switching it to 1 when the particle hits the nearshore
for the first time, is convenient for estimates of the probability of coastal hits (An-
drejev et al. 2010, 2011). A zero value of pij mk at the end of a time window means
that the particle has only been offshore or has left the test area through the open
boundary.
The time during which the particle has drifted in the basin without touching the
coast can be measured by assigning to another variable aijmk the time elapsed from
the beginning of simulations until the particle hits a nearshore for the first time. If the
particle does not hit the coast, aijmk is assigned the length of the time window. The
results of this method applied for sea areas with an open boundary may substantially
depend on how exactly the particles that drift out of the computational area are
handled (Soomere et al. 2011c).
(k)
The cell-wise probabilities of coastal hits Pi,j and particle age A(k)
i,j were calcu-
lated for each time window k as the average of the relevant values pij mk , a mk over M
ij
particles released into a particular cell (i, j ):
1 mk 1 mk
M M
(k) (k)
Pi,j = pij , Ai,j = aij . (10.2)
M M
m=1 m=1
The cumulative average probability P̄i,j for the coastal hits and the cumulative av-
erage age Āi,j in a particular cell for the first n time windows were defined in the
classical manner:
1 (k) 1 (k)
n n
P̄i,j (n) = Pi,j , Āi,j (n) = Ai,j . (10.3)
n n
k=1 k=1
Setting the parameter n in (10.3) equal to the total number of time windows Nmax
gives the estimates for the probability of a coastal hit and the particle age at the
particular grid cell over the entire simulation period. The resulting values for sin-
gle cells pij = P̄i,j (Nmax ) and aij = Āi,j (Nmax ) constitute the pixels of the spatial
maps reflecting two possible quantifications of the offshore areas. Similarly, one can
define the average probability for coastal hits (or particle age) for any sea domain
of interest by further averaging over the resulting values pij or aij for cells in this
area, or, alternatively, for a selected season or other time interval.
The resulting average time for the coastal hit (particle age) is similar to com-
monly used water age (Deleersnijder et al. 2001) that has been extensively studied
for the Baltic Sea (Engqvist et al. 2006) and Gulf of Finland (Andrejev et al. 2004a,
2004b). The difference is that here the time is defined following the Lagrangian
framework for limited-duration trajectories ending at the nearshore while the ma-
jority of earlier studies use Eulerian velocities and other specifications of the age.
Fig. 10.3 Mean probability

P (k) for particular time
windows (blue) and
cumulative average
probability P̄ (n) (red) (upper
panel); mean particle age
A(k) for particular time
windows (blue) and
cumulative average particle
age Ā(n) (red) (lower panel)
in the Gulf of Finland based
on the 1 nm OAAS model for
May 1987–December 1991
(Andrejev et al. 2011)
10.5.3 Long-Term Course of Probability and Particle Age

(k)
The temporal behaviour of the average probability P (k) = Pi,j
and particle age
(k)
A(k) = Ai,j
over all particles for a particular time window k (Fig. 10.3) allows
a rough estimate of the length of the time interval to be covered in order to reach
climatologically valid results. The angled brackets denote the arithmetic mean over
all sea points in the calculation area. Their behaviour was analysed in Andrejev
et al. (2011) using the 1 nm OAAS model and above-discussed on-line trajectory
calculation scheme. The modelling period of 1 May 1987–31 December 1991 was
divided into 170 consecutive 10-day long time windows. At the beginning of each
window ten particles were released into each surface grid cell and locked in the
uppermost layer.
The fluctuations in the cumulative values P̄ (n) and Ā(n) rapidly decreased when
n increased. Their changes only partially followed the patterns in forcing factors.
During the first months of calculations (the calm period of May–June 1987) the
values of P̄ (n) were about 0.48 (Fig. 10.3). Over the subsequent windy season, until
February 1988, the kinetic energy density of surface wind increased by a factor of
three (Soomere et al. 2011a) but P̄ (n) only increased by about 30 % (Fig. 10.3).
The seasonal course of the mean particle age A(k) was, as expected, in antiphase
with P (k) (Fig. 10.3). The largest probabilities for coastal hits occurred roughly
simultaneously with the smallest particle age during autumn and winter storms.
Both the quantities P (k) and A(k) exhibited substantial short-term variations
with amplitudes comparable with or even larger than their seasonal variations
(Fig. 10.3). In particular, P (k) revealed especially pronounced seasonal variation,
338 T. Soomere
from about 0.4 (mostly in calm seasons) to values very close to 1. Their relative
fluctuations were, however, much smaller than similar variations in the forcing fac-
tors. For example, monthly averaged kinetic energy density of surface wind varied
by more than 10 times in the Gulf of Finland (Andrejev et al. 2011).
Both the cumulative quantities already closely matched their asymptotic values
P̄ (Nmax ) ≈ 0.67, Ā(Nmax ) ≈ 5.3 days after the first half-year of simulations and
revealed moderate variations after about two years of calculations. Their short-term
and seasonal fluctuations were very small (about 3 % from the average) from the
third year on and were almost negligible (below 1 %) after the fifth year. Therefore,
the cumulative measures in question calculated over >3 years reflect well long-term
internal properties of current-driven transport patterns even in a domain hosting
extensive seasonal variations in the forcing pattern such as the Gulf of Finland. This
time scale is close to the average ‘classical’ water age (equivalently, the time scale
of renewal of water masses) in the Gulf of Finland (Andrejev et al. 2004a) that also
revealed very small changes after three years of simulations.
The asymptotic values P̄ (Nmax ) and Ā(Nmax ) can be used to roughly evaluate
whether the parameters of the developed procedure are sensible. The overall proba-
bility of coastal hits had a reasonable level P̄ (Nmax ) ≈ 0.67, not very close to 1 and
also not small. It signifies that about 2/3 of the released particles have contributed
to the estimates of the environmental risk. The limit value of the particle age was
somewhat smaller than might be expected based on Viikmäe et al. (2010) where typ-
ically only 30–40 % of the particles released in the central part of the gulf reached
the nearshore within 10 days. Finally, we note that both P (Nmax ) and A(Nmax ) may
depend on the length of the time window tw in use. This dependence is usually much
stronger for A(Nmax ), as tw is assumed to be the age of all particles that have not
reached the coast (Soomere et al. 2011c, 2011d).
10.5.4 Spatial Distributions of Probability and Particle Age
The most important intermediate products of the technology are spatial distributions
of the measures of the environmental risk pij (x, y) and aij (x, y) associated with the
release of pollution. These maps can serve as starting points of various engineering
applications and decision support systems.
The average probability of a coastal hit for single cells pij over the entire simu-
lation period shows a rich spatial structure for the test areas (Fig. 10.4). The proba-
bilities to the west of Paldiski (Fig. 10.2) in the Gulf of Finland are strongly affected
by the presence of the open boundary. The areas with smallest pij < 0.4 are located
far offshore. The regions of small pij have relatively small gradients. This is par-
tially caused by the choice of the length of the time window (10 days) an increase
in which eventually would lead to larger contrasts in pij (Soomere et al. 2011c).
A wide area where pij ∼ 0.5–0.6 extends from Lahemaa almost to the island of
Gogland. Its bend to the south into Narva Bay obviously reflects the presence of
islands (that are ignored in the 2 nm simulations of Soomere et al. 2011c). Smaller
Fig. 10.4 Probabilities of hitting the coast (above) and particle age (days, below) for May
1987–December 1991 calculated using the 1 nm OAAS model and 10-day long trajectories
(Soomere et al. 2011c)
areas of this probability level are in the easternmost part of the gulf and between
Tallinn and Helsinki.
The particles released at a distance <3–4 km from the coast almost surely
(pij > 0.9) hit the coast within 10 days. The probability for coastal hit is high
(pij > 0.8) in the entire Neva Bay. A relatively wide belt of high pij character-
izes the coastline sections with complex geometry such as Lahemaa or the vicinity
of Tallinn Bay. Its dimensions might be interpreted as the width of the coastal zone.
This zone is frequently specified based on a high residence time of water near the
coast (see, for example, Lessin et al. 2009). A large probability pij in the nearshore
areas obviously signifies low persistency of water masses, while a small pij may
simply mean that the net transport is directed to the offshore (Soomere et al. 2011a).
A wide coastal zone exists in this sense along the Finnish coast, NE coast of the gulf
340 T. Soomere
and from Lahemaa to Paldiski along the southern coast of the gulf. On the contrary,
an intense exchange of water with the offshore is typical for Narva Bay and NE
coast of Estonia.
The areas with relatively large particle age are located, as expected, far from the
coasts and islands. Their spatial distribution (Fig. 10.4) mirrors the properties of a
similar distribution for the probabilities such as a rich internal structure and a certain
asymmetry in the North-South direction. The distribution of the particle age in areas
with aij > 6 days is obviously smoothed to some extent owing to the use of 10-day
long time windows. The largest values of particle age and the smallest probabilities
are located in areas considerably shifted to the north from the geometrical axis of
the gulf in the narrowest part of the gulf between Tallinn and Helsinki. These areas
do not necessarily coincide with domains that have the longest distance to the coast.
This implicitly reflects the anisotropy of the surface currents and signifies that the
entire approach leads to nontrivial results for the Gulf of Finland.
10.5.5 Dependence on Spatial Resolution
The discussed properties obviously depend on the ability of the circulation model
to represent the geometry and bathymetry of the basin and to resolve mesoscale
motions. The dependence of the distributions of probabilities and of the particle age
on the model resolution was studied using the OAAS model with spatial resolutions
of 2, 1 and 0.5 nm and otherwise identical setup, forcing and boundary conditions.
The number of sea points in the calculation area was 2270 for the 2 nm model, 8810
for the 1 nm model and 31,838 for the 0.5 nm model (Andrejev et al. 2010).
The temporal behaviour of the mean probability P (k) and particle age A(k) for
particular time windows, and their cumulative values P̄ (n) and Ā(n) was very sim-
ilar for all horizontal resolutions (Fig. 10.5). The time series of P (k) and A(k) are
highly correlated (for example, r = 0.98 for the 1 nm and 0.5 nm simulations) and
have small bias, mean and root-mean-square deviation, and also a very low level of
spread. The standard deviations for the pointwise values from the average values of
P (k) and A(k) are almost the same for different resolutions (Andrejev et al. 2011).
This suggests that the variations in these fields with respect to the mean values have,
on average, a similar structure.
The estimates of the asymptotic values P̄ (Nmax ) and Ā(Nmax ) are practically
insensitive to the resolution of the ocean model. Therefore, these quantities are to
some extent similar to certain scalar fields such as salinity and temperature that are
also well represented by eddy-permitting models starting from a 1 nm resolution.
The resolution of the ocean model affects to some extent the 2D maps of the
probability pij and particle age aij (Fig. 10.6). The overall appearance of these
maps, the location of the isolines and the areas of low probabilities and high particle
age largely coincide for all resolutions. The largest differences between the relevant
maps at the resolutions of 1 nm and 0.5 nm are in the size of the areas of the smallest
Fig. 10.5 Difference

between the mean probability
P (k) and the cumulative
average probability P̄ (n)
(upper panel) and the mean
particle age A(k) and its
cumulative value Ā(n) (lower
panel) calculated at different
horizontal resolutions of the
OAAS model. Green and red:
difference between the results
of the 1 nm and 2 nm models;
blue and magenta: difference
between the results of the
1 nm and 0.5 nm models
(Andrejev et al. 2011). The
results for the 1 nm model are
shown in Fig. 10.4
probabilities (pij < 0.4) and the areas of the largest particle age (aij > 8 days): do-
mains of very small probability or of large particle age are larger in the calculations
with the 1 nm model.
The accuracy of the representation of islands and bathymetry obviously is most
important with respect to these differences. Only the few largest islands are rep-
resented in the 2 nm OAAS model. An increase in the resolution evidently leads
to a much better reproduction of mesoscale eddies and smaller-scale flow features
The presence of the Finnish archipelago and many small islands, such as Keri to
the north of Prangli, Sommers (Someri) between Gogland and Vyborg, and Malyj
Tjuters (Pieni Tytärsaari, also Väike Tütarsaar) to the NE of Kunda at the 0.5 nm
resolution (Fig. 10.2) obviously leads to a reshaping of the areas with high pij and
low particle age.
The presence of islands and the more exact representation of the bathymetry of
the Gulf of Finland (Andrejev et al. 2010) are evidently responsible for a relatively
high level of local fluctuations in the fields of pij and aij at the 0.5 nm resolution.
Although a significant increase in the complexity of the fine structure of the result-
ing fields is evident in Fig. 10.7 for the 0.5 nm model, the shape and location of
the isolines are almost the same. The increase in resolution is accompanied by an
increase from 4 to 16 times in the number of evaluated trajectories, which definitely
improves the accuracy of the estimates for sea domains of fixed size.
The areas of minimal probability (maximal age) correspond well with sea areas
hosting either relatively intense westward subsurface transport or with domains with
quasi-steady eddies (cf. Fig. 11 of Andrejev et al. 2004a). This match suggests that
these eddies reflect the gulf’s bathymetry rather than dynamic mesoscale features.
Such a ‘geometric’ determination of the location of eddies evidently enhances the
342 T. Soomere
Fig. 10.6 Probability of a coastal hit in the Gulf of Finland simulated with the use of the OAAS
model with a resolution of 2 nm (upper panel), 1 nm (middle panel) and 0.5 nm models (lower
panel) for the period 1987–1991 (Andrejev et al. 2011)
Fig. 10.7 Particle age in the Gulf of Finland simulated with the use of the OAAS model with a
resolution of 2 nm (upper panel), 1 nm (middle panel) and 0.5 nm (lower panel) for the period
1987–1991 (Andrejev et al. 2011)
344 T. Soomere
Fig. 10.8 Distribution of p̂

characterizing the probability
of hitting the northern and
southern coasts using the
smoothing method and
tw = 20 days. Black and red
lines indicate the
equiprobability lines for the
direct and smoothing
methods, respectively
similarity of the results obtained with the models at 1 nm and 0.5 nm resolution as
both these models reasonably reproduce the bottom shape.
10.6 Applications for Decision-Making
10.6.1 Sharing the Costs: the Equiprobability Line
Many elongated sea areas are divided between different countries. In such cases it
is of interest to specify the (equiprobability) line so that the probability of transport
of the pollution released at this line to the opposite coasts would be equal. This
situation has clear practical importance for the Gulf of Finland: Estonia is located
on the southern coast and Finland on the northern coast so that the dividing line
largely follows the centreline of the water body. An appropriate quantification of the
offshore domains in this framework is possible by assuming the ‘costs’ of hitting the
opposite coasts equal in the magnitude of ‘badness’ but having opposite signs.
This approach was employed in Soomere et al. (2010, 2011c) using the 2 nm
RCO model data for 1987–1991, the TRACMASS code, 10 or 20 day long trajecto-
ries, the time lag between realizations of 1 day and the specification of the nearshore
as a 3 grid cell (about 11 km) wide belt (Viikmäe et al. 2010). Differently from the
above calculations with the OAAS model, the presence of islands was entirely ig-
nored.
The problem was addressed using two methods that only differed from each other
by the number and the location of the particles. The direct method (called Method
I below) tracked four particles released in each grid cell. A counter for a cell was
initially set to 0 and switched to ±1, when at least three out of these reached one
and the same nearshore (−1 for the northern and +1 for the southern coast). The
averaging procedure was exactly the same as described above. The resulting quan-
tity p̂ was in the range of −1 ≤ p̂ ≤ 1 (Fig. 10.8) where p̂ ≈ 1 and p̂ ≈ −1 stood
for a high probability of hitting the southern and the northern coast, respectively.
Another (smoothing) method (Method II) accounted for the drift of particles in a
cluster of nine cells. The counter was set to ±1 when ≥5 particles released into the
centres of the cell of interest and into the eight cells surrounding it reached one and
the same coast. In early simulations using 10-day long trajectories (Soomere et al.
2010) the separation point of the opposite coasts was set at the mainland border
between Estonia and Russia at the River Narva mouth in the SE of the Gulf of
Finland (28◦ 2 E, 59◦ 27.5 N). In later simulations with the same model and 20
day long trajectories, the separation point was set at the mouth of the River Neva
(Soomere et al. 2011c).
The equiprobability line followed the gulf axis in the narrowest part of this water
body, deviated substantially to the north of it at the gulf entrance and in the widest
part of the gulf, and came quite close to the southern coast of the easternmost narrow
part of the gulf. While the latter feature evidently mirrors the flow of the voluminous
runoff from River Neva that merges with the overall cyclonic circulation in the gulf
in this area (Chap. 6), the two other deviations of the equiprobability line from the
centreline of the gulf apparently signify certain systematic anisotropic features of
the transport by surface currents.
10.6.2 Cross-Sections
Different cross-sections of the distributions of the probability, the particle age and
the quantity p̂ characterizing the probability of hitting the opposite coasts of the
Gulf of Finland reveal several important features of the underlying fields (Fig. 10.9).
For relatively narrow parts of the gulf all the fields have a clearly defined extremum
or zero-crossing. A deviation from these points by only a few km is associated with
a considerable change in the values of the underlying quantities (equivalently, in the
environmental risks). Interestingly, the exact locations for the optimum values (po-
tential locations for the sailing line) not necessarily coincide for different measures
even in the narrowest part of the gulf.
While the zero-crossings of p̂ are usually located very close to the gulf axis,
the extrema of other fields may be substantially shifted to either side. The largest
deviations are shown by the locations for the longest particle age for a particular
longitude. The particularly large deviation of these locations for the two ways of
interpreting the fate of particles that drift out of the gulf demonstrates the sensibility
of the entire approach on the underlying value system.
Another significant feature of the cross-sections of relatively wide sections of the
gulf is the presence of a wide area of low cross-gulf gradients in which the values of
the underlying distributions reveal quite limited changes. This area can be naturally
identified as an area of reduced risk. In most occasions this area ends abruptly, and
relocation of a potentially dangerous activity closer to the coast is associated with a
rapid increase in the relevant risks. The presence of a well-defined bend in the cross-
sections in Fig. 10.9 indicates a possibility for systematically selecting a reasonable
distance from the coast for possible sources of environmental risks. This feature can
be particularly useful for the specification of the minimum acceptable distance from
the coast for tanker routes along open ocean coasts.
346 T. Soomere
Fig. 10.9 Examples of cross-sections of probabilities for coastal hits (green), particle age (Method
IVa: red, Method IVb: magenta; see below Sect. 10.7.3 for the explanations) and for the quantity
characterizing the probability of hitting the northern and southern coasts (direct method: black;
smoothing method: cyan, both shown as (p̂ + 1)/2: for the gulf entrance (a), western narrow
part (b), middle of the gulf (c) and for the wider part of the gulf between Narva and Kotka (d). Cir-
cles indicate the optimum location of the potentially dangerous activities (Soomere et al. 2011c).
The location of cross-sections is shown in Fig. 10.10
10.6.3 Simplest Optimum Fairways
The calculated distributions can be used as a support for decision-making in a vari-

ety of problems. The global minimum of the probability of coastal hits and similarly
the maximum of the particle age indicate a climatologically valid approximate so-
lution for a stationary location of a potentially dangerous activity, for example, a
drilling rig, an oil platform, or a crossing point of ship traffic.
An important application is the environmental management of ship traffic in
terms of minimizing the consequences of potential release of contaminants. Exam-
ples of the quantity p̂, the probabilities of coastal hit and the particle age have been
used in different applications. Regardless of which particular measure for the quan-
tification of the offshore has been used, the goal was to determine the location of the
optimum fairway (more generally, the pathway for a moving source of pollution)
based on the obtained distributions.
The technique will be discussed here using an example of environmental risks as-
sociated with current-driven coastal pollution but it is obviously applicable to con-
struct an optimum for virtually any set of cost distributions of possible accidents
generated by the definition of valuable or vulnerable areas in terms of any conve-
nient measure of the environmental or economic risk. We start from an idealized
case when the length of the optimum fairway does not play a role and the basic goal
is to minimize the average probability of coastal hit or to maximize the particle age
along the fairway or to share costs equally.
If the intention is to divide costs equally, an inherent approximation to the opti-
mum solution is the equiprobability line (Soomere et al. 2010). Sailing along this
line naturally results in dividing the potential costs equally between the opposite
coasts: a deviation from it to either side increases the projected costs of conse-
quences for this side. It is heuristically clear that the equiprobability line generally
does not provide a global minimum for environmental risks. This solution is only
suitable for elongated basins and accounting for the presence of islands may be
complicated.
For a sea area of elongated shape the maps of probability of coastal pollution and
of particle age, ideally, provide an elongated ‘trough’ of low values for probabilities
or, equivalently, a similar ‘crest’ of particle age, and a more or less gently sloping
pattern of the measure −1 ≤ p̂ ≤ 1 between the opposite coasts. Figures 10.7–10.9
suggest that the resulting fields generally have one local extremum of the probability
and age (and also one zero-crossing point of the quantity p̂) for all cross-sections of
the Gulf of Finland except for the entrance and for the easternmost part of the gulf.
It is natural to assume that the optimum fairway goes roughly along this trough or
crest, or follows the zero-crossing line of p̂.
In practical applications the underlying data for Figs. 10.7–10.8 were first cor-
rected manually in order to adjust a few clearly erroneous values (zeros of the prob-
ability or very large values of the particle age) for several bays deeply cut into main-
land. In the model resolution, currents in these areas were very small and almost no
particle propagation occurred in the RCO model that worked without local spread-
ing. The latitude for the points of the optimum fairway was found from the analysis
of the cross-section of the relevant field along each longitude (Fig. 10.9). It was set
at the centre of the grid cell corresponding to either the minimum probability pij or
the maximum particle age pij (Soomere et al. 2011c; Lu et al. 2012). The obtained
estimates for the optimum fairway were confined to discrete values of the centres of
grid cells and sometimes contained abrupt shifts by 2–3 cells to the south or north.
Optionally the curves were smoothed over five neighbouring values in the east–west
direction (Fig. 10.10).
Another estimate of the optimum fairway roughly following the equiprobabil-
ity line was constructed from the zero-crossing points of the measure p̂ (Soomere
et al. 2010). These points were well defined in the relatively narrow parts of the gulf
that hosted comparatively large north–south gradients of p̂. The small-scale fluc-
tuations in the distribution of p̂ in some parts of the bay (especially in the eastern
part where its north–south slope was small) were smoothed by averaging the cross-
section over five neighbouring grid points. An estimate for the zero-crossing of p̂
for each longitude was found using a linear approximation between its smoothed
values at adjacent grid points. Doing so led to a single clearly defined zero-crossing
point for each longitude. The curves consisting of the zero-crossing points required
no additional smoothing in the east–west direction (Soomere et al. 2011c).
348 T. Soomere
Fig. 10.10 The location of the optimum fairways based on Methods I–IV (see Sect. 10.7.3). The
colour code is the same as for Fig. 10.9. The dotted black line indicates the geometrical centreline
of the gulf for the 2 nm grid and the vertical lines correspond to the cross-sections in Fig. 10.9.
Note that the location of smoothed optimum fairways may slightly deviate from those indicated in
Fig. 10.9 (Soomere et al. 2011c)
Fig. 10.11 Trajectories of ships in the Gulf of Finland over two weeks in November 2009
(Soomere et al. 2010). Data courtesy HELCOM
10.6.4 Following Local Decisions
For regions with complicated geometry and strongly heterogeneous transport pat-
terns the shape of the relevant distributions may be quite complicated; for example,
certain cross-sections may host several extrema. A simple algorithm to obtain a first
approximation for the optimum fairway was introduced in Andrejev et al. (2011). It
accounts for a specific pattern of the ship traffic in the Gulf of Finland (Figs. 10.2,
10.11) and is based on local properties of the underlying measures. The majority of
cargo ships (including tankers that carry the largest potential for pollution) sail from
the Baltic Proper to large harbours in the eastern part of the gulf (HELCOM 2009).
The easternmost point of the fairway for ships sailing to the Saint Petersburg area
was chosen near the destination harbour. The next point was selected as the one cor-
responding to the minimal probability of a coastal hit (or the maximal particle age,
Fig. 10.12 The simple

algorithm for calculating an
optimum fairway. The next
fairway point is chosen from
among the centres of cells
1–5 or 2–4 adjacent to the
instantaneous location of the
ship (marked by the arrow)
or the smallest |p̂| among the five adjacent grid points in the main sailing direction
(Fig. 10.12). The process was repeated until the fairway reached the Baltic Proper.
The resulting fairway may head to the north or south for some periods of time but
cannot go back.
Fairways constructed using this algorithm and based on the probability for the
coastal hits (Fig. 10.13) follow the identical route along most of the Gulf of Finland
and diverge to the different destinations about half-way from the Helsinki–Tallinn
line to Gogland. The fairway to Vyborg is located asymmetrically, mostly to the
north of the centreline of the gulf, suggesting that a southward surface drift prevails
almost everywhere except possibly to the north of Lahemaa. This feature suggests
that the anticyclonic gyre identified for the surface layer of the Gulf of Finland in
the 2 nm RCO simulations (Soomere et al. 2010) is implicitly evident also in the
simulations with a higher resolution.
As the general appearance of the distributions for the probabilities (Fig. 10.4)
and the particle age (Fig. 10.7) is fairly similar, it is not unexpected that the corre-
sponding optimum fairways quite closely follow each other (Fig. 10.13). They are
mostly located to the north of the gulf axis (by 2–8 km on average) and meander
substantially in some sections. They almost overlap in the relatively narrow part of
the gulf between Naissaar and Porkkala and in narrow passages between islands,
e.g., to the south of Gogland (Fig. 10.2). They deviate up to 20 km from each other
in the widest sections of the gulf where the gradients of the underlying fields are
small (cf. Fig. 10.9).
Surprisingly, the two optimum lines also considerably deviate from the axis of
the gulf in the very narrow area between Tallinn and Helsinki that hosts extremely
heavy crossing cargo and passenger traffic (HELCOM 2009). If the fairways were
allowed to exert local turns by up to 135◦ , they may follow a northern route between
Gogland and the Finnish archipelago. Such a large difference between the results of
quite similar methods of fairway construction highlights the influence of the local
decision at each grid cell on the overall shape and parameters of the optimum sailing
line and certainly calls for the implementation of more elaborated approaches to
specify a globally optimized fairway.
In essence, the described procedure is a discrete variant of the method of the least
steep gradient for finding crests or troughs on a 2D map of elevations in which all the
decisions are made locally. If the underlying measure (pij , aij or p̂ij ) has exactly
one minimum, maximum or zero-crossing across the gulf, the process obviously
finds and follows it. The template for the next point in Fig. 10.12 can be rotated in
order to fit with the general direction of the fairway, which for example goes to the
350 T. Soomere
Fig. 10.13 Upper panel: Optimum fairways to Vyborg, Saint Petersburg, Ust-Luga and Sillamäe
based on the probabilities for the coastal hits calculated for 1987–1991 using the 1 nm OAAS
model (Fig. 10.6) and an additional restriction that the local turn of the fairway should not exceed
90◦ at each point (Andrejev et al. 2011). The fairways are smoothed over 7 subsequent points. The
underlying bathymetry (depth in metres) is from Andrejev et al. (2010). Lower panel: optimum
fairways from Vyborg to the Baltic Proper based on spatial distributions of probabilities for the
coastal hits (blue) and particle age (red) with no restrictions to the local turn of the fairway. The
black line indicates the centreline of the gulf (Soomere et al. 2011b)
north after passing through the Great Belt (Fig. 10.14). The obvious shortcoming is
that the rotation of the template to force the optimum fairway to head to the desired
destination can only be made in 45◦ steps.
The process frequently leads to extensive excursions of the resulting path across
the main sailing direction (Fig. 10.14). The resulting fluctuations may be sup-
pressed, for example, iteratively after the entire fairway has been specified (Soomere
et al. 2011b). Another way of receiving more smooth optimum sailing lines ab initio
is to restrict the local turns in the sailing direction. A simple way of doing so is to
search the next point among three adjacent cells located strictly in the main sailing
direction (cells 2–4 in Fig. 10.12) (Andrejev et al. 2011).
Fig. 10.14 A selection of

optimum fairways in the SW
Baltic Sea based on the
probability for coastal hits
(red) and particle age (white)
for the period of 1990–1994
on the background of
bathymetry (depth in metres)
(Lu et al. 2012, modelling
and visualization by X. Lu
and E.V. Stanev)
A less obvious feature is that the process is not symmetric with respect to the
change in the search direction. In the case of the Gulf of Finland it generally fails to
establish the optimum fairway for ships sailing eastwards because the optimum line
not necessarily approaches the destination harbour. In general, the systematic use of
spatio-temporal variations in these distributions in order to minimize environmental
risks is a complicated 2D optimization problem.
There exist much more elaborated methods for finding an optimum line over a
2D field such as Dijkstra’s algorithm (used in Höglund and Meier 2012 for solving
a similar problem based on Eulerian transport), discrete versions of the variational
method, or methods based on Monte Carlo simulations (see Chap. 11 for a descrip-
tion of some alternatives). The main purpose of the presented exercises is, however,
to demonstrate that even the use of a very rough estimate for the optimum fairway
has a clear potential for a substantial benefit in certain environmental terms, ex-
pressed either as a considerably smaller probability for a coastal hit or as a clearly
larger time for reaching the coast.
10.7 The Benefit and Uncertainties
One consequence from the vast variety of ecological values and indicators of en-
vironmental well-being (Rice 2003) is that the estimates of the benefit enormously
depend on the particular choice of the value system. In this context an instructive
feature of the presented exercise is not only the invariance of some estimates of the
gain from options in calculations but, more importantly, the potential increase in the
basic quantitative measures of environmental risks.
352 T. Soomere
10.7.1 Quantification of the Benefit
The simplest estimates of the benefit for a stationary location of a potentially dan-
gerous activity that has to be carried out in the open sea are the ratio of the mini-
mum probability min(pij ) and the average probability P̄ (Nmax ), and the difference
between the maximum max(aij ) and average particle age Ā(Nmax ). The typical val-
ues of P̄ (Nmax ) are in the range from 0.62 to 0.67 in the Gulf of Finland (Andrejev
et al. 2011; Soomere et al. 2011c) whereas the minima of the probability extend
well below 0.3. In realistic conditions there exist many constraints for the location
of such spots and the benefit has to be estimated by comparing the extrema of the
particular measure with its average values over the technically suitable sea domain.
For certain choices of measures only the additional ‘costs’ of the consequences
can be estimated. An example is that the equiprobability line only divides equally
the potential loss between the opposite coasts. The relevant measure p̂ij contains
very limited information about the actual probability of the coastal hit. The benefit
from its use can be measured to some extent as the average value of p̂ (either for a
particular location or for a fairway roughly following this line) showing the added
risk for one side.
Similarly the potential benefit from the use of an optimum sailing line substan-
tially depends on the underlying measure, the value system and additional con-
straints. The simplest way to estimate the benefit from the use of the sailing line
that provides a reduced level of probability of coastal hits in case of possible acci-
dents can be roughly evaluated in terms of the decrease in the probability of such a
hit compared with a ‘random’ location of the fairway. A convenient measure is the
ratio popt /P̄ (Nmax ) of this probability popt calculated along the optimum fairway
over the average probability for the entire sea. In these terms the benefit is quite sub-
stantial. For example, for the 2 nm RCO model the average probability P̄ (Nmax ) for
a coastal hit in the Gulf of Finland within 20 days is 0.62. The similar average value
along the optimum fairway is 0.347, giving a gain of 44 % (Soomere et al. 2011c).
For the 1 nm OAAS model and 10 days long trajectories P̄ ≈ 0.67 and popt ≈ 0.4
(that is, about 40 % lower) along the optimum fairways shown in Fig. 10.13.
In the same vein the benefit can be expressed in terms of an increase in the time
elapsed from the release of the adverse impact until it reaches the coast. The average
particle age is 8.7 days in the Gulf of Finland for the 2 nm RCO model and 20-
day long trajectories (Soomere et al. 2011c). The average over the points lying on
the optimum fairway to the eastern end of the gulf is 12.4 days. The gain is thus
up to 42.5 %. The values of aopt and Ā(Nmax ) are much shorter (5.3 and 9 days,
respectively) for calculations using 10-day long trajectories and the 1 nm OAAS
model (Andrejev et al. 2011).
It is remarkable that the potential in increase in the time available for combating
the pollution is almost the same, slightly below 4 days, for radically different model
setups. This invariance could suggest that the potential increase in time is largely
independent on the particular length of the trajectories (provided the number of hits
to the vulnerable area leads to adequate statistics) and possibly is an intrinsic feature
of the current-driven dynamics of the particular domain.
The provided statistical estimates of the gain from the systematic use of an op-
timum sailing line (almost halving the probability for coastal hits or buying almost
four days to combat the pollution in the open sea) are obviously considerably overes-
timated. Such a high benefit is evidently not possible in realistic conditions because
a large part of the ship traffic uses anyway deep offshore areas that have relatively
low probabilities for coastal hits and high values of the particle age (Andrejev et al.
2011).
10.7.2 Critically Questioning the Recommendations
The computed 2D maps inevitably contain some noise. Given also the intrinsic un-
certainties of the technology, it is obvious that the presented solutions are just ap-
proximations for the best ones. The above has also shown that they may substantially
differ when based on different criteria used in building the maps and may depend
on the details and options of the technique. The results for their use in engineering
applications, however, have to be robust: new achievements (e.g., an increase in the
resolution of the ocean model) may perhaps add more details but should in no way
override the principal outcome of the first approximation.
The constructed fairways represent approximations of the climatologically valid
solution as they are based on a multi-year average of the underlying fields of prob-
ability and particle age from which the potential seasonal effects evidently have
been filtered out. Substantial seasonal variations in net and bulk transport patterns
calculated using a similar technology (Soomere et al. 2011d), similar variations in
the basin-average values of wind stress (Andrejev et al. 2011), significant seasonal
patterns of the predominant wind direction (Soomere and Keevallik 2003) and dis-
tributions of upwelling regions (Lehmann and Myrberg 2008) suggest that proper-
ties of surface transport (and, consequently, the location of the optimum fairway)
apparently show extensive seasonal and possibly also interannual variations in the
Gulf of Finland. Some of the variations in the dynamical regimes corresponding to
the inflow and outflow conditions in the SW Baltic Sea have been addressed in Lu
et al. (2012). A much longer time period has to be covered by simulations in or-
der to establish the presence and extension of the related variations in the optimum
solutions.
The quality of the recommendations for the location of stationary activities can
be roughly estimated as the level of invariance of (the extrema of) the maps of
probability, particle age and p̂ with respect to different options (Fig. 10.7). There
exists, however, no single measure to characterize the quality of the solutions for the
optimum fairway. Technically, the robustness of the outcome of the presented ap-
proach can be quantified as the typical root-mean-square deviation (rmsd) between
the optimum fairways. The difference between various optimum sailing lines can be
interpreted as a measure of the uncertainty of the entire approach (Soomere et al.
2010). Uncertainty has not necessarily a negative meaning here: strictly following
the optimum path is not always the best solution as ships do need some freedom
354 T. Soomere
Fig. 10.15 Probabilities (left panel, %) of hitting the coast and particle age (right panel, days) for
the years 1987–1991 calculated with the 2 nm RCO model and TRACMASS code with the time
window of 20 days for the Gulf of Finland (Soomere et al. 2011c). Note that the largest values of
particle age considerably exceed those calculated in Andrejev et al. (2011) using tw = 10 days
for manoeuvring. For practical purposes it is thus imperative to estimate how sen-
sitive the environmental risks are (e.g., how rapidly the probability increases or the
particle age decreases) with respect to small variations in the ships’ routes from the
theoretical optima.
10.7.3 Robustness and Uncertainty
An attempt to shed light on the questions raised in the previous subsection is de-
scribed in Soomere et al. (2011c). They used the same ocean model (the 2 nm RCO
model), forcing conditions, particle tracking scheme (off-line TRACMASS code),
set of 20-day long trajectories of particles released into the centres of each of 3131
sea grid cells in the Gulf of Finland and the 3-cell wide nearshore to consider five
different ways to define the optimum fairway. The direct (Method I) and smoothing
(II) method were used to specify the equiprobability line as discussed above.
The optimum fairways were calculated using the probability for coastal hits (III)
and two methods to handle the age of particles that leave the gulf. Firstly, the age
counter was stopped when the particle left the gulf for the first time (IVa). This
assumption leads to an obvious underestimation of particle age at the entrance of the
gulf (Fig. 10.15). Alternatively, the age for particles leaving the gulf was assigned
the length of the time window (IVb). Doing so apparently overestimates particle age
for regions of intense water exchange with the Baltic Proper.
The methods basically differ from each other in the definition of the benefit. Only
Method I relies on a larger set of trajectories (4 particles in each cell, see above).
The optimum fairways were constructed from the extrema or zero-crossing points
of the north-south cross-sections of the resulting 2D distributions along 110 discrete
longitudes between 21◦ 41 E and 28◦ 57 E. The cross-sections of p̂ and the resulting
fairways based on other measures were smoothed over five subsequent points. The
nearshore of all islands in the gulf was ignored and only hits to the mainland coasts
were accounted for in order to make the calculations of the equiprobability line and
the optimum fairways directly comparable.
Fig. 10.16 Deviation of the optimum fairways based on Methods I–IV from the centreline of
the Gulf of Finland. The positive direction is to the north. The centreline is determined using the
2 nm RCO model grid and not necessarily exactly follows the geographical centreline of the gulf
The resulting variations in the optimum sailing line implicitly characterize the
intrinsic uncertainties (and thus robustness) of the entire approach. The typical rmsd
between the optimum fairways specified from different criteria was 6–16 km. This
estimate averages out several interesting patterns of the deviations. All five lines
were located in a narrow corridor between latitudes 23◦ 40 E and 26◦ 20 E where
they deviated from each other by <10 km (Fig. 10.16). Such a concentration of the
solutions apparently results from the narrowness of this part of the Gulf of Finland
where less than ten grid cells represent the offshore dynamics of trajectories and the
underlying fields have steep gradients in the north-south direction (Fig. 10.9). The
two versions of the equiprobability line were located relatively close to each other
(bias 1.6 km, rmsd 6.3 km). The line representing the optimum of the probability of
coastal hits follows these lines very closely.
Some of the lines substantially deviated from each other (by up to 50 km) at
the entrance to the gulf and in wider parts of this water body where the north–
south gradients of the underlying fields were considerably smaller. The systematic
deviation of the optimum fairways to the north of the centreline of the gulf (around
27◦ 20 E) and a smaller deviation to the south around 25◦ 30 E may be associated
with a slow anticyclonic gyre (Soomere et al. 2011d). The large deviation to the
south around 28◦ E evidently reflects the predominant transport to the north in this
area jointly by the voluminous runoff of the River Neva and the general cyclonic
circulation pattern in the Gulf of Finland (Andrejev et al. 2004a).
Three estimates of optimum fairways out of five were on average located 2–4 km
to the north of the centreline of the gulf. Interestingly, the methods IVa and IVb led
to quite different locations for the optimum fairway not only at the entrance to the
Gulf of Finland but in the entire gulf. The fairway corresponding to Method IVb
was located almost 8 km to the north of the axis whereas the one for Method IVa
was almost 4 km to the south of the axis (Soomere et al. 2011c). The large deviation
of these lines in the entire gulf indicates that particles in the entire northern part
of the Gulf of Finland have a large chance to leave the gulf. This example vividly
356 T. Soomere
demonstrates how sensitive the resulting optimum fairway may be with respect to
seemingly small variations in the method.
10.7.4 Nearly Optimal Solutions
The issues of robustness and possible uncertainties are closely related to the question
of sensitivity of the optimum fairways calculated using different options (Soomere
et al. 2011c). A convenient way to quantify the sensitivity of the environmental risks
associated with optimum solutions with respect to small variations is to construct a
‘corridor’ surrounding the optimum sailing line. In such a corridor the underlying
measure of risk (probability for a coastal hit, particle age or p̂) is allowed to vary to
some extent compared to its optimum value. The shape of such corridors together
with the values for the gradients of the underlying fields provides information about
the flexibility of shipping and characterizes the degree of freedom for captains (or
for the decision-making in general) to choose the sailing line with a reasonable
increase in the environmental risks. The along-gulf variations in the width of such
corridors may also highlight sea regions with different internal dynamics.
The spatial variability of such corridors with a predefined total area (equivalently,
with a fixed average width of 15 km in the Gulf of Finland) is analysed in Soomere
et al. (2011c) using the 2 nm RCO model, the TRACMASS code and the five spec-
ifications of the optimum solution found from the cross-sections of the underlying
distributions as described in the previous section. Within such a corridor the proba-
bility of coastal hits (or the particle age) is allowed to differ to some extent from the
relevant minimum (maximum).
A corridor with an average width of 15 km (8.1 nm) corresponds to an increase
in the probability of coastal hits by up to 0.0293 (or to a decrease in the particle
age by up to 0.54136 days) for each longitude compared to the optimum value.
The corridor’s border lines were determined from an analysis of the cross-sections
of the underlying distributions as described above. The locations of the probability
p = pmin + 0.293 (pmin is the minimum probability for the particular longitude) or
the particle age of a = amax − 0.54136 days were found for each latitude. The curve
consisting of such points was smoothed over five neighbouring points.
The resulting corridors were quite wide at the entrance to the Gulf of Finland,
relatively narrow in the Tallinn–Helsinki region (Fig. 10.17), widened substantially
in the eastern part of the gulf in the vicinity of Gogland and become very narrow
again in the area to the north of Luga Bay. The minimum and maximum widths
were 7.35/9.6 km and 29.1/42 km, respectively, for Methods III/IVa.
Similar corridors were defined as the sea area between the equiprobability lines
corresponding to p̂ = ±0.11189 for the direct method I and to p̂ = ±0.09542 for
the smoothing method II. The fields of p̂ had a very steep north–south slope in the
central part of the bay in both cases. The corridor for the direct method (Fig. 10.18)
had a minimum width of only 2.1 km. The slope was very small at the entrance of the
gulf where the resulting corridor was up to 48.6 km wide. The equiprobability line
Fig. 10.17 Optimum fairway based on probabilities for coastal hit (green) and particle age (red,
Method IVa) and the relevant ‘corridors’ with the average width of 15 km in the Gulf of Finland.
The dotted line shows the geometrical centreline of the gulf (Soomere et al. 2011c). The back-
ground colour scale shows water depth (m) (Andrejev et al. 2010)
Fig. 10.18 Sea area in which the quantity p̂ found using the direct method I is |p̂| < 0.11189
(thin cyan lines surrounding the yellow area) and in which the similar quantity found using the
smoothing method II is |p̂| < 0.09542 (thin black lines) in the Gulf of Finland. Solid cyan and
black lines indicate the relevant equiprobability lines and the dotted line—the centreline of the
gulf (Soomere et al. 2011c)
was mostly located near to its northern border. The corridor widened up to 26.2 km
near Gogland and narrowed considerably at the entrance to the Neva Bay.
The similar corridor for the smoothing method corresponds to p̂ = ±0.09542.
It is even narrower (down to 1.19 km) than the one for the direct method in the
vicinity of the Tallinn–Helsinki line, reaches a width of over 28 km near Gogland
and 37.6 km at the entrance to the Gulf of Finland. Both corridors largely follow
the centreline in the narrowest part of the gulf but deviate considerably to the north
of it in the vicinity of Gogland. The two equiprobability lines and corridors almost
coincide near Hanko (where they are located up to 10 km to the north from the
centreline) but deviate from each other in the entrance area of the Gulf of Finland.
The drastic variations (from 1 or 2 km up to 30 km) of their local width signify
358 T. Soomere
that the sensitivity of the results with respect to small changes in the environmental
criteria may largely vary in different parts of the gulf (Soomere et al. 2011c).
10.7.5 Limitations of Modelling
The construction of different maps using the presented technology and specifically
studies of its reliability and associated uncertainties are computationally expensive
and time-consuming. It is thus natural to address potential simplifications of the cal-
culations involving a minimum loss of accuracy but retaining the reliability of the
results. A generic way of reducing the computational efforts is to decrease the res-
olution of the underlying hydrodynamic model. Doing so is only feasible until the
model still adequately reproduces the mesoscale dynamics (cf. Albretsen and Røed
2010). As computational costs increase rapidly when increasing the 3D model reso-
lution, such a reduction is still highly desired for the practical use of the technique.
A natural limit is that the model should be at least eddy-permitting. This is a se-
vere condition in the test areas of the Baltic Sea where the baroclinic Rossby radius
is usually about 2–4 km, and may be even down to 0.5–1 km at specific locations
(Fennel et al. 1991; Osiński et al. 2010; Nekrasov and Lebedeva 2002).
Experience from numerical physical oceanography suggests that although the in-
stantaneous simulated currents are quite different for different resolutions and even
the statistics of currents may exhibit substantial changes for different model resolu-
tions (Albretsen and Røed 2010), some other fields such as salinity or temperature
can be reasonably replicated using models that poorly resolve the mesoscale dynam-
ics. Moreover, these fields may remain practically the same at different resolutions
(Andrejev et al. 2010; Myrberg et al. 2010b). For example, the appearance of sim-
ulated currents in the Gulf of Finland may change abruptly when the resolution is
increased from 0.5 to 0.25 nm but the salinity and temperature fields are almost the
same as for a resolution of 1 nm (Andrejev et al. 2010).
The overall location of the lowest probabilities of coastal hits in the meridional
cross-sections generally corresponds to the location of the equiprobability lines in
the western and central parts of the gulf in 2 nm simulations that ignored the pres-
ence of islands. The increase in the resolution from 2 nm to 1 nm and/or the more
exact representation of islands leads to several obvious differences in the results for
the eastern part of the gulf, especially between Gogland and Kotka (cf. Fig. 10.7 and
Fig. 10.14). In this context, it is natural to ask whether the above-mentioned maps of
environmental risks (reflecting, in essence, long-term statistics of the current-driven
transport), or at least certain parts of their integral features, belong to the family of
those characteristics that are largely insensitive to changes in the resolution of the
underlying ocean model. A related task is to identify an optimum spatial resolution
for the ocean model for different applications in a particular basin.
10.7.6 Sensitivity to the Resolution of the Ocean Model
The sensitivity of the technology in this respect is analysed in Andrejev et al. (2011)
based on on-line simulations of 10-day long Lagrangian trajectories calculated for
the period of 1 May 1987–31 December 1991 for the Gulf of Finland. The relevant
maps for the probability of coastal hits and for the particle age are evaluated using
the OAAS model at three different horizontal resolutions of 2, 1 and 0.5 nm and with
otherwise identical vertical resolution, initial, boundary and forcing conditions, and
trajectory calculation scheme. This range of resolutions characterizes a transition
from a poor representation of mesoscale effects by an eddy-permitting model to one
which is expected to adequately resolve statistical features of the field of mesoscale
eddies.
The dynamics of water masses in the Gulf of Finland is extremely complicated
and the resolution of even the 0.5 nm model does not perfectly resolve all the
small-scale features of water motion. The impact of subgrid-scale turbulence on
the spreading of trajectories was parameterized by the addition of a random distur-
bance containing a strong rotational component as described above. The resulting
set of 2D maps of cell-wise probability of hitting the coast and particle age is pre-
sented in Fig. 10.7. The general appearance of the distributions for the probabilities
and particle age are fairly similar in different resolutions. Particularly close are the
fields for 1 nm and 0.5 nm resolutions. The correlation between pointwise values
of probability and particle age is 0.98, the bias for p is 0.024 and for age as low
as 0.012. The mean absolute deviation is 0.037 and 0.21, respectively. The rmsd
between the two sets of fields is even smaller, 0.0021 and 0.081, respectively.
Also several integral measures almost coincide for the three resolutions. The
spatially averaged limiting values of the probability P̄ (Nmax ) and the particle age
Ā(Nmax ) also display hardly any dependence on the resolution. They both reach
almost identical stationary levels (0.66–0.69 and 5.25–5.33 days, respectively; the
standard deviations vary from 0.16 to 0.19 and 1.06 to 1.30) after five years of sim-
ulations (Andrejev et al. 2011).
The resulting solutions for fairways may be much more strongly affected by the
particular horizontal resolution of the ocean model than the integral variables and
2D maps discussed above. In spite of the almost perfect match of the integral char-
acteristics the optimum fairways only match each other in certain parts of the gulf
(Fig. 10.19). They only overlap in a short section of the narrowest part of the gulf
between Naissaar and Porkkala and converge at a certain point to the north of Lahe-
maa. The fairways for the 1 nm and 0.5 nm resolutions also converge in the narrow
passages between the islands to the south of Gogland (Soomere et al. 2011b, 2011c).
In other parts of the gulf the optimum fairways calculated using different reso-
lutions show a complicated pattern of behaviour. While all the optimum lines are
concentrated into a narrow corridor to the west of Naissaar, they deviate consider-
ably from each other even in the narrow area between Tallinn and Helsinki. The
fairways to Vyborg calculated at different resolutions visit completely different ar-
eas of the Gulf of Finland (Fig. 10.18). While the differences between the fairways
at the 1 nm and 0.5 nm resolutions are moderate, the fairway for the 2 nm model
360 T. Soomere
Fig. 10.19 Optimum fairways to Vyborg according to the spatial distributions of the probability
for coastal hits (solid lines) and the particle age (dashed lines) at resolutions of 2 nm (red and black
respectively), 1 nm (green and cyan) and 0.5 nm (yellow and white). The depth scale to the right of
the map is given in metres (Andrejev et al. 2011)
apparently reflects a completely different pattern of underlying dynamics, especially

in the central part of the Gulf of Finland.
This example vividly illustrates the importance of the choice of an adequate hor-
izontal resolution on the resulting location of the optimum fairway. Although the
spatial distributions of the relevant fields are qualitatively similar for all resolutions,
the optimum locations for fairways depend substantially on the resolution. It is re-
markable that the results for the 2 nm model differ considerably from those obtained
using finer-resolution models. As mentioned above, it is not possible to judge on the
basis of our simulations whether or not this difference is driven by the accuracy of
simulations of hydrodynamics and/or whether the accuracy of the representation of
the Finnish archipelago plays a role.
Interestingly, the potential benefit from the use of the optimum fairway (esti-
mated, e.g., as the difference between the mean probability of coastal hits and the
average value of this probability along the optimum fairway) very little depends on
the resolution of the ocean model. The benefit is also not directly connected with the
length of the resulting fairway. The longest of the three optima in Fig. 10.19 offers
the largest benefit. It leads to a decrease in the relevant probability from the mean
value of 0.67 to 0.46 along the fairway (Andrejev et al. 2011). Somewhat surpris-
ingly, the best formal benefit (mean probability over the optimum fairway 0.432) is
provided by the 2 nm model. The highest-resolution model provides a fairway with
the mean probability of 0.536 and with a very modest gain.

The proposed technique combines several advanced applications (a 3D circulation
model, a Lagrangian trajectory tracking method, a statistical analysis of certain
probabilities, and a method for 2D optimization) into one entity. Although con-
struction of a single modelling system is possible, we have intentionally highlighted
the interfaces and connections between the separate basic steps so that the user may
easily replace each component with a more convenient or more elaborated one. The
essence of the method does not depend on the properties of a particular model for
each single step. The presented approaches highlight the potential of such a multi-
step modelling system for the preventive planning of maritime activities. They also
reveal several instructive features of the outcome of the technique and particularly
the importance of a feasible choice of the target function.
It is important to emphasize that implementations of the technology described in
this chapter are based on several simplifications, and the validity of some of them
is not obvious. They mostly account for the impact of surface-current-driven propa-
gation of pollution on the optimum solution. Realistic applications have to account
for not only the impact of wave- and wind-induced transport of contaminants but
also complicated physical and chemical processes affecting their fate (ASCE 1996).
These questions will be partially addressed in Chap. 11.
The constructed example fairways are only conditionally optimized. The pre-
sented idealized examples (in which only the current-driven propagation is ac-
counted for) signify, however, that the impact of current-driven transport may add
a significant contribution to the marine-induced hazards in certain coastal sections.
A major simplification that might not be entirely justified is neglecting the poten-
tial impact of subgrid-scale processes in applications using the RCO model and the
DMI/BSHcmod model. Doing so may considerably alter the statistics of trajectories,
especially for longer time intervals.
The presented material suggests that eddy-permitting models with a grid step
exceeding half the local baroclinic Rossby radius are suitable for a quick check of
whether or not any potential gain from the presented method is feasible. However,
higher-resolution simulations using eddy-resolving models are definitely necessary
for detailed planning.
Although the instantaneous values of the basin-wide probability of coastal hits
and particle age reveal substantial spatio-temporal variations in different sea areas
and/or in different years and seasons (Fig. 10.3), their cumulative values converge
quite rapidly to a fixed level for a wide range of resolutions of the ocean model.
This convergence suggests that these asymptotic values characterize certain intrinsic
combinations of the geometry of the basin and internal current structure, and per-
haps even could be interpreted as new measures of the vulnerability of the nearshore
of water bodies to offshore traffic accidents for the chosen length of time window.
The (spatial) standard deviation of the probability and particle age indicates the
magnitude of their variations across the water body and thus serves as a natural in-
dicator implicitly characterizing the potential benefit from the smart positioning of
dangerous activities for the particular sea area.
The very limited dependence of several discussed measures and properties on
the resolution of the underlying hydrodynamic models needs some attention. The
probable reason is the multi-step averaging procedure of short-term transport fea-
tures (over time intervals exceeding the typical turnover time of mesoscale eddies
362 T. Soomere
and over the 5-year time interval) that filters out many short-term features of the
circulation. This filtering apparently affects the results of simulations that satisfac-
torily capture the mesoscale features to an almost equal extent. The overall match
of the optimum fairways for the 1 nm and 0.5 nm models suggests that the decisive
aspects of the mesoscale circulation of the Gulf of Finland (Andrejev et al. 2004a),
including the combined effects of the prevailing SW winds, the general structure
of the density field, the bathymetry and geometry of the gulf, etc., have been ade-
quately obtained using the 1 nm OAAS model. In other words, one might speculate
that the 1 nm model reproduces the statistical properties of current-driven transport
in the Gulf of Finland quite well. This conjecture is not completely unexpected but
is nevertheless interesting.
Acknowledgements This study was supported by the funding from the Estonian Science Foun-
dation and the European Community’s Seventh Framework Programme (2007–2013) under grant
agreement No. 217246 made with the joint Baltic Sea research and development programme
BONUS. The BalticWay project attempted to identify the regions in the Baltic Sea that are as-
sociated with increased risk compared to other sea areas and to propose ways to reduce the risk
of them being polluted by placing activities in other areas that may provide less risk. The research
was also partially supported by the Marie Curie RTN SEAMOCS (MRTN-CT-2005-019374), the
Marie Curie Transfer of Knowledge project CENS-CMA (MC-TK-013909), Marie Curie Rein-
tegration Grant ESTSpline (PERG02-GA-2007-224819), the targeted financing by the Estonian
Ministry of Education and Science (grants SF0140077s08 and SF0140007s14) and the Estonian
Science Foundation (grant No. 7413). The contribution of Anders Anbo towards the application of
the TRACMASS model in the Institute of Cybernetics and the help from Kristofer Döös during
its use are gratefully acknowledged. The Wave Engineering Laboratory team is deeply grateful to
Markus Meier and Anders Höglund (SMHI) who provided the RCO model data and meteorological
forcing in the framework of the BONUS cooperation.
References
Abascal AJ, Castanedo S, Medina R, Liste M (2010) Analysis of the reliability of a statistical oil
spill response model. Mar Pollut Bull 60:2099–2110
Albretsen J, Røed LP (2010) Decadal simulations of mesoscale structures in the northern North
Sea/Skagerrak using two ocean models. Ocean Dyn 60:933–955
Ambjörn C (2007) Seatrack web, forecasts of oil spills, a new version. Environ Res Eng Manag
3(41):60–66
Ambjörn C (2008) Seatrack web forecasts and backtracking of oil spills—an efficient tool to find
illegal spills using AIS. In: IEEE/OES US/EU-Baltic international symposium, Tallinn, Estonia,
May 27–29, 2008. IEEE Press, New York, pp 168–176
Andrejev O, Sokolov A (1989) Numerical modelling of the water dynamics and passive pollutant
transport in the Neva inlet. Meteorol Hydrol 12:75–85 (in Russian)
Andrejev O, Sokolov A (1990) 3D baroclinic hydrodynamic model and its applications to Sk-
agerrak circulation modelling. In: Proc 17th conf Baltic oceanographers, Norrköping, Sweden,
pp 38–46
9:1–16
Andrejev O, Sokolov A, Soomere T, Värv R, Viikmäe B (2010) The use of high-resolution bathy-
metry for circulation modelling in the Gulf of Finland. Est J Eng 16:187–210
334
ASCE Task Committee on modeling of oil spills, Water Resources Engineering Division [ASCE]
(1996) State-of-the-art review of modeling transport and fate of oil spills. J Hydraul Eng
122:594–609
Blanke B, Raynard S (1997) Kinematics of the Pacific Equatorial Undercurrent: an Eulerian and
Lagangian approach from GCM results. J Phys Oceanogr 27:1038–1053
Blumberg A, Mellor G (1987) A description of a three-dimensional coastal ocean circulation
model. In: Heaps NS (ed) Three-dimensional Coastal Ocean models. Coastal and estuarine sci-
ence series, vol 4. American Geophysical Union, Washington, DC, pp 1–16
Bowden-Kerby A (2001) The original MPAs (Marine protected areas). MPA News 3:5–6
Deleersnijder E, Campin J-M, Delhez EJM (2001) The concept of age in marine modelling. I.
Theory and preliminary model results. J Mar Syst 28:229–267
13,514
597
Döös K, Nycander J, Coward AC (2008) Lagrangian decomposition of the Deacon Cell. J Geophys
Res—Oceans 113:C07028
Eide MS, Endresen Ø, Brett PE, Ervik JL, Røang K (2007) Intelligent ship traffic monitoring for
oil spill prevention: risk based decision support building on AIS. Mar Pollut Bull 54:145–148
Engqvist A, Andrejev O (2003) Water exchange of the Stockholm archipelago—a cascade frame-
work modelling approach. J Sea Res 49:275–294
Engqvist A, Döös K, Andrejev O (2006) Modeling water exchange and contaminant transport
through a Baltic coastal region. Ambio 35:435–447
Res 11:23–36
French DP, Rines H, Masciangioli P (1997) Validation of an orimulsion spill fates model using
observations from field test spill. In: Proceedings of the 20th Arctic and Marine oil spill program
(AMOP) technical seminar, environment, Canada, Ottawa, Ontario, pp 933–961
364 T. Soomere
French-McCay DP (2004) Oil spill impact modeling: development and validation. Environ Toxicol
Chem 23:2441–2456
Funkquist L (2001) HIROMB: an operational eddy-resolving model for the Baltic Sea. Bull Mar
Inst, Gdańsk 28:7–16
Glasby GP (1997) Disposal of chemical weapons in the Baltic Sea. Sci Total Environ 208:145–147
Gollasch S, Leppäkoski E (2007) Risk assessment and management scenarios for ballast water
mediated species introduction into the Baltic Sea. Aquat Invasions 2:313–340
Gray JS (1997) Marine biodiversity: patterns, threats and conservation needs. Biodivers Conserv
6:153–175
Gräwe U, Wolff J-O (2010) Suspended particulate matter dynamics in a particle framework. Envi-
ron Fluid Mech 10:21–39
HELCOM (2009) In: Stankiewicz M, Vlasov N (eds) Ensuring safe shipping in the Baltic. Helsinki
Commission, Helsinki, 18 pp
Sweden, 29 pp
Jönsson B, Lundberg P, Döös K (2004) Baltic sub-basin turnover times examined using the Rossby
Centre Ocean model. Ambio 23:257–260
Kachel MJ (2008) Particularly sensitive sea areas. Hamburg studies on Maritime Affairs, vol 13.
Springer, Berlin, pp 376
Kleine E (1994) Das operationelle Modell des BSH für Nordsee und Ostsee. Bundesamt für
Seeschifffart und Hydrographie, 126 pp (in German)
Kokkonen T, Ihaksi T, Jolma A, Kuikka S (2010) Dynamic mapping of nature values to support
prioritization of coastal oil combating. Environ Model Softw 25:248–257
Kurennoy D, Parnell KE, Soomere T (2011) Fast-ferry generated waves in South-West Tallinn Bay.
J Coast Res 64(Special Issue):165–169
Lehmann A (1995) A three-dimensional baroclinic eddy-resolving model of the Baltic Sea. Tel-
lus A 47:1013–1031
Lekien F, Coulliette C, Marsden J (2003) Lagrangian structures in very high-frequency radar data
and optimal pollution timing. In: Experimental chaos. AIP conference proceedings, vol 676,
pp 162–168
378 pp
Lessin G, Ossipova V, Lips I, Raudsepp U (2009) Identification of the coastal zone of the cen-
tral and eastern Gulf of Finland by numerical modeling, measurements, and remote sensing of
chlorophyll a. Hydrobiologia 692:187–198
Lidskog R, Elander I (2012) Sweden and the Baltic Sea pipeline: between ecology and economy.
Mar Policy 36:333–338
Lilover M-J, Pavelson J, Kõuts T (2011) Wind forced currents over the shallow Naissaar Bank in
the Gulf of Finland. Boreal Environ Res 16(Suppl A):164–174
Lindow H (1997) Experimentelle Simulationen windangeregter dynamischer Muster in hochau-
flösenden numerischen Modellen. Meereswissenschaftliche Berichte No 22, Institut für Ost-
Liu S-K, Leendertse J (1978) Multidimensional numerical modelling of estuaries and coastal seas.
Adv Hydrosci 11:95–164
Matthäus W, Lass HU (1995) The recent salt inflow into the Baltic Sea. J Phys Oceanogr 25:280–
286
Millero F, Kremling I (1976) The densities of the Baltic Sea deep waters. Deep-Sea Res 23:611–
622
Myrberg K, Andrejev O, Lehmann A (2010a) Dynamic features of successive upwelling events in
the Baltic Sea—a numerical case study. Oceanologia 52:77–99
Söderkvist J, Sokolov A, Kuosa H, Anderson TR, Lehmann A, Skogen MD (2010b) Validation
BFU Res Bull 4(5):89–93
Ollitrault M, Gabillet C, Colin de Verdiere A (2005) Open ocean regimes of relative dispersion.
J Fluid Mech 533:381–407
Osiński R, Piechura J (2009) Latest findings about circulation of upper layer in the Baltic Proper.
In: BSSC 2009 abstract book, Tallinn, August 17–21, 2009, p 103
Parnell KE, Delpeche N, Didenkulova I, Dolphin T, Erm A, Kask A, Kelpšaitė L, Kurennoy D,
Quak E, Räämet A, Soomere T, Terentjeva A, Torsvik T, Zaitseva-Pärnaste I (2008) Far-field
vessel wakes in Tallinn Bay. Est J Eng 14:273–302
Rice J (2003) Environmental health indicators. Ocean Coast Manag 46:235–259
Richardson LF (1926) Atmospheric diffusion shown on a distance-neighbour graph. Proc R Soc A
110:709–737
Nikulin G, Wyser K (2011) The Rossby Centre Regional Climate model RCA3: model descrip-
tion and performance. Tellus A 63:4–23
2nd edition. In: Baltic Sea Science Congress, Stockholm, 25–29 November 2001. Poster #147.
Sokolov A, Andrejev O, Wulff F, Rodriguez Medina M (1997) The data asssimilation system for
data analysis in the Baltic Sea. System Ecology Contributions No 3. Stockholm University,
Sweden, 66 pp
366 T. Soomere
Soomere T, Andrejev O, Sokolov A, Myrberg K (2011a) The use of Lagrangian trajectories for
identification the environmentally safe fairway. Mar Pollut Bull 62:1410–1420
Soomere T, Andrejev O, Sokolov A, Quak E (2011b) Management of coastal pollution by means
of smart placement of human activities. J Coast Res 64(Special Issue):951–955
ways in elongated basins using Lagrangian trajectories: a case study for the Gulf of Finland, the
Baltic Sea. Ocean Dyn 61:1669–1680
Soomere T, Delpeche N, Viikmäe B, Quak E, Meier HEM, Döös K (2011d) Patterns of current-
induced transport in the surface layer of the gulf of Finland. Boreal Environ Res 16(Suppl
A):49–63
Soomere T, Viidebaum M, Kalda J (2011e) On dispersion properties of surface motions in the Gulf
Viikmäe B, Soomere T (2013) Spatial pattern of hits to the nearshore from a major marine highway
in the Gulf of Finland. J Mar Syst
Viikmäe B, Soomere T, Parnell KE, Delpeche N (2011) Spatial planning of shipping and offshore
activities in the Baltic Sea using Lagrangian trajectories. J Coast Res 64(Special Issue):956–960
WWF (2010) Future trends in the Baltic Sea. WWF Baltic Ecoregion Programme. Solna, Sweden,
40 pp
Zhurbas V, Laanemets J, Vahtera E (2008) Modeling of the mesoscale structure of coupled up-
welling/downwelling events and the related input of nutrients to the upper mixed layer in the
Gulf of Finland, Baltic Sea. J Geophys Res—Oceans 113:C05004
Chapter 11
Applications of an Oil Drift and Fate Model
for Fairway Design
Jens Murawski and Jacob Woge Nielsen
Abstract Pollutants at sea undergo weathering related changes in the composition

and are transported by winds, waves and currents. Methods are needed to study the
impact of changing weather and ocean conditions on oil chemistry and physics and
to estimate all the driving forces of oil transport at sea. The present study highlights
the potential of operational oil drift and fate models for the analysis of systematic
drift patterns that has been prevented so far by their numerical complexity. A large
number of drift simulations are performed to cover a three years period 1992–1994,
with 8 releases per month on a grid of 528 release positions in the Gulf of Fin-
land. Although sparse in resolution, the results are still characteristic. Large-scale
wind induced drift has been found to dominate the surface transport during windy
seasons, and to be of similar magnitude as the current drift during calm seasons.
Currents dominate the subsurface transport regime, and become more and more sig-
nificant at later stages of the spill development, when more and more oil droplets
are mixed down by waves. To estimate the combined effects of oil transport and oil
weathering, seasonal drift patterns are analysed and alternative fairway designs are
derived using a Monte Carlo technique. The quality of the designs is evaluated under
environmental and economical viewpoints using professional ship routing software.
We demonstrate that it is possible to find balanced solutions that are economically
beneficial and that follow green goals like low fuel consumption and reduced risk
for coastal oil pollution. Wave-driven advection of oil spills has been studied for the
one-year period of 1992, to cover a seasonal cycle. It has been found to impact oil
propagation in areas where the wave field exhibits strong gradients, e.g., at locations
with depth gradients and in the wind shadow of islands.
J. Murawski (B) · J. Woge Nielsen

Danish Meteorological Institute, Lyngbyvej 100, 2100 Copenhagen E, Denmark
e-mail: jmu@dmi.dk
J. Woge Nielsen
e-mail: jw@dmi.dk

DOI 10.1007/978-3-319-00440-2_11,
368 J. Murawski and J. Woge Nielsen
11.1 Introduction
The ever-increasing density of maritime traffic and especially tanker traffic con-
tributes strongly to the possibility of shipping accidents with the consequence of
near-coastal oil spills. As described in previous chapters, model-based investigations
of various risk factors have been used successfully to minimize the consequences of
shipping disasters and to increase the efficiency of fairway designs. Economical
and environmental factors are thereby fundamentally non-conflicting. In the process
of making shipping routes safer and environmentally friendlier by adaptive design,
they often become more beneficial in terms of faster travels (e.g., going with the
currents instead of against them, Lo and McCord 1995; McCord et al. 1999) and
lower fuel consumption (Lo and McCord 1998; Christiansen et al. 2004). Environ-
mental and economical considerations are, however, rarely entering the planning
stage of fairway designs (Stokstad 2009) although model-developed design crite-
ria can obviously contribute to optimized sea traffic, e.g., by effective route plan-
ning, emission reduction, improved disaster management or increased performance
of transport vessels.
Disaster management in the marine environment is largely understood as a re-
sponse action to accidents. In particular, the partial goal to manage maritime pol-
lution still is to develop remedial action plans (e.g., Keramitsoglou et al. 2003).
There is, however, an increasing tendency towards planning and developing various
preventive measures (Delpeche-Ellmann and Soomere 2013). These efforts aim at
bridging the gap between the economical and environmental factors by means of in-
cluding potential environmental losses into the risk equation (see Chap. 1). The final
goal is to build a combined platform of coupled forecasting models for the atmo-
sphere, waves and ocean circulation, which feed the relevant information concerning
the drift and fate of pollutants at sea into the disaster impact model, and develop a
suitable quantification of the relevant risks and costs for ship routing services and
fairway performance analysis.
The efforts in this direction described in the previous chapters were targeted at
the evaluation of fairway and sailing line design under safety, risk, disaster impact
and ship performance viewpoints combined with certain environmental criteria. The
analysis was based on the idea of highlighting the potential of semi-persistent (but
still statistically significant) current patterns and current-driven oil drift characteris-
tics in the risk for coastal areas. In regions where the oil propagation is heteroge-
neous and locally varying currents play a decisive role in the oil drift, it is possible
to exploit the drift statistics for the planning of the optimal sailing line or for safer
fairway design. In this context favourable ship routes feature a statistically lower
possibility for certain cost-intensive consequences. The environmental potential of
adaptive fairway design was addressed in terms of the reduction of risk of coastal
hits by the oil pollution (to be more precise, in terms of the drift of oil spills into
near-coastal regions). The key idea was to minimize the cost of the consequences,
i.e., the impact of accidents that may occur at sea.
The developed technology for fairway design makes use of well-defined local
minima or maxima of certain risk factors (such as the probability of hitting vul-
nerable areas or the time it takes for the pollution to hit such areas) related to the
11 Applications of an Oil Drift and Fate Model for Fairway Design 369
unfavourable pathways of the oil drift (Soomere et al. 2011a; Höglund and Meier
2012). The idea was to design environmentally safe fairways so that semi-persistent
current patterns crossing the fairway would transport the potential oil spill mainly to
offshore areas and in this way effectively increase the oil combating time (Andrejev
et al. 2011). Based on the resulting quantification of the offshore regions, various
methods (such as Monte-Carlo-type optimization, Dijkstra’s algorithm, or simply
following the local maxima or minima of the resulting ‘cost function’) (Soomere
et al. 2011b; Höglund and Meier 2012) can be used to minimize certain estimates
of the cumulated risks along the routes. The typically used measures, with a clear
physical interpretation, are the average residence time of potential adverse impacts
in the offshore (also called particle age, Andrejev et al. 2011; Soomere et al. 2011a)
or the probability of landing (or beaching) of these impacts (called also the proba-
bility for coastal hits in Chap. 10). A decent measure of risk should also take into
account the time spent by the vessel at sea (its increase owing to the increase in the
length of the sailing line represents a risk factor in itself) and economical factors as
well.
While the results presented in this book so far are mostly based on theoretical
considerations, the practical use of these emerging fairway designs is only feasi-
ble if they are carefully evaluated under economical viewpoints using a classical
ship routing–performance evaluation systems. The final decision about the choice
of the optimum sailing line should therefore account for both the environmental risk
estimates and economical factors. The combined evaluation should thus be based
on a joint measure that incorporates the increased oil residence time at sea (or the
reduced probability of coastal oil landings) along the environmentally optimum fair-
ways, and evaluates this gain against faster sea travel, reduced amount of fuel con-
sumption, and possibly increased other risks (of groundings and collisions).
The analysis presented in the previous chapters provides the user with a mani-
fold of comparable technologies and possibilities to evaluate the data and fairway
designs under different viewpoints. The analysis involved a multitude of pollutant
transport and fate simulations with the goal of statistical evaluation of the charac-
teristic drift patterns. A cluster of fairway design studies (Soomere et al. 2011a,
2011b, 2011c; Lu et al. 2012) were accompanied by high-resolution Lagrangian
surface drifter experiments (Kjelsson and Döös 2012).
A major limitation of the above studies is their focus on the current-driven oil
drift. This viewpoint, although suitable to highlight the key ideas of the technology
of fairway design based on approximately solving the inverse problem of pollution
propagation, has largely ignored other factors affecting the drift and fate of an oil
spill. The pollutant transport has been studied by means of neutrally buoyant (pas-
sive) particle drift assessments, where each particle represented a drifting volume
of persistent contaminant and was passively advected. In other words, the pathways
and fate of realistic oil spills were replaced by a cluster of trajectories of small ide-
alized persistent oil drops that were locked in the uppermost layer. Equivalently,
realistic oil spills were substituted by clouds of water parcels (more generally, pas-
sive tracers). This means inter alia that all changes to the oil itself as well as wind
and wave effects were completely neglected and exclusively current-driven transport
has been accounted for in the solution. Although current-induced pollutant transport
normally dominates its subsurface propagation, wind-induced transport is at least of
the same order of magnitude at the surface (Chap. 4) and wave impact definitely
plays a role in the transport at the sea surface (Ardhuin et al. 2009).
In this chapter we shall expand this analysis towards accounting for these basic
factors that affect the oil spills in the real marine environment and in many occasions
govern the oil drift characteristics. Namely, we shall employ an operational oil drift
and fate model (from now on referred to as the drift model) for approximately solv-
ing the inverse problem of oil spill propagation. The model in use has been proven
to be largely reliable, given the local topography is well resolved and an adequate
forcing is applied. The general idea is, however, the same as in Chap. 10: to iden-
tify offshore areas, from where the propagation of an oil spill to vulnerable areas is
unlikely or takes a long time.
This approach to the fairway design is not only quite new but also computation-
ally demanding as it requires numerous simulations of a highly complicated model
that handles not only the oil drift (as the above research has been doing) but also
the complex physics and chemistry of oil at sea. To our knowledge no similar at-
tempts have been undertaken before. The option of using realistic oil drift and fate
simulations to estimate the effects of oil weathering, winds and waves in combina-
tion with the technology developed in the previous chapters provides a synthesis of
the different processes that determine the fate of oil at sea, and makes it also possi-
ble to evaluate their relative importance for accurate forecasts and low-risk fairway
designs.
The chapter starts with an introduction into the characteristic processes that de-
termine the drift (Sect. 11.2.2) and fate of oil at sea (Sect. 11.2.3). These sections
introduce the basic concepts of operational oil spill models and sketch the basic the-
oretical ideas of how to describe the behaviour of oil in sea water. The reader is next
provided with an overview of the implementation of the oil drift model (Sect. 11.3),
with focus on the particular model used in this study and methods of calculations of
drift trajectories and a subsequent overview of the setup of the model for the target
area, the Gulf of Finland (Sect. 11.3.2). This basin is chosen as study area because
its highly variable hydrodynamic regime is partially constrained by Baltic Sea in-
flows and also strongly impacted by the riverine input. The estimates of the general
features of oil propagation in this water body (Sect. 11.4) are based on a sequence
of simulations in which a grid of potential oil spill sites has been stretched across
the gulf. The simulations are repeated many times with the same pattern of spills in
order to estimate characteristic drift patterns from a large number of single trajecto-
ries of oil parcels. The calculated trajectories, pathways and characteristic oil drift
patterns serve as the basis for the further analysis. Methods have been devised to
develop optimum fairways, similar to those described above and in Andrejev et al.
(2011).
Following the concept developed in the previous chapters, we identify and eval-
uate various optimum fairway designs that minimize the exposure of coastal areas
to pollution from specific sites along the fairway. The gridded risk maps are used to
identify environmentally safe pathways through the gulf. Optimum fairways mini-
mize aggregated risks along the routes and also account for the potential increase
in the sailing time (that represents a risk factor in itself and an economical factor
as well). The result is a design and evaluation of optimum fairways under envi-
ronmental and economical viewpoints (Sect. 11.5). The remainder of the chapter
is devoted to the analysis of surface wave impacts on the current-driven oil propa-
gation (Sect. 11.6). Along with the use of an advanced oil drift model, accounting
for the impact of surface waves is another key development here as waves have not
been taken into account during the fairway design process described in the above
chapters. To complete the picture, wave related changes in the regional and local oil
drift patterns are studied in some detail.
11.2 Oil at Sea
Oil is a toxic chemical substance that—when handled in large quantities at sea—

puts the marine and coastal environment at extreme risk of being rendered severely
polluted or inhabitable for a considerable span of time. Oil may be released into the
environment from drilling (production) platforms, from pipelines, or from vessels.
An oil spill is usually the consequence of some sort of accident. It may take place
abruptly, or during an extended time span, potentially of the order of months.
11.2.1 Oil Spills
Although the largest single oil spills released to the marine environment stem from
offshore oil platforms and spills from underwater pipelines also form a notable part
of the oil releases (Burgherr 2007), definitely the largest number of oil spills in the
marine environment result from transporting oil at sea. The size of spills originating
from a vessel varies several orders of magnitude (Dalton and Jin 2010; Jerneloev
2010), depending mainly on whether oil is the cargo, or the vessel’s driving agency.
In the first case, several 100,000 tonnes of crude oil may be lost to the environment,
in the second case, of the order of 1,000–10,000 tonnes of refined (fuel) oil may
be released. Since a vessel contains a limited amount of oil, spills from vessels are
often, but not always, of relatively short duration, and can sometimes be regarded
as instantaneous. The damage to the environment, however, may last for decades
or even be permanent (Monson et al. 2011). Potentially, an oil spill may wipe out
entire species.
Until now, the Baltic Sea has been spared of major sized oil spills, which could be
disastrous in this almost enclosed marine environment. The Gulf of Finland has been
particularly lucky, with only two sizeable accidents or misbehaviours within the last
decade (Soomere and Quak 2007; Wang et al. 2008). The largest spill so far to
take place was the Globe Asimi accident, outside Klaipėda, Lithuania in 1981, when
16,000 tonnes of oil were lost to the environment (Midboe and Persson 2004). More
Fig. 11.1 Fate of an oil spill at sea. The image is kindly provided by the International Tanker
Owners Pollution Federation Ltd. http://www.itopf.com/fate.html
recent spills include the Baltic Carrier (Christensen 2002) and Fu Shan Hai colli-
sions (2001 and 2003, respectively). On both occasions, a vessel with her 1,000–
3,000 tonnes of fuel oil was lost due to improper navigation (Pålsson 2012). The
case of the MV Prestige in 2002 was a near miss: she started out from Saint Pe-
tersburg, Russia (Delpeche-Ellmann and Soomere 2013) and on reaching Galicia,
Spain, she leaked 63,000 tons of heavy fuel oil.
The chemical composition of the oil that is released at sea is also often unknown,
apart from the basic distinction between fuel and crude oil. Oil is composed of a
number of hydrocarbon compounds, ranging from light, volatile kerosene, to heavy,
stable tar. The exact composition constitutes an oil type. The possible number of
crude oil types is almost infinite (Børresen 1993). In oil drift models hydrocarbon
compounds are grouped into fractions, defined by chemical structure and molecu-
lar weight. The relative weight of each fraction determines distillation properties,
density, saturation vapour pressure, and viscosity at a reference temperature.
The composition and the physical properties of the oil change over time due to
weathering processes at sea (Fig. 11.1) and also later on the beach (Christensen
2002). Volatile components (kerosene) quickly evaporate when exposed to air. On a
time scale of a few days, oil mixes with sea water under the action of surface waves
and forms a relatively stable emulsion. In such a mixture, the water content is two to
four times larger than the oil content, and the viscosity of such a mixture increases
by several orders of magnitude compared with pure oil (e.g., Daling et al. 2003).
The consequences of the described processes are crucial for the combating effort to
be successful as the resulting volumes of the oil or mixture, as well as changes in the
fluid properties (that affect the pumping) must be taken into account in planning the
cleaning operations. Only some of these processes are treated in operational oil drift
models: evaporation, water-in-oil emulsification, spreading and horizontal and ver-
tical dispersion. More elaborate methods include oxidation, dissolution, interaction
with suspended matter and others. See Dick and Soetje (1990), Børresen (1993),
Reed et al. (1999) for a comprehensive review.
Oil spills are most efficiently combated at sea, before any shoreline is polluted.
If weather conditions are favourable, barricades may be set up to contain the spill,
and oil is collected (skimmed and pumped) from the sea surface or combated chem-
ically (Fingas 2011). At the coast, conditions for combating are usually less perfect,
and a major, lengthy, labour-expensive and costly clean-up operation, with usually
doubtful results, is commonly at hand even if the amount of spilled oil is moderate
(Bi et al. 2011). In nearshore waters, prevailing weather conditions (first of all the
direction of wind and waves) and the distance from spill site to shore, decide which
scenario (open water or coast) unfolds.
For offshore accidents the third major player is the system of currents and the
fourth—the time. Minor spills disperse into the open waters, provided enough time
passes before the bulk of the spill reaches the nearshore. Even this process is not
without consequences. For example, in semi-enclosed seas like the Baltic Sea the
cumulative effects of small spills have been detected as a systematic increase in the
background hydrocarbon concentration (Pikkarainen and Lemponen 2005). Still at
a moderate level, this diffusive pollution is considered less harmful than soiling the
ecologically sensitive shoreline.
11.2.2 Oil Drift
Oil spill modelling has been an extremely valuable tool for the management of and
combating with oil pollution (Fingas 2011) as well as a key contributer to the rele-
vant decision-making process for about thirty years (Reed et al. 1999). A standard
oil drift model produces information on the oil’s location, slick size, spreading, com-
position, and drift either in a hindcast mode for past events or, more importantly,
forecasts these parameters and properties for some time in the nearest future based
on the available or forecast metocean information (e.g., Janeiro et al. 2012). It is
normally used as an operational stand-by tool, to be activated in case of an incident.
An emergency drift calculation is performed in real time, tracking a spill that has
taken place at a given point at a certain time instant, or within a limited span in time
and space, using available information on the actual and future weather and ocean
conditions. Preventive measures can then be taken at sea, weather, wave and ocean
current (met-ocean) conditions permitting.
An oil drift model estimates the evolution in time and space of a hypothetic
or a real oil spill. Given a set of initial conditions describing the spill itself (time
and location of release, magnitude, duration, oil type), and present and upcoming
metocean conditions, a suite of information about the oil spill (called spill state
vector) is updated using a time stepping algorithm and using certain equations and
relationships governing the oil spill evolution. The state vector contains all time-
dependent physical and chemical information required for the next step forward in
time to be evaluated similarly to the process of solving of any differential equation
described in Chap. 3. Based on it, the geographical position and extent of the spill,
the amount of shoreline or bottom contamination, toxic levels, slick layer thickness,
etc., may be derived.
In the forecast mode, an oil drift simulation is a valuable decision tool for spill
response: how, when and where to combat a spill and thereby reduce the environ-
mental impact. In scenario studies, drift simulations are a tool for risk analysis and
assessment.
The basis for an oil drift simulation is a simple displacement-with-time calcula-
tion
t1

s1 = s0 + u s(t), t dt, s(t0 ) = s 0 , (11.1)
t0
which maps the initial particle position s 0 at time t0 to a later position s(t1 ) at time
t1 similarly to the upgrade of the position of passive tracers or water parcels in cal-
culations of their Lagrangian trajectory in Chap. 7. Here u stands for the advection
velocity. Basically, this procedure applies to both objects (oil parcels) floating on
the sea surface and oil drops that are fully or partially submerged. In both occasions
the drift velocity is estimated at the object’s position. The same method is applied
to a chemical substance or pollutant: it is treated as a ‘particle’ that drifts passively
with the ocean current.
The major difference compared to the calculation of Eulerian or Lagrangian tra-
jectories of tracers in previous chapters is that a particle here only resembles a solid
object in the sense that it cannot be split. Importantly, it has additional inherent prop-
erties that change with time: size, shape and composition. Some of these properties
feed back on the advection process. Still, the position of each particle is updated
using standard time stepping algorithms. Differently from the analysis in Andrejev
et al. (2011), Soomere et al. (2011a), Lu et al. (2012) where only advection by cur-
rents was accounted for, in many advanced oil drift models a much more proper
representation of the advection velocity u is achieved. Since inaccuracies accumu-
late along the particle track, high-precision methods and/or high temporal resolution
(of the order of ten minutes) are crucial for trajectory calculations.
Each single trajectory is a function of the flow field, hydrography (understood
here as the properties of the sea water at site) and initial properties and conditions of
the spill. This means that all trajectories of oil particles that start from a small region
often experience quite limited spreading and tend to propagate in a close vicinity of
each other, similarly to the trajectories of idealized passive tracers calculated from
the current field only (see Chap. 9). This feature requires the use of some spread-
ing mechanism to simulate the impact of small-scale (subgrid-scale) turbulence on
trajectories similarly to various methods used to match the statistics of spreading
of idealized and realistic trajectories of drifters (Andrejev et al. 2010; Kjelsson and
Döös 2012).
Commonly, an ensemble of trajectories s n , n = 1, . . . , N , where N is the number
of particles, is calculated using some perturbation that is applied either to the forcing
fields, e.g., ocean currents (Andrejev et al. 2010) or properties of wind and waves,
or to the instantaneous site (as described in Chap. 7) or the release time of the spill.
The resulting trajectory ensemble is represented at any future instant t1 by a particle

‘cloud’ as typical in ensemble modelling (Vandenbulcke et al. 2009). Summing up
and averaging each particle’s state vector over the particle cloud gives information
about the overall spill state: total volume, mean thickness, average position and
composition at that point in time.
The atmospheric and ocean models only resolve motions, the horizontal exten-
sion of which exceeds their grid size, but water particles and oil parcels are also
impacted by small-scale (turbulent) processes that are not resolved by the mod-
elled basic forcing fields. There are many possible means to simulate the impact
of subgrid-scale processes on the statistics of calculated trajectories as described in
Chaps. 3, 7 or 10. Dick and Soetje (1990) introduced a pseudo-random, properly
scaled perturbation to the advection velocity to simulate the contribution of local
turbulence to the forcing fields. The perturbation is a vector sum of the impact of
the three-dimensional (3D) ocean current, wind drift (typically 3–3.5 % of the wind
speed at the surface, decreasing with depth), downward wave-induced mixing, and
a non-isotropic horizontal spreading velocity. The latter depends on the wind veloc-
ity and acts to elongate the particle cloud in the wind direction. A substitute of the
Brownian motion is also√added to replicate the slow horizontal spreading of a spilled
substance with time as t. This perturbation is active even in the total absence of
external forcing. The trajectories of oil parcels are three-dimensional and may reach
the sea bottom. When an oil particle sinks to the sea bed or reaches the shore, it
is assumed to settle and leave the water phase as in Höglund and Meier (2012). In
particular, all chemical processes come to a halt as the particle reaches the sea bed
or the shore.
For object tracking, the density of trajectories in such an ensemble may be in-
terpreted as a probability distribution of the occurrence of this object at a particu-
lar location or time instant. For distributed substances, this ensemble is commonly
interpreted as representing natural spreading (or even splitting up) of the spill as
particles separate from each other. Each single trajectory may still be regarded as
being deterministic, pertaining to a certain fraction of the spill.
The trajectory approach reduces the ensemble prediction method to a manage-
able-size problem in real-time applications, where the calculation speed is an issue.
The key simplification is that it is only necessary to insert perturbations into the
advection velocity along trajectories. Weather or ocean model ensembles, in which
the entire flow field is perturbed and updated, typically have 20–50 members. Tra-
jectory ensembles produced by drift models may be by 1–2 orders of magnitude
larger.
When the pollutant stems from a point or a line source, and within limited time,
the calculation of trajectories is much faster than the so-called grid point method
in which the spatio-temporal changes in the pollutant concentration are evaluated
basically for the entire basin. Apart from the replication of the general properties of
water masses (temperature, salinity), the grid point method is normally used to sim-
ulate spreading from more or less permanent point sources, e.g., river outlets which
supply the potential pollutants basically continuously, and these pollutants may in-
fluence the entire water body. The relevant calculations of advection and fate of, say,
nutrients or tracers, give complete concentration maps as a direct result. The trajec-
tory method, in which the computational effort is limited to the actually affected
region, requires a post-processing step to transfer the spill state vector (particle po-
sitions and properties) into the information about the concentration of the pollutant.
The study presented below is mostly limited to the trajectory analysis.
11.2.3 Oil Fate
For passive tracers, the combination of advection and spreading fully describes the
time evolution of a spill. The resulting information is, however, not sufficient for
an adequate reproduction of the time history of an oil spill. Oil is a chemically and
physically active substance. It interacts not only with sea water but also with air
and (suspended) sediment. On top of that, it reacts on sunlight, impacts the biolog-
ical environment, and, on a longer time scale, is impacted through biodegradation.
These interactions result in major transformations of the oil, leading to changes in
its composition and thereby in its chemical and physical properties (Fingas 2011).
To replicate the described processes to a certain extent, a superstructure is ap-
plied to the routines for trajectory calculations to simulate short-to-medium time
range chemical processes, weathering, horizontal spreading as a surface film, and
buoyant motion affecting the oil particles along their trajectories. These processes
exert a certain feedback on the advection, mainly through changes in the density
of oil particles, equivalently, their buoyancy (which is generally non-neutral any-
way) relative to sea water and their tendency to form a surface film. A part of an
oil spill is often ‘mixed down’ by the action of waves in rough seas (see below) and
is located at a certain depth in the water column. This also happens when the spill
occurs at a certain depth. The water column in the seas and oceans often possess
vertically sheared systems of currents (see, e.g., the description of the Ekman spiral
in Chap. 2, Sect. 2.3.5). In these cases subsurface oil moves at a different speed and
direction than oil at the surface. The submerged part of the spill typically lags behind
the surface part. This mechanism, which in turn depends on oil properties (density,
viscosity), ambient water properties and on certain metocean conditions, elongates
the spill in the flow direction and may even split it up into isolated patches.
The spreading of oil over the sea surface is an anisotropic process, with different
mechanisms responsible for horizontal and vertical spreading. It is particularly im-
portant for the early stages of an oil spill. This anisotropy makes it appropriate to
assign to an oil particle a disc-like shape, characterized by its coordinates, volume
and thickness. Such overlapping discs may be added up to determine the thickness
of the surface film.
Horizontal spreading is an advective–diffusive process that undergoes three
phases until it ultimately results in the terminal thickness—the formation of a
micron-thin surface film. The initial spill radius r first increases by buoyancy-inertia
forces. As the slick grows thinner, the spreading moves into a buoyancy–viscous
regime. The final stage is governed by viscosity and surface tension (Fay 1971).
The oil slick thickness decides which spreading phase is active. The last two phases
depend strongly on oil viscosity and thus on the oil composition and ambient tem-
perature. The ability of oil to form a quasi-stable emulsion with sea water, with
viscosity increase by several orders of magnitude (Børresen 1993), plays a key role
in these processes. One cubic metre of oil can be spread over 0.1 km2 , so that the
terminal slick thickness is just a few tens of microns, depending on the outcome of
the emulsification process.
Vertical spreading is governed by buoyancy forces and wave-induced mixing:
downward mixing occurs due to turbulence generated by surface waves in the up-
per layer whereas buoyancy forces may resurface the oil. The mixing depth mainly
depends on the wave height and wind speed. When mixed down, the surface film is
more or less evenly distributed over the mixing depth. This results in a reduction of
its toxic level (in units of oil volume per water volume) to some extent. For exam-
ple, if an oil slick of thickness 10 µm with zero water content (corresponding to a
volume of 10 m3 /km2 ) is mixed down to 10 m depth, the resulting concentration is
1,000 ppb (part per billion) total hydrocarbon.
Similar ensemble particle tracking methods, with a chemico-physical superstruc-
ture built upon a basic advection scheme, are used in the modelling of atmospheric
pollution and in simulations of three-dimensional spreading of, e.g., radioactive sub-
stances or volcanic ash (Sørensen 1998; Sørensen et al. 2007). This approach has
proved valuable in several incidents with large-scale impact and duration of weeks
to months such as Chernobyl in 1986 (Brandt et al. 1998, 2002) and the Fukushima
accidents in 2011 (e.g., Dietze and Kriest 2012, among many others), or the Eyjaf-
jallajökull eruption in 2010 (e.g., Webster et al. 2012).
In real-time applications, a large part of the information required to carry out an
oil spill simulation is often based on rough estimates. The amount of spilled oil or
the rate of its release are largely unknown. The ambient sea water density and tem-
perature at the accident site may vary in quite a large range but their characteristics
at the accident site and along the pathways of the spill are frequently estimated from
the relevant climatological values. Sea ice conditions may also not be known with
great precision and wave conditions are often parameterized using some simple rela-
tionships. Trajectories may still be calculated with a high degree of accuracy, but the
description of the chemical and physical evolution of the spill, which is important
for oil combating, will suffer as a consequence.
Oil spill modelling in ice covered water is not widely described in the literature
(Reed and Aamo 1994) although there exists a certain pool of experiments (Yapa
and Belaskas 1993) and observations (Wang et al. 2008). The wind drift is reduced
in the presence of sea ice depending on the ice concentration (Leppäranta 2010).
The process is discussed in Johansen et al. (2005), Drozdowski et al. (2011). Sea ice
influences oil drift in dense ice (ice concentration larger than 70 %) while there is no
influence in low ice concentrations (less than 30 %). For oil at the surface, the drift
velocity is a combination of current speed and ice drift velocity, with proper weights
assigned depending on the ice concentration. This is a rather crude approach, limit-
ing the effect of the ice to the drift velocity. weathering processes are significantly
slower in ice leads, but the spatial scale is of the order of meters, which is typically
not resolved by numerical ice models (Reed et al. 1999). This remains an unresolved
problem for high-latitude oil drift modelling.
11.3 Modelling of Circulation and Oil Spill

Oil drift modelling has been developed largely since the 1980s, and a number of
models are now on the market (Reed et al. 1999). Differences in forecasting quality
are mainly related to externalities, i.e., model set-up and forcing. Statistical results
of drift pattern analysis and ship routing design studies are expected to be resilient
to the choice of the oil drift and fate model.
In this study we use the operational oil dispersion model BSH dmod (Dick and
Soetje 1990). Both the hindcast and forecast of the oil drift substantially rely on the
information about various hydrophysical fields that are usually calculated using a
‘classical’ ocean circulation model. The Danish Meteorological Institute (DMI) uses
for this purpose an operational three-dimensional, free-surface, baroclinic ocean cir-
culation and sea ice model that solves the primitive (Navier–Stokes) equations for
horizontal momentum and mass, and budget equations for salinity and heat on a
spherical grid that co-moves with the Earth’s rotation.
11.3.1 DMI’s Ocean Circulation and Oil Spill Modelling System
The DMI’s circulation model used to produce the necessary information for oil
spill tracking was originally developed by the Federal Maritime and Hydrographic
Agency (BSH) in Hamburg (called BSH cmod then), and has been applied to the
North Sea and Baltic Sea since the early 1990s (Kleine 1994; Dick et al. 2001).
Since the initial implementation at the DMI, the model (called DMI/BSHcmod) has
been continuously developed further (Larsen et al. 2007; Liu et al. 2009; She et al.
2007). The model code has been completely rewritten and redesigned conceptually,
which has led to an increase in numerical efficiency, in terms of correctness, porta-
bility and speed. Among the added components are, e.g., the implementation of a
k–ω turbulence scheme, and the development of a new mixing scheme (Berg 2012).
The development follows Fortran 95 standards and strict code styling rules (Berg
and Weismann Poulsen 2012; Weismann Poulsen and Berg 2012). The model be-
came the official Baltic Sea model in the MyOcean framework and is called HBM
(HIROMB–BOOS Model)1 since then.
The HBM model shares many features with the Rossby Centre Ocean Model
(RCO) that is described in detail in Chap. 4. It reproduces the hydrodynamic pro-
cesses on spatial scales from local scales of some hundred meters to regional scales
1 HIROMB is an abbreviation for High Resolution Oceanographic Model for the Baltic. As the
name suggests, it is a circulation model with high resolution intended to be used for the Baltic Sea
(e.g., Funkquist 2001). BOOS stands for the Baltic Operational Oceanographic System.
of several kilometers. The vertical transport assumes hydrostatic balance and in-
compressibility of sea water. Horizontal transport is modelled using the Boussi-
nesq approximation, where density differences are neglected in all but gravity terms.
Higher order contributions to the dynamics are parameterized following Smagorin-
sky (1963) in the horizontal direction, and using a k–ω turbulence closure, which
has been extended for buoyancy-affected geophysical flows (Umlauf et al. 2003) in
the vertical direction. The turbulence model includes a parameterization of breaking
surface and internal waves. Stability functions from Canuto et al. (2001, 2002) for
the vertical eddy diffusivities of salinity, temperature and momentum have been ap-
plied. The numerical implementation uses a staggered Arakawa C-grid and z-level
coordinates, a flux-corrected horizontal advection scheme and free-slip conditions
along the coastlines.
Boundary conditions at the sea surface and sea bed use the quadratic dependence
of the wind induced shear stress and frictional bottom stress on the wind speed
and current speed, respectively. Following Andrejev et al. (2004) the wind drag
coefficient was set to CD = (0.63 + 0.066U10 ) × 10−3 , where U10 is the wind speed
at a standard 10 m height. Unrealistic strong wind forcing during storms is avoided
by making the wind stress parameterization dependent on the effective wind velocity
in relation to the current velocity. At the sea bed, a constant friction coefficient R =
0.0021 has been applied. The model is two-way coupled with a Hibler-type sea ice
model (Hibler 1979) that handles both ice dynamics and thermodynamics (Kleine
1994; Dick et al. 2001). To account for the penetration of short-wave radiation into
the subsurface layers a proper parameterization of the insolation suited for the Baltic
Sea (Meier 2001) is implemented.
The DMI’s operational model for oil drift and fate simulation has been origi-
nally developed by the BSH (Dick and Soetje 1990) for the German Bight. The
model was successfully applied to the North Sea and Baltic Sea by the DMI and
BSH, and to West Greenland waters by the DMI. This model has been developed
for tide-dominated seas that occasionally host very rough wave conditions such
as the North Sea, with shallow near coastal waters and limited impact of sea ice
on oil drift and fate. The model has still been successful in the Baltic Sea condi-
tions. The presence of ice cover in a part of this basin during a substantial frac-
tion of the year is accounted for through dynamics of the sea that depends on sea
ice.
The model is able to deal with ship driving fuel oils and a variety of natural crude
oils. The classes of crude oils include extra heavy and heavy crude oils (IFO 450,
Crude Venezuela) with high solidification points and low evaporation, heavy fuel
oils (Bunker C), conventional crude oils (Crude Statfjord, Crude Ekofisk, Crude
Nigeria), light crude oils (Crude No. 2) and light volatile petroleum products with
high evaporation (such as kerosene). Each oil class is composed of 8 fractions with
different boiling points and molecular weight. For the Gulf of Finland simulations
conventional crude oils (Crude Statfjord) were selected.
As discussed above, oil in sea water undergoes several transformations. The
model considers atmospheric interaction processes (evaporation), wave-driven
ocean surface processes (water-in-oil emulsification), horizontal dispersion and
spreading of oil, and subsurface processes (vertical dispersion). Evaporation is

treated fraction-wise as a function of mass, molecular weight, vapour pressure and
wind velocity according to Audunson (1980), Dippner (1984). More volatile frac-
tions eventually evaporate completely within a relatively short time interval. The
loss in the oil mass is compensated by entrainment of sea water into water-in-oil-
emulsion. The intensity of this process depends on wind speed and is modelled as
a saturation process leading eventually to a maximum uptake of ∼70 % of water
(Mackay et al. 1982, 1983).
Oil slicks in sea water undergo advection, spreading and dispersion due to cur-
rents and winds. First of all, oil parcels are carried by ocean currents (current ve-
locity u) and also impacted by the wind. These (advection) processes are modelled
using an Eulerian time stepping scheme ∂s = u · ∂t. The direct impact of wind is
accounted for by means of adding 3.0 % of the wind velocity in the direction of
the wind at the surface to the surface current (Reed and Aamo 1994). Spreading is
modelled using modified Fay equations (Fay 1969, 1971) where spreading in the
wind direction is enhanced by setting the spill radius in this direction 1 + 0.5U10
times larger than the radius in the cross-wind direction.
Horizontal dispersion and turbulence are treated as a random subgrid process by
adding to u a random component Ru and a component equivalent to the Brown-
ian motion RB . Vertical dispersion is modelled as an interplay of upwards buoyant
motion according to Stokes’ law and wave-induced downwards mixing, treated as a
turbulent diffusive process with wind- and wave-dependent eddy diffusivity (Sver-
drup et al. 1942; Leibovich 1975). Turbulence effects are included as random droplet
size distributions of buoyant oil parcels (Johansen 1982; Forrester 1971) and random
contributions to the eddy diffusivity. Submerged oil parcels are staying submerged
in most cases and usually strand at the sea bottom. These parcels are removed from
the calculations and do not enter the drift analysis below and only those near-surface
oil parcels that reach the shoreline are accounted for.
11.3.2 Oil Spill Modelling in the Gulf of Finland
Circulation simulations for the subsequent analysis of oil drift in the Baltic Sea
employed DMI’s operational circulation model for storm surge warning and ocean
forecasts (the code is tagged as MyOcean V1). The applied set-up (Fig. 11.2) covers
the North Sea and Baltic Sea with a spatial resolution of 3 nautical miles (nm)
constructed using the 1 nm bathymetry of Seifert et al. (2001), and the transition
area in a higher resolution of about 0.5 nm. The target area of the Gulf of Finland
was also covered with a 0.5 nm grid that was two-way nested into the Baltic Sea
grid. Doing so made it possible to model 3D ocean currents with a high spatial
resolution for oil drift simulations in this area. The bathymetry of the Gulf of Finland
was constructed from the gridded depth information with a resolution of 0.25 nm
(Andrejev et al. 2010) and thus adequately replicates the actual small-scale features
of the seabed in this basin, except in the central deep area of the gulf. Owing to
Fig. 11.2 The computation domains of the 2D model NOAMOD, the HBM model covering the
North Sea and the Baltic Sea (NS/BS HBM), the transition area between the North Sea and Baltic
Sea covered by a high-resolution model, and the nested oil drift model computational area in the
Gulf of Finland
the substantial decoupling of the dynamics of the upper and the lower layers in the
Gulf of Finland (see Chap. 6 for a detailed overview) ignoring seabed details in
the deepest part of the gulf apparently does not strongly influence the drift in the
uppermost layer.
Earlier circulation studies in the Gulf of Finland (Andrejev et al. 2011) have
indicated that 0.5 nm (1 km) is an adequate resolution for the simulation of the
essential features of currents in this basin, of which the subsurface flow driven by
the River Neva outflow at depths from 3 m to 10 m is the most prominent one. To
account for this, the vertical resolution in the upper 10 m of the Gulf of Finland
model is set to 3 m at the surface and 1 m below. In the North Sea and the rest of the
Baltic Sea the thickness of the uppermost layer was set to 8 m and the subsurface
layers had a thickness of 2 m. Below the 10 m level the vertical resolution varied
gradually from 2 m in the upper 80 m to 25 m at the maximum model depth of the
Gulf of Finland (125 m). Below this depth the layer thickness increases to 50 m at
the maximum model depth of 550 m in the Norwegian trench.
The model was forced with Denmark’s Climate Centre (DKC) 50 years reanaly-
sis (1960–2010, Christensen et al. 2006) for the North Atlantic, North Sea and Baltic
Sea. The spatial and temporal resolution of the forcing are 0.11◦ and 1 hour, respec-
tively. Daily river discharge data from the hydrological model of the Swedish Mete-
orological and Hydrological Institute (SMHI) were used at river mouths (Bergström
1976, 1992). At open model boundaries between Scotland and Norway and in the
English Channel, tides composed of 17 constituents and pre-calculated surges of
DMI’s North Atlantic Model (NOAMOD) were applied. This 2D barotropic surge
model covers large parts of the north-eastern Atlantic with a resolution of 6 nm.
Salinity and temperature boundary conditions were extracted from the monthly cli-
matology. The 4.5-year-long run was initialized with operational archive data in
August 1990. The first 1.5 years constituted a spin-up period, to ensure that the den-
sity field was in balance with the velocity field. The following 3 years, 1992–1994,
were used for the oil drift and fate studies.
The oil drift and fate model has been developed for individual spill prediction,
not for systematic studies of metocean forced and topographically steered pollution
propagation. Therefore its advantages in terms of the enhanced physical description
of weathering and transport processes (e.g., the ability to adequately account for the
wind drift, wave-induced vertical mixing and horizontal dispersion) may become a
disadvantage if the goal is the production of a large pool of simulation results as the
total runtime increases almost linearily with the number of simulations. In order to
keep the computational efforts for the identification of characteristic drift patterns
at a reasonable level, we restricted the spatial resolution of the release sites of oil
spills to 528 locations in the Gulf of Finland and the covered timeframe to three
years (1992–1994).
To match the results with the outcome of purely surface-current-driven trajectory
simulations described in Chaps. 9 and 10 (Soomere et al. 2010, 2011c), the length
of each simulation was set to 10 days. The simulations were restarted every 4th day
with the same configuration of oil spills. This sums up to in total 152,064 simula-
tions of single oil spills. The total physical duration of the runs (about 5 minutes
each run at a single node) would have taken 528 days on a serial computer. The suc-
cess of the presented study was reached owing to the portability of the drift model
code in a supercomputer facility. All in all 22 runs on 2 computational nodes were
executed in parallel to reduce the physical runtime to about 1 month. This was the
first time when an operational drift model was employed to calculate the statistics
for a whole region, the Gulf of Finland.
11.4 Characteristic Drift Patterns
Oil propagation in the marine environment is often non-uniform and follows re-
gional and seasonal patterns that are enforced by the cumulative effects of the cur-
rents and winds. The analysis presented in the previous chapters and in Soomere
et al. (2011d) suggests that in the long-term mean, particles often follow represen-
tative pathways that determine when and where they, on average, arrive at the coast.
If all sections of the shoreline are considered to be of equal value (Andrejev et al.
2011; Soomere et al. 2011a), the classical question—where and when the oil hits—
might be reverted to where did the earliest (often the fastest) particle came from and
when it started its motion.
11.4.1 Calculation of Drift Patterns
An analysis of these questions in terms of purely current-driven motions of passive

particles has been presented in the previous chapters. It provides certain spatial dis-
tributions of the probability for different offshore areas to serve as starting points
of pollution that hits a coast within a certain time limit and the time it takes for
the pollution to reach the coast. The analogues of such spatial fields are called drift
patterns below. To construct these patterns, the area of interest is covered by a grid
of release positions of realistic oil spills. Similarly to the above, the resulting en-
sembles of drift of oil spills are analysed statistically. The ‘where’ is then referring
to a marker (a certain grid point) on a spatial map, and the ‘when’ determines the
residence time of oil at sea. In a similar way, the probability of oil landing is derived
from an ensemble of oil drift simulations.
The exposure of coastal stretches to the pollution released at a specific site de-
pends on the drift pattern, i.e., on the landing probability and residence time. The
anisotropy of oil propagation in many sea areas makes these factors a complex func-
tion of topography, metocean forcing and flow regime. In other words, they are not
just geometrical functions depending on the ‘downwind’ or ‘downwave’ distance to
the shoreline. Moreover, they apparently vary within the year and are definitely dif-
ferent for different seasons (Soomere et al. 2011d). Therefore, there is little hope to
identify certain dynamical, well-defined links between the oil release sites and po-
tentially affected coastal sections. A feasible solution is to use a statistical approach.
Similarly to efforts described in Chaps. 4, 9 and 10, a large number of simulations of
oil propagation should be performed and analysed to establish statistical links. The
results are statistically significant spatial distributions of certain measures charac-
terizing environmental risk factors, which reflect certain frequently occurring drift
patterns. The obvious next step towards practically useful results is the merging of
risk factors into one (risk) parameter φ (see Eq. (11.2) below) for the assessment of
environmentally safe fairways in Sect. 11.5.
Model based investigations of oil drift patterns involve two kinds of simulations:
high-resolution weather and ocean circulation model runs with a goal to evaluate the
forcing fields impacting oil drift and fate, and oil drift model runs for drift assess-
ments. All processes2 along the chain from the atmosphere to the ocean, and further
on to the drift and decay of oil at sea have to be well represented by the models that
are applied. The quality and resolution of the forcing data is crucial for an adequate
replication of the drift and fate of oil, especially in near coastal waters (Broström
et al. 2011).
To ensure the necessary accuracy of the drift predictions, high-quality operational
ocean model and atmosphere reanalysis results were selected for the simulations.
Special focus was put on the accurate modelling of the atmospheric and hydrody-
namic conditions in the Gulf of Finland. As discussed in Sect. 11.3, the operational
ocean circulation model HBM, normally used for the Baltic Sea in a resolution of
2 Currents and winds drive oil advection and affect oil chemistry and physics in sea water.
Fig. 11.3 Histogram of residence times of oil particles at sea at three positions: 60.0◦ N (brown),
59.75◦ N (ochre), 59.93◦ N (gray) of longitude 26.2◦ E. The results of 288 simulations (1000
particles), each 10 days long, for the years 1992–1994 are presented. The inset shows the spatial
distribution of oil release sites (blue dots)
3 nm, was extended to a resolution of 0.5 nm for the Gulf of Finland, the main test
area for the preventive risk mitigation technology described in this book. The oil
drift model (Sect. 11.3) was forced with winds extracted from the DKC 50 years
reanalysis (1960–2010) (Christensen et al. 2006) and with the results of circulation
model runs that were also forced with DKC reanalysis fields. The three year period
of 1992–1994 was selected for the study of oil drift patterns and the evaluation of
ship routing designs. The set-up focuses on the fate of oil release at offshore po-
sitions, particularly on the onshore oil propagation. The 528 offshore oil release
positions had a more dense coverage (3 × 22 km) at the northern side of the Gulf
of Finland where the drift patterns were expected to favour outflow conditions and
westward transport. At the southern side of the gulf a lower resolution of about
6 × 22 km was selected for the release sites of oil spills (Fig. 11.3).
The analysis of characteristic oil drift patterns is based on two parameters: (a) the
oil residence time at sea and (b) the probability for coastal oil contamination. Mod-
elled are continuous tracks of oil volumina, associated with passively advected par-
ticles. The landing probability (called the probability for coastal hits in Chap. 10) is
defined as the ratio of landed to released particles. As a statistical measure it is well
defined for an ensemble of oil spill simulations. Note that the basic assumption (to
focus on the analysis of particles rather than volumes) implies that the amount of
landing oil is overestimated because typically just a fraction of the initial amount of
oil in each particle will land. This approach is still reasonable in the case when each
accident is treated with equal significance.
The residence time of oil particles at sea is defined as the time between their re-
lease and landing (and thus is equivalent to the particle age in simulations described
in Chap. 10 and in Andrejev et al. 2011; Soomere et al. 2011c, among others). It is a
measure for each individual oil particle, equivalently, each particle has its own res-
idence time. This measure is highly variable (Viikmäe et al. 2010). As we consider
not just ensembles of particles, but multi-ensembles of 1000 particles per release,
it is necessary to estimate a characteristic residence time that adequately represents
all the simulations.3 We use an appropriate statistical measure derived from the his-
togram of the residence time—the distribution of the amount of landed oil particles
as a function of time (cf. Chap. 9 and Viikmäe et al. 2010).
The oil drift model tracks the position of each particle until it leaves the Gulf of
Finland or settles on the shore, or on the seabed. As mentioned above, oil parcels
ending up on the seabed are ignored and only all those particles that land at a depth
shallower than 0.5 m are counted. This excludes sea bed fixation and ensures that
only near coastal contamination is studied. The analysis of the residence time at sea
focuses on the identification of cluster points, i.e., times at which large numbers of
particles land. The relevant histograms have a clearly defined peak (Fig. 11.3). The
location of this peak is a natural choice for the characteristic residence time. For
each release position, the first major peak in the histogram was chosen to represent
the typical residence time of oil at sea.
11.4.2 Seasonal Drift Patterns
The Gulf of Finland provides a good example for drift model based investigations of
characteristic oil drift patterns. Situated in the north-eastern extension of the Baltic
Sea (Fig. 11.2), it is a semi-enclosed sub-basin with a strong salinity gradient that
is supported by relative saline water inflow from the deeper layers of the Baltic Sea
and by river discharge from the largest regional fresh water source, the River Neva
(see Chap. 6). Its general circulation is strongly wind driven, but density driven cur-
rents play a role as well (Andrejev et al. 2004). Wind driven inflow of more saline
water at the Estonian (southern) side and outflow of somewhat less saline water at
the Finnish (northern) side of the gulf produce an extremely complicated circula-
tion pattern that still contains several persistent features that might be used for the
design of optimum fairways (Soomere and Quak 2007). Seasonal variations of river
inflow, solar radiation and wind forcing make the stratification and current patterns
also highly variable in space and time. To derive climatologically valid fairway de-
signs that serve all seasons, large ensembles of realistic oil drift simulations have to
be analysed. Counting 8 drift simulations per month, for a period of 3 years and for
528 release positions, we use a total of 152,064 drift simulations. This pool of sim-
ulations allows also for the evaluation of the relative importance of the direct wind
impact and current-induced drift patterns, as the oil drift model include parametriza-
tions for both.
3 Ensemble averages are not appropriate, because some of the particles travel for a long time, espe-
cially while submerged and solely current driven. Ensemble minima are problematic as well. They
often represent nothing more than the drift geometry and the shortest distances to the coast.
Fig. 11.4 Monthly mean wind drift 0.03U10 (upper panel, blue), speed (upper panel, red) and
direction of the mean current (lower panel, magenta) and wind (lower panel, green) for the years
1992–1994
Oil drift patterns at the sea surface of the Gulf of Finland are subject to seasonal
variations and depend strongly on dominant meteorological conditions and river
runoff. Below the surface, wind drift is reduced and currents play the major role in
the oil advection. The result is a strong seasonal variation of the mean drift patterns
and drift velocities (Soomere et al. 2010) and, consequently, of the distribution of
oil landing positions at the coast. The buoyancy effects are often unable to overcome
wave-induced downwards mixing at later stages of the spill evolution when weath-
ering has removed the more volatile and dissolvable oil components. As a result,
submerged particles remain submerged in most of the cases. As only particles at the
surface (depth <0.5 m) are counted, submerged oil is removed from the analysis.
The predominant driving force for oil advection at offshore locations of the Gulf
of Finland varies: it is the direct wind impact during stormy seasons and a combina-
tion of the system of drift currents and the direct wind impact during calm seasons.
It could be argued that 3 % of the wind speed in the direction of the wind (Reed and
Aamo 1994) is too high a value for the direct wind-driven oil drift in semi-enclosed
regions like the Gulf of Finland where the wind-driven surface current speed is only
about 1.4 % of the wind speed (Chap. 6, Hela 1952). The value of 3 % was estimated
on the basis of drift experiments under off-shore conditions. It holds well for light
winds and non-breaking waves, conditions that are typical for the Gulf of Finland
during a large part of the year (Soomere et al. 2008). Moreover, this value represents
an average of several estimates found in the literature (0.8–5.8 % , Pahlke 1985) and
also matches well the values (3–3.5 %) that are typically used for the operational oil
drift prediction.
Two major patterns can be identified that characterize oil drift in the Gulf of Fin-
land. During two seasons—winter with mainly western and south-western winds,
Fig. 11.5 Spatial distribution of the probability of oil landing (colour scale). Solid lines indi-
cate streamlines of average 10 m winds, vectors reflect drift currents (sum of wind-induced drift
0.035U10 and ocean currents) for summer and winter months (upper panel) and transitional sea-
sons (spring-autumn, lower panel) in 1992–1994
and summer periods with low winds (Fig. 11.4)—longshore oil advection predom-
inates at the Finnish coast. It is driven by winds in the wintertime and apparently
mostly by surface currents in summer. In both cases the (wind- or current-driven)
drift is directed mostly parallel to the coastline towards the east or even to offshore.
As a consequence, the onshore oil advection at the Finnish coast in winter and in
summer is not likely and the probability for oil landing is relatively low (Fig. 11.5)
compared to this probability for the transitional seasons. The values of the probabil-
ity of oil landing for 10-day long simulations range from below 20 % to a maximum
slightly exceeding 40 % at the northernmost part of the Gulf of Finland, at the north-
eastern (Russian) coast and in a small section to the east of the Finnish seaport
Kotka. As the panels in Fig. 11.5 show spatially integrated values, the areas very
near to the coast might be underrepresented. This probability varies strongly across
the gulf. The probabilities of landing lower than 40 % predominate at the Finnish
coast and at the north-eastern (Russian) coast of the gulf. They are confronted with
generally higher than 50 % landing probabilities in sea domains closer to the Es-
tonian coast and to the south-eastern (Russian) coast of the gulf. A broad band of
Fig. 11.6 Spatial distribution of the average oil residence time at sea (colour scale). Solid lines
indicate streamlines of average 10 m winds, vectors reflect drift currents (sum of wind-induced
drift 0.035U10 and ocean currents) for summer and winter months (upper panel) and transitional
seasons (spring-autumn, lower panel) in 1992–1994
probabilities larger than 60 % extends along the southern coast of the gulf. In more
than half of the cases, a released oil spill is drifting ashore within a short time.
The average residence time (the time between the oil release at position X(t) and
the landing at the shore at time t + T ) (Fig. 11.6) reflects the time available for oil
combating. For release sites at the southern coast the average combating time before
a oil spill reaches the nearshore is less than 3 days. This time is much larger for the
more northern release sites. The residence time is generally larger than 3.5 days
and exceeds 5 days in a large domain that starts almost at the entrance of the gulf at
24.3◦ E and extends along the northern coast all the way towards 26.6◦ E. Therefore,
a shift of the release position northwards across the gulf would give an advantage
of more than 2 days in oil combating time. As described before, the gain is even
increased by a reduction of the coastal landing probability from values of more than
50 % to values below 30 %.
Another clearly distinguishable pattern becomes evident for the transitional sea-
sons: spring, late summer and autumn. The monthly mean wind directions change
from the west-south-west to south-south-west or even south (Fig. 11.4) in these
months. Consequently, the predominant wind drift intensifies the onshore oil ad-
vection in the northern region of the gulf. As a result, the spatial patterns of the
probability and residence time are reversed compared to their appearance in winter
and summer. Whereas the summer and winter patterns show relatively low values
(<30 %) for the probability of oil landing in the northern regions of the Gulf of
Finland, the corresponding spring and autumn patterns reveal maximum values of
more than 50 % in this domain (Fig. 11.5). A broad band of low probabilities for
coastal landing (<20 %) extends along the Estonian coast that is located upwind
from the sea in these seasons.
The spatial distributions of the average oil residence time mirror the analogous
distribution for the landing probabilities. In particular, the distribution of the res-
idence time for summer/winter and for the transitional seasons are almost exactly
reversed with respect to each other. Namely, oil in areas where if released in sum-
mer and winter, it spent a long times at sea, exhibits generally faster coastal landing
in autumn and spring (Fig. 11.6). Almost everywhere along the southern (Estonian)
coast of the gulf the released oil spills spend more than 4 days at sea before they hit
the coast whereas at the northern (Finnish) and north-eastern (Russian) coast the oil
resides less than 2 days at sea. Therefore, in winter and summer there will be, on
average, about 3 days more time than in the transitional seasons to combat the po-
tential oil pollution released near the Finnish coast. Correspondingly, about 3 extra
days will be available for oil combating in winter and summer off the Estonian coast.
The performed drift pattern analysis results in a set of distinct maps that are
suitable for fairway design similarly to the efforts described in Chap. 10. There is
always a dilemma of which map to choose as probably no unambiguous measure
of the environmental benefit of employing one of the two results can be defined.
Moreover, there are certain ambiguities in the patterns that arise from frequently
changing wind conditions. The two distinct and substantially different patterns for
the summer-winter (SW) and spring-autumn (SA) seasons seem to be quite stable
over the 3 years of the modelled period. From the 36 months in total, 18 months
feature the SW pattern and 15 months the SA pattern. No distinct pattern seems to
exist during the remaining three months. As none of these SW and SA drift patterns
seems to be dominant or significantly overlapping enough to form the background
for fairway design decisions, a smart decision would be to design different optimum
fairways for different seasons.
11.5 Fairway Design
Fairways are usually designed to comply with topographical conditions, traffic reg-
ulations, national and international agreements, and basic economical standards. In
the focus of the design process are the vessel’s operational conditions. Environmen-
tal factors (such as atmospheric conditions, wave properties or ocean circulation)
that might considerably influence the travel have so far been of minor importance.
The optimum fairways are usually as short as possible, contain as few straight lines
as possible and involve a minimum number of turns. The obvious reason behind
such a structure is that ship engines reach maximum efficiency under certain con-
stant conditions and that ship construction typically makes course changes fuel-
consuming.
In this section we continue the analysis of drift patterns from the viewpoint of
the design of optimum fairways and evaluation of the potential gain from their use.
Along with the continuing practice of path optimization in shipping operations with
respect to a steady ocean and atmosphere, we make an attempt to incorporate into the
fairway design the aggregated information about the transport of pollutants under
realistic metocean conditions. The basic fairway designs are derived by Monte Carlo
techniques accounting for both the increase in oil combating time (expressed via the
residence time of oil at sea) and the decrease in the probability of oil landing. Such
a design of the favourable fairway makes use of both the derived distributions and
especially of the location of low probabilities and long residence times.
The constraint of travel distance reduction is imposed by a path length optimiza-
tion routine. The efforts result in separate fairway designs for seasonal and com-
bined drift patterns. The ‘winning’ design is identified by using of a ship routing
analysis software and a comparison of the economical performance of various fair-
ways in terms of travel time and fuel consumption. The basic designs represent
indicators on which the specific design process could be based. Ship routing evalua-
tions might serve as a tool for further investigation. To the authors’ knowledge there
has been no attempt yet to explore the possibility of route evaluations during the
planning process of fairways. Neither environmental viewpoints, in terms of pollu-
tant impact reduction, nor economical viewpoints, in terms of fuel and travel time
reduction due to ocean circulation have been taken into consideration.
11.5.1 The Design Method
In its essence the problem of deriving optimum fairways from model results is an
inverse one. The available information about the oil residence time in open waters
and of the probability of coastal hits for individual runs is related to global distribu-
tions of risk parameters and cost functions φ(x, t) for the optimization procedures.
The output are pathways of low aggregated risk specified by path integrals of φ(x, t)
over certain curves that are identified as optimum fairways. While the cost functions
are specified by statistical ensembles of model results and derived risk measures, the
fairways not only follow the local risk minima (which may lead to extensive and un-
justified detours, see Soomere et al. 2011b; Lu et al. 2012) but also optimize the path
length to some extent. Longer ship tracks are not only fuel and cost intensive, but
also less preferable environmentally, because of, e.g., a larger amount of emissions.
Furthermore, longer pathways represent a larger risk per se, as the time spent at sea
and potential accident risks increase.
Optimum routes thus combine minimums of the path length and risk values inte-
grated over these paths. This sort of problem is typically addressed by Monte Carlo
techniques. Its straightforward use together with combined risk and path length
cost functions is problematic because of the different nature of the two parameters.
The process of risk minimization tends to connect local minima and to produce me-
andering tracks, whereas path length minimization tends to generate straight lines.
A reduction of the problem to one single meaningful optimum cost function that
combines the two characteristics is usually not possible. For that reason, a sequen-
tial Monte Carlo approach is used, where the first priority is to minimize the cost
function and the path length minimization is of secondary order. During such a pro-
cess the risk factors and the pathway length need not be weighted against each other
within one optimization cycle.
An aggregated cost function φ(X) is defined here at each release position as a
joint measure of the relative time spent at sea T (X)/TM and the probability for
coastal oil landings NL (X)/NT = nr=1 r
NLr /(nr · 1000):

f1 T (X) f2 NL (X)
φ(X) = + 1− . (11.2)
f 1 + f 2 TM f1 + f2 NT
Here TM = 10 days is the period that each drift simulation covers, nr is the number
of releases and NT = 1000 particles have been released at the beginning of each
simulation. The two ad hoc weighting factors f1 = 2 and f2 = 1 are introduced
to strengthen the relative significance of the oil combating time (expressed as the
offshore residence time) with respect to the probability of landing. The inclusion
of these factors enhances the weight of the possibility of successful oil combating
against the severeness of an oil spill as the probability factor refers directly to the
amount of oil that reaches the shoreline.
In a first step, the function φ(X) is spatially interpolated to φ(x) to increase the
resolution (x marks all model grid points) and random sampling of φ(x) along paths
p(x) is used to maximize the path integral of φ(x) along p(x). The paths p(x) are
initialized with lines connecting the maxima of φ(x) at the longitudes of the release
grid. The optimization criterion is the maximization of Ψ , the integrated local cost
function along paths s(x):

φ(s(x)) ds
Ψ= . (11.3)
ds
The optimization during the first phase focused on the maximization of the cost
function. The algorithm used regular random re-initializations to avoid local max-
ima of φ(x) and simulated annealing to gradually restrict the random modifications
and to freeze the pathways into their final configuration. This process often leads
to quite complicated fairways that are unsuitable for navigation as in Soomere et al.
(2011b), Lu et al. (2012). The second step aims at minimizing the length of the path-
way. To avoid straight lines as solutions, a few random locations along the pathways
were fixed in regions where the cost function φ(X) was at least 0.7. Monte Carlo
techniques use random variations that follow the spatial grid representation of the
circulation model x to find a global minimum of the path length. At each model
longitude the pathway is perturbed along the latitude so that, for example, going
eastwards a random shift occurs in a latitude environment of the western neighbour
point.
The resulting pathways thus represent a certain balance between attempts to max-
imize the integrated cost function and to minimize the path length. The resulting
optimum is generally relative rather than a global or absolute one. The gain (relative
performance) from the use of an optimum pathway can be evaluated in terms of the
ratio of the properly normalized integral of the cost function over its path and the
corresponding cost function for other pathways.
11.5.2 Environmentally Friendly Fairways
The four most important oil harbours in the eastern Gulf of Finland were selected for
the evaluation of the performance of the described technique: Saint Petersburg, Vy-
borg and Ust-Luga in Russia, and Sillamäe in Estonia. Given the extensive variabil-
ity of the patterns of the oil residence time at sea and of the probability for landing,
we chose to develop the Monte Carlo based fairway optimizations for three basic
situations. The oil drift patterns for SW and SA seasons (discussed in Sect. 11.4.2)
are first analysed separately. These patterns are also combined for a joint analysis.
Seasonal ensembles of the cost function φ are formed for each release position X.
Ensemble averages are studied for the SW and SA designs (Fig. 11.7). The third
pattern, the basis for the combined (CMB) fairway design, was constructed to cover
all seasons (Fig. 11.7).
The two seasonal patterns can be merged in many ways. A simple way was to
produce annual averages that cover the complete set of monthly drift patterns. The
result was rather homogeneous and lacked the heterogeneity of the seasonal pat-
terns and was thus unsuitable for the identification of distinct fairways. Instead, the
distribution of ensemble maxima was used for the development of CMB fairway
designs that can be used for all seasons. This distribution covers larger areas with
substantial values of the cost function as even single values at a certain instant may
contribute to this map. Namely, the distribution of ensemble maxima needs only
one value to highlight a particular release site whereas an ensemble average nor-
mally needs a much larger amount of high values to become significant. The results
in Fig. 11.7 confirm that the constructed annual pattern is a suitable basis for the
fairway optimization.
The results of the basic design evaluation in terms of the path length, path integral
of the cost function φ and their ratio (the normalized path integral of φ) (Table 11.1)
suggest that the SA design performs a little better than the SW design. Although the
gain of using a SA design is somewhat reduced by the longer path length, it is fully
compensated by much lower levels of environmental risks (expressed as the val-
ues of the integrated cost function). The performance of the SA design is, however,
rather low outside the spring and autumn seasons. A cross-seasonal evaluation re-
veals that the performance of the SW design is better balanced across the seasons.
The variability of the SA performance is larger, but the cross seasonal difference
is low. This means that the annual average performance of SW and SA designs are
nearly equal. The difference is generally less than ±0.01, except for Sillamäe, where
it is 0.06.
Fig. 11.7 Optimized summer-winter (SW), spring-autumn (SA) and all-seasons (CMB) fairways
to Vyborg, Saint Petersburg, Ust-Luga and Sillamäe corresponding to cost function φ maxima, i.e.,
the minima of the probability of coastal hit and the maxima of oil combating time
The combined design (CMB) performs, on average, better for all seasons than the
seasonal designs. As it represents the locations of the maxima of the cost function
of the entire drift simulation ensemble, it is expected to have better performance.
A more realistic evaluation (Table 11.2) cross-examines the fairway designs under
equal conditions by looking at the performance values of the designs for seasonal
and combined maximum patterns. The combined design has generally a better over-
Table 11.1 Cross-comparison of optimum fairways for summer-winter (SW), spring-autumn (SA)
and all seasons (CMB). Fairway
efficiency is quantified as the ratio of the path integral of the cost
function φ along a fairway φ ds and the length of the fairway ds, the maximum values of which
represent the most efficient fairways
Design Destination Fairway Summer-winter SW Spring-autumn SA
length, km Path integral, Quotient Path integral, Quotient

km φ ds/ ds km φ ds/ ds
SW(north) Saint 287.1 124.2 0.43 100.1 0.35

SW(south) Petersburg 288.1 122.2 0.42 101.6 0.35
SA/CMB 290.0/289.2 80.1/99.4 0.28/0.34 149.0/134.3 0.51/0.46
SW Vyborg 231.6 111.5 0.48 77.5 0.34

SA/CMB 243.7/242.3 72.2/87.2 0.30/0.36 124.9/110.7 0.51/0.45
SW Ust-Luga 251.1 110.6 0.44 89.2 0.36

SA/CMB 253.0/252.3 69.0/89.3 0.27/0.35 134.9/123.4 0.53/0.49
SW Sillamäe 224.0 101.3 0.45 96.7 0.43

SA/CMB 223.6/222.7 57.6/79.8 0.26/0.36 126.0/113.0 0.56/0.50
Table 11.2 Cross-comparison of fairway designs for summer-winter (SW), spring-autumn (SA)
and combined oil drift pattern (CMB) representing local maxima of the seasonal drift pattern en-
semble of the cost function

Design Destination φ ds/ ds φ ds/ ds φ ds/ ds Average
SW pattern SA pattern CMB pattern
SW(north) Saint 0.74 0.60 0.58 0.64

SW(south) Petersburg 0.72 0.69 0.64 0.68
SA CMB 0.53/0.74 0.87/0.85 0.69/0.78 0.70/0.79
SW Vyborg 0.80 0.54 0.55 0.63

SA/CMB 0.60/0.74 0.86/0.76 0.72/0.80 0.73/0.77
SW Ust-Luga 0.72 0.68 0.65 0.68

SA/CMB 0.54/0.73 0.89/0.89 0.71/0.80 0.71/0.81
SW Sillamäe 0.76 0.75 0.71 0.74

SA/CMB 0.50/0.75 0.89/0.90 0.72/0.82 0.70/0.82
all performance. For Sillamäe and Ust-Luga it performs equally good or even better
than the seasonal designs. This is surprising as the SW and SA designs are specifi-
cally based on drift patterns for the relevant seasons. But it should be remembered
that CMB designs were generated using a combined ensemble maximum pattern,
and therefore they are to some extend adapted to the seasonal maximum pattern as
well.
The choice between the SW/SA and the CMB designs is, in essence, the choice
between specialists and all-rounders. Ultimately, it is the task of decision-makers
to pick one or the other fairway, or to choose between an all-seasons solution or a
specialized solution for one season. We only note here that the use of the average
of drift patterns seems to be more critical and challenging for the design process
compared to the pattern of maxima. The latter seems to cover all the design options
and to make comparisons possible. It is not unexpected that combined all-season
designs perform best in the all-year-round comparison because they simply cover
better the seasonal features of the drift. The choice is still not obvious and probably
requires a clear design strategy. If a seasonal design should be picked for an all-
year-round application, the SW design is preferable.
11.5.3 Maritime Services for Performance Analysis
Marine transportation is lucrative, but is heavily loaded with demands for ever in-
creasing cargo quantities and shipping efficiency to support a very competitive mar-
ket. At the same time, ship accidents are among the most disastrous environmental
catastrophes, leaving whole coastlines polluted. What makes ship accidents so dev-
astating and spill combating so challenging is the difficulty to access the source of
a spill in due time, especially in case of heavy weather or subsurface releases. As
marine navigation will alway be risky, management tools like ship routing software
and well developed maritime services might help to reduce the risk of accidents and
the consequences of consecutive spills. As explained in Chap. 1, a proper measure
of risk is the product of the probability of an accident and the cost of its conse-
quences. In the context of ship routing usually only the first factor of this product is
addressed while in the present book the focus is on the possibilities to minimize the
consequences, should an accident happen somewhere. Combining these two view-
points may jointly address both disaster prevention and impact minimization and
thus together aim at a harmonization of shipping demands and safety requirements.
Maritime services are addressing this issue in two ways: weather routing is mak-
ing seaborne transport safer, and ship routing contributes to the efficiency of ship-
ping. Statistical analysis of weather-related ship casualties of 150,000 North At-
lantic and North Pacific crossings of 26,000 vessels in 1978–1982 indicates that
ship routing applications reduce the casualty rates by 15 %. The International Mar-
itime Organization (IMO) acknowledged this already in 1983 and recommended the
use of weather routing services to operators.
Weather routing and ship routing are two terms that are often used synonymously.
In the context of this chapter, weather routing is dealing with the safety aspects
of maritime traffic, i.e., it is guiding vessels through or around dangerous weather
situations, whereas ship routing is applied to reach direct economical benefit through
planning and use of high-performance routes and the analysis of performance of
existing routes.
Ship routing is a well established maritime service and widely used for ship per-
formance estimations, voyage analysis and route optimizations in relation to travel
time, fuel consumption and CO2 emissions. Its main objectives are to reduce the
operating cost and to increase the efficiency of maritime traffic. Ship routing links
the ship’s position with its destination so that the travel time and costs are optimized
whereas the journey still remains safe and efficient. The quality of route analysis
and optimization depends critically on the forecast range of high-quality predic-
tions. The choice of the best route might easily depend on the metocean conditions
10 days ahead. Winds, wind waves, swell, ocean currents and water depths are input
parameters for the performance calculations.
Vessels are monitored and in case of an alert, duty analysts develop course strate-
gies and recommend course changes to meet weather conditions effectively, for ex-
ample, to avoid heavy weather and critical ocean conditions like tropical cyclones,
deepening storms and freak waves. Each vessel is represented by an individual pro-
file of construction and performance parameters: engine type and efficiency, friction
drag by the ocean and wind, main speed and fuel consumption. The design speed
of transport vessels is defined by its construction, but the optimum speed may vary
within a certain range. This allows certain adaptation strategies to market prices (low
rates–slow steaming) and fuel expenses (Ronen 1982, 1986; Molinero and Mitsis
1984; Evans and Marlow 1990). The optimum speed, often referred to as the eco-
nomical speed, is most likely seen as the speed that maximizes profits, but route
optimization allows also to address green goals like emission reduction and con-
sumption limitation. Route scheduling allows to assume economic and fuel saving
speed whenever there is time on a crossing. Up to 8 % fuel savings are reported.
In this context it is of large interest to compare the ‘performance’ of environmen-
tally optimized fairways with similar evaluations resulting from commercial voyage
planning and weather routing. We shall compare the fairways designed above with
the ones derived using the software SeaPlanner, which is a cooperative development
of FORCE Technology and the Danish Meteorological Institute. As an onboard de-
cision support system it links the operator at the DMI with the navigator on board
and provides necessary information for weather guiding and voyage optimizations.
Calculated routes are combinations of rhumb lines and great circle routes. They have
been optimized according to the total fuel consumption, minimum enroute time,
motor speed (revolutions per minute, RPM) and vessel speed for a fixed arrival time
(ETA) or with respect to the calm water speed.
Operational route predictions depend on frequently updated metocean forecasts
(every 6 hours), state-of-the-art optimization tools and advanced and flexible propul-
sion and propulsion plant models. The software includes resistance calculations
based on speed, draught-dependent still water resistance and additional components
for waves, shallow water and winds. A semi-empirical model calculates resistance
on the basis of the vessel’s ability for seakeeping (calculated and model tested effi-
ciency to handle high sea states), shallow water friction (CFD calculated and model
tested drag parameterization) and the vessel’s wind friction drag (estimated from
a database of wind tunnel tests). Beside the resistance parameter, the propulsion
model includes a detailed description of propeller characteristics for fixed or con-
trollable pitch and engine configurations. The software is able to distinguish be-
tween single and multiple engine set-ups, diesel electric propulsion, the use of a
Waste Heat Recovery system, and includes many other options.
11.5.4 Economical Aspects of Fairway Design
The performance of shipping operations is measured in terms of the voyage speed

and fuel consumption. Evaluations of alternative fairway designs provide estimates
for the economical gain of applying seasonal or combined designs depending on a
particular ship and metocean conditions. To stay close to reality, a medium range
tanker Torm Ragnhild with a length of 183 m, a deadweight of 46,187 tons and
maximum and average speed of 15.2 and 13.8 knots, respectively, was selected to
evaluate the above-described fairways. Torm Ragnhild is a typical tanker for the
Baltic Sea that is suitable for sailing through the relatively shallow Danish Straits.
Its draught (12.247 m) is too deep for some of the shallow passages in the Gulf
of Finland and some of the fairways calculated above were slightly modified (the
cruises were moved offshore in the vicinity of some islands and peninsula).
Geography plays an important role for route evaluations in relatively small water
bodies with strong demands for coastal navigation. The Gulf of Finland is a char-
acteristic example with rather steep coasts in the south-west, shallower waters in
the east and an archipelago at the northern coast. As a rule of thumb, depths less
than 100 m affect the tankers’ resistance and have to be taken into consideration.
The ship propulsion model calculates stronger resistances for shallower water. The
ultimate limit for the water depth is the vessels draught. Cruises along the shallower
northern shores of the gulf are subject to stronger sea bed interactions and ship hull
resistances than the southern cruises. This aspect has not been considered during the
fairway design above but it most likely affects the economical performance.
The analysis presented here is based on the annual mean wind, wave and current
fields that were used to specify weather and sea conditions. Seasonal variations, al-
though of strong consequence for the maritime transportation were not taken into
consideration. The ship routing analysis is based on three years of mean wind, wave
and current data (1992–1994). The data has been used for pollution transport studies
before (Sects. 11.4 and 11.6), but it has been averaged over time for the economical
evaluations. Fairways of different designs are compared with each other according
to their departure/destination harbour. East and west going routes are analysed to-
gether. The focus lies on the fuel consumption and the average speed, which have
been studied under conditions of either fixed planning speed or fixed travel time.
The total consumption and speed are dependent on the preset calm water speed and
the metocean conditions. For Torm Ragnhild with an average velocity of 13.8 knots,
a calm water speed of 13 knots was selected. Under conditions of deep water, low
winds, insignificant waves and zero currents, the calm water speed is equal to the
velocity of the tanker. Under realistic conditions, the speed is often lower because
winds, waves and shallow water add to the resistance.
The economical gain associated with the relocation of existing fairways has been
estimated by cross-performance analysis. The existing routes and new designs have
been evaluated together, which allows performance assessment under equal condi-
tions (Table 11.3). The existing routes favour transport from the main harbour in the
Gulf of Finland region, Saint Petersburg. The economical benefit of employing alter-
native routes in terms of fuel consumption and travel time will therefore be minimal
Table 11.3 Comparison of fairways from the route atlas (RA), summer-winter (SW), spring-
autumn pattern (SA) and combined designs (CMB), and routes optimized using the SeaPlanner
software (SPO) on east to west crossings of the Gulf of Finland. The fairways to each destination
are ordered from the north to the south. Time is given in hours:minutes, speed in knots and fuel
consumption in metric tons (mt)
Design Harbour Westwards going Eastwards going
Time Speed Fuel Time Speed Fuel
SW(north) Saint 13:56 11.4 9.04 13:22 11.9 8.681

CMB Petersburg 13:34 11.8 8.81 12:41 12.6 8.239
SPO(north) 13:16 11.9 8.608 12:28 12.6 8.088
RA 13:32 11.9 8.784 12:43 12.6 8.26
SW(south) 13:36 11.8 8.827 12:48 12.5 8.314
SPO(south) 13:33 11.7 8.79 12:36 12.6 8.181
SA 14:00 11.7 9.092 13:00 12.6 8.444
SW Vyborg 11:50 11.3 7.681 11:24 11.7 7.406

SPO(north) 11:27 11.8 7.438 10:52 12.4 7.054
SPO(south) 11:45 11.8 7.635 11:08 12.7 7.225
RA 11:49 11.9 7.675 11:04 12.6 7.189
CMB 11:52 11.9 7.702 11:07 12.7 7.217
SA 12:14 11.9 7.945 11:21 12.8 7.373
RA Ust-Luga 14:04 11.9 9.132 13:22 12.6 8.674

SW 12:02 11.8 7.817 11:26 12.5 7.423
SPO(north) 11:57 11.8 7.76 11:09 12.6 7.239
CMB 12:05 11.8 7.848 11:23 12.6 7.395
SPO(south) 12:08 11.8 7.878 11:13 12.7 7.285
SA 12:32 11.8 8.136 11:35 12.6 7.515
RA Sillamäe 11:46 12.0 7.64 11:13 12.6 7.285

SW 11:14 12.0 7.293 10:35 12.7 6.872
CMB 11:07 12.0 7.215 10:28 12.7 6.801
SPO(north) 10:24 12.0 6.751 09:49 12.8 6.374
SPO(south) 10:40 11.9 6.931 10:01 12.8 6.498
SA 10:42 11.9 6.95 10:02 12.8 6.512
for this harbour. But having only one route to follow and many harbours to serve,
the existing network apparently fails to provide cost effective solutions for other
harbours in the Gulf of Finland such as Vyborg in the northern part of the gulf and
Ust-Luga and Sillamäe at the southern coast. The existing preset fairway system that
serves all destinations has the advantage of minimum maintenance costs, but also
the drawback of limited flexibility and low efficiency. Adaptive management ap-
proaches would favour networks of alternative routes that serve different harbours
similarly and would give operators the possibility to choose an optimum solution.
But these ideas seem to be out of the scope of currently realizable solutions.
The pool of fairways to Saint Petersburg (not shown) is unique in the sense that
their length is rather similar, although they cover nearly the entire width of the gulf.
Ship performance in terms of travel time, speed and fuel consumption is therefore
not simply a measure of the length of the crossing, but is determined by water depth
and metocean conditions. The fastest route is provided by the route atlas (RA), com-
bined designs (CMB) and the SeaPlanner optimization routines (SPO), which out-
perform most of the other candidates in terms of fuel consumption as well. Perfor-
mance variations are the results of the strong River Neva outflow and the dominant
wind directions that generate heterogenous metocean conditions and have a strong
impact on the shipping conditions in one direction or another.
Whereas westward going cruises favour more northern fairways to use the ad-
vancing power of the River Neva outflow, eastward going cruises prefer southern
fairways to avoid it. SeaPlanner optimization adapts to the situation and produces
different sets of fairways for eastward and westward going cruises. For the analysis,
they serve just as reference value. Leaving them aside, the fairways from the route
atlas and combined designs are competing for the top economical performance, the
first being better on westward going cruises and the other being better on eastward
going ones. From an economical point of view it would make sense to keep the
twofold set of routes, because comparisons with SeaPlanner optimized fairways
(SPO) show that performance could be gained by moving RA and CMB routes even
further apart (RA routes should be moved northwards and CMB routes southwards).
From the viewpoint of performance, the open sea fairways SPO, RA and CMB
are superior to more coastal fairways such as SW and SA (Table 11.3). That sur-
prises at first because the SW design follows the River Neva outflow more clearly
than any other design. The low performance can be explained by the lower water
depth and increased shallow water resistance that has been taken into account by
the ship propulsion model but was ignored in the SW and SA design. An evidence
for the positive impact of the River Neva outflow on west going cruises is provided
by the fact that the SeaPlanner optimized route, which follows the northern SW de-
sign but stays more offshore, is the fastest route with the lowest fuel consumption.
Analysing the consumption parameter for a fixed travel time of 13 hours east-
to-west provides a consistent picture. The SeaPlanner optimized route (SPO), fol-
lowing the River Neva outflow, is reaching the time goal with the lowest required
average speed and fuel consumption.
Vyborg is an important harbour for oil transportation in the Gulf of Finland. The
fairways that connect its premises to the Baltic Proper4 use three different sets of
routes. The first and shortest route follows the SW designs and runs along the north-
ern shore of the Gulf of Finland in a direct line to Vyborg. Although by far the
shortest route, it is also by far the most uneconomical one. Shallow water resistance
in near coastal waters is again rendering it inefficient, pushing engine load and fuel
4 This notion is used here to denote the Eastern, Northern and Western Gotland Basin, Bornholm
Basin and Gdańsk Bay (cf. Chap. 2, Fig. 2.1).

consumption to the highest, and developed average speed to the lowest level. The
best fairway designs combine short distance with sufficiently deep water and low
hull friction. The second set of routes for the candidates (SA, CMB and the estab-
lished fairway RA) partially achieves this by following the gulf’s axis until passing
by the island Gogland (see Fig. 10.2 in Chap. 10) in the south. Although staying
offshore all the time, their performance is reduced by the length of the crossing. The
‘winning’ one is the SeaPlanner optimized fairway that runs through the northern
part of the gulf and passes by Gogland in the north. This route is shorter, which
results in shorter travel times and lower fuel consumption, by even lower or compa-
rable average speed. It is worth mentioning that the optimal SPO routes are identical
for the east- and westward crossings. The results are therefore uniform and consis-
tent.
The example of routes to Vyborg and Saint Petersburg demonstrate that it is
probably impossible to find an optimum solution that would fit for all harbours,
ships, and constraints. The performance analysis shows that a northward shift of the
Vyborg route (passing by Gogland in the north) would improve travel time and fuel
consumption measurably. On the other hand, such a shift would separate the Vyborg
route from the Saint Petersburg route and would generate a decomposed system of
fairways. Optimization is configuration dependent and pursues individual solutions
for small or single-member networks with higher efficiency. This becomes apparent
for harbours like Sillamäe and Ust-Luga. The established network with its tree-like
structure, from which the different routes branch off, forces the maritime traffic to
take uneconomical detours with up to 25 nm longer fairways, which amount to about
15 % of the length.
The port of Ust-Luga is situated in Luga Bay in the south-eastern part of the
gulf. Although most of the fairway designs choose for economical reasons a more
or less direct route out of the bay and towards the west, the route atlas orders vessels
to take a detour to the north, to meet the Saint Petersburg fairway. This amounts
to about 2 hours additional travel time, and more than 1 metric ton (mt) more fuel
consumption, which is significant for the relatively small distances along the Gulf of
Finland. The existing fairway, as specified by the route atlas, is about 25 nm longer
than any other fairway. As this fairway is situated in the southern part of the gulf, its
performance is not much affected by the River Neva outflow. The path length and
integrated effects of the shallow water resistance are again the main performance
limiting factors.
It is not surprising that near coastal fairways, which are most favoured by SA
designs, are strongly affected by shallow water hull resistance. Other fairways show
quite small performance variations. The best solution is provided by the SeaPlan-
ner optimization routines, which feature short distances and courses that directly
target the Baltic Proper. The optimized Ust-Luga fairways are mainly defined by
geography and are identical for eastward and westward crossings.
The port of Sillamäe is situated in north-eastern Estonia. Its connection to the
Baltic Sea fairway network also runs along the Saint Petersburg line, which meets
the Sillamäe fairway near Gogland. The existing RA route is on average about 11 nm
longer than the alternative designs, which amounts to up to 1 hour longer travel time
and 0.75 mt more fuel consumption. The distance is again the major performance
defining factor. SPO optimal fairways follow the southern coast closely, but stay out
of shallow waters.
The data presented in Table 11.3 vividly demonstrates that the effects of shallow
water resistance greatly reduced the performance of near-coastal designs and that the
distance is the main performance factor for the considered harbours, some of which
are better accessed over the northern route (Vyborg), some others over the southern
route (Sillamäe, Ust-Luga), whereas the optimum to Saint Petersburg probably goes
along an axial route. The presence of a heterogenous circulation pattern has a strong
effect on the performance of fairways, mostly in the northern part of the Gulf of
Finland where the impact of the River Neva outflow is significant.
The result is a spatial separation of the eastern and western going optimal fair-
ways, which either try to make use of or to avoid the predominant circulation. This is
most apparent for Saint Petersburg where the optimum routes (SPO north) are either
passing by Gogland in the north, turning gradually to the south as they approach the
entrance to the Baltic Proper, or they are coming from the west and move gradually
northwards as they pass by Gogland in the deep South. Therefore, economically
optimized solutions may be provided by specially tailored fairways for individual
harbours and transport directions.
The fairways stemming from the combined design (CMB) seem to be closer to
the performance of economically optimized designs than any of the seasonal de-
signs. For the southern harbours Sillamäe and Ust-Luga, they perform far better
than the fairways from the route atlas, and for northern and eastern harbours their
performance is comparable.
11.6 Wave-Induced Oil Propagation
Waves mix the upper ocean and have tremendous effects on surface oil slicks. The
additional layer of oil is efficiently damping the waves, by reducing the wind en-
ergy uptake and by extracting energy for downwards mixing of buoyant oil droplets
and for intermixing of water with oil. Emulsions are formed which increase the vis-
cosity enormously, intensifying the damping effect. The Braer accident in January
1993 which happened off the coast of the Shetland Islands provides a good example
for this (Edgell 1994). The accident was followed by a storm and heavy seas which
mixed down the oil completely, so that it vanished from the view of the observers.
Later on, when winds and waves subsided, the oil re-emerged as emulsified patches.
Contemporary oil drift and fate models include efficient parameterizations for mix-
ing and emulsification. However, wave-induced oil propagation has been largely
neglected because of the intense computational efforts and the widespread belief
that the wave-induced drift is relatively small compared to the drift caused by ocean
circulation and wind. The estimates given below demonstrate that wave-induced
(Stokes) drift can locally reach the magnitude of the wind- and current-induced
drift.
Most of the published studies of wave-driven effects on the oil drift focus on vari-
ous scenarios, particular shipping accidents or specific storms. Systematic inclusion
of wave-induced drift effects in oil spill models has been problematic mostly be-
cause of the associated large computational efforts. For example, in basically calm
local conditions but in sea areas affected by swell it is necessary to run a global
wave model.
We focus on the analysis of the magnitude of wave-induced drift effects under
different weather conditions. For this purpose, 50,688 drift simulations at 528 re-
lease sites were conducted in the Gulf of Finland to estimate average effects of
waves on the propagation pattern of oil at sea. The simulations cover the year 1992,
which represents an average year. Unlike the above-described fairway design stud-
ies that covered three years of data, we skip the analysis of seasonality of the wave-
induced oil drift as the simulation period is too short to analyse drift patterns on a
sub-annual scale. Instead average wave effects are studied.
11.6.1 Calculation of Wave-Induced Oil Drift
One of the major issues in the coupling of wave and 3D circulation models is that
these models usually do not account for the vertical structure of the counterpart
(Mellor 2003). The equations that govern the surface wave motion are often first
integrated from the bottom to the water surface, and then are phase averaged. Also,
the horizontal velocities in the upper layer of the sea are often assumed to be ho-
mogeneous in the vertical direction. While this assumption can be partially justified
by the problems of replicating accurately the vertical structure of the upper-layer
flow in large-scale circulation models, the result is a mismatch of the treatment of
z-dependent horizontal velocities and other properties between the wave models and
3D circulation models. This mismatch is apparently largest just at the sea surface,
that is, in the flow region that directly affects the oil propagation.
A derivation of the depth-dependent equations that account for the vertical struc-
ture of the wave motion follows the analysis of Mellor (2003). As the procedure
is technically quite complicated, we omit the details and only present the final re-
sult. The goal is to derive a prognostic equation for the depth-dependent mean cur-
rent u(x, z, t) in the presence of waves, expressed as a sum of the ambient (e.g.,
modelled) current and wave-induced depth-dependent drift in a sea area of finite
depth D. The starting point is the momentum equation du/dt = F − ∇δp for an
incompressible fluid, where d/dt denotes the total derivative, F is the sum of all
wave-independent external forces and body forces and ∇δp is the pressure gradi-
ent that specifically involves the impact of the oscillating wave motion on the mean
current. The derivation relies on the linear (Airy) wave theory but includes the anal-
ysis of second-order terms in the wave height. The resulting equation, in which the
integration is performed from the bottom to a certain surface h, is:
∂ u
∂ u
∂
= F
− u
· − · S, (11.4)
∂t ∂x ∂x
where angle brackets mean averaging over the wave period, u

= u0 + usd , usd =
ωka 2 /(2 sinh2 kD) is the ‘classical’ Stokes drift, u0 has the meaning of the ambient
(e.g., modelled) Eulerian current, calculated without accounting for the impact of
waves, and

cf k ⊗ k 1
S=E +δ A−δ B , (11.5)
cg k2 2
where
sinh 2kh + 2kh cosh 2kh − 1
A= , B= . (11.6)
sinh 2kD + 2kD 2 sinh2 kD
Here S denotes the radiation stress tensor, E = (1/2)ga 2 is the wave energy, g
is the acceleration due to gravity, a is the wave amplitude, cf is the phase speed
(celerity), cg is the group speed, k is the wave vector, k = |k|, h = D(1 + ζ ) is
the local depth of layer ζ and δ is Kronecker’s delta (generally defined as δij = 1
for i = j and δij = 0 for i = j but here indicating the horizontal coordinates). The
resulting expressions (11.4) together with (11.5) reveal that, to the second order,
the only non-vanishing contribution to the hydrodynamic transport equations is the
divergence of radiation stress in the momentum equation. Therefore, waves act on
the surface flow by introducing an accelerating force, which in the equation above,
becomes apparent as a divergence term of the surface stress. Note that turbulent
mixing and nonlinear dissipation have been excluded from these considerations.
The wave energy equation is easily constructed from the momentum equation
du/dt = F − ∇ · S by scalar multiplication from the right with the velocity vector u
and following phase averaging. The outcome is a prognostic equation for the kinetic
energy Ekin = 12 u2
of which the waves’ contribution is Ekinw = 1 δu2
. Its deriva-
2
tion makes use of the identity ∂(x,t) u · u = 2 ∂(x,t) u2 , where ∂(x,t) denotes partial
1
derivatives with respect to x or t. The resulting equation is:

∂Ekin ∂Ekin ∂
= F
· u
− u
· − · S · u
. (11.7)
∂t ∂x ∂x
The radiation stress appears as a sink term on the right-hand side of Eq. (11.7).
Its contribution that is basically equivalent to the change in the momentum carried
by waves is closely linked to the wave energy gradient and thereby to the amount
of energy that is dissipated through wave breaking. Its effect is perhaps most well-
known in the theory of wave set-up, a process that may lead to substantial increase
in the local water level in selected coastal areas (Dean and Bender 2006). In wave
modelling this effect is rarely considered.
As wave- and wind-induced stress are treated similarly by the model, it is
straightforward to evaluate the level of wind stress, the impact of which on sur-
face currents would be equivalent to the wave-driven forcing. For this purpose, the
forcing term ∂x · S + ∂z τ (U10 ) is rewritten to a function for U10 in case of a balance
of forces. Doing so makes it possible to express the magnitude of the wave-induced
forcing in terms of the wind speed, i.e., to calculate the properties of wind that would
drive the same currents as the waves do.
Fig. 11.8 Wave induced momentum contribution in units of wind speed
For simplified geometries, the wave-induced drift and other effects related to the
radiation stress can be quantified analytically (Longuet-Higgins and Stewart 1964).
An example is the one-dimensional case of unidirectional long-crested linear (Airy)
waves propagating along a sloping bottom with constant slope in the x-direction.
For this situation the term (k ⊗ k)/k 2 = 1 and the wave-induced surface drift can be
calculated directly from the wave energy gradient and the quotient of the group and
phase speeds cg /cp (that is close to 1 in shallow water conditions). From Eq. (11.4)
it follows that the wave drift in a sea area of constant depth h = D is governed
solely by the wave energy gradient ∂x E = (ρg/8)HS ∂x HS , where HS stands for
the significant wave height. If the total depth integrated wave input is considered
(that is, Eq. (11.5) is integrated over all the water column so that h = D), Eq. (11.5)
simplifies considerably, as the factors A, B become simply A = 1 and B = 1.
The driving force of waves equals the driving force of wind if the wave en-
ergy gradient ∂x E equals the vertical wind induced current shear stress gradient
∂τ/∂z. For the Gulf of Finland model set-up, with a grid resolution of 0.5 nautical
miles ≈921 m, a surface layer thickness of 3 m and a drag coefficient of CD =
(0.63 + 0.066 · U10 ) × 10−3 , the wind- and wave-driven forces are equal if the wave
height change over one grid cell is
2
HS = 3.93 × 10−4 + 4.12 × 10−5 U10 · U10 /HS . (11.8)
Under certain conditions the wave induced radiation stress might become as strong
as the wind induced drift (Fig. 11.8).
The spatial distribution of the mean wave-induced surface stress (the magnitude
of the wave-induced force that drives the surface current but expressed in terms of
wind speed) for the Gulf of Finland averaged over the years 1991–1994 suggests
Fig. 11.9 Mean wave induced surface stress 1991–1994 in units of wind speed, i.e., solutions
of the discretized equation /x(wave-stress) = −/z(wind-stress) with x ≈ 921 m and
z = 3 m
that the local maxima at near coastal locations represent an average driving force
equivalent to 17 m/s wind (Fig. 11.9). In case of strong storms, like the 2005 win-
ter event (when wind speed of 37.5 m/s was measured in Estonia, Suursaar et al.
2006; Soomere et al. 2008), the wind equivalent of the wave-induced surface stress
increases significantly and might locally reach values of 25 m/s. The spatial distribu-
tion of the regions where the wave-driven surface currents are the strongest depends
primarily on wave propagation and wave energy gradients, and is only weakly de-
pendent on the local wind strength. Note that the results may substantially depend
on the particular model set-up and wind information as here the gradients of physi-
cal fields are compared. It is still remarkable that wave-driven effects in near-coastal
domains (Fig. 11.9) may locally exert significant impact on the onshore oil propa-
gation and cannot be ignored. This is even more the case because topographically
steered nearshore currents usually carry the pollutants alongshore, whereas wind
waves often cause onshore transport in the surface layer.
11.6.2 Modelled Wave-Induced Transport
Assessments of wave-induced momentum and transport often estimate the Stokes

drift diagnostically and add it to the Eulerian currents. This is usually done based
on spectrally integrated wave parameters such as the significant wave height, pe-
riod and direction that are employed to calculate the representative values of the
wave number and energy, and finally to calculate the Stokes drift. In order to get a
better representation of the physical processes, we use a coupled ocean wave and
circulation model that makes it possible to calculate directly the wave-driven drift.
For applications in regions with low tides and weak mean flows such as the Baltic
Sea and the Gulf of Finland, a one-way coupled system can be applied, in which the
Fig. 11.10 Difference of the 1992 annual mean oil residence time (upper panel) and the probabil-
ity of landing (lower panel) with and without waves. Red and blue contours represent positive and
negative values of the differences. The boxes show the locations of the panels in Fig. 11.11
wave model is feeding the circulation model. The additional wave contribution to
the mean drift currents is implemented as an additional force at the surface. Because
wave effects are largely driven by the energy gradient, their impact depends strongly
on topography and geometry, and varies from region to region. The overall picture
is rather patchy (Fig. 11.10) and we shall focus on regional differences (Fig. 11.11).
The indicators are, as above, the probability of landing and oil residence time at sea.
Differently from the above, we present results in terms of the average wave-induced
differences to their values for the year 1992.
The results show that oil residence times in near coastal waters in the presence of
waves differ considerably from the results presented above even in terms of annual
averages. The modelled local differences of the residence time vary from −2.2 days
to 1.3 days (Fig. 11.10). The average effect is noticeable everywhere in the gulf
and amounts to up ±10 % of the probability of landing (Fig. 11.10). Interestingly,
the largest wave-driven effects occur in very localized areas. This feature may par-
tially result from the spatial interpolation of the evaluated differences (that basically
represent point values) into the model grid. A local maximum might therefore indi-
cate the oil propagation properties at a single release site. Note also that Figs. 11.10
and 11.11 show the combined effect of geometry and topography of the study area,
Fig. 11.11 Difference of the 1992 annual mean oil residence time (left) and landing probabilities
(right) with and without waves for areas indicated in Fig. 11.10. Red and blue contours represent
positive and negative values of the differences in the probability of landing (left) and residence
time (right). Dark contours represent the 30 m water depth and the model coastline
and meteorological forcing. Therefore, it is not unexpected that the largest wave-
driven differences occur close to the coast, around islands and at locations with
considerable offshore depth gradients. All the listed regions host extensive wave
refraction and the release of momentum carried by waves (Fig. 11.9).
A closer view at the results for the middle of the Gulf of Finland (panels A1,
B1 of Fig. 11.11) reveals the importance of wave-driven effects on the formation of
currents and oil propagation. Offshore wave-driven drift has been usually considered
to be of minor influence on both the current regime and on the average transport
quantities. This assumption seems to be true for the overall current regime. The
local transport properties, however, show significant dependence on the wave fields.
Namely, local wave energy gradients (dE/dx in Eqs. (11.4), (11.5) and (11.6)) due
to the topography and wind shadowing effects may produce significant drift currents
and may have a substantial impact on the drift pattern in the study area.
The performed numerical experiments provide evidence for local differences of
the probability of oil landing of up to ±12 % near certain islands. As mentioned
above, the residence time increases by up to 1.3 days at certain locations when wave-
driven transport is accounted for. Although the depth gradients that appear at larger
depths exert a minor impact on waves and currents, the persistency of the wave-
induced drift and the longer interaction ranges lead to noticeable results. Here, the
interaction range means the length of a drift path on which the drifter experiences
considerable depth gradients. The impact on the probability of landing and on the
residence time increases with the interaction range as oil experiences more and more
wave impetus on the way. Small wave-induced differences of the drift path after a
short time might grow with distance from the release position and may alter the final
drift path significantly.
In the surf zone and close to the shoreline, the release of momentum carried by
beaching waves often produces coastal jets (Longuet-Higgins and Stewart 1964).
The strongest wave-driven currents occur at incidence angles of about 45◦ . Such
coastal jets might prevent onshore oil propagation or carry it further to other coastal
sections (Fig. 11.11 A3, B3). The magnitude of the wave-driven drift depends on
whether the waves can exert a cumulative effect along the pathways of oil propaga-
tion. In general, the effect is the strongest in the nearshore where (although featured
by strong interaction forces through large wave energy gradients) it is character-
ized by a rather small interaction range (which is the distance between the seaward
border of the surf zone and the landing point).
The presence of wind waves may increase the probability for oil landing by about
10–15 %. The residence time decreases moderately, by about 4 hours, in these cases.
The decrease in the residence time is twice as large in the upstream direction of the
coastal jet where the number of landing particles is smaller. The range of probability
changes is about 30 % (from +10 % to −20 %). Some particles reach the coast faster
and are counted in the histogram for travel time (Fig. 11.3), but others are travelling
with the coastal jet for a while and are too much diverted from their original path to
hit the shoreline. It should be remembered that wave-induced currents and coastal
jets not necessarily contribute to the onshore oil propagation. Although wave energy
gradients and contributions of their momentum to the currents are mainly following
the depth gradients, often in a direction towards the coast, nearshore jets are influ-
enced by the land boundaries and more or less follow the coastline. It is unclear
why waves lead to an increase in the residence time by 7–8 hours in some offshore
regions. The reasons may be related to specific pathways of oil transport.
Additionally to the wave breaking and changes in the water depth, there is yet
another process that may modify the impact of waves. At the lee-side of islands,
the wave energy gradients and drift currents are caused by wind shadowing effects.
Considerable fetch length may be required for waves in the island’s shadow to grow
until they match their freely propagating neighbours. For example, the annual mean
wave heights in 1992 at the lee-side of Gogland (it is exposed to waves with rather
long fetch, some even generated in the Baltic Proper) indicate that the shadow ex-
tends to about 21 km. For waves coming from the west, the strongest wave energy
gradients occur at the northern and the southern cape of the island and generate
northerly and southerly currents at the southern and the northern end of the island,
respectively. Depending on the local topography, wave interaction creates systems
of eddies in up- and downwave direction, clearly seen in panels A2, B2 of Fig. 11.11.
The eddies carry away pollutants approaching from several coastal stretches where
the wave-induced currents point offshore. There is also a high chance for the en-
hanced onshore transport at some locations. Their impact amounts to significant
differences in the residence time of up to 1.3 days and in the probability of landing
of about 10 %.
This example study of oil propagation and weathering shows that the impact of
wind waves on oil at sea may be very strong at selected locations. It still remains
open how this affects the fairway design. Although the developed fairways ignore
wave-induced transport, they naturally avoid the vicinity of islands and normally
do not enter the nearshore regions where these effects are the strongest, and thus
avoid the most critical zones in this respect. It would be highly interesting (although
computationally expensive) to perform a closer evaluation of the properties of wave-
induced drift along the presented fairways and an assessment of the impact of waves
on the fairway performance.
11.7 Summary and Conclusions
The presented results expand the outcome of the previous chapters of this book to-
wards a systematic inclusion of the realistic properties of oil drift into the preventive
technology for environmental management of semi-enclosed seas. Additionally to
the above-described effects of currents on the transport of ‘ideal’ pollutants, we have
estimated, to a first approximation, the joint influence of realistic systems of winds,
waves and currents on the optimum fairway design and have also examined seasonal
variations of the characteristic drift patterns for the Gulf of Finland.
Consistently with the estimates made for the ideal pollution solely driven by sur-
face currents, it has been found that a relocation of release positions to regions with
favourable drift properties leads to an increase in the average residence time of oil
at sea by at least 3 days and to a decrease of the probability of landing by more than
30 %. As a key new development, we established the presence of substantial and to a
certain extent counter-intuitive patterns of seasonal variability in the location of the
favourable fairways. Its major source is the seasonal variation of the wind proper-
ties, first of all wind speed and direction. Its presence demonstrates the importance
of wind-driven dynamics of oil drift and associated drift patterns in the study area
and apparently in other similar regions.
Surface currents play a dominant role in oil propagation during relatively calm
periods. The reason for the strong influence of wind is the de facto limitation of
the analysis to oil in the surface layer. Oil that is mixed down in the water column
is solely driven by currents. It is usually not buoyant enough to rise to the sur-
face again, especially at later stages of the spill development when up to 75 % of
oil have been taken up by emulsification and volatile components have evaporated
completely.
Both wind and waves play an important role, to some extent complementary to
each other, in the drift and fate of oil spills. The oil parcels that reach the shoreline
first have normally travelled at the surface all the time and have been exposed to
the wind forces for the duration of the travel time. Although oil moves fastest at the
surface due to the wind, at the same time oil parcels at the surface are continuously
exposed to the impact of waves that successively mix them down into the water
column. The fastest-propagating oil parcels thus reflect a delicate balance of the
impact of wind and waves.
The performed analysis of oil drift patterns provides the basis for the critical
evaluation of the existing ship routes. Monte Carlo techniques have been used to
analyse seasonal and annual drift patterns and to generate season-specific optimum
fairways. A key development is the evidence that greatly different seasons such
as winter and summer may require the same fairway design whereas a drastically
different design fits much better for both transitional seasons, spring and autumn.
The combined summer and winter designs have the best overall performance in
terms of path integrals of the cost function along fairways and therefore at best
represent the average patterns of joint propagation of oil owing to sea currents and
winds.
As each design represents a set of seasons, it is crucial for the overall estimation
to compare the cross-seasonal performance. Summer-winter designs perform just a
little worse in spring and autumn. By contrast, spring and autumn designs perform
much better in their season, but much worse outside their season. The drawbacks of
seasonal designs can be to some extent compensated by using annual mean designs.
Such combined designs represent neither maxima nor minima of the seasonal con-
figurations, they are universal and for that reason performance all-rounders. It is not
surprising that such designs have something of everything and that they combine
a good annual mean performance with a small cross-seasonal performance drop.
The true value of combined design becomes apparent when one compares all three
designs to each other.
The choice of the final design depends on the long-term strategy of maritime
management and policies, and environmental as well as economical performance.
If much focus is put on the all-season, climatologically valid performance, then
combined designs are preferable, but also summer-winter designs are worthwile to
look at. Spring-autumn designs might be disregarded, because of a substantial drop
in their performance in summer and winter. Combined designs seem to be a good
option, because their economical performance in terms of travel time, speed and
fuel consumption is in the range of optimized economical designs of the existing
ship routing software packages.
Cost efficient navigation in coastal waters depends on direct, short distance
courses. It should also account for the particular water depth to avoid resistance
effects at the ship’s hull, and advancing or adverse circulation conditions. In this
sense the combined all-year-round designs outperform their seasonal counterparts
which favour more near-coastal waters along the northern and southern coasts of the
Gulf of Finland. They mostly represent offshore designs which are less hampered
by resistance effects. They are both economically beneficial and environmentally
advantageous. That makes them recommendable for future fairway design studies
of cost efficient networks in the Gulf of Finland.
Our studies have also revealed some aspects of the joint impact of wind, waves
and currents on the oil drift and fate. In order to highlight the potential impact of
waves on oil advection one year of drift simulations (1992) were repeated using
the wave-driven drift currents. Interestingly, the impact of waves is very localized.
Both its sign and amplitude vary over sea areas on a scale of 10 km. This effect
in the study area (the Gulf of Finland) is apparently caused by the complicated
bathymetric and geometrical features that lead to strong local wave energy gradients
and variations in other wave properties owing to shadowing effects. Wave energy
gradients that actually drive the spatially varying currents lead to differences of
up to ≈1 cm/s, at instances even ≈3 cm/s compared to the cases when waves are
ignored. These additional currents may prevent or advance onshore oil propagation,
according to particular topography and wind and wave conditions. Under average
conditions, local residence time differences of up to −2.2 to 1.3 days and differences
in the probability of oil landing of up to ±20 % have been found.
An obvious drawback of this method is that a maritime accident, and, by conse-
quence, an oil spill, is supposed to take place with the same likelihood, independent
of geographical position or weather conditions. In reality, some weather and ocean
conditions (high winds and waves, fog, sea ice, or icebergs in other parts of the
world) are riskier than others, and some waters are more difficult to navigate than
others. Merging sea traffic towards the same pathways would in itself enhance the
risk of collision. It would be highly interesting to continue efforts in this direction,
for example, towards introduction of realistic collision and accident risk into the
mode, equivalently, to introduce site- or weather-dependent weights to the spill sim-
ulations.
Acknowledgements The underlying studies were performed in the framework of the BalticWay
project, in which the DMI was jointly supported by the funding from the Danish Ministry of
Science, Innovation and Higher Education and the European Commission’s Seventh Framework
Programme (FP 2007–2013) under grant agreement No. 217246 made with the joint Baltic Sea re-
search and development programme BONUS. The models were run using meteorological forcing
kindly provided by the Danish Climate Centre (DKC).
References
9:1–16
334
Oceanogr 39:2820–2838
Audunson T (1980) The fate and weathering of surface oil from Bravo blowout. Mar Environ Res
3:35–61
Berg P (2012) Mixing in HBM. Scientific Report 12-03, Danish Meteorological Institute. http://
www.dmi.dk/dmi/sr12-03.pdf
Berg P, Weismann Poulsen J (2012) Implementation details for HBM. Technical Report 12–11,
Danish Meteorological Institute. http://www.dmi.dk/dmi/tr12-11.pdf
Bergström S (1976) Development and application of a conceptual runoff model for Scandinavian
catchments. SMHI Reports RHO, No 7, Norrköping
Bergström S (1992) The HBV model—its structure and applications. SMHI Reports RH, No 4,
Norrköping
Bi H, Rissik D, Macova M, Hearn L, Mueller JF, Escher B (2011) Recovery of a freshwater wetland
from chemical contamination after an oil spill. J Environ Monit 13:713–720
Børresen JA (1993). Olje på havet (Oil on the sea), Gyldendal
Brandt J, Christensen JH, Frohn LM, Zlatev Z (1998) Numerical modelling of transport, dispersion,
and deposition—validation against ETEX-1, ETEX-2 and Chernobyl. Environ Model Softw
15:521–531
Brandt J, Christensen JH, Frohn LM (2002) Modelling transport and deposition of caesium and
iodine from the Chernobyl accident using the DREAM model. Atmos Chem Phys 2:397–417
Broström G, Carrasco A, Hole LR, Dick S, Janssen F, Mattsson J, Berger S (2011) Usefulness of
high resolution coastal models for operational oil spill forecast: the Full City accident. Ocean
Sci 7:805–820
Burgherr P (2007) In-depth analysis of accidental oil spills from tankers in the context of global
spill trends from all sources. J Hazard Mater 140:245–256
Canuto VM, Howard A, Cheng Y, Dubrovikov MS (2001) Ocean Turbulence. Part I: One-point
closure model-momentum and heat vertical diffusivities. J Phys Oceanogr 31:1413–1426
Canuto VM, Howard A, Cheng Y, Dubrovikov MS (2002) Ocean turbulence. Part II: Vertical
diffusivities of momentum, heat, salt, mass and passive scalars. J Phys Oceanogr 32:240–264
Christensen JH (2002) Application of multivariate data analysis for assessing the early fate of
petrogenic compounds in the marine environment following the Baltic Carrier oil spill. Polycycl
Aromat Compd 22:703–714
Christensen OB, Drews M, Christensen JH, Dethloff K, Ketelsen K, Hebestadt I, Rinke A (2006)
The HIRHAM Regional Climate Model. Version 5. DMI technical report No 06-17. http://www.
dmi.dk/dmi/tr06-17.pdf
Christiansen M, Fagerholt K, Ronen D (2004) Ship routing and scheduling: status and perspectives.
Daling PS, Moldestad MO, Johansen O, Lewis A, Rodal J (2003) Norwegian testing of emulsion
properties at sea—the importance of oil type and release conditions. Spill Sci Technol Bull
8:123–136
Dalton T, Jin D (2010) Extent and frequency of vessel oil spills in US marine protected areas. Mar
Pollut Bull 60:1939–1945
Dean RG, Bender CJ (2006) Static wave setup with emphasis on damping effects by vegetation
and bottom friction. Coast Eng 53:149–165
Dick S, Soetje KC (1990) An operational oil dispersion model for the German Bight, Dtsch Hy-
drogr Z, Ergänz.heft, Reihe A 16, 43 pp
Dietze H, Kriest I (2012) Cs-137 off Fukushima Dai-ichi, Japan—model based estimates of dilu-
tion and fate. Ocean Sci 8:319–332
Dippner JW (1984) Ein Strömungs- und Öldriftmodel für die Deutsche Bucht. Veröff Inst Meeres-
forsch Bremerhaven, Suppl B8, 188 pp
Drozdowski A, Nudds S, Hannah CG, Niu H, Peterson IK, Perrie WA WA (2011) Review of
oil spill trajectory modelling in the presence of ice. Canadian Technical Report of Hydrogra-
phy and Ocean Sciences 274. Fisheries and Oceans Canada, 84 pp. http://www.dfo-mpo.gc.ca/
Library/344774.pdf
Edgell N (1994) The Braer tanker incident—some lessons from the Shetland Islands. Mar Pollut
Bull 29:361–367
Evans JJ, Marlow P (1990) Quantitative methods in maritime economics, 2nd ed. Fairplay Publi-
cations, Coulsdon
Fay JA (1969) The spread of oil slicks on a calm sea. In: Hoult DP (ed) Oil on the sea, pp 53–63
Fay JA (1971) Physical processes in the spread of oil on a water surface. In: Proceedings of
joint conference on prevention and control of oil spills, USA, Washington, DC, June 15–17,
pp 463–467
Fingas MF (ed) (2011) Oil spill science and technology. Prevention, response and cleanup. Else-
vier, Amsterdam, 1156 pp
Forrester WD (1971) Distribution of suspended oil particles following the grounding of the tanker
Arrow. J Mar Res 29:151–170
Funkquist L (2001) HIROMB: an operational eddy-resolving model for the Baltic Sea. Bull Mar
Inst Gdańsk 28:7–16
Hela I (1952) Drift currents and permanent flow. Commentationes physico-mathematicae, vol XVI.
Societas Scientarium Fennica, Helsinki, 14, 27 pp
Hibler WD III (1979) A dynamic thermodynamic sea ice model. J Phys Oceanogr 9:817–846
Janeiro J, Martins F, Relvas P (2012) Towards the development of an operational tool for oil spills
management in the Algarve coast. J Coast Conserv 16:449–460
Jerneloev A (2010) The threats from oil spills: now, then, and in the future. Ambio 39:353–366
Johansen Ø (1982) Drift of submerged oil at sea. Continental Shelf Institute Report P 319/1 Trond-
heim, 51 pp
Johansen Ø et al (2005) Simulations of oil drift and spreading and oil spill response analysis. Arctic
Operational Platform project (ARCOP)
Kleine E (1994) Das operationelle Modell des BSH für Nordsee und Ostsee, Konzeption und Über-
sicht. Bundesamt für Seeschifffart und Hydrographie, 126 pp (in German)
Keramitsoglou I, Cartalis C, Kassomenos P (2003) Decision support system for managing oil spill
events. Environ Manag 32:290–298
Kjelsson J, Döös K (2012) Surface drifters and model trajectories in the Baltic Sea. Boreal Environ
Res 17:447–459
Larsen J, Høyer JL, She J (2007) Validation of a hybrid optimal interpolation and Kalman filter
scheme of sea surface temperature assimilation. J Mar Syst 65:122–133
Leibovich S (1975) A natural limit to the containment and removal of oil spills at sea. Ocean Eng
3:29–36
Liu Y, Zhu J, She J, Zhuang S, Fu W, Gao J (2009) Assimilating temperature and salinity profile
observations using an anisotropic recursive filter in a coastal ocean model. Ocean Model 30:75–
87
Lo HK, McCord MR (1995) Routing through dynamic ocean currents—general heuristics and
empirical results in the Gulf-stream region. Transp Res, Part B, Methodol 29:109–124
Lo HK, McCord MR (1998) Adaptive ship routing through stochastic ocean currents: general
formulations and empirical results. Transp Res, Part A, Policy Pract 32:547–561
Longuet-Higgins MS, Stewart RW (1964) Radiation stresses in water waves: a physical discussion
with applications. Deep-Sea Res 11:529–562
McCord MR, Lee YK, Lo HK (1999) Ship routing through altimetry-derived ocean currents.
Mackay D, Shiu WY, Hossain K, Stiver W, McCurdy D (1982) Development and calibration of an
oil spill behavior model. Department of Chemical Engineering and Applied Chemistry, Toronto
University (Ontario), 94 pp
Mackay D, Stiver W, Tebeau PA (1983) Testing of crude oils and petroleum products for environ-
mental purposes. In: Proceedings of the 1983 oil spill conference, pp 331–337
Mellor G (2003) The three-dimensional current and surface wave equations. J Phys Oceanogr
33:1978–1989
Midboe F, Persson H (2004) Oljeutsläpp och dess miljökonsekvens i Östersjön, Institute of Geo-
sciences, Uppsala University
Molinero CM, Mitsis SN (1984) Budgeting fuel consumption in a cruise liner. Eur J Oper Res
18:172–183
Monson DH, Doak DF, Ballachey BE, Bodkin JL (2011) Could residual oil from the Exxon Valdez
spill create a long-term population “sink” for sea otters in Alaska? Ecol Appl 21:2917–2932
Pahlke H (1985) Physikalische Grundlagen der mechanischen Ölbehandlung. Teil 3: Verhalten von
Öllachen auf Wasser, Umweltforschungsplan des Bundesministers des Inneren. Forschungs-
bericht 10203 204/01, 185 pp (in German)
Pikkarainen A-L, Lemponen P (2005) Petroleum hydrocarbon concentrations in Baltic Sea subsur-
face water. Boreal Environ Res 10:125–134
Pålsson J (2012) Oil spill preparedness in the Baltic Sea countries. Report written for the Baltic
Master II project, World Maritime University
Reed M, Aamo OM (1994) Real time oil spill forecasting during an experimental oil spill in the
Arctic ice. Spill SciTechnol Bull 5:203–209
Ronen D (1982) The effect of oil price in the optimal speed of ships. J Oper Res Soc 33:1035–1040
Ronen D (1986) Short term scheduling of vessels for shipping bulk or semi-bulk commodities
originating in a single area. Oper Res 34:164–173
She J, Berg P, Berg J (2007) Bathymetry impacts on water exchange modelling through the Danish
Straits. J Mar Syst 65:450–459
2nd edition. In: Baltic Sea science congress, Stockholm, 25–29 November 2001, Poster #147.
Smagorinsky J (1963) General circulation experiments with the primitive equations: I. The basic
experiment. Mon Weather Rev 91:99–164
Soomere T, Andrejev O, Sokolov A, Myrberg K (2011a) The use of Lagrangian trajectories for
identification the environmentally safe fairway. Mar Pollut Bull 62:1410–1420
Soomere T, Andrejev O, Sokolov A, Quak E (2011b) Management of coastal pollution by means

of smart placement of human activities. J Coast Res 64(Special Issue):951–955
ways in elongated basins using Lagrangian trajectories: a case study for the Gulf of Finland, the
Baltic Sea. Ocean Dyn 61:1669–1680
Soomere T, Delpeche N, Viikmäe B, Quak E, Meier HEM, Döös K (2011d) Patterns of current-
A):49–63
Sørensen JH (1998) Sensitivity of the DERMA long-range Gaussian dispersion model to meteoro-
logical input and diffusion parameters. Atmos Environ 32:4195–4206
Sørensen JH, Baklanov A, Hoe S (2007) The Danish emergency response model of the atmosphere
(DERMA). J Environ Radioact 96:122–129
Suursaar Ü, Kullas K, Otsmann M, Saaremäe I, Kuik J, Merilain M (2006) Hurricane Gudrun and
modelling its hydrodynamic consequences in the Estonian coastal waters. Boreal Environ Res
11:143–159
Sverdrup HU, Johnson MW, Fleming RH (1942) The oceans. Prentice-Hall, New York
Umlauf L, Burchard H, Hutter K (2003) Extending the k–ω turbulence model towards oceanic
applications. Ocean Model 5:195–218
Vandenbulcke L, Beckers J-M, Lenartz F, Barth A, Poulain P-M, Aidonidis M, Meyrat J,
Ardhuin F, Tonani M, Fratianni C, Torrisi L, Pallela D, Chiggiato J, Tudor M, Book JW,
Martin P, Peggion G, Rixen M (2009) Super-ensemble techniques: application to surface drift
prediction. Prog Oceanogr 82:149–167
Yapa PD, Belaskas DP (1993) Radial spreading of oil under and over broken ice—an experimental-
study. Can J Civ Eng 20:910–922
Wang K, Leppäranta M, Gästgifvars M, Vainio J, Wang C (2008) The drift and spreading of the
Runner 4 oil spill and the ice conditions in the Gulf of Finland, winter, 2006. Est J Earth Sci
57:181–191
Webster HN, Thomson DJ, Johnson BT, Heard IPC, Turnbull K, Marenco F, Kristiansen NI,
Dorsey J, Minikin A, Weinzierl B, Schumann U, Sparks RSJ, Loughlin SC, Hort MC, Leadbet-
ter SJ, Devenish BJ, Manning AJ, Witham CS, Haywood JM, Golding BW (2012) Operational
prediction of ash concentrations in the distal volcanic cloud from the 2010 Eyjafjallajökull erup-
tion. J Geophys Res—Atmos 117:D00U08
Weismann Poulsen J, Berg P (2012) More details on HBM—general modelling theory and survey
of recent studies. Technical Report 12-16, Danish Meteorological Institute. http://www.dmi.
dk/dmi/tr12-16.pdf
Concluding Remarks
Tarmo Soomere
The central driving force of the proposed technique is the contemporary paradigm
that the fundamental benefit in environmental aspects should be achieved by means
of reasonable investments and minimum interference to the course of things. This
paradigm has motivated the BONUS BalticWay team to address the possibilities for
avoiding beforehand potentially extensive damage rather than allocating large in-
vestments to mitigate the already occurred damages. The key idea is to quantify the
potential of different (offshore) areas to serve as a source of remote danger to the
environment in case an adverse impact is released in such an area. An estimate is
obtained by means of solving the relevant inverse problem approximately. The ap-
plications have focused on the evaluation of the potential of current-driven transport
of neutrally buoyant pollution or contaminants to the nearshore (which is tagged
with a fixed price label). The key elements are Lagrangian trajectories of persistent
pollution or contaminant particles that are passively carried by surface currents. The
final step of minimizing environmental risks makes use of the spatial variability of
the resulting quantification.
The potential of each sea point to pose a danger to vulnerable regions is quan-
tified with the use of the probabilities of the pollution reaching the vulnerable area
and with the use of the time it takes for the pollution to arrive at such an area.
The latter is much richer in content and characterizes the response time available
after the accident before the impact reaches the vulnerable areas. The spatial dis-
tributions of the two measures are generally not identical and they both carry some
important (albeit not completely independent) information. Alternatively, local so-
lutions offering an equidistribution of the costs (or responsibilities) between the
T. Soomere (B)
T. Soomere

DOI 10.1007/978-3-319-00440-2,
418 T. Soomere
neighbours on different coasts may serve as politically correct options. More com-
prehensive solutions can be found by globally minimizing the probability of pol-
lution or by maximizing the time in question. All demonstrated solutions can be
applied for the optimization of (climatologically or seasonally valid) stationary lo-
cations as well as for the optimization of risks associated with moving sources
of danger. The presented approaches have used the nearshore as an example of
vulnerable areas and ship traffic as a source of danger. A generalization of the
technique to virtually any set of vulnerable areas and their values is straightfor-
ward.
The presence of semi-persistent current patterns as discussed in Chap. 9 is a
strong argument for the feasibility for this approach. The presented analysis pro-
vides a hint for a more general criterion based on the ability of certain introduced
measures to characterize the internal dynamics of surface currents. First of all, a
combination of the asymptotic values for the cumulative probability of coastal hits
and particle age (equivalently, the residence time of oil at sea in Chap. 11) with their
(spatial) standard deviation deserves attention as all these are virtually independent
of the ocean model resolution. Although these measures obviously depend on the
size of the water body, they implicitly characterize certain properties of current-
driven surface transport that cannot be extracted directly from classical Eulerian
velocity fields. The listed properties evidently differ considerably, for example, for
tide-dominated and microtidal basins of similar size. It is also natural to expect that
basins hosting mostly jet-like currents differ in this sense from regions with predom-
inantly eddy-like dynamics. Most probably, relatively low-resolution (but at least
eddy-permitting) hydrodynamic models may be used for checking whether or not a
substantial decrease in environmental risks is possible from the discussed technique
in a particular region.
The importance of the model resolution becomes evident in attempts to construct
practical tools for decision-making. The presented material confirms that the derived
solutions can only be reliable if the model is eddy-resolving. The resulting optimum
locations for potentially dangerous activities depend, not surprisingly, substantially
on the specific target. The optimum solutions show also highly nontrivial depen-
dence on the seasonal forcing patterns (in the Gulf of Finland, Chap. 11) or on the
overall flow regime in the transition zone between the Baltic Sea and the North
Sea (Chap. 10) It is now evident that simple heuristic solutions based on qualitative
arguments (e.g., a uniform shift of the entire fairway towards a less dangerous po-
sition) do not work. Instead, a largely meandering fairway may lead to a substantial
decrease in environmental risks.
An obvious measure of the overall potential of environmental benefit is the min-
imum value of the probability of a coastal hit, equivalently, the maximum particle
age. The areas hosting the extreme values are natural candidates for activities with
potentially large environmental consequences. It might be beneficial, for example,
to relocate to such an area a leaking vessel that cannot be brought into a harbour
because of rough weather or tough ice conditions. There might exist several minima
for the relevant distributions and thus also multiple options to minimize the conse-
quences of accidents. A similar decision-making process can also be used when a
Concluding Remarks 419
ship has to be towed into a harbour with minimum potential danger to the surround-
ing environment.
The spreading of the optimum locations based on different approaches implic-
itly characterizes the applicability and robustness of the entire approach. The out-
come strongly depends on the nature of the underlying measures. If their distribu-
tions have a well-defined extremum (or zero-crossing), the outcome of the tech-
nique is fairly robust. If the spatial gradients of these distributions are small, even
extensive variations in the optimum solutions may lead to a marginal increase in
the environmental risk. In particular, extensive variations in the optimum loca-
tions for small changes of the model parameters and selection criteria signify that
the domain in question may contain areas with quite different internal dynamics.
These areas can be separated based on the magnitude of gradients of the fields
characterizing the ‘ability’ of different domains to serve as a source of coastal
pollution. The proposed technique, therefore, may be used as a tool for identify-
ing regions with different internal properties for a use in maritime spatial plan-
ning.
Probably the most promising application of the discussed technique is ship rout-
ing based on environmental criteria. There is obvious potential for greening of the
marine transport by means of very limited investments into scientific research (oper-
ational) modelling and development of decision support systems. The average prob-
ability of coastal pollution can be made substantially smaller, at least theoretically,
by means of a quite insignificant increase in the length of the sailing line. A sub-
stantial increase in the potential reaction time to oil pollution is the most promising
aspect as it also provides an option of natural weathering and dispersion of some
types of pollution.
The proposed technique can also be used as a tool supporting decisions about
how far the fairway should be located from the coast (or from vulnerable areas)
facing the open ocean. The probability of coastal pollution for open ocean coasts
can be reduced by shifting ship routes farther offshore. Such a shift creates addi-
tional direct costs and increases threats via offshore dangers such as considerably
higher wave loads. Ideally, such decisions should reflect a balance of environmental
concerns and related additional costs.
The entire method is computationally relatively expensive because credible
results require high-resolution eddy-resolving simulations. The complexity of
the problem increases drastically when also wave-induced drift is accounted for
(Chap. 11). At present, its use seems to be only feasible for relatively small sea
areas. However, if the calculations have been performed once for a particular re-
gion with the necessary resolution and accuracy, the resulting maps have a long
term value and can be used for various exercises in marine spatial planning. The
key contribution to the needs of society is the possibility of extracting and visual-
izing important information for decision-making that usually remains concealed in
classical methods of the analysis of ocean currents.
The above discussion has demonstrated the overall sensitivity of the optimum
location of fairways on several details of the entire approach such as the resolution of
the underlying ocean model. The sensitivity of the results on different choices of the
420 T. Soomere
constituents (in particular, on the missing or ignored aspects of ocean dynamics such
as subgrid turbulence) definitely merits more detailed attention. Further research is
obviously necessary to adequately quantify the measures of the potential benefit and
to understand the robustness and sensitivity of the technique with respect to changes
in the model options.
Index
A Age, see particle age and water age

Accident, 5–9, 12, 14, 15, 23, 284, 288, 293, Agriculture, 2, 196
321–323, 325, 327, 346, 352, 368, Agulhas Current, 17, 89
371, 377, 384, 390, 395, 401, 411, Air density, 98, 109
417, 418 Air flow, 7, 59, 198
consequences of, 1, 3–6, 8, 9, 12, 15, 285, Air mass, 226, 227, 244, 247
323, 325 Air temperature, 38, 99, 108, 109, 114, 115,
marine or offshore, 4, 7, 361, 373, 411 161, 192, 197, 295, 330
probability of, 1, 4, 8, 12, 16, 22, 24, 395 Air–sea exchange, 164, 166
ship, vii, 4, 5, 14, 15, 17, 22, 125, 321, 326, Air–sea flux, 39, 114, 160, 202, 295
368, 395, 402 Air–sea interaction, 16, 37, 191, 197
Accident site, 7, 8, 15, 19, 20, 322, 323, 377 Airy waves, 402, 404
Acidification, 2 Åland Sea, 34, 55
Acoustic Doppler Current Profiler, ADCP, 147, Albedo, 60, 330
208 Alborán gyres, 148
Adriatic Deep Water (ADW), 161 Alborán Sea, 148, 149, 160
Adriatic Sea, 21, 158, 160, 161, 168 Algae, 105, 294, 296
Advection, 23, 32, 35–38, 41, 42, 52, 113, 116, Algerian basin, 148
121, 124, 125, 144, 166, 167, Algerian Current, 137, 148
189–192, 195, 201–203, 210, 216, Algerian eddy, 148
227, 253, 256, 261, 267, 268, 332, Algorithm, 9, 10, 65, 68, 238, 252, 253, 349,
374–377, 379, 380, 383, 384, 386, 391
411 root-solving, 238
vertical, 55, 56, 123 Altimeter data, 136–138, 143
Advection equation, operator, 107, 125 Analytical solution, 78, 225, 226, 229, 230,
Advection terms, 42, 203 236, 238–240, 243, 247
Advection velocity, 256 Analytical time integration method, 235, 247
Advection–diffusion equation, 102, 106, 117, Angular velocity, 47, 87, 107
125, 241 Anisotropic, anisotropy, 15, 16, 42, 103, 182,
Advective–diffusive process, 376 197, 204, 213, 292, 320, 340, 345,
Adverse impact, 6–8, 15, 18, 23, 24, 285, 288, 375, 376, 383
292, 293, 298, 299, 306, 313, 319, Annual cycle, 33, 37, 41, 57, 137, 201
323–326, 352, 417 Antarctica, Antarctic waters, 3, 99
transport of, 321–323 Anthropogenic pressure, 2, 18, 21, 285
Aegean Sea, 135, 145, 160, 161, 168–170 Antikithira sill, 170
Aerospace industry, 66 Approximate solution, 1, 17, 20, 51, 319, 325,
Africa, African coast, 148, 160, 161 346

DOI 10.1007/978-3-319-00440-2,
422 Index
Approximation error, 227 287, 300, 307, 328, 329, 332, 334,
Archipelago, 34, 41, 186, 397 335, 348–350, 354, 399–401, 409
Archipelago Sea, 34, 55, 158 Baltic Sea, vii, 13, 14, 16, 18, 21, 22, 24,
Arctic region, 3, 92, 99 32–39, 41–51, 53–61, 85, 102, 103,
Area of reduced risk, 18, 19, 284–286, 319, 110–119, 125, 131, 133–139,
321, 323, 324, 345 141–144, 150, 154–158, 161–163,
Argo (float), 138 172, 181–185, 187–189, 191, 193,
Argos (satellite), 138, 253, 254 195, 198, 199, 201, 203, 208, 213,
Ariane (algorithm), 226, 238–240, 247 215–217, 226, 247, 251–256, 261,
Arkona Basin, 22, 34, 35, 37, 55, 143, 144, 262, 269, 270, 273, 274, 276–278,
150, 151, 156, 157, 162, 286–288, 283–288, 294, 295, 319, 320,
328 325–332, 336, 370, 371, 373,
ARPEGE-Climate model, 160 378–381, 383, 385, 397, 400, 405,
Asia, 2 418
Asymmetry, asymmetric, 9, 10, 12, 53, 200, dynamics, 42–44, 46, 53, 57, 59, 61, 106,
340, 349, 351 181, 182
Asymptotic value, 338, 340, 361, 418 geometry, 32, 33, 45, 134
Atlantic, Atlantic Ocean, 13, 16, 21, 32, 53, sub-basins, 32–34, 37, 38, 44, 45, 59, 61,
85, 99, 134–137, 148, 149, 156, 113, 144, 158, 182, 185, 199, 284,
160, 198, 331, 382 385
Atlantic water, 148, 149, 157, 160, 161, 169 SW region of, 22, 23, 34, 44, 183, 283,
Atmospheric conditions, 60, 157, 287, 389 285–287, 290, 300, 304, 313, 326,
Atmospheric data, 114, 115, 159, 161 327, 329, 331, 332, 335, 351, 353,
Atmospheric forcing, 32, 37, 57, 112, 358
114–116, 138, 159, 160, 169, 189, volume, 34, 35, 185
191, 197, 201, 286, 290 Baltic Sea Action Plan, viii
Atmospheric model, 66, 68, 114, 146, 225, Baltic Sea Index, BSI, 57
255, 295, 327, 375, 383 Baltic Sea protected areas, BSPA, 14
regional, 59, 112, 115, 256, 295, 330 BalticWay, xi
Atmospheric pressure, 57, 108, 135, 146, 198, Barents Sea, 21
253, 259, 330, 331 Baroclinic eddy, 152, 158
Autocorrelation function, 258, 264, 265 Baroclinic instability, 148, 214
Automotive industry, 66 Baroclinic mode, 43, 44, 86
Avalanche, 66 Baroclinic motion, 43, 51, 56, 86, 107, 138,
AVISO, 143 146, 154, 214
Azores high, 198 Baroclinic response, effects, 43, 51, 99, 206,
Azov Sea, 134 214
Barometer, 251, 253
B Barometric pressure, 146
Backward difference scheme, 75–77, 79–82 Barotropic forcing, 142
Backward Euler method, 76, 79, 82 Barotropic mode, 43, 44, 51, 52, 86
Ballistic law, 271, 273 Barotropic motion, 43, 51, 58, 86, 99, 107,
BALTEX, 59, 116 114, 152, 158, 206, 213, 214, 244
BALTEX Hydrological Data Centre, BHDC, Barotropic pressure, 135, 150
59, 116, 161 Barotropic surge, 381
Baltic Carrier, 372 Bathymetry, 32, 33, 45, 56, 61, 113, 122, 134,
Baltic Environment Database, BED, 59 142, 147, 151, 162, 183, 185, 186,
Baltic Marine Environment Protection 211, 216, 254, 288–290, 299, 329,
Commission, HELCOM, viii, 14, 330, 340, 341, 350, 351, 362, 380
331, 348 Batumi eddy, 137, 158
Baltic Operational Oceanographic System, Belt, see Great Belt
BOOS, 378 Belt Sea, 22, 34, 37, 55, 118, 144, 287, 288,
Baltic Proper, 22, 23, 34, 102, 116–118, 121, 290, 291, 328
143, 150, 182, 261, 270, 274, 275, Benthic invertebrate, 124
Index 423
Beta-effect, beta-plane, 43, 203, 214 bottom, 92, 98

Biharmonic friction, 110 turbulent, 96
Biodegradation, 376 upper, 92
Biogeochemical processes, state, model, 38, Boussinesq approximation, 41, 89, 106, 330,
117, 132, 149, 171, 192, 209 379
Biscay, Bay of, 137 Brackish (sea), 21, 35, 134
Black Sea, 13, 21, 131–139, 142, 145–148, Brackish (water), 39, 84, 97, 188, 287, 288
151, 153–156, 158, 160, 163–166, Braer, 401
168–170 Breton National Wildlife Refuge, 2
Black Sea inflow, 160 Brownian motion, 375, 380
Black Sea water, 151 Brunt–Väisälä frequency, 111, 136, 155, 197
Blocking event (of flow), 147 Bryan–Cox primitive equation model, 159
Bloom, 142, 145 Bryan–Cox–Semtner type model, 106, 295,
Boltzmann, Stefan, see Stefan Boltzmann 329
Bornholm, 332 BSHcmod model, 155, 378
Bornholm Basin, 34, 35, 55, 102, 118, 143, Bulk formulae, 295
150, 157, 161, 261, 287, 328, 399 aerodynamic, 161, 165
Bornholm Channel, 55, 156, 157 empirical, 108, 114
Bornholm Deep, 55, 117 Buoy, 59, 197, 212, 253, 254, 331, 334
Bosphorus, see Bosporus Buoyancy, 92, 125, 135, 151, 153, 166, 183,
Bosporus, 145–147, 150–152 376, 386
Bosporus plume, 152 neutral, 56, 189
Bosporus Straits, 133, 134, 141, 142, 145, 146, Buoyancy current, 205, 214
151, 153, 161, 164 Buoyancy effect, force, 150, 286, 376, 377
Bosporus undercurrent, 153 Buoyancy flux, 150, 164–166, 169, 170
Boston Harbour, 10, 11, 17, 325 haline, 165, 166, 169
Bothnia thermal, 165, 166, 169
Bay of, 33, 34, 38, 50, 53, 118 Buoyancy gradient, 204
Gulf of, 32, 34, 36, 37, 55, 117, 182, 193 Buoyancy–viscous regime, 376
Sea of, 34, 53, 55, 118, 158, 199, 284
Bottom drag coefficient, 48, 98, 111, 112 C
Bottom friction, 32, 42, 45, 48, 49, 51, 83, 93, Cádiz, Gulf of, 154
98, 111, 116, 141, 150, 151, 203, Calm conditions, 8, 38, 273, 286
216 Calm water speed, 397
Bottom roughness, 111 Camarinal Sill, 148, 149
Bottom sediment, 2, 124 Cargo, 5, 10, 21, 283, 326, 348, 395
Bottom slope, 287, 289 hazardous, 12, 102, 103
Bottom stress, 98, 108, 379 Cartesian coordinates, 87, 330, 336
Bottom water, 36, 37, 83, 117, 143, 144, 151, Catchment area, 112, 114, 115, 132, 134, 185
183, 213 Cauchy problem, 296
Boundary condition, 43, 48, 71, 72, 77, 78, 97, Celerity, 403
98, 107, 109, 113, 114, 116, 117, Central difference method, 75
203, 233, 293, 295, 296, 331, 332, Centrifugal force, 84
340, 359, 379, 382 Centripetal acceleration, 87
bottom, 97, 108, 232 CFD, see computational fluid dynamics, CFD
flux, 110 Characteristic current, 206
free-slip, 97, 379 Chemical model, 151
kinematic, 97, 98 Chemical oceanography, 200
lateral, 99, 107, 113 Chemical processes with oil, 286, 375–377
no-flux, 97, 99, 118 Chemical warfare agents, 327
no-slip, 98, 107, 330 Chemicals, chemical pollution, 3, 103, 288,
open, 59, 60, 99, 106, 113, 295, 329 319
Boundary layer, 89, 116 Chernobyl, 377
atmospheric planetary, 108 Chlorofluorocarbons, CFC, 153
424 Index
CIL, see cold intermediate layer, CIL Climatologically valid solution, 299, 302, 337,
Circularly polarized motions, 334 346, 353, 385, 418
Circulation, 16, 18, 32, 54, 59, 135–139, 144, Cloudiness, 108, 114, 161, 295, 330
148, 154–156, 158, 159, 169, 181, Coast, 12, 13, 15, 37, 42, 47, 49, 53, 55, 68,
183, 184, 188, 190, 200, 204–208, 121–123, 149, 158, 195, 196, 198,
210, 216, 217, 291, 294, 310, 329 200, 203, 205, 206, 208, 211, 303
anticyclonic, 209 Coastal area, 158, 159, 197, 200, 204, 216,
atmospheric, 57, 143, 169, 226, 227, 247 284, 285
baroclinic, 43, 45, 46, 56, 57, 217 Coastal crash, 8, 12, 13
barotropic, 43 Coastal eddy, 158
basin-wide, 151, 210 Coastal engineering, vii, 4
bottom, 56, 158 Coastal environment, 19, 371
counter-clockwise, 53, 156, 205 Coastal hit, 19, 121, 123, 299, 302–308, 311,
cyclonic, 32, 37, 53, 54, 57, 135, 136, 150, 313, 332, 334–339, 342, 344,
164, 184, 205, 208, 209, 211, 215, 346–361, 368, 369, 383, 384, 390,
216, 288, 289, 305, 310, 345, 355 418, 419
deep water, 43, 55 Coastal jet, 162, 208, 289, 302, 408
large-scale, 118, 198, 269 Coastal pollution, 5, 13, 17, 22, 306
layered, 217 current-driven, 346
mean, 46, 53, 54, 57, 102, 184, 205, 206, Coastal water, 2, 6, 8, 135
211, 215–217, 288, 302, 308, 331 Coastal zone, 2, 8, 12, 13, 34, 35, 41, 44, 45,
quasi-stationary, 214 60, 102, 195, 196, 307, 339
residual, 204, 207 Coastline, 12, 33, 140, 186, 195, 232, 325, 339
secondary (wind-generated), 45 Coccolithophorids, 145
subsurface, 138 Cold intermediate layer, CIL, 164, 166, 168
surface, 45, 50, 53, 56, 102, 148, 156, 158, Cold intermediate water, CIW, 164, 165
169, 205, 217 Collision (of ships), 4, 5, 9, 12, 369, 372, 411
thermohaline, 83, 84, 99, 131, 135, Coloured dissolved organic matter, CDOM,
139–141, 148, 154, 156, 159–161, 144
163, 168, 184 Computational fluid dynamics, CFD, 65–71,
vertical, 55, 118, 148, 153, 155, 159 396
wind-driven, 45, 46, 56, 57, 188, 216, 217 Concentration (lagoon) type basin, 131, 145
Circulation dynamics, 181, 185, 197, 202, 204 Conservation of mass, 41, 56, 70, 87, 106, 225,
Circulation model, 15, 18, 20–23, 32, 49, 227, 232, 233, 244
56–60, 65, 96, 99, 102, 104–106, Conservation of momentum, 41, 70, 87, 106
117, 121, 123, 149, 184, 204, 217, Contaminant, 196, 206, 217, 286, 294, 296,
225, 226, 230, 241, 252, 253, 255, 346, 417
272, 273, 277, 283, 293–298, Continental slope, 134–137, 151, 158, 167,
300–302, 306, 307, 309, 313, 168
325–335, 340, 360, 378, 380, 383, Continuity equation, 41, 42, 90, 232, 233, 330
391, 402, 405, 406 Convection, 38, 39, 55, 90, 152, 155, 157, 159,
eddy-permitting, 288, 328, 329, 340, 358, 168–170, 190, 192
359, 361, 418 thermal, 32
eddy-resolving, 305, 319, 329, 361, 418, winter, 160, 161, 169
419 Coriolis acceleration, 32, 41–43, 45–47, 50,
high-resolution, 1, 21, 334, 378, 381 87, 88, 202, 203, 216
operational, 383 Coriolis effect, force, 84, 89, 106, 203
Circulation pattern, 99, 158, 181, 183, 184, Coriolis frequency, 88
187, 198, 204, 205, 211, 216, 335 Coriolis parameter, 42–44, 107, 136, 202–204,
CIW, see cold intermediate water, CIW 289
Climate model, 66, 85, 98, 160 Correlation time, 244
Climatological data, 36, 98, 99, 110, 113, 159, Corsica, Strait of, 168
160 Cost function, 5, 334, 369, 390–394, 410
Index 425
Cost of the consequences of an accident, 4, 5, semi-persistent, vii, viii, 1, 9, 17, 18, 21,
8, 11, 13–15, 19, 286, 321–325, 23, 283, 286, 290–294, 299, 302,
344, 346, 347, 352, 368, 395, 418 308, 309, 312, 313, 321, 323, 327,
Counter-current, 132, 162 368, 369, 418
Coupled ocean wave and circulation model, ‘Current rose’, 53, 55
405 Current speed, 47–49, 53, 55, 57, 103, 147,
Courant number, 83 206–208, 214, 277, 289, 301, 308,
Courant–Friedrichs–Lewy (CFL) condition, 309, 377, 386
83, 86 Current Spy, 212
Cretan Arc straits, 168, 170 Current velocity, 41, 44, 46, 49, 52, 53, 103,
Cretan Deep Water, CDW, 168 111, 209, 211, 294
Cretan Intermediate Water, 161
Cretan Passage, 168 D
Crete, 169 Damage to the environment, 5, 7, 11, 13, 17
Crimea Peninsula, 166 Danger to the environment, 12, 322, 419
Crude oil, 371, 372, 379 Danish Islands, 143
Current, 5, 9, 10, 16–18, 21, 22, 24, 41, 48, 50, Danish Meteorological Institute, DMI, 23, 155,
51, 53, 54, 60, 66, 83, 84, 89, 99, 331, 378, 379, 396
123, 145–147, 160, 183, 184, 202, Danish Straits, 32, 34, 35, 37, 43, 59, 111, 113,
205–209, 212, 214, 216, 227, 243, 114, 139, 142, 151, 213, 286, 287,
252, 256, 259, 261, 270, 274, 276, 290, 291, 295, 302, 328, 331, 397
Dardanelles, 133, 141, 145, 146
278, 283–286, 288, 289, 291, 296,
Darss Sill, 118, 119, 156, 157, 163, 291
302
Data assimilation, 98, 115, 159, 171
bottom, 35, 55, 92, 97
Data Assimilation System, DAS, 59
buoyancy-driven, 329
Dawson’s integral, 236
coastal, 17, 53, 148, 160, 200, 289, 291,
Dead Sea, 109
307, 312
Debris, 286, 296
deep, 84
Decision support system, 9, 313, 338, 396, 419
density-driven, 133, 146, 207, 385 Decoupling of layers, 189, 191, 201, 207, 213,
drift, 45, 206 216, 310, 381
jet, 17, 135, 285, 289, 299, 309, 320, 333, Deep basin, 134
418 Deep ocean, 48, 51, 92, 134–136, 148, 154,
large-scale, 43, 135 155, 203, 216, 290
mean, 53, 103, 208 Deep water, 35, 42, 55, 56, 99, 113, 117, 138,
quasi-stationary, 21 143, 144, 148, 149, 153, 155, 157,
subsurface, 253, 254, 274, 275, 289, 292, 164, 168, 169, 189, 192, 230
311, 327 Deep water formation, 131, 152, 159
surface, 8, 16, 20, 23, 45–49, 52, 53, 84, Deepwater Horizon, 6
102, 124, 132, 140, 147, 156, 157, Delay parameter, 152
200, 203, 205, 206, 208, 210, 212, Delta Works, 4
215, 216, 283, 284, 286, 290–293, Denmark, 32
301, 320, 323, 324, 327, 332, 340, Denmark’s Climate Centre, DKC, 381, 384
345, 417, 418 Density, 32, 35–37, 39–41, 43, 45, 51, 67, 84,
transient, 45, 216 86, 89, 91–93, 96, 97, 99, 104, 107,
unidirectional, 208 108, 111, 135, 141, 164–166, 188,
wave-driven, 405, 408, 409 193, 197, 202, 216, 246, 329, 362,
wind-driven, 53, 207, 275, 386 376, 377, 379, 382, 385
Current measurements, 43, 46, 149, 184, 196, Density gradient, 45, 59, 86, 97, 113, 144, 188,
206, 208, 209, 216 195, 204, 216
Current pattern, 285, 310 Density jump, 192
concealed or hidden, 18, 21, 23, 285, 312, Density of oil, 372
321, 323 Depth gradient, 407, 408
favourable, 17, 18 Design speed, 396
426 Index
Detrainment, 153 Drag coefficient, 98, 141, 404

Diapycnal mixing, 92, 135, 149, 213 Draught, 396, 397
Dicothermal layer, 37–39, 192 Drift, 13, 16, 49, 56, 92, 106, 138, 200, 207,
Dietrich Center for Air Sea Technology, 239, 285, 292
DieCAST, 155, 159 current-driven, 385, 387, 401
Difference equation, scheme, see finite ice, 41, 49, 199, 200, 253
difference scheme subsurface, 22
Differential equation, 67, 69, 70, 95, 225, 226, surface, 23, 291, 296, 331, 349
229, 231, 234, 236, 296, 373 wave-driven, 296, 401, 402, 404, 405,
partial, PDE, 65, 69, 71, 73, 74, 78–80, 83, 407–409, 411
107, 111 wind-driven, 212, 387, 401, 404
Diffusion, 56, 92, 113, 121, 155, 202, 227, Drift speed, 49, 200
240–242, 244, 256, 332 Drifter, 196, 197, 200, 212, 252–254, 257–259,
horizontal, 116, 121, 195, 295 264, 268, 270, 272, 273, 275–278,
Laplacian, 243 297, 300, 307, 320, 333, 374
molecular, 92 GPS/GSM, 251, 253, 270–274, 277, 278
numerical, 78, 80, 227, 243 pair of, 244, 251–254, 257, 259, 264–266,
vertical, 155 269–275, 277
Diffusion coefficient, 155 simulated, 251, 252, 255, 256, 260–269,
Diffusion equation, 125 275, 276, 278
Diffusivity coefficient, 202 surface, 53, 241, 244, 251–253, 258, 261,
Dijkstra’s algorithm, 351, 369 270, 278, 300
Dilution (estuarine) type basin, 131 SVP, 251–270, 273–278
Direct problem, 322 triplet of, 254, 264
of the propagation of passive tracers, 283, Drifter experiments, 212
286, 293 Drifting buoy, 18, 293
Direct problem of oil propagation, 323 Drogden Sill, 118, 119, 156, 163
Disaster impact model, 368
Drogue, 212, 251, 253, 274, 276
Dispersion, 92, 252, 259, 273, 372, 378, 380
Dumping (of material into sea), 3, 7, 284, 327
absolute, 244, 251, 252, 257, 262, 263,
DWD model, 59
266–268, 275–277, 333
Dye, 197
horizontal, 379, 380, 382
Dynamical equations, 42
Lagrangian, 225
Dynamical relocation of tugboats, 13
numerical, 116
Dynamics, 143, 181, 190, 200, 206, 210
relative, 241, 244, 251, 252, 257, 259, 265,
basin-scale, 136, 152, 206, 216
266, 268–270, 275–277
turbulent, 244 current-driven, 352
vertical, 380 eddy-scale, 136
Dispersion of radionuclides, 247 multi-layered, 278, 291
Dissipation, 84, 85, 110, 150, 156, 196, 403
Dissolved substance, matter, 8, 91, 286, 296 E
Divergence (of radiation stress), 403 e-folding time, 47
Divergence of trajectories, 297, 332 Earth, 83, 87, 99, 107, 136
Divergence (of turbulent flux), 107 Earth Sciences (GES) Data and Information
DMI, see Danish Meteorological Institute, Services Center (DISC), 145
DMI Earth’s rotation, 41, 84, 87, 102, 151, 202, 205
DMI/BSHcmod model, 155, 161, 331, 332, Eastern Mediterranean basin, 160
361, 378 Eastern Mediterranean Deep Water, EMDW,
DMI’s operational circulation model, 380 168
Double diffusion, 92 Eastern Mediterranean Transient, EMT, 131,
Downscaling, 59, 85, 115, 160, 277 139, 145, 160, 168–170
Downwave direction, 18 ECMWF, 160, 165
Downwelling, 45, 195, 196, 209, 216, 302 Ecological risk, 103
Downwind direction, 18 Ecologically sensitive area, 3, 102, 125
Index 427
Ecosystem, 3, 5, 6, 135, 182, 213, 285 differential, see differential equation

marine, 2, 3, 5, 21 Equation for the inertial motion, 47
Ecosystem modelling, 171, 196 Equation of motion, 42, 43, 47, 52, 69, 70, 86,
Eddy, 136, 150, 299, 309, 341, 409 226, 227, 239
anticyclonic, 158, 169 Equation of state, 41, 91, 107, 108, 112, 202,
Eddy diffusivity, 95, 107, 110, 112, 118, 150, 330
196, 197, 201, 331, 379, 380 Equations for heat and salt budget, 41, 330
Eddy viscosity, 42, 48, 49, 107, 110, 112, 202, Equator, 83, 88, 99
203, 241, 243 Equidistribution of costs, 19
Eigen-oscillation, 146 Equilibrium state, 99, 206
Ekman, V.W., 47 Equiprobability line, 325, 326, 344, 345, 347,
Ekman depth, 48, 49 352, 354–358
Ekman drift, 45, 47, 206, 211, 214, 216, 306 ERA-40 reanalysis, 59, 99, 115, 160, 256, 275,
Ekman layer, 48–50 277, 295, 330
Ekman number, 43 ERA-75 reanalysis, 277
Ekman pumping, 153, 164, 165 ERA-Interim reanalysis, 99, 115, 233, 234,
Ekman spiral, 48–50, 53, 376 277
Ekman theory, solution, 47, 49, 52 Error function, 236
Ekman transport, 48, 49, 88, 195, 289, 310 Error (in the model), 60, 66, 67, 171
Elasticity theory, 70 Error sensitivity, 70
EMT, see Eastern Mediterranean Transient Espartel Sill, 148, 149
Emulsification, 372, 377, 379, 401, 410 Estonia, 33, 103, 182, 185, 197, 212, 259, 340,
Emulsion, 372, 377, 380, 401 344, 345, 385, 392, 400
Energy budget, 39, 60, 188, 191, 201, 216 Estonian archipelago, 253
Energy dissipation, 273 Estonian coast, 186, 195, 200, 208–210, 311,
Energy input scale, 269 387, 389
Engineering solution, 1, 320, 323, 324, 338, Estuarine circulation, dynamics, transport, 35,
353 55, 142, 158, 161, 213, 305, 328
English Channel, 381 Estuary, 68, 161, 187, 188, 328
Ensemble approach, 102, 121, 294 Estuary-type basin, 131, 138, 145, 155, 166,
Ensemble average, 121, 123, 263, 265, 269 181, 210
Enstrophy cascade, 269 EU Strategy for the Baltic Sea Region, viii
Entrainment, 151, 153, 157, 216 Euler Forward method, 238
Environmental benefit, 18, 351–354, 360, 389, Eulerian current, 308, 311, 403, 405
417, 418 Eulerian time stepping scheme, 380
Environmental criterion, 1, 10, 15, 321, 353, Eulerian tracking of pollution propagation, 327
358, 368, 419 Eulerian velocity, 18, 21–23, 226, 286, 290,
Environmental damage, loss, 5, 10, 15, 368, 291, 293, 299, 304, 308, 310, 312,
374 326, 336, 418
Environmental impact assessment, 10 Eulerian (viewpoint, specification,
Environmental management, 1, 9, 16, 19, 21, description), 18, 67, 86, 102–106,
183, 285, 312, 313, 319, 321, 324, 120, 121, 125, 225, 226, 286, 293,
346 294, 312
Environmental pressure, vii Eulerian–Lagrangian method, 125
Environmental risk, vii, 1, 5, 8, 12, 13, 23, 24, Europe, European, 2, 21, 22, 24, 131, 132,
278, 283, 286, 302, 303, 319, 320, 135–137, 139, 140, 142, 151, 154,
322–325, 338, 345–347, 351, 352, 171, 295, 330
354, 356, 358, 369, 383, 392, European Centre for Medium-Range Weather
417–419 Forecasts, ECMWF, 98, 234
minimizing of, 293 European Space Agency, ESA, 7
remote, 1, 285 European Union, 2
Envisat, 136 Evaporation, 16, 60, 83, 109, 114, 124, 131,
EOS-80, 190 133, 134, 146, 169, 187, 188, 330,
Equation 372, 379, 380, 410
428 Index
Exchange flow, 135, 142, 143, 149, 150 Finnish Institute of Marine Research, 40, 208
Exhaust emission, 5, 6, 9, 390, 396 Fish eggs, larvae, 285, 293
reduction of, 368, 396 Fish stock, 5, 13
Exponential law, 268 Fishing ground, 24
Extinction length, 110, 112 Fishing industry, 2
Eyjafjallajökull, 377 Fixed arrival time, 396
Fixed mooring, 18, 149, 293
F Floating object, 16, 49, 211, 296
f -plane approximation, 43, 203 Flow
Fairway, 5, 6, 8, 10, 12, 13, 17, 19, 23, 24, 102, cardiovascular, 66
121, 123, 254, 288, 319, 323, 325, critical, 141
347, 352, 390, 391, 397–401, 409, eddy-containing, 299, 304
410, 418, 419 eddy-dominated, 299, 309, 312
environmentally optimized, 368–370, 383, gravity, 66
392, 396 hydraulically controlled, 141
optimization of, 102, 103 inviscid, 140
optimum, 1, 9, 10, 12, 15, 16, 18–22, 24, isotropic, 53, 306
125, 286, 319, 322, 323, 326, laminar, 65, 91
345–357, 359–362, 368–371, 385, layered, 50, 209
389, 390, 392, 394, 400, 401, 409, nonrotating, 140
410, 419 planetary-scale, 44
Fairway design, 6, 9, 10, 12, 22–24, 139, 319, pyroclastic, 66
325, 349, 368–371, 385, 389, 390, rectilinear, 299
392–402, 409–411
rotating, 244
adaptive, 368
subcritical, 141
Fairway performance, 368, 399, 401, 409
subsonic, 65
Fairway relocation, 10, 11, 397
supercritical, 141
Fårö Deep, 55
supersonic, 65
Fast ferry, 12, 288
turbulent, 65, 92, 244, 269, 276
Fast Fourier Transform, 259
two-layer, 145
Fatigue damage, 24
zonal, 289
Fay equations, 380
Federal Maritime and Hydrographic Agency, Flow direction, 47, 51, 208, 209, 211, 246,
BSH, 378, 379 290, 307, 310
Fehmarn Belt, 143, 151, 156 Flow relaxation, 99
Ferrel cell, 234 Flow speed, 51, 143, 208
Fetch length, 408 Fluid dynamics, 20, 65, 67, 69, 86, 87, 226
Filament, 136 Flux-corrected, monotonicity preserving
upwelling, 195 transport (FCT) scheme, 116
Finite difference (FD) method, 69–73, 75, 76, Food web, food chain model, 124
83, 95 Forcing conditions, 182, 290, 293, 295, 296,
Finite difference scheme, 74–76, 78, 79, 354, 359
226–228, 232, 247, 257, 258, 302 Forcing factors, 15, 16, 32, 183, 204
Finite element (FE) method, 69–71, 95, 97, Forcing field, 70, 184, 290, 338, 374, 375
104 Forcing mechanism, 83, 97, 142, 166
Finite Volume Coastal Ocean Model, FVCOM, Formal safety assessment methodology, 17
97 Fortran 77, Fortran 95, 226, 227
Finite volume (FV) method, 69–71, 95, 97 Forward difference method, 74–76, 79, 80, 82
Finland, 182, 185, 200, 212, 339, 344, 385, Forward Euler method, 76, 77, 80, 81, 226
387 Fraction (of oil), 372, 379, 380
Finnish archipelago, 341, 349, 360 FRAM ocean model, 226
Finnish coast, 195, 198, 205, 208, 209, 214, Free surface, 96, 98, 99, 234, 295
291, 387, 389 Freezing, 16, 37, 199, 200
Finnish Environment Institute, SYKE, 189 Freezing point, 37, 39, 192, 330
Index 429
Fresh water, 35, 36, 39, 43, 45, 83, 86, 115, Gibraltar, Strait of, 133, 134, 141, 142,
152, 160, 164, 165, 169, 181, 183, 148–151, 161
187, 188, 193, 203, 205, 216, 297, Glacier, 83
305, 328 Glider, 138, 171
Fresh water budget, 32, 45, 57, 134, 137, 216 Globe Asimi, 371
Fresh water deficit, surplus, 134, 187, 189 Gogland, 186, 194, 338, 341, 349, 356–359,
Fresh water flux, 109, 132, 160, 188, 196, 290 400, 401, 409
Friction, 32, 45, 47, 141, 144, 146, 203, 216 Gotland, 13, 34, 259
horizontal, 42, 43, 52, 110, 203 Gotland Basin, 36, 37, 55, 118, 144, 215
interfacial, 150, 200 Eastern, 34, 55, 102, 118, 143, 157, 193,
internal, 42 261, 287, 328, 399
shallow water, 396 Northern, 34, 55, 56, 102, 143, 158,
vertical, 42, 43, 203 182–186, 192, 197, 200, 201, 203,
Friction coefficient, 47, 379 207, 209, 213–215, 246, 253, 261,
linear, 239 287, 288, 300, 305, 328, 399
Front, 32, 44, 181, 204, 209, 210, 213, 290 Western, 34, 55, 102, 118, 143, 261, 328,
density, 136 399
Frontal system, 143 Gotland Deep, 40, 55, 117, 157, 189
Froude number, 43, 140, 141 Gotland Sea, 34, 35, 46, 53, 55, 185, 188, 189,
Fu Shan Hai, 372 201, 206
Fuel consumption, 9, 10, 16, 24, 103, 368, 369, GPS sensors, 253, 254, 271
390, 396–401, 410 Gradient operator, 42, 202
Fukushima, 377 Graph (mathematical), 55
Grashof number, 150
G Gravity acceleration, 41–43, 85, 87, 89, 107,
Galicia, 372 164, 202, 403
Gauss elimination, 82 Gravity current, 90, 140, 148, 152, 157, 164,
Gaussian distribution, 242, 244 171
GCM, see General Circulation Model, GCM Gravity force, 83, 89, 90
Gdańsk Bay, 34, 102, 143, 261, 287, 328, 399 Great Belt, 32, 51, 52, 133, 141, 142, 144, 156,
Gdańsk Deep, 55 157, 161–163, 294, 350
General Circulation Model, GCM, 66, 70, 83, Greenland, 379
85, 226–228, 230–232, 234, 235, Grid
239, 240, 243, 247 adaptive, 104, 121, 153, 154, 171
Genetic connectivity of marine populations, B, Arakawa B, 107, 228
247 C, Arakawa C, 228, 232, 330, 379
Geographical coordinate system, 68 Cartesian, rectangular, rectilinear, 68, 225,
GeoHydrodynamics and Environment 227, 228, 230
Research, GHER, 154 curvilinear, 68, 225, 228, 230
Geophysical flow, 44, 134, 150, 270, 379 nested, 60, 68, 69, 380
Geophysical fluid dynamics, 44, 68, 247 non-rectangular, 227, 230
Geopotential coordinates, 95 regular, 68, 104, 114, 124
Geopotential height, 246 spherical, 230, 378
Geostrophic balance, 50, 52, 89, 150, 209 structured, 68–70, 151
Geostrophic eddy, 84, 209 uniform, 72, 73
Geostrophic flow, 50–52, 89, 239 unstructured, 68, 70, 97, 104, 121, 153,
Geostrophic transport, 137 154, 171
Geostrophic velocity, 239 Grid box, 226, 228–232, 235–237, 245, 256,
Geostrophic wind, 58, 114 261, 265, 276
Geothermal heat, 83 Grid cell, 68, 96, 116, 226, 297, 299, 300, 303,
German Bight, 144, 379 310
German Federal Maritime and Hydrographic Grid resolution, 68, 69, 77, 85, 97, 114, 154
Agency, BSH, 155, 331 Grid size, 121, 168, 204, 301
430 Index
Grid step, 44, 228, 278, 288, 328, 329, 333, HELCOM, see Baltic Marine Environment
361 Protection Commission
Grid wall, 229, 231, 234 Helsinki, 206–208, 288, 291, 328, 339, 340,
Grid-based method, 67, 68 349, 356, 357, 359
Grid-free method, 67 Hemisphere
Grounding, 4, 5, 9, 13, 369 northern, 47, 49–51, 88, 205, 289
Group speed, 403, 404 southern, 88
Gulf of Cádiz, see Cádiz, Gulf of Hibler-type sea ice model, 295, 379
Gulf of Finland, 20, 22, 23, 33–35, 37, 43, 44, High Resolution Limited Area Model,
53, 55, 61, 117, 139, 158, 181–201, HIRLAM, 59, 161
203–217, 246, 251–254, 261, 268, High Resolution Oceanographic Model of the
270, 271, 277, 278, 283–288, 291, Baltic Sea, HIROMB-BOOS, 139,
292, 294–313, 319, 325–333, 155, 253, 278, 300, 331, 378
335–345, 347–349, 351–360, 362, High-speed vessel, 5, 12
368, 370, 371, 379–389, 392, Hiiumaa, 186
397–402, 404–407, 409, 411, 418 Himmerfjärden Estuary, 117
geometry, 198, 288 Hindcast, 57, 59, 60, 256
volume, 185 HIRLAM, see High Resolution Limited Area
Gulf of Lions, 159–161 Model
Gulf of Mexico, 12 HIROMB, see High Resolution Oceanographic
Gulf of Riga, 34, 35, 37, 53, 55, 183, 199, 284 Model of the Baltic Sea,
Gulf Stream, 17, 53, 89, 205 HIROMB-BOOS
Gyre, 84, 88, 89, 135, 136, 158, 159, 288, 290, HIROMB–BOOS Model, HBM, 378, 381, 383
305 Humidity, 161, 240
anticyclonic, 50, 160, 184, 205, 306, 310, relative, 114, 330
349, 355 specific, 108, 114, 233, 246, 295, 330
cyclonic, 102, 160 Hybrid approach, 121, 125
subtropical, 88 Hybrid Coordinate Ocean Model, HYCOM,
97, 154
H Hybrid coordinates, 97, 154, 232, 233
Hadley cell, 234 Hybrid method, 95, 125
Haline conveyor belt, 55, 56, 157 Hydraulic assumption, 140, 144, 147
Halocline, 32, 35–39, 41, 48, 55–57, 59, 102, Hydraulic control, 147, 149, 150, 152
116, 119, 152, 153, 188–192, 213, Hydraulic jump, 141, 149
214, 216 Hydraulic resistance, 201
permanent, 189 Hydrocarbons, 372, 373, 377
secondary, 36, 38 Hydrography, 32, 59, 113, 158, 181–183, 216
Halskov Rev, 52 Hydrological model, forcing, 116
Hamburg, 378 Hydrostatic approximation, 42, 90, 106, 202,
Hanko, 185, 212 330
Hanko peninsula, 208 Hydrostatic balance, 50, 90, 96, 140, 379
Harbour, 24, 102, 103, 287 Hydrostatic model, 155
Harmonic diffusivity coefficient, 110 Hydrostatic pressure, 35, 213
Harmonic viscosity coefficient, 110 Hypersaline, 109
HBM, see HIROMB–BOOS Model, HBM Hypoxia, 143
Headland, 289
Heat budget, 201 I
Heat flux, 39, 57, 60, 91, 92, 99, 108, 109, 112, Ice conditions, 197, 199–201
166, 192, 197, 201, 202, 331 Ice cover, 8, 12, 39, 41, 49, 50, 57, 83, 111,
latent, 109, 112, 114, 201 181–183, 193, 199–202, 206, 259,
net, 166, 201 261, 284, 286, 292, 297, 306, 329,
sensible, 112, 114, 201 330, 377, 379
surface, 39, 108, 137, 161, 165 Ice formation, 200
vertical, 330 Ice melting, 16, 37, 190, 192, 200, 201, 330
Index 431
Iceberg, 16, 411 Joint Panel on Oceanographic Tables and

Icelandic low, 198 Standards, JPOTS, 91, 108
Ill-posed problem, 322 Jokioinen, 201
Impact of adverse substances, 284, 292 Jump layer, 32, 195
Impact of wind and waves, 1, 6, 15, 17, 20, Juvenile fresh water, 188
278, 285, 296, 302, 320, 332,
409–411 K
Incompressibility, 41, 85, 90, 232, 233, 379 k–l scheme, 95
Indo-Pacific, 230 k–ε turbulence scheme, 95, 197, 295, 329
Inertial circle, 47 k–ω turbulence scheme, 378, 379
Inertial force, 42, 46, 84, 141, 203 Kapu, tabu, 2
Inertial oscillation, 43, 45–47, 102, 203, 205, Karpathos sill, 170
214, 216, 225, 238–240, 247, 261, Kassos sill, 170
262, 264, 270, 275, 295, 299 Kattegat, 21, 32, 34, 37, 55, 103, 106,
Inertial period, 46, 47 112–114, 118, 134, 141–144, 157,
Inflow, 32, 35, 37, 43–45, 55–57, 60, 99, 111, 162, 193, 255, 286–288, 290, 295,
113, 114, 117–119, 131, 133, 140, 329
143, 144, 146, 148, 150, 152, 156, northern, 113, 144
161–164, 183, 187, 188, 190, 203, southern, 144, 291, 310
207, 209, 211–216, 287, 290, 302, Kemi, 33
311, 313, 326, 335, 353, 385 Kerch, 166
Initial condition, 59, 71, 72, 77, 78, 97, 111, Kerch Strait, 166
116, 117, 229, 231, 236, 238 Keri, 341
Insolation, 379 Kerosene, 372, 379
Instability, 92, 136, 140 Kiel Channel, 286, 287
Institute of Oceanographic Sciences, Deacon
Kinematic viscosity, 91
Laboratory, IOSDL, 226
Kinetic energy, 92, 197, 337, 338
Inter-basin exchange, 132, 134, 148
eddy, 252
Intermediate water, 92, 131, 139, 148, 159,
turbulent, 91, 94, 110, 150, 196
161, 163, 164, 166
Knorr, 153
International Gulf of Bothnia Year 1991, 55
Knudsen’s formula, 207
International Maritime Organization, IMO, 2,
Kolmogorov length scale, 84
3, 13, 17, 21, 182, 284, 395
Kolmogorov’s law, 269
Intrusion, 153, 162
Korsör, 52
Inverse modelling, 149
Inverse problem, 20, 24, 312, 319, 322, 417 Kotka, 185, 186, 199, 206, 208, 346, 358, 387
Inverse problem of oil spill propagation, 20, Kotlin, 186, 201
370 Kronecker’s delta, 93, 403
Inverse problem of pollution propagation, 1, Kunda, 341
17, 139, 293, 319, 369 Kuroshio, 17, 53
Invertible model, 20, 322 Kymi, River, 187
Ionian basin, 138, 168, 169
Iridium (satellite), 253, 254 L
Irreversibility, 92 Lagrangian correlation, 244
Isobar, 43, 51, 89 Lagrangian (eddy) diffusion parameterization,
Isobath, 151, 152, 200, 209 256, 261, 265
Isohaline, 190 Lagrangian (eddy) diffusivity, 244
Isopycnal, 43, 95–97, 164, 214, 232 Lagrangian flow, 18
Isopycnal coordinates, 96, 97, 154 Lagrangian integration, 243
Isopycnal surface, 84, 92, 166 Lagrangian model, 253, 268, 276, 278
Lagrangian particle, 125
J Lagrangian propagation, 18, 293
Japan, Sea of, 132 Lagrangian statistics, 257
Jason-2, 136 Lagrangian surface drifter, 369
432 Index
Lagrangian time scale, 257–260, 264, 265, Layered structure, 32, 37, 56, 192, 207
267, 268, 276 Least-squares fit, 108, 110, 146
Lagrangian trajectory, 21–23, 120, 139, 226, Lee wave, 149
227, 243, 244, 252, 255, 278, 286, Length scale, 44, 84, 90–95, 136
293, 319, 323, 326, 332, 359, 360, Levantine basin, 148, 160, 161, 168, 169
369, 374, 417 Levantine Intermediate Water, 159, 161, 169
Lagrangian trajectory model, 225 Lifetime, 18, 92, 195, 196, 285, 292, 313
Lagrangian transport, 1, 18, 20, 283, 296, 309, Ligurian Sea, 300
313 Linear approximation, 297
Lagrangian velocity, 18, 257, 258, 262, 264, Linear interpolation, 70, 226, 229, 231, 235,
275, 293, 299 239, 243
Lagrangian velocity autocorrelation, 257, 258, Linear regression, 206
262, 264, 268, 276 Linear stress formula, 98
Lagrangian (viewpoint, specification, Linear trend, 310
description), 18, 67, 86, 102, Linear wave theory, 402
104–106, 124, 125, 197, 225, 226, Lines of characteristics, 71, 72, 78, 80, 81
241, 252, 278, 293, 296, 320, 332, Lin’s law, 269, 270
336 Litter, 286, 293, 296
Lahemaa, 291, 338–340, 349, 359 Little Belt, 32, 141, 142, 156
Lake Michigan, 310 Local jet, 44, 204, 214, 308
Landsort Deep, 34, 55, 117 Local truncation error, LTE, 74–76
Langmuir current, 275 Longitude, 68, 99, 107, 114, 185, 186, 228,
Large-eddy simulation (LES) scheme, 95 246, 261, 345, 347, 356, 384, 391
Latitude, 39, 42, 45, 47, 68, 88, 107, 114, 197, Loss of propulsion or steering, 13
202, 228, 239, 261, 295, 347, 355, Lost container, 16, 285, 286, 296, 319
356, 384, 391 Louisiana, 2
Lax equivalence theorem, 79 Luga, River, 187
Layer Luga Bay, 356, 400
bottom, 32, 35–37, 39, 51, 56, 155, 161,
168, 188–191, 208, 213, 216 M
deep, 49, 151, 153, 190 M.I.T. general circulation model, MITgcm, 96
lower, 37, 38, 45, 50, 51, 54–56, 147, 150, Major Baltic Inflow, MBI, 113, 143, 157, 161
184, 189, 191, 195, 201, 209, 212, Marine ecosystem services, vii
213, 216, 381 Marine environment, vii, 2, 3, 6, 8, 13, 16–18,
subsurface, 157, 200, 207, 211, 216, 277, 43, 60, 182, 293, 294, 320, 332,
291, 292, 305, 312, 321, 379, 381 368, 370, 371, 382
surface, 8, 15, 21, 32, 36–39, 42, 44, 45, 47, Marine optics, 184
49, 51, 55, 56, 89, 92, 113, 118, 121, Marine pollution, 6, 13
151, 161, 164, 166, 170, 188–193, Marine protected area, MPA, 2, 7, 13, 14, 324
195, 200–203, 206, 208, 213, 214, Marine safety, 145, 182
216, 247, 273, 296, 309–311, 313, Marine traffic, 285, 348, 368, 395, 396, 400,
321, 323, 331, 332, 349, 404, 405 411
upper, 32, 35–39, 45, 48–50, 54–56, 143, Marine transportation, 2, 5, 13, 283, 328, 395,
145–147, 150, 155, 157, 159, 184, 397
188, 189, 191, 192, 200–202, 209, Maritime route, 103, 121, 123
210, 216, 381, 402 Maritime spatial planning, 419
upper mixed, 35, 37–39, 155, 189, Markov model, process, 241, 244, 247, 276
193–195, 208, 310 Marmara, Sea of, 21, 145–147, 153
uppermost, 8, 21, 49, 56, 59, 121, 147, 197, Marsili (Marsigli), Count Luigi Ferdinando,
207, 209, 211, 212, 214, 216, 253, 132, 145
254, 268, 270, 272, 274, 277, 278, Mass flux, 233, 256
283, 285, 286, 292, 296, 297, 323, Matrix, matrix inverse, 82, 88
329, 330, 332, 337, 369, 381 Maximal exchange, 141
Layer interface, 233 Maximum depth, 34, 96, 134
Index 433
Mediterranean, 10, 135 eddy-resolving, 116, 154, 204, 244

Mediterranean deep layers, 168 high-resolution, 32, 60, 69, 95, 114, 115,
Mediterranean outflow, 154 140, 151, 181, 183, 184, 210, 211,
Mediterranean plume, 151 278
Mediterranean Sea, 10, 13, 131–139, 145, 148, isopycnal, 232
149, 151, 154–156, 159–161, 164, large-scale, 60
166, 168, 169, 226 non-hydrostatic, 151
Mediterranean water, 132, 147, 149, 151 numerical, 44, 56–59, 65–67, 69, 71, 77,
Mersa–Matruh gyre, 159 82–84, 86, 95, 97, 99, 100, 131,
Mesh, see grid 138–140, 144, 146, 147, 150–152,
Mesoscale dynamics, 16, 17, 44, 135, 182, 154, 156, 157, 159–161, 168, 171,
202–205, 210, 214, 216, 217, 244, 181–184, 190, 204, 207, 210, 216,
277, 289, 295, 299, 313, 327, 328, 226, 300
340, 341, 358, 359, 362 operational, 13, 15, 60, 151, 155, 212, 325,
Mesoscale eddy, 32, 45, 53, 92, 102, 105, 136, 419
137, 143, 168, 181, 204, 209, 211, primitive equation, 106
216, 270, 272, 289, 295, 305, stochastic, 244
308–310, 329, 341, 359 three-dimensional, 50, 56, 58, 59, 61, 102,
turnover time, 203, 292, 300, 308, 309, 150
312, 313, 361 trajectory, 253, 264, 266, 275, 301, 302,
Mesoscale feature, 21, 155, 183, 184, 203, 306
207, 211, 214, 272, 291, 313 off-line, 283
Meteorological conditions, 44, 182, 197 Model calibration, 66, 153
Meteorological data, 58 Model output, 162, 275, 295, 297
Meteorological forcing, 59, 60, 146, 149, 161, Model validation, 58, 66, 124, 153, 171, 184,
183, 197, 386, 407 253, 295, 329
Metocean (drivers, factors, conditions), 9, 10, Model verification, 59, 184
12, 20, 286, 320, 373, 383, 390 MODIS, 145
Metocean information, 373 Modular Ocean Model, MOM, 96, 139,
Miami Isopycnic Coordinate Ocean Model, 152–155, 165, 166
MICOM, 97 Moisture flux, 60
Mixing, 32, 37–39, 41, 56, 68, 92, 115, 117, Molecular viscosity, 41, 87, 202
135, 142, 143, 147, 149–152, 155, Molecular weight, 379, 380
156, 158, 167, 168, 171, 181–184, MOM, see Modular Ocean Model, MOM
188, 189, 192, 195, 196, 216, 243, Momentum equation, 50, 90, 93, 94, 98, 240,
255, 329, 380, 401 255, 402, 403
convective, 35, 36, 189 Momentum exchange, 155, 197, 200
diapycnal, 92, 94, 97, 113 Momentum flux, 108, 112, 114, 116, 164
horizontal, 94, 105, 243 Monaco Underwater Reserve, 2
isopycnal, 92, 94 Monte Carlo technique, 20, 351, 368, 369,
mechanical, 38, 55, 192 390–392, 410
subgrid, 105, 110, 295, 329 Morphology, 33, 200
turbulent, 32, 92, 105, 153, 196, 240, 403 MOSM, see Multiphase Oil Spill Model,
vertical, 36, 94, 105, 110, 116, 123, 124, MOSM
155, 157, 166, 210, 213, 243, 328 Multi-tracer approach, 124
wave-driven, 375, 377, 382, 386 Multiphase Oil Spill Model, MOSM, 124
wind-induced, 35, 36, 38, 39, 192 Muuga Bay, 209, 255
Mixing depth, 195, 377
Mixing length, 42 N
Mode-split method, 86 Naissaar, 208, 255, 270, 349, 359
Model Nansen, Fridtjof, 49
basin-scale, 140 NAO index, see North Atlantic Oscillation,
coupled atmosphere-ocean, 115 NAO
eddy-permitting, 56, 105, 116, 160, 295 Narva, 186, 199, 346
434 Index
Narva, River, 187, 345 Nucleous of European Modelling, NEMO,

Narva Bay, 185, 338, 340 139, 146, 155, 160, 277
NASA’s Science Mission Directorate, 145 Numerical error, 80
National Centres for Environmental Prediction Numerical instability, 79, 80, 97
and for Atmospheric Research, Numerical method, 65, 67–71, 80
NCEP/NCAR, 99 Numerical scheme, 80, 83, 184, 227
Naval Oceanographic Office, NAVOCEANO, consistency, 65, 79
253 convergence, 65, 79
Navier–Stokes equations, 20, 41, 87, 90, 94, stability, 65, 79, 80
106, 124, 378 Numerical solution, 70, 72, 80, 81
Navigation, vii, 4, 5, 12, 284 Numerical stability, 82, 83, 110, 116, 227
Navy Layered Ocean Model, NLOM, 97 Nutrients, 18, 117, 149, 195, 196, 285, 376
Nearshore area, 2, 6, 13, 14, 23, 185, 186, 195,
205, 291, 292, 299, 302–307, 319, O
321, 322, 324, 332–334, 336, 338, OAAS model, 300, 330–333, 335, 337,
339, 344, 354, 361, 373, 383, 409, 339–344, 350, 352, 359, 362
417, 418
Ocean Circulation and Climate Advanced
NEMO, see Nucleous of European Modelling,
Model, OCCAM, 99, 106, 230, 255,
NEMO
329
NEMOMED8 model, 160
Ocean conveyor belt, 55, 84, 230
Netherlands, 4
Ocean dynamics, 41, 42, 83, 84, 91, 93, 115,
Neugrund, 186
117, 163, 202
Neural network technique, 147
Ocean General Circulation Model, OGCM,
Neutrally buoyant substances, 286
243, 247, 265
Neva, River, 35, 181, 187, 190, 210, 211, 246,
291, 294, 312, 328, 345, 355, 381, Ocean model, 22, 49, 65, 66, 68, 83, 86, 95,
385, 399–401 97, 99, 100, 104–106, 111,
Neva Bay, 187, 188, 190, 199–201, 210, 211, 114–117, 152, 154, 155, 188, 211,
291, 339, 357 217, 226, 251, 253, 255, 275, 277,
Neva Bight, see Neva Bay 278, 335, 340, 353, 358, 359, 361,
New South Wales, 2 375, 378, 383, 418
Newton, Isaac, 67 Ocean-atmosphere interface, 96
Newtonian fluid, 87 Offshore activity, vii, 1, 6–8, 284, 325, 326
Newton’s laws, 84 Offshore area, 1–3, 11, 13–15, 18–21, 23, 24,
Non-invertible, 20 103, 186, 195, 196, 209, 210, 214,
Non-spherical coordinates, 99 284, 285, 289–291, 299, 302, 304,
Norrköping Deep, 55 306, 319, 321–324, 326, 327, 335,
North America, 2 336, 338–340, 344, 346, 353, 355,
North Atlantic, 9, 10, 156, 230, 252, 381, 395 369, 370, 384, 386, 397, 400, 408,
North Atlantic Deep Water, 230 417, 419
North Atlantic Model, NOAMOD, 381 Offshore engineering, industry, vii, 7, 323
North Atlantic Oscillation, NAO, 57, 135, 198 Offshore structure, 4, 19, 23, 293, 327
North Pacific, 9, 395 Oil
North Sea, 21, 22, 32, 35, 37, 45, 55, 60, 113, dissolution, 372
131, 137, 139, 141–143, 150, 151, dissolved, 124
161, 162, 188, 189, 191, 215, 287, emulsified, 124
288, 325, 328, 331, 332, 334, oxidation of, 372
378–381, 418 particulate, 124
North Sea model, 59 Oil chemistry and physics, 368, 370
North Sea water, 328 Oil drift, 15, 103, 296, 320, 368–370, 373, 377,
Norway, 13, 57, 381 378, 380, 383, 386, 408, 410, 411
Norwegian Sea, 21 current-driven, 368
Norwegian trench, 381 wave-driven, 402
Nuclear power, 7, 66 wind-driven, 386, 409
Index 435
Oil drift and fate, 8, 121, 252, 277, 296, 321, Oxygen, 143, 144, 151, 213, 285, 288
368–370, 379, 383, 409–411
Oil drift and fate model, 20, 23, 188, 368, 370, P
371, 373, 374, 378, 381–385, 401, Pacific, Pacific Ocean, 2, 53, 252
411 Pair separation, 254, 257, 259, 265, 266,
operational, 20, 368, 370, 372, 379, 382, 268–272, 277
386 Paldiski, 185, 338, 340
Oil drift pattern, 371, 384 Paldiski Deep, 185
characteristic, 370, 382–384, 386, 409 Paradigm, 1, 3, 5, 12, 131, 417
Oil drift pattern characteristic, 385 Particle age, 19, 215, 295, 319, 324, 325,
Oil film, 296, 376, 377 336–341, 343, 345–354, 356, 357,
Oil platform, 3, 6–8, 284, 322, 327, 346, 371 359–361, 384, 418
Oil pollution, 6–8, 13–15, 18, 21, 23, 24, 286, Particularly sensitive sea area, PSSA, 3, 7, 182,
288, 304, 320, 322–324, 331, 419 284, 321, 327
Oil propagation, 321, 324, 368, 370, 382, 383, Path, 12, 67, 84, 147, 169, 226, 245, 298, 300,
402, 405–409 306, 307
current-driven, 371 Path integral, 390, 392
simulation of, 383 Path length minimization, 391
wave-driven, 371, 401 Path optimization, 390
Oil spill, 1, 3, 4, 6–8, 13, 15, 16, 20, 21, 102, Pathway, 153, 278, 283, 292, 293, 300, 311,
103, 105, 118, 120, 121, 123–125, 332, 346, 369, 370, 377, 390–392,
254, 277, 285, 291, 292, 296, 302, 408, 411
320, 321, 323, 325, 326, 368–378, Pathway of rapid transport, 308, 312
380, 382–384, 386, 388, 389, 391, Persian Gulf, 132
395, 401, 402, 410, 411 Persistence, 53, 54, 88, 205, 207, 208, 210,
beaching of, 302 211, 214, 289, 291, 292, 307, 312,
illegal, vii, 322 313, 339
source of, 321 Petroleum industry, 66
Oil spill combating, 325, 374, 395 Phase speed, 44, 403, 404
Oil spill combating time, 369 Physical oceanography, 13, 21, 46, 51, 56, 132,
Oil spill model, 20, 120, 121, 123–125, 211, 133, 139, 140, 144, 145, 171, 182,
253, 322, 377, 380 184, 200, 288, 358
operational, 370 Phytoplankton, 124
Oil spill model (OSMS), 123 Pipeline, 3, 6, 7, 284, 371
Oil spill propagation, 8, 20, 22, 49, 123, 294, Plume, 151–154, 162
319 Poland, 57, 259
Oil spreading, 103, 106, 125, 372 Pole (geographical), 83, 88, 99
Oil transportation, 2, 12, 283, 371, 399 Pollution, vii, 1, 3–8, 14, 18, 19, 49, 183, 285,
Operational oceanography, 23, 171 292, 294, 307, 313, 338
Øresund, 32, 34, 55, 133, 141, 142, 144, 156, Pollution control, 182
157, 162, 163, 294 Pollution propagation, 1, 3, 15, 17, 18, 20, 188,
Ornstein–Uhlenbeck process or model, 244 278, 285, 293, 319–324, 331, 361,
Oslo, 57 397
Osmussaar, 185, 186, 212 Polynomial approximation, 73, 74, 108
Otranto, Strait of, 168 Pomeranian Bay, 13, 288, 291, 328
Outflow, 133, 134, 140, 143, 144, 147, Põõsaspea, 185
149–151, 153, 154, 161–163, 169, Porkkala, 185, 349, 359
171, 183, 187, 190, 203, 207, Potential density, 170
209–212, 214, 215, 287, 290, 302, Potential energy, 92, 210
305, 306, 313, 326, 328, 335, 353, Potential temperature, 99, 106, 107
385 Power law, 251, 269, 272–274, 277
Overflow error, 79 Practical Salinity, 190
Overturning, 55, 92, 152, 157, 163, 246 Prandtl number, 107, 111, 112, 150
Oxidation of H2 S, 151 Prangli, 341
436 Index
Precipitation, 16, 60, 83, 108, 109, 114, 131, Radiation stress, 403, 404
133, 134, 146, 187, 188, 201, 215, Radioactive substance, 8, 117, 286, 296, 377
295, 330 Rail, rail transport, 2, 7, 12
Predictor–corrector method, 82 Random disturbance, 301
Pressure anomaly, 115 Random noise, 241
Pressure gradient, 32, 42, 43, 45, 50, 51, 89, Random perturbation, 240–243, 256, 276
90, 96, 97, 137, 188, 203, 216, 402 Random transport model, 207
horizontal, 42, 83, 89 Random velocity, 240, 241
Pressure (physical), 41, 51, 67, 83, 88–91, 107, Random walk, 125, 243, 244, 272
111, 246 Ratio of net and bulk transport, 283, 309, 310,
Prestige, 6, 7, 372 312, 313
Preventive maritime planning strategy, 6 Rauma, 55
Preventive reduction of environmental risks, 13 Rayleigh–Taylor instability, 67
Primitive equation model, 155, 295, 329, 330, RCA model, 255, 256, 275, 277
378 RCAO model, 255
Primitive equations, 90, 106, 111, 114, 139, RCO, see Rossby Centre Ocean Model, RCO
168 Reanalysis (meteorological), 98, 115, 233
Princeton Ocean Model, POM, 154 Red Sea, 132
Probability for different offshore areas to serve Reduced gravity, 141, 151
as starting points of pollution that Reference density, 86, 89, 107
hits a coast, 383 Regime shift, 57
Probability of coastal pollution, 19, 292, 294, Regional geometry, 134
303, 306, 307, 319, 325, 336–342, Regional Ocean Modelling System (ROMS),
344–361, 418, 419 96
Probability of current-driven transport of Relaxation time, 47
pollution to a vulnerable area, 1, 15, Remote danger, 12, 319, 417
16, 19, 21, 319, 320, 322, 324, 325, Remote impact, 321, 322
368 Remote influence, 1, 6, 12, 14
Probability of oil landing, 369, 383, 384, Remote sensing, 184, 209
387–393, 406–409, 411 Renewal, 117, 189, 215, 338
Prognostic equation, 107, 110 Renewal index, 215
Propagation scenario, 294 Renewal time, 335
Propagation time, 299 Rescue boat, 16, 285, 319
Propulsion, 396 Rescue operation, 252
Propulsion model, 396, 397, 399 Residence time of oil at sea, 295, 369,
Pycnocline, 84, 158, 164, 166, 168 383–385, 388–392, 406–409
secondary, 210 characteristic, 385
Residence time of water, 144, 184, 215, 339
Q Residual flow, 146, 205, 206, 210
Quadratic friction law, 111 Resistance, 396, 397, 400, 410, 411
Quadratic stress formula, 98, 379 shallow water, 399–401
Quantification of offshore, 1, 15, 18, 19 still water, 396
Resolution
R horizontal, 58, 60, 82, 83, 85, 111–113,
Radar, 171 115, 116, 151, 159–161, 197, 204,
Radiation, 60, 201 211, 255, 288, 295, 298, 326, 327,
long-wave, 109, 112, 114, 330 329–333, 340–344, 347, 349, 353,
short-wave, 109, 110, 114, 330, 379 358–361, 368, 381–384, 391, 404,
solar, 37, 38, 83, 112, 192, 197, 201, 329, 418, 419
385 spatial, 60, 84, 252, 265, 277, 278, 296,
terrestrial, 201, 202 298, 300, 301, 380
total, 201 temporal, 21, 60, 82, 114, 166, 168, 214,
Radiation budget, 37, 191, 201 261, 295–299, 326, 330–332, 334,
Radiation condition, 99, 118 374, 381
Index 437
vertical, 50, 56, 96, 111, 112, 116, 188, Russia, 103, 182, 185, 345, 372, 387, 389, 392
211, 328–330, 332, 359, 381
Reversal, 35, 213 S
Reversible, 92 Saint Petersburg, 22, 187, 287, 348, 350, 372,
Reynolds, 93 392–394, 397, 399–401
Reynolds number, 91, 92 Saint Petersburg flood barrier, 195, 210
Reynolds stress, 41, 42, 94 Saline contraction coefficient, 165
Reynolds–Averaged Navier–Stokes (RANS)
Salinity, 35–37, 39–41, 45, 83, 84, 91, 92, 99,
equation, 94
106–109, 111, 113, 118, 119,
Rhodes gyre, 160, 161
133–135, 138, 143, 144, 146, 147,
Richardson number, 112, 197
151, 153, 154, 157, 160–164,
Richardson’s law, 251, 269–271, 273, 274, 277
166–169, 181, 187–191, 204, 209,
Riga, 254, 278
226, 240, 246, 255, 256, 275, 329,
Rigid lid approximation, 98
331, 340, 358, 375, 378, 379, 382
Rim Current, 158, 166
Ringhals, 114 Salinity anomaly, 157
Risk equation, 4, 5, 11, 12, 19, 368 Salinity budget, 134, 146
Risk of coastal pollution, 334 Salinity gradient, 162, 182, 188, 216, 288, 328,
Risk parameter, 383 385
River, 41, 43, 45, 68, 83, 116–118, 120, 134, horizontal, 188, 206
135, 152, 160, 161, 166, 187, 193, Salinity unit, 36, 190
196, 203, 205, 207, 215, 290, 331, Salt diffusivity, 93
370, 375, 381, 385 Salt exchange, 140
River discharge, 16, 59, 112, 133, 183, 186, Salt flux, 108, 109
187, 215, 297, 331, 381, 385 Sardinia, Strait of, 168
River inflow, 118, 131, 133, 187, 295 Satellite information, 136–138, 145, 195, 196
River runoff, 32, 35, 36, 45, 60, 109, 112, 115, Saturation process, 380
116, 133, 134, 138, 146, 160, 161, Scandinavia, 255
164, 169, 185, 187, 189, 190, 196, Scattering ellips, 210
206, 207, 209, 211, 214, 216, 256, Scotland, 381
291, 312, 345, 355, 386 Sea
Road, road transport, 2, 7, 8, 12 semi-enclosed, 2, 3, 15, 17, 23, 113, 131,
Rossby Centre Ocean Model, RCO, 86, 106, 132, 135, 137, 139, 142, 151, 154,
112–118, 255, 256, 260–262, 164, 171, 183, 322, 373, 409
264–267, 275–277, 283, 294, 295, shelf, 2, 32, 50, 136, 216
297, 299, 301, 303, 307–310,
Sea breeze, 198
329–333, 335, 344, 347, 349, 352,
Sea level, 33, 60, 96, 104, 108, 111–114, 133,
354–356, 361, 362, 378
135–138, 143, 145, 146, 158, 159,
Rossby number, 43
197, 198
Rossby radius, 43, 44, 202, 204
Sea level anomaly, 136, 143, 146
baroclinic, 32, 44, 60, 102, 116, 136, 141,
158, 181, 204, 209–211, 217, 272, Sea level gradient, 51
288, 289, 291, 295, 309, 313, 328, Sea level pressure, 114, 161, 295
358, 361 Sea level rise, 2, 32, 33
barotropic, 44 Sea of Marmara, see Marmara, Sea of
Rotating reference frame, 41, 84, 86, 87 Sea surface, 2, 5, 7, 21, 23, 32, 37, 39, 42, 45,
Rotation, 48, 50, 87, 142, 350 47–49, 51, 98, 103, 107–109,
clockwise, 140 117–119, 135, 137, 138, 152, 153,
Rotational effect, 46, 136, 141, 142 163, 165, 191, 195, 200, 201, 203,
Roundoff, 80 204, 211, 215, 232, 255, 277, 288,
Royal National Park, 2 296, 319
Rügen, 291 Sea surface elevation, 147, 234
Runge–Kutta method, 226, 238, 247 Sea surface height, 32, 45, 160, 216, 256
Running mean, 262, 264, 267, 275 Sea surface salinity, 143, 162–164, 190, 207
438 Index
Sea surface temperature, 38, 60, 109, 117, 149, Sill, 55, 90, 140–143, 145, 147, 148, 158, 161,
161, 165, 192, 193, 201, 253, 259, 168, 170, 183, 185, 188, 207, 328
260 Sill depth, 35, 141, 148, 158
Sea surface tilt, 42, 51, 52, 203 Sillamäe, 350, 392–394, 398, 400, 401
SeaPlanner software, 396, 399, 400 Simulations
Search-and-rescue, 182 high-resolution, 59, 95, 207, 301, 335
Season long-term, 110, 114, 116, 169
calm, 310–312, 326, 338, 368, 386 numerical, 139, 144, 146, 150, 153, 157,
summer, 39, 387–389, 393, 394, 398, 410 159, 160, 166, 169, 171, 183, 184,
transitional, 283, 311, 387–389, 394, 410 205, 207, 214, 216
windy, 310, 311, 326, 337, 368, 386 Skagen, 332
winter, 254, 386–389, 393, 394, 398, 410
Skagerrak, 143, 144, 161–163, 287
Seasonal cycle, 158, 190, 191
Skären-type coast, 185
Seatrack Web, 253, 278, 300, 303, 331
Skillful scale, 85
Sebastopol, see Sevastopol
Seiche, 45, 206, 208, 216 Slipshavn, 52
Seiskari, see Seskar Smagorinsky model, 95, 110, 150
Sensitivity experiments, 113 SMHI, 115, 116
Separation rate, 244, 261, 271–274, 276, 334 Smögen, 112, 114
Serial correlation, 301 Smoothed-particle hydrodynamics, SPH, 67,
Seskar, Seskär, 186 68
Seskar (Seskär) basin, 186 Solar forcing, 201
Sevastopol, 253 Solar insolation, 107–109
Sevastopol eddy, 137, 158 Sommers (Someri), 341
Shallow area, 12, 36, 48, 85, 119, 144, 186, Sound, see Oresund@Øresund, see
187, 253, 288 Oresund@Øresund
Shallow water, 12, 106, 186, 396, 397, 401, South Pole, 99
404 Spain, 7, 372
Shear dispersion, 273 Spatial scale, 44, 102, 294, 302
Shear instability, 150 Spawning area, 7, 23
Shear stress, 47, 83, 94, 379, 404 Specific heat capacity, 107
Shelf, 134, 148, 151, 152, 164, 167 Specific weight, 89
Shetland Islands, 401 Speed
Ship routing, 1, 8–10, 12, 23, 24, 211, 285, net transport, 202, 308, 309, 311
292, 323, 325, 368, 369, 378, 384,
Speed of sound, 85
390, 395–397, 419
Spherical coordinates, 43
environmental, 10
Spill state vector, 373, 375, 376
weather, 10, 395, 396
Ship routing software, 368 Spin-up, 47, 100, 113, 119
Ship traffic, 1–11, 17, 18, 21, 22, 253, SPLIT-QUICK advection scheme, 116
284–286, 288, 292, 323, 325, 326, Sponge layer, 331
328, 346, 348, 349, 353, 418 Spreading, 58–60, 105, 154, 182, 186, 196,
Shipping, vii, 1, 2, 7, 9, 10, 13, 19, 102, 278, 251–253, 269–271, 276–278, 297,
284, 324, 326, 356, 390, 395, 397, 300–302, 306, 333, 334, 347,
399 373–377, 380
environmental management of, 10 artificial, 303
environmentally friendly, 10 horizontal, 375, 376
Shipping industry, 5, 20, 323 of trajectories, 298
Shipping route, 10, 103 vertical, 376, 377
Shoreline, 32, 53 Spreading rate, 196, 197, 251, 268, 270,
Sicily, Strait of, 168 272–274, 277, 278, 307, 334
Sigma-coordinates, 96, 97, 232 Spreading scheme, 301
Significant wave height, 404, 405 Standard deviation, 241, 243, 340, 359, 361,
Silja Festival, 254, 278 418
Index 439
Statistical analysis, 1, 20, 21, 23, 139, 171, Surface tension, 376
278, 283, 286, 300, 319, 322, 323, Surface Velocity Program, SVP, 251–253, 263
335, 360, 395 Surface water, 8, 35, 37–39, 45, 55, 56, 83, 88,
Statistical approach, 4, 16, 17, 383 93, 109, 143, 144, 152, 159, 163,
Statistical characteristics, 264, 276 166, 188, 192, 195, 196, 213, 299,
Statistical feature, 205, 217, 260, 300, 301, 332 305–307, 311
Statistical measure, 121, 123, 253, 384, 385 Suspended matter, 293
Stefan Boltzmann’s constant, 109 Sustainable development, viii
Stefan Boltzmann’s law, 109 Suursaari, see Gogland
Stencil, 77, 78, 80–82 Sweden, 32, 275, 329
Steric effect, 136 Swedish Meteorological and Hydrological
Stochastic dynamic programming, 10 Institute, SMHI, 58, 114, 329, 381
Stockholm, 34, 254, 278 Synoptic eddy, 44, 210, 214, 289, 299, 302,
Stokes drift, 275, 296, 401, 403, 405 308, 329, 333
Stokes law, 380 Synoptic scale, 42, 206, 285, 289
Stolpe Channel, 55, 144, 156, 157 Szczecin, 57
Storm, 4, 9, 10, 53, 195, 198, 284, 303, 396,
402 T
Storm surge, 68, 380 Tallinn, 185, 206, 288, 328, 339, 340, 349,
Strait, vii, 51, 53, 111, 131–135, 137, 138,
356, 357, 359
140–146, 148–153, 156, 157,
Tanker, vii, 125, 325, 326, 345, 348, 397
161–163, 168, 171, 291
double-hull, 4
Strait control, 138
Tanker traffic, 4, 22, 368
Strait exchange, 131, 133, 146, 147, 150, 152
Tar, 372
Straits’ dynamics, 140, 141, 171
Tarifa Narrows, 148
Stratification, 8, 35–38, 41, 44, 55, 56, 59, 60,
Taylor series, 73–75
84, 92, 115, 117, 136, 142, 151, 153,
155, 157, 158, 161, 162, 166, 183, Temperature, 32, 36–41, 56, 83, 84, 91, 92, 99,
187–189, 195, 197, 200, 204, 210, 106, 108, 109, 111, 113, 115, 118,
213, 216, 270, 286, 291, 328, 385 138, 149, 151, 153, 154, 160, 164,
four-layer, 39 181, 188, 191–193, 195, 201, 202,
three-layer, 36 204, 209, 226, 233, 240, 246, 255,
two-layer, 32, 35, 37, 51, 158, 189, 191 256, 259, 275, 276, 330, 331, 340,
Stream function, 245, 246 358, 375, 377, 379, 382
Lagrangian, 225, 244–247 Temperature gradient, 32, 37, 39, 188, 202
barotropic, 246 horizontal, 188, 193, 195
overturning, 246 Temperature of maximum density, 37, 39, 41,
Structural analysis, 70 192
Structural reliability analysis, 17 TEOS-10, 36, 190
Subgrid mixing, 94 Terrain-following coordinates, 96, 97, 154,
Subgrid processes, 93, 95, 116, 121, 252, 256, 232, 234
268, 300, 301, 334, 375, 380 Thermal expansion coefficient, 165
parameterization of, 93–95, 105, 116, Thermal wind law, 51
225–227, 240, 241, 243, 244, 247, Thermocline, 32, 36, 38, 39, 47, 51, 192, 193,
252, 276–278, 329 208, 209, 276
Subgrid scale, 227, 243, 261, 276 inverse, 37, 39, 192
Subgrid turbulence, 240, 241, 243, 252, 255, seasonal, 37, 38, 164, 165, 192
267–269, 374 summer, 38, 192
parameterization of, 240–242, 267, 268 winter, 39, 41
Submarine passage, 156 Thermodiffusivity, 93
Super-inertial frequency, 197 Thermohaline cell, 148
Surface dynamics, 292, 304, 306 Thermohaline circulation, 57
Surface pressure, 233 Thermohaline gradient, 32
Surface roughness, 60, 197, 198 Tidal current, 146, 148, 150, 208, 216
440 Index
Tidal forcing, 16, 32, 45, 46, 137, 142, passive, 106, 116–118, 124, 125, 153, 227,
148–150, 216, 381 283, 286, 293, 326, 369
Tide gauge, 59 Tracer age, 144
Time available for combating with the oil Tracer concentration, 121, 123
pollution, 388–391 Tracer equation, 227
Time it takes for the pollution to reach a Tracer model, 105, 106
vulnerable area, 1, 15, 19, 21, 320, Tracking scheme, 300
324, 325, 336, 368, 417 TRACMASS, 22, 225–228, 232, 234, 235,
Time it takes for the pollution to reach the 238–241, 243, 244, 246, 247,
coast, 19, 23, 304, 306, 307, 311, 255–257, 267, 268, 276, 283, 294,
322, 336, 351, 352, 383 297, 299–303, 307–309, 322, 332,
Time lag, 207, 258, 264, 299, 301, 304, 308, 333, 344, 354, 356, 362
310, 335, 344 Traditional approximation, 106
Time scale, 84, 85, 102, 105, 113, 114, 118, Trajectory, 1, 8, 16, 20, 21, 105, 225–235,
123, 184, 203, 206, 207, 210, 216, 237–247, 251–253, 256–260, 262,
217, 261, 264, 275, 276, 283, 286, 264–272, 275, 276, 278, 279, 294,
292, 294, 299, 300, 302, 307, 310, 296–298, 300–304, 307, 309, 320,
312, 313, 324, 326, 338 322, 323, 332–336, 339, 341, 344,
advective, 113 345, 352, 354, 355, 359, 361, 370,
climatological, 57 374–377
overturning, 113, 114 Lagrangian, 256, 283, 286, 293, 294,
seasonal, 313 296–309, 311–313
Trajectory calculation, 226, 227, 234, 240,
synoptic, 313
297, 298, 309, 313, 332, 333, 374,
Time splitting scheme, 329, 330
382
Time step, 72, 77–81, 83, 85, 86, 93, 104, 105,
off-line, 105, 226, 256, 297, 332, 334
226, 235, 236, 240, 241, 243, 247,
on-line, 226, 227, 235, 256, 333–335, 337
256–258, 268, 297, 300–302, 330,
Trajectory equation, 334
332, 334
Trajectory model, 225, 226, 243, 246, 307,
baroclinic, 112, 116, 295, 330
329, 332
barotropic, 295, 330 Trajectory solution, 232, 240
Time window, 298, 299, 305–310, 312, Transit time, 9, 117, 229, 231, 236, 237
335–338, 340, 354, 361 Transition zone, 139, 142, 143, 150, 156, 161,
Time-stepping method, 234, 236, 238, 240, 162, 186, 188, 287, 332, 380, 381,
247 418
Tjuters, 186 Transoceanic crossing, 10
Tjuters, Malyi (Pieni Tytärsaari, also Väike Transport, 57, 77, 137, 151
Tütarsaar), 341 barotropic, 143, 152
Topographic control, 32, 113, 131, 134, 144, bulk, 283, 307–310, 312, 313, 326, 353
148, 151, 157, 162, 189, 205, 209, counter-estuarine, 158
213, 214, 216, 302, 382, 405 current-driven, 16, 17, 23, 37, 60, 206, 216,
Topography, 32, 33, 45, 56, 57, 59–61, 84, 96, 217, 285, 286, 289, 292, 293, 296,
113, 116, 134, 135, 140, 142, 148, 298, 302, 307, 313, 319–321, 323,
154, 157, 161, 181, 183, 188, 200, 326, 338, 358, 361, 362, 368, 370,
328, 370, 383, 389, 406, 408, 409, 417, 418
411 Eulerian, 18, 22, 299
Torm Ragnhild, 397 horizontal, 158, 379
Tourist industry, 13 Lagrangian, 22, 286, 292, 293, 298–300,
Toxic algae, 294 307, 310, 311, 313, 322
Toxic level of oil, 374, 377 meridional, 245, 310, 312
Tracer, 15, 16, 106, 117–124, 144, 152, 153, net, 53, 134, 207, 283, 307–313, 326, 339,
226–228, 232, 243, 244, 255, 300, 353
310–312, 376 onshore, 405
Eulerian, 102, 105, 117, 121, 123–125, 227 semi-persistent, 53, 289, 307
Index 441
Transport (cont.) Unique solution for trajectories, 226, 230

subsurface, 341, 368 United Nations, UN, viii
surface, 8, 16, 45, 132, 159, 289, 290, 298, United States, 10, 12
309, 353, 370, 418 Univariate optimum interpolation, 114
vertical, 146, 245, 379 Upwelling, 45, 49, 50, 60, 149, 188, 193, 195,
wave-driven, 15, 16, 285, 303, 320, 361, 196, 209, 216, 331, 353
368, 405, 408, 409 Ust-Luga, 350, 392–394, 398, 400, 401
wind-driven, 16, 45, 285, 303, 320, 368, Utö, 36, 190, 192, 207
370
zonal, 229, 231 V
Transport equation, 71, 72, 76, 79–82 Van Dantzig, David, 4
Transport of heat, 143 Varberg, 114
Transport of nutrients, 186 Variability, 188, 205, 208, 252, 259, 261, 262,
Transport of pollution, 203, 232 264, 267
Transport of salt, 143, 163 interannual, 138, 160, 188, 312, 353
Transport pathway, 283, 289, 312, 313 seasonal, 16, 36, 37, 99, 138, 146, 152,
Transport pattern, 18, 150, 183, 247, 283, 289, 159, 166, 188–191, 302–304, 310,
293, 300, 307, 311, 313, 348 313, 335, 337, 338, 353, 385, 386,
409
Trapped motion, 290
spatial, 10, 16, 102, 273, 312
Travel distance, 390, 400, 401, 410
temporal, 102, 114, 137, 138, 165, 181,
Travel time, 390, 396, 397, 399, 400, 408, 410
197, 204, 209, 215, 234, 261, 306
Truncation error, 85, 227
Velocity, 47, 48, 67, 86, 87, 97, 160, 162, 163,
Turbulence, 36, 84, 91, 92, 95, 103, 110, 116,
197, 200, 208, 214, 229, 231, 234,
121, 125, 141, 195, 196, 201, 202,
235, 238, 240, 241, 243, 245, 247,
244, 252, 256, 268–270, 295, 300,
255–259, 261, 267, 269, 276, 297,
332, 375, 377, 380, 420
298, 301, 302, 305, 308, 309
fully developed, 269
compressibility of, 270
horizontal, 196, 197
horizontal, 42, 202, 203, 256, 261, 262,
small-scale, 272
268, 283, 289, 296, 402
subgrid, 256, 267, 268, 276, 297, 306, 307,
meridional, 161, 163, 228, 262, 264
332, 333, 359 potential component of, 270
three-dimensional, 273, 274 power spectrum of, 244, 261, 262
two-dimensional, 269 solenoidal component of, 270
vertical, 60, 197, 217 surface, 48, 49, 158, 293, 294
Turbulence closure, 110, 111, 197, 295, 329, turbulent, 263
379 vertical, 41, 42, 106, 195, 202, 203, 228,
Turbulence parameter, 112 233, 261, 296
Turbulence parameterization, 60, 94, 95, 217, zonal, 228, 229, 262, 264
240, 244, 252, 256, 261, 265, 268, Velocity autocovariance, 258
276 Velocity gradient, 91
Turbulence scheme, 155, 256, 267, 276 vertical, 92
Turbulent flux, 39, 201 Ventilation, 143, 144, 153, 157, 166, 168
Turbulent friction, 47, 107, 111 Vessel, 5–7, 13, 198, 285, 286, 288, 322, 368,
Tvärminne, 208 369, 371, 372, 389, 395–397, 400,
Two-layer exchange, 140, 142, 148, 152 418
Two-layer model, 119 Vessel traffic systems, 4
Tytärsaari, see Tjuters Vessel wake waves, 1, 12, 15
Viscosity
U numerical, 227
Ukraine, 253 Viscosity of oil, 372, 376, 377, 401
Uncertainty, 3, 60, 296, 303, 307, 324, 351, Viscous dissipation, 91
353–356, 358 Viscous-advective-diffusive limit, 150
Underwater mining, 3, 284 Volcanic ash, 321, 377
442 Index
Volcano eruption, 321 coastally trapped, 209

Volume transport, 230, 232, 233, 245 freak, 396
Voyage speed, 10, 397 internal, 44, 84–86, 90, 92, 110, 116, 141,
Vulnerable area, vii, 1–3, 6, 8, 13–15, 18, 19, 155, 183, 197, 204, 213
21, 24, 285, 299, 302–304, 307, Kelvin, 45, 216
313, 319–322, 324–326, 346, 352, Rossby, 44
368, 370, 417–419 shallow-water, 45, 85
Vulnerable environment, 283 surface, 44, 66, 84–86, 92, 103, 110, 124,
Vyborg, 199, 200, 341, 349, 350, 359, 360, 125, 183, 184, 186, 295
392–394, 398–401 tidal, 66, 85
Weathering, 286, 324, 368, 370, 372, 376, 377,
W 382, 386, 409, 419
Waste Heat Recovery system, 396 Wedge (of saline water), 35, 183, 188, 213,
Wastewater management, 66 215, 328
Water age, 102, 117–120, 144, 215, 324, 331, Western Mediterranean Deep Water, 161
336, 338 Whale, 10, 11, 13, 17, 321, 325
vertical gradient of, 118 Wiener process, 242
Water budget, 38, 60, 131, 146, 187–189, 214 Wind drag, 296, 302, 320, 332, 379, 396
net, 131 Wind drift, 375, 382, 386
Water cascading, 152, 171 Wind forcing, 16, 38, 45, 57, 60, 144, 188,
Water exchange, 43, 57, 100, 113, 114, 192, 197, 201, 205, 213, 216, 252,
132–134, 139–142, 144, 150, 164, 275, 277, 292, 385
181, 184, 188, 207, 213–215, 287, Wind gustiness, 115, 256, 275, 295, 330
295, 305, 328, 329, 331, 340, 354
Wind impact, 49, 83, 227, 273, 275, 277, 386
inter-ocean, 247
Wind set-up, 135
Water flux, 36, 133, 135, 138, 152
Wind speed, 35, 49, 53, 98, 103, 104, 115, 144,
Water mass, 2, 16, 18, 32, 35–37, 45, 56, 83,
198–201, 206, 212, 213, 275, 303,
85, 89, 92, 97, 99, 102, 117, 118,
386, 403, 405
135, 139, 144, 149, 152–154, 158,
Wind speed extremes, 115, 295
159, 161, 163, 164, 168, 182, 185,
Wind stress, 32, 42, 45, 48, 92, 93, 98, 108,
187–189, 192, 195, 196, 201, 206,
112, 146, 159, 164, 165, 203, 204,
209, 210, 213, 216, 225–227, 244,
206, 210, 216, 330, 353, 379, 403
247, 252, 274, 278, 285, 287–289,
Wind velocity, 212, 234, 375, 379, 380
293, 295, 296, 306, 310, 320, 328,
329, 331, 338, 339, 359 World Ocean, 2, 3, 17, 32, 60, 136, 182, 188,
Water mass formation, 131, 156–158, 160, 230, 252, 259, 276, 296
163, 166, 168 World Ocean Circulation Experiment, WOCE,
Water renewal, 215 253, 254
Water surplus, 115 Wormley, 226
Wave action, 183, 186, 376
Wave breaking, 67, 92, 103, 110, 155 Y
Wave energy gradient, 403, 404, 406, 408, 409 Year of the Gulf of Finland 1996, 182
Wave kinetic energy, 197 Year of the Gulf of Finland 2014, 182
Wave speed, 44, 85, 86, 141, 204
Wave stress, 403–405 Z
Wave(s) z-coordinates, 95–98, 154, 295, 329, 330, 379
acoustic, 85, 89 Zero-crossing, 345, 347, 349, 354, 419
baroclinic, 116 Zooplankton, 124

2013 Book PreventiveMethodsForCoastalPro

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2013 Book PreventiveMethodsForCoastalPro

Uploaded by

Copyright:

Available Formats

Tarmo Soomere

Ewald Quak Editors

Towards the Use of Ocean Dynamics

ISBN 978-3-319-00439-6 ISBN 978-3-319-00440-2 (eBook)

Library of Congress Control Number: 2013942501

© Springer International Publishing Switzerland 2013

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Project coordinator: Professor Tarmo Soomere, email: soomere@cs.ioc.ee

Oleg Andrejev (1941–2012)

1 Towards Mitigation of Environmental Risks . . . . . . . . . . . . . . 1

Part I Modelling the Underlying Dynamics

Part II Lagrangian Dynamics and Inverse Problems

10 Applications of the Inverse Problem of Pollution Propagation . . . . 319

Abstract This introductory chapter describes the background of a preventive tech-

T. Soomere, E. Quak (eds.), Preventive Methods for Coastal Protection, 1

1.2 The Notion of Risk, Probability and Consequences

1.3 Changing Paradigms of Local and Remote Risks

Kofoed-Hansen 2001; PIANC 2003; Soomere 2005). Some of these by-products

1.4 Managing Propagation of Contaminants and Oil Spills

Typically, the potentially dangerous by-products are represented by several non-

Fig. 1.1 The Prestige oil

shipping routes (Schwehr and McGillivary 2007), or proper designation of marine

1.5 Ship Routing Systems

1.6 Environmental Management of Ship Routes

1.7 Minimizing the Remote Impact of Ship Traffic

The first operational model targeted at a preventive reduction of environmental

1.8 Site-Specific Cost of the Consequences

1.9 Separation of the Impact of Wind, Waves and Currents

1.10 Dynamical and Statistical Forecast

A consequence of the intrinsic complexity of the reproduction of the details of ocean

societies’ rules is mainly prescriptive, without any transparent link to an overall

1.11 Favourable Patterns of Currents and Current-Induced

1.12 Quantification of Offshore Domains

1.13 A Preventive Technology of Environmental Management

1.14 From Classical Physical Oceanography to Environmental

1.15 Concluding Remarks

Kai Myrberg and Andreas Lehmann

T. Soomere, E. Quak (eds.), Preventive Methods for Coastal Protection, 31

2.1 Introduction to the Baltic Sea Geometry and Topography

Kattegat 22,287 23 130 515

South-western Baltic Sea

(Table 2.1), Bornholm Basin and Gdańsk Bay.

2.2 Basic Hydrography

Table 2.2 The main characteristics of the Baltic Sea stratification

Upper layer 40–80 m Wind, convection All year

2.3 Circulation Dynamics

2.3.1 Basic Principles

The system of ocean currents is three-dimensional (3D). Their dynamics is gov-

where u = (u, v, w) is the current velocity, is the Earth’s rotation rate (Ω =

4 After Joseph Valentin Boussinesq (1842–1929).

U/T U 2 /L fU ρ −1 H p/L AH U/L2 Av U/H 2

2.3.2 Barotropic and Baroclinic Flows and Rossby Radii

2.3.3 Dynamics of Surface Currents

Fig. 2.5 Inertial oscillations

2.3.4 Inertial Oscillations

In physical oceanography inertial oscillations refer to the circular motions of water

2.3.5 Ekman Drift

The Swedish oceanographer V.W. Ekman explained theoretically the formation of

Fig. 2.6 The Ekman profile

documented much later for the open ocean.

Fig. 2.7 The wind-driven

2.3.6 Geostrophic Flow

U/T U 2 /L fU ρ −1 H p/L AH U/L2 Av U/H 2

For the function value at (x = xm , t = nt), we will use the notation

f (xm , nt) = fm(n) .

t (n) (n) (n)

ρ(x, y, z, t) = ρ0 + ρ(x, y, z, t), where ρ0 ρ.