Coastal Engineering 134 (2018) 212–228

Contents lists available at ScienceDirect

Coastal Engineering
journal homepage: www.elsevier.com/locate/coastaleng

Implementation and validation of a multi-domain coastal hazard forecasting

system in an open bay
Nikolay Valchev *, Petya Eftimova, Nataliya Andreeva
Institute of Oceanology, Bulgarian Academy of Sciences, P.O. Box 152, 9000 Varna, Bulgaria

A R T I C L E I N F O A B S T R A C T

Keywords: European coasts are increasingly threatened by hazards related to low-probability and high-impact hydro-
Storms meteorological events. Uncertainties in hazard prediction and capabilities to cope with their impact lie in both
Surge future storm pattern and increasing coastal development. Therefore, adaptation to future conditions requires a re-
Waves
evaluation of coastal Disaster Risk Reduction (DRR) strategies and introduction of a more efficient mix of pre-
Flooding
vention, mitigation and preparedness measures. The latter presumes that development of tools, which can manage
Erosion
Delft-FEWS the complex process of merging data and models and generate products on the current and expected hydro-and
morpho-dynamic states of the coasts, such as forecasting system of flooding and erosion hazards at vulnerable
coastal locations (hotspots), is of vital importance. Output of such system can be of an utmost value for coastal
stakeholders and the entire coastal community. In response to these challenges, Delft-FEWS (Flood Early Warning
System) provides a state-of-the-art framework for implementation of such system with vast capabilities to trigger
the early warning process. In addition, this framework is highly customizable to the specific requirements of any
individual coastal hotspot. Since its release, many Delft-FEWS based forecasting system related to inland flooding
have been developed.
In this paper, a set-up of Delft-FEWS based forecasting system for Varna Bay (Bulgaria) and a coastal hotspot,
which includes a sandy beach and port infrastructure, is presented. The system output generated in hindcast mode
is validated with available observations of surge levels, wave and morphodynamic parameters for a sequence of
three short-duration and relatively weak storm events occurred during February 4–12, 2015. Generally, while a
longer term system operation/validation is needed for the model skills to be affirmed, the results obtained
indicate a reliable prediction of coastal hazards, thus giving a sound basis for estimation of hazard impact.

1. Introduction
Storms are one of the most significant hydro-meteorological phe-
nomena producing coastal flooding and endangering human life and
occupation on European coasts. Moreover, storms cause diverse
morphological changes some of which could have significant environ-
mental and/or economic consequences. Therefore, model predictions on
current and expected near-shore and coastal hydro- and morphodynamic
states related (but not exclusively) to storm occurrence can be used to
support decision making on Disaster Risk Reduction (DRR) response.
Using this information, the impact of coastal hazards, such as coastal
flooding and beach erosion, can be predicted and eventually mitigated.
One of the tools, which can manage the complex process of merging
data and models and to generate products on current and expected
hydro- and morpho-dynamic state of the coasts, is a coastal forecasting
system (Bogaard et al., 2016). Such systems form a special class of
environmental decision support systems as they operate in real time,
rather than as a tool in support of strategic planning (Matthies et al.,
2007). Output of such system can be valuable for coastal stakeholders
and the entire coastal community with coastal stakeholders being all
interested parties engaged in management and profitable utilization of
the hotspot resources and infrastructure, while coastal community in-
cludes the local population related to the sea.
This paper aims to present the process of implementation and vali-
dation of a coastal forecasting system for a hotspot of national impor-
tance located in Varna Bay (Bulgaria), which is based on Delft-FEWS
(Flood Early Warning System).
The foundation of Delft-FEWS is data-centric with a common data-
model interface, through which all components interact. The main pur-
pose of this framework is to provide a platform for building of an

* Corresponding author.
E-mail address: valchev@io-bas.bg (N. Valchev).

http://dx.doi.org/10.1016/j.coastaleng.2017.08.008
Received 28 January 2017; Received in revised form 14 July 2017; Accepted 7 August 2017
Available online 23 August 2017
0378-3839/© 2017 Elsevier B.V. All rights reserved.
N. Valchev et al.
operational forecasting system that allows flexible integration of data
and models (Werner et al., 2013). Initially developed for operational
forecasting of river flow and potential inland flooding (Burnash and
Singh, 1995; Werner et al., 2005), Delft-FEWS has evolved to cover
marine and coastal hazards (De Kleermaeker et al., 2012). Nowadays,
through activities within the RISC-KIT (Resilience-Increasing Strategies for
Coasts - toolKIT) project, this platform is being further developed by
extending its functionality to serve coastal early warning systems
(Bogaard et al., 2017).
Unlike other well-developed systems focused on forecasting of near-
shore circulation, for instance (Stanev et al., 2011), or waves (Inghilesi
et al., 2012) – the coastal FEWS concept goes beyond by translating
near-shore (boundary) conditions to onshore coastal hazards, such as
flooding, wave overtopping and erosion. This makes estimation of ex-
pected hazard extents and intensities in terms of, for instance, inundation
depth or beach erosion possible.
Application of similar near-shore model systems has been demon-
strated previously. For example, Verlaan et al. (2005) make use of the
hydrodynamic model DCSM (Dutch Continental Shelf Model) in order to
predict storm surge and tidal amplitude at the Dutch coast. Aimed at
coastal monitoring along the Dutch coast, Baart et al. (2009) introduced a
real time forecasting operational system in order to predict and hindcast
morphological storm impacts. In line with this trend, Van Ormondt et al.
(Van Ormondt et al., 2012) demonstrated the potential of the forecasting
system CoSMoS (Coastal Storm Modelling System), a prototype version
of the coastal FEWS, to predict dune and beach erosion during storms, as
well as hazardous swimming conditions during calmer marine weather.
Furthermore, Sembiring et al. (2015) reported another application of the
generic CoSMoS system, which is not limited only to storm impact pre-
diction but can be used to assess other coastal hazards, such as rip cur-
rents and flooding.
A near-real time coastal operational system was also established in
the Black Sea. It provides nowcast and forecast of off-shore sea level and
circulation (Kubryakov et al., 2012). Data originating from this system
together with wave forecast, generated using operational meteorological
data, were employed to test an early warning system prototype for a
stretch of sandy Shkorpilovtsi beach in Bulgaria (Valchev et al., 2014).
This event-based system predicted the storm impact on the beach using a
set of indicators, which convert model results to coastal hazards, such as
depth-velocity, flooding and erosion, and resulted in adequate warnings
and series of actions to be undertaken by responsible authorities. How-
ever, this system could not be sustained due to lack of continuous oper-
ational forcing and boundary conditions data and, more importantly, lack
of a suitable platform to guide and monitor the forecasting process – is-
sues that are resolved in the current application.
As shown by Harley and Ciavola (2013), local coastal inundation risk
can be managed by a combination of a forecasting system and suitable
DRR measures such as artificial dune placement. Clearly, hazard evalu-
ation is not the pinnacle which a coastal early warning system can reach.
It can be complemented by additional modules, bridging hazard and
impact assessment, and ultimately providing guidance for operation of
possible response and preparedness measures.
Aiming at such sophisticated system, RISC-KIT made a progress to-
wards impact-oriented forecasting by developing a Bayesian Decision
Support System (DSS) (J€ ager et al., 2017), which connects site specific
geomorphic setting, hazard extent and intensities, and coastal vulnera-
bility indicators, such as flood depth damage and erosion, in order to
estimate socio-economic impact. Moreover, it can predict to what extent
the impact can be reduced by implementation of various DRR measures.
Similarly to all RISC-KIT case study sites, Bayesian DSS was also devel-
oped for Varna. However, discussion on this module is not considered
herein. Some details can be found in (Cumiskey et al., 2017).
This paper provides an overview of a coastal forecasting system
implementation on four nested domains (Section 3). The study site set-
tings are described in Section 2. Validation of the system through com-
parison of model output with field observations (surge, waves and
213
Coastal Engineering 134 (2018) 212–228
onshore hazard modeling) together with visualization examples are
presented in Section 4. Finally, in Section 5, an insight on obstacles,
limitations and future challenges is presented.
2. Study site
The study site is part of Varna regional coastline (Bulgaria), which is
located on the western Black Sea. The Black Sea is an elliptical semi-
closed basin with weak tides (about 8 cm) with a complex orography
located in south-eastern Europe. The western coastal region is relatively
low in comparison to eastern and southern ones (Fig. 1A). The area of the
Black Sea shelf is about 24% of the total basin area and its outer edge can
be traced along the 100–150 m depth contour (Fig. 1B). The Black Sea
can be considered a fairly busy sea comprising various economic activ-
ities such as ship traffic, fisheries, aquafarming, gas exploration and
recreation.
Varna regional coastline is approximately 70 km-long and includes
Varna Bay, which is an open bay stretching from cape St. George to the
north to cape Galata to the south (Fig. 1C). Its northern arm is entirely
protected by coastal defence structures – rocky revetments, groynes and
jetties – that resulted in formation of several beaches. In the innermost
part of Varna Bay there is a sandy spit – low elevation area, nowadays cut
by two artificial navigable canals connecting Varna Bay to Varna Lake. It
holds the largest transport and port agglomeration industrial complex in
Bulgaria. Another important regional economic activity is tourism, which
supports large seaside resorts, as well as small restaurants, beach bars,
shops, sport and leisure facilities scattered along the coastline. Varna is
the only coastal city in Bulgaria that marks a population growth (2.8%)
(National Statistical Institute, 2011).
The selected hotspot (Fig. 2) – a highly vulnerable coastal sector ac-
cording to Valchev et al. (2016) – includes Varna beach and the break-
water of Port of Varna (herein referred to as port wall). Varna beach is a
1.2 km-long exposed to ESE sand body of variable width – from 5 m in the
northernmost to 110 m in the southernmost part. It is backed by a 10 to
18 m-high cliff and is bordered by a Y-head groyne to the north and by
the port wall to the south. Beach slope varies between 0.09 and 0.18 and
sediment grain size – between 0.28 and 0.81 mm. The port wall is a part
of the port agglomeration industrial complex. It is a 1.18 km-long con-
crete structure with maximum height of 7.2 m. The port wall embank-
ment consists of rocks and tetrapods.
Varna Bay has experienced extreme historical storms propagating in
from the eastern half with the main sources of hazard being high storm
surges, large waves and coastal erosion (Galabov et al., 2015; Trifonova
et al., 2012; Valchev et al., 2012). These events caused damages to
beaches, along with port and other coastal structures, touristic and pri-
vate property (Andreeva et al., 2011; Stakev, 1980).
3. Coastal forecasting system set-up
The primary objective of the coastal forecasting system is to provide
predictions, with sufficient lead time, of coastal hazards resulting from
short-term hydro-meteorological conditions. These hazard predictions
can be then used for impact assessment and eventually as a basis for
making decision to take warning and response actions. This is ensured by
the ability of the Delft-FEWS platform to integrate real-time data from
various observation and forecasting networks, to generate predictions
and to disseminate results through appropriate products to the warning
process (Werner et al., 2013).
3.1. Structure and models
The system structure is designed to guarantee a smooth forecasting
process that is a sequence of steps starting with data import, continuing
with a number of data processing and modelling steps, and resulting in
generation of products to be disseminated through user interface (Werner
et al., 2013). In the current application, hydrodynamic and
N. Valchev et al.
Fig. 1. Schematic map of the study site indicating the model domains: A) Black Sea; B) Bulgarian (BG) shelf; C) Varna Bay and hotspot (port wall and Varna beach). Blue dot denotes
position of the meteorological station, green dot – tide gauge; yellow dot – ADCP. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
morphodynamic models are applied to develop a prediction. They are
organized in a chain so that flooding and erosion at the study site (hot-
spot) can be simulated. The models employed are Delft3D FLOW-WAVE
(Deltares et al., 2014) and XBeach (Roelvink et al., 2009). In particular,
the use of a sequence of coupled Delft3D FLOW-WAVE models is meant to
provide conditions at the hotspot boundary by downscaling the meteo-
rological forcing from basin-wide scale (Black Sea) to the Varna hotspot
domain (port wall and beach), while XBeach is used to model the coastal
hazards – flooding and erosion. The system supports hindcast mode,
through which past storm events can be analyzed, while models are
calibrated and validated. The hindcast runs also generate initial states
for each domain in the modelled system. The logical course of the
forecasting process is outlined in a workflow, which lists functional
steps and the order in which these are to be run. A schematic represen-
tation of the coastal forecasting system and data/work flow can be seen
in Fig. 3.
To increase the lead time, meteorological forecast data from Nu-
merical Weather Prediction models are used. To this end, the system
gathers global winds and atmospheric pressure data of 1/4
horizontal
and 3 h temporal resolution (1/2
and 6 h for the analysis) having a
forecast horizon of 8 days. The data originator is the Global Forecast
System (GFS), which is a weather forecast model produced by the Na-
tional Centre for Environmental Prediction (NCEP) (Yang et al., 2006).
Data come in NetCDF (forecast) and GRIB2 (hindcast/analysis) formats.
Once parsed, all data are stored within the Delft-FEWS database,
which contains both configuration and dynamic time series data. These
data are either imported from external sources (in the presented case
214
Coastal Engineering 134 (2018) 212–228
meteorological forecast solely) or produced by internal data manipula-
tion methods or models run within the system. This is necessary because
most of the data that are imported from external sources are not at the
appropriate temporal and spatial scale to be applied as an input to a
forecasting model. These tools are organized in General Purpose modules
(transformation, interpolation, etc.) and are used for instance to inter-
polate data on grids of different resolution or to derive scalar time series
out of the gridded output (Deltares and Delft-docu, 2015).
The core of the system is the forecasting module the key element of
which is the General Adapter. It controls the forecasting process by
communicating with external models via specific model adapters. In
other words, the external models Delft3D FLOW-WAVE and XBeach are
incorporated in Delft-FEWS using their own model adaptors.
3.2. Forecasting process
The forecasting process is designed as sequential Delft3D FLOW-
WAVE model runs on three nested domains starting from the Black Sea
(Fig. 1A). This procedure ensures a smooth transition of boundary con-
ditions to the area of interest (hotspot) and is necessary due to lack of
available boundary conditions for that location. The aim of modeling on
the first domain is to generate hydrodynamic conditions in terms of water
levels, flow velocities and 2D wave spectra on a basin scale that are used
to feed the open boundaries of the second domain - Bulgarian shelf
(Fig. 1B). In its turn, the shelf model output provides hydrodynamic
conditions at the open boundary of the highest resolution Delft3D model
domain – Varna Bay (Fig. 1C). This model output feeds the final model in
N. Valchev et al.
Fig. 2. View of the hotspot (Photo by George Palov; http://panomagic.eu/en/3919).
the chain of models– XBeach, as the computational domain of the latter
covers the selected hotspot (Fig. 4).
The forecast generated by XBeach model, consisting of time series of
maximum inundation depth and sedimentation-erosion pattern, among
others, are subsequently fed into the Bayesian DSS and eventually used
for impact assessment and warning release. All resulting time series data
(scalar, grids and spectra) are stored in the database.
The hydrodynamic model Delft3D-FLOW is implemented to simulate
2D depth-averaged, unsteady flow and transport phenomena resulting
from meteorological forcing (Deltares et al., 2014). The effect of density
differences due to a non-uniform temperature and salinity distribution
(density-driven flow) are not considered. For the two nested domains,
seaward boundary is assumed to be controlled by water level, while the
other two water boundaries (north and south) – by normal component of
the current velocity. The wave model Delft3D-WAVE solves the wave
Fig. 3. Schematic representation of the coastal forecasting system structure and data/work flow (thick grey lines indicate data export, dotted grey line – data import, shaded thicker grey
lines – GFS meteo data).
215
Coastal Engineering 134 (2018) 212–228
action balance equation and is coupled online with Delft3D-FLOW. The
wave model is run in the third-generation mode accounting for the
following processes - generation by wind; dissipation by whitecapping,
bottom friction and depth-induced breaking; and non-linear wave-wave
interaction (quadruplets and triads) (Deltares et al., 2016).
The 2DH morphodynamic model XBeach is used for calculation of
wave propagation, long waves and mean flow, sediment transport and
morphological changes of the near-shore area and Varna beach (Roelvink
et al., 2015). It is implemented in ‘surfbeat’ mode, which includes run-up
and run-down of long waves (swash), and uses morphological accelera-
tion factor of 10 (Ranasinghe et al., 2011). The hard structures located
within the hotspot area are presented as non-erodible layer. Beach sed-
iments are defined as a single sediment fraction with D50 ¼ 0.41 mm and
D90 ¼ 0.67 mm, which are the mean values of those properties for all
sediment samples gathered within the project.
N. Valchev et al. Coastal Engineering 134 (2018) 212–228

Fig. 4. Bathymetry and topography of Varna hotspot (port wall and beach) and XBeach model grid: Cartesian, curvilinear with variable cross-shore (1 m at the beach face and 30 m off-
shore) and long-shore (from 9 m to 68 m NE to SW along the hotspot) resolution. Blue dot denotes location for which boundary conditions from Delft3D FLOW-WAVE are provided.

Detailed information about the extent, grid, time step of all individual
models within the forecasting module together with related input (ba-
thymetry, topography, forcing, and boundary conditions) and output
data is presented in Table 1.
4. Coastal forecasting system validation
The coastal forecasting system was tested and subsequently validated
for a sequence of three short-duration relatively weak storm events,
which occurred during February 4–12, 2015, with the second one the
most energetic. They arose by the interaction between a deep cyclone
over the Mediterranean Sea (absolute minimum 900 hPa) and a large
area of high atmospheric pressure spreading over the Northern Europe
(absolute maximum 1032 hPa) (Fig. 5A). The frontal zone of strong
pressure gradient between these two atmospheric circulation systems
passed over the Black Sea resulting in predominantly N to NE winds with
speed of up to 14 m/s (historic data are available either on www.
wunderground.com or www.stringmeteo.com for weather station
15552). Delft-FEWS visualization of GFS wind data is presented in
Fig. 5B. The three events were divided by transitions in wind speed and/
or direction, which predetermined to a great extent the model behavior.
From regional climate viewpoint the wind speed magnitude represents

216
N. Valchev et al. Coastal Engineering 134 (2018) 212–228

Table 1
Overview of models and specific set-up details at each domain.

DOMAIN

Black Sea Bulgarian shelf Varna Bay Varna hotspot

Model D3D FLOW-WAVE D3D FLOW-WAVE D3D FLOW-WAVE XBeach

Extent (lon  lat) 16.3
 7.3
210  195 km 23  7 km 2.35  2.75 km
Grid type Spherical, regular Spherical, regular Spherical, regular Cartesian, curvilinear
Grid resolution 5 km 1 km 111 m variable
Time step 10 min 5 min 1 min automatic based on Courant criterion
Bathymetry EMODnet, 1/8’ (EMODnet EMODnet, 1/80 Echo sounding/maps, 20 m Echo sounding/navigation charts, 1 m
Bathymetry Consortium, 2016)
Topography N.A. N.A. N.A. DEM, horizontal 1 m/vertical 10 cm
Forcing GFS wind and pressure GFS wind and pressure GFS wind and pressure N.A.
Boundary conditions No external boundary water levels, flow velocities, water levels, flow velocities, water levels, 2D wave spectra
2D wave spectra 2D wave spectra
Output H, flow velocities, 2D wave H, flow velocities, 2D wave H, flow velocities, 2D wave H, Hrms, Dm,, water depth, flow velocities, discharge,
spectra, Hs, Tp and Dm spectra, Hs, Tp and Dm spectra, Hs, Tp and Dm bed level, sedimentation-erosion pattern

moderate to strong winds related to typical (high-probability) winter

storms. Additionally, it is to be noted that storms hitting from the north CB ¼ bx  by (2)
have become increasingly frequent during the latest decade (Valchev
et al., 2012). bx ¼ arctan ðRÞ (3)

4.1. Data and methods
In order to validate the system, the model output was compared with
measurements gathered during the same period, which are expected to
represent the ground truth. Validation data consist of:
 Sea level measurements by tide gauge (VEGAPULS 61 manufactured
by VEGA Grieshaber KG, Germany) located in the Port of Varna. Data
precision is ±5 mm.
 Offshore waves registered at depth of 22.5 m by Acoustic Doppler
Current Profiler (ADCP) Workhorse Sentinel 600 kHz manufactured
by Teledyne RD Instruments, USA. Wave time series with 20 min
duration were registered once per hour. Data accuracy is ±1% (lo-
cations of the tide gauge and the ADCP are indicated in Fig. 1B).
 Pre- and post-storm beach topography measurements using an
autonomous Trimble® Site Positioning System with RTK corrections
(0.05 m horizontal and 0.07 vertical precision).
 Sediment properties (D50 and D90) derived from a pre-storm sediment
sampling.
The output hydrodynamic parameters of interest were water level,
significant wave height (Hs), peak wave period (Tp) and mean wave di-
rection (Dm), as well as 2D wave spectra. The validation aimed at proving
the quality of boundary conditions generated by Delft3d FLOW-WAVE
model, which feed the hazard model. It was done by direct comparison
between modelled and observed series and estimation of their overall
bias and root-mean square errors. For directional data, circular correla-
tion and circular bias are used as statistical error metrics (Berens, 2009).
The circular bias is defined by subtracting the mean angular of the
computed parameter from the mean angular of the observed. Herein, the
mean angular is computed by transforming the directional data into two
vector components with magnitude of unity, and then taking the four
quadrant inverse tangent of the resultant of the vectors as the mean
angular. Circular correlation is computed by defining correlation coeffi-
cient of the directional data by making use of the mean angular measures.
Circular correlation (CC) and circular bias (CB) are defined as follows
(Berens, 2009):
P
N illustrate the result of implementation of the forecasting system for the
sin ðxi  bx Þ sin ðyi  by Þ
CC ¼ si¼1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN
sin2 ðxi  bx Þ sin2 ðyi  by Þ
i¼1
N  
1 X cos ðmi Þ
R ¼ (4)
N i¼1 sin ðni Þ
In Eq. (1)–(4), b
x and by are the mean angular of the simulated and
observed parameters respectively, ðm; nÞ is the plane component of the
directional data (unit vector), and R is the mean resultant vector.
As for morphodynamics, the validation process was focused on the
changes of sub-aerial part (above the mean sea level) of Varna beach in
terms of erosion-sedimentation pattern and eroded volumes. This
approach was preferred since beach erosion is considered as one of the
most important hazards to be forecast. To this end, XBeach model pre-
storm bed was built by modifying the default bed with pre-storm beach
topography (measured on January 30, 2015). Since there were no data
on pre- and post-storm bathymetry, the most dynamic part of the bottom
(0–5 m depths) was replaced by a surface generated using Dean formu-
lation (Dean, 1977) for sand properties of samples collected during the
pre-storm field survey. Following the same approach, a post-storm bed
was created on the basis of topography data measured on February 12,
2015. Thus, the difference between the pre- and post-storm test beds
reflect the morphological changes that took place on the beach as a result
of the considered events. The morphological analysis is supported by
calculation of Brier Skill Score (BSS) (Sutherland et al., 2004). Addi-
tionally, positions of measured and calculated maximum wave run-up,
representing the flooding hazard, were compared.
4.2. Visualization of predictions
The first event took place on February 4–5, during which SE-ESE
winds had maximum speed of about 9 m/s. The transition between the
first and second event was marked by a sharp change of wind direction to
NW. The second event lasted three days (February 6–8). During this time-
span wind remained almost stable blowing from NE as its speed reached
14 m/s. In the course of the storm decay wind started turning to NW.
Between the second and the third event there was a transitional one-day
episode marked by weak NNW winds. After that, wind got steady both in
direction (N) and speed (about 12 m/s) during the course of the third
event that lasted for three days (February 10–12). The following figures
most energetic instance of the considered period.
(1) Fig. 6A shows that surge was predicted for the western and north-
western Black Sea shelf that measured up to 27 cm. During the same
time there was an area of high waves spreading all over the western half

217
N. Valchev et al.
Fig. 5. Mean sea level pressure and wind vector map over Europe (A) and wind field over the Black Sea (B) on February 6, 2015 18:00 GMT.
of the basin with a spatial maximum near the Bulgarian shelf (Fig. 6B),
where significant wave height was predicted to reach 3.5 m, which is
quite moderate, given that it can be three times larger in case of a severe
storm (Valchev et al., 2012).
Downscaling to the Bulgarian shelf predicted higher surges of 24 cm
for the northern half (Fig. 7A), whereas maximum flow velocity was
0.4 m/s (Fig. 7B). Storm waves affected to a greater extent the same
region as the orographic features sheltered the two large bays – Varna
and Burgas. Waves with maximum significant wave height of 3 m and
peak periods of 7.5 s was predicted for the area off-shore of Varna Bay
(Fig. 7C and D) but due to the wave direction they arrived carrying
considerably less energy reaching only 1.8 m at the bay (Fig. 8B) and
1.67 m at the hotspot (Fig. 9B) offshore limits.
218
Coastal Engineering 134 (2018) 212–228
An interesting feature can be seen in Fig. 9A, which shows the pre-
dominant storm direction of coastal currents (in this case the flow ve-
locity is 0.7 m/s) causing sediments to be carried away from the northern
part of the beach and accumulated on the southern one. This flow feature
can explain the present-day aspect of Varna beach. In addition, Fig. 9C
and D displays bottom level and erosion pattern at the hotspot that will
be discussed further in the paper.
4.3. Validation results and discussion
4.3.1. Meteorological forcing
As mentioned above the forcing data are provided by the GFS analysis
products. Generally, they show good agreement with observations, in
N. Valchev et al.
Fig. 6. Visualization of forecasting system results for water level (A) and significant wave height (B) for the Black Sea domain on February 6, 2015 21:00 GMT.
particular with respect to mean sea level pressure (MSLP) (Fig. 10).
However, some discrepancies are noted that could explain the models'
performance. Generally, GFS underpredicts the wind speed magnitude
except for the short transitional periods between the identified events. In
case of the February 9 episode this is particularly discernable. For that
period, GFS predicts SW to W winds, while in fact the wind direction was
45–50
closer to N – W to NNW. A large discrepancy in direction can be
seen for the first event as well. The agreement is best for the most en-
ergetic event, while for the last one it is weaker in particular with respect
to wind direction – NNW (GFS analysis) versus NNE (observations).
219
Coastal Engineering 134 (2018) 212–228
4.3.2. Surge
Prevailing winds from the northern quarter did not allow for a
considerable elevation of water level. Besides, time-spans of steady winds
were rather short, which prevented a continuous rise in water level
during the whole period of simulation. Delft3D FLOW-WAVE model
performance for surge level is analyzed by comparing simulated surge
with observations at a location in the Port of Varna. Fig. 11 presents
water levels' plot for that location. It shows that simulated surge levels
are in very good agreement with observations, except for the second
storm, for which the difference between surge level maxima is 0.07 m.
N. Valchev et al.
Fig. 7. Visualization of forecasting system results for water level (A), flow velocity (B), significant wave height (C) and peak wave period (D) for the Bulgarian shelf domain on February 6,
2015 21:00 GMT.
The simulated surge level maximum is 0.27 m. For the rest, predicted
surge level local maxima correspond to the observed ones in terms of
magnitude: 0.08 m (first storm), 0.10 m (transitional episode) and 0.13 m
(third storm). However, they were reached with a certain delay, mark-
edly during the second half of the validation period.
Table 2 lists error statistics of considered parameters for the entire
validation period, as well as for the three identified events. Bias values
for surge level are found mostly negative varying between 4 cm and
6 cm. Moreover, the observed surges exceed at most the modelled ones
when the entire period is considered. Surge root-mean-square error is
relatively higher and its maximum coincides with the second storm
(0.07 cm), while for the rest of the time it is constant (0.05 cm).
4.3.3. Waves
To large extent, the nature of wave model performance is attributed to
the quality of available forcing conditions. As reported in other studies,
the quality of the wave model decreases substantially when the wind
conditions show strong temporal and spatial gradients. This is particu-
larly valid for enclosed basins, where underestimation of wind speed by
the atmospheric model is often found (Akpinar and Ponce de Le on, 2016;
Davidan et al., 2006).
As in case of the water level, predominantly northern and moderate
winds did not allow for generation of large waves. In such conditions,
sheltering effect of jutted capes Kaliakra and St. George resulted in strong
refraction of wave field. Besides, as already mentioned, the analysis tends
to hindcast a larger northern or even western component of wind vector
that hinders the model and causes it to predict less amount of energy
entering the bay with quite a pronounced delay of several hours with
respect to the observations. This is deemed to be directly connected with
the wind direction since the larger the bias between predicted and real,
the bigger the delay. Nevertheless, results show that simulated wave
parameters are in good agreement with observations in particular during
the second storm. Comparison between modelled and observed time
series of wave parameters at the location of ADCP is presented in Fig. 12.
The difference between modelled and observed Hs maxima for the second
220
Coastal Engineering 134 (2018) 212–228
storm is 25 cm, while for the last one it is considerably less (6 cm). The
largest is difference for the first event (40 cm).
Biases of Hs and Tp are all negative and the smallest one is estimated
for the third event (Table 2). Root-mean-square errors are in the range
0.2–0.35 m for Hs and 0.93–1.15 s for Tp. A positive circular bias is
calculated only for the first event (21.7
), which is due to the larger
southern component of the hindcast wave direction; while the negative
bias is strongest for the last one (13.8
), for which the model predicted
NE direction, while in reality it was ENE. Logically, the circular corre-
lation is not very high but decreases in the course of the events. It is
highest for the period as a whole (0.73) (Table 2). In terms of wave di-
rection, the best prediction is achieved for the second and most ener-
getic storm.
Scatter plots presented in Fig. 13 confirm that in most of the cases
model underpredicts wave parameters. They also suggest moderate cor-
relation between model and observations except for Tp due to the larger
scatter. However, the linear fit for this property is closest to the perfect
one because the data cloud is uniformly distributed around it. The pre-
diction error increases as wave parameters grow but the error is highest
for surge (since considerable deviation is noted already for the bin
0 < H < 0.15 cm) and lowest for Tp.
Described wave model behavior conforms to the findings of Cher-
neva et al. (2008), who reported an underestimation of significant
wave heights by WAM Cycle 4 model in the case of low wind en-
ergy input.
As for 2D wave spectrum, comparison of observed and modelled
spectra (Fig. 14) reflects the influence of difference between the real and
predicted wind forcing as described above. As a result, a bias of almost
20
between mean wave directions of propagation is noted. Moreover,
the directional spreading of modelled waves is much broader – N-ESE – as
opposed to the observed one – NE-ESE. It is closely followed by spreading
of frequencies, which is particularly noticeable for wave components
with spectral energy higher than 20 J/m2/Hz/deg. It is in the range of
0.16–0.25 Hz as predicted by the model in comparison to 0.12–0.17 Hz as
calculated from the field data.
N. Valchev et al.
Fig. 8. Visualization of forecasting system results for water level (A) and significant wave height (B) for the Varna Bay domain on February 6, 2015 21:00 GMT.
Thus, due to the stronger refraction, the modelled wave field shows a
more irregular relatively short-crested waves with periods varying be-
tween 4.0 and 6.3 s as opposed to the observed one, which is depicted as
more regular with periods 5.9–8.3 s. This affects directly the amount of
wave energy (37.27 vs. 45 J/m2/Hz/deg) that enters the bay, and ulti-
mately the magnitude of Hs (1.47 vs. 2.08 m).
Since the validation results are obtained for low-energy and mark-
edly fetch-limited conditions, it can be claimed that the quality of
predictions would be much higher for high-energy (low-probability)
events with greater wind fetch. Hence, for the current configuration,
221
Coastal Engineering 134 (2018) 212–228
Delft3D FLOW-WAVE model predictions are of acceptable accuracy
capable of providing the best possible boundary conditions to the
hazard model.
4.3.4. Onshore hazard modeling
The aim of validation is to assess the ability of the XBeach model to
reliably simulate the main hazards of interest: erosion and flooding.
Since there are no pre- and post-storm bathymetries available, it is not
possible to validate the sedimentation pattern. This is not considered a
major problem as, on one hand, the reconstructed equilibrium bed is
N. Valchev et al. Coastal Engineering 134 (2018) 212–228

Fig. 9. Visualization of forecasting system results for mean flow velocity (A) and mean root-mean-square wave height (B), bottom level (C) and beach erosion pattern (D) for the hotspot
domain on February 6, 2015 21:00 GMT.

believed to represent sufficiently well the initial morphological condi- on the beach and hinterland area. Moreover, the system in its decision
tions and, on the other, it is logical to focus on the beach face and berm, support module is oriented towards assessment of feasibility of DRR
since their state is of key importance for evaluation of coastal impacts measures to be implemented for the area above the water level.

Fig. 10. GFS analysis vs. observed times series of wind parameters for February 4–12, 2015: (from top to bottom) wind direction, wind speed and mean sea level atmospheric pressure at
Varna Bay (blue dot in Fig. 1B). Shaded areas represent transitions between identified events.

222
N. Valchev et al. Coastal Engineering 134 (2018) 212–228

Fig. 11. Model vs. observed water levels for February 4–12, 2015 at Varna Bay (green dot in Fig. 1B). Shaded areas represent transitions between identified events.

The simulated morphological changes during the validation period

Table 2
Error statistics of surge level (H), significant wave height (Hs), peak wave period (Tp) and
show non-homogeneous pattern of erosion and accumulation along
mean wave directions (Dm) for whole validation period as well as for the three identified Varna beach. The overall picture presented in Fig. 15 reveals that the
events within it. erosion is concentrated in a narrow strip along the beach face, while the
Event Parameter Bias RMS Circular Circular accumulation occurs at a wider spatial extent especially in the most
error correlation bias southern part of the submerged beach. This considerable accumulation
Whole period H 0.06 m 0.06 – – is attributed to the collision of two opposing long-shore wave currents
4-12/02 Hs 0.25 m 0.25 – – (as shown in Fig. 9A), which creates conditions for local sedimentation.
Tp 1.15 s 1.15 – – On the northern end of the beach, the accumulation is minor due to the
Dm – – 0.73 10.0
direction of long-shore current and presence of groyne cutting the
First H 0.002 m 0.05 – –
sediment supply. Thus, it can be stated that the local currents are
4-5/02 Hs 0.18 m 0.22 – –
Tp 0.87 s 1.14 – – determined by the configuration of port wall resulting in different
Dm – – 0.62 21.7
accumulation along the beach, which in its turn influences the flow
Second H 0.04 m 0.07 – – pattern and erosion rate.
6-8/02 Hs 0.11 m 0.35 – –
In order to avoid misinterpretation, it should be noted that the un-
Tp 0.57 s 1.13 – –
Dm – – 0.59 8.4

derlying orthophoto image was taken in December, 2013 and therefore
Third H 0.02 m 0.05 – – does not represent the beach morphology at the time of simulations.
10-12/02 Hs 0.02 m 0.20 – – Fig. 16 presents a comparison between calculated and measured
Tp 0.22 s 0.93 – – morphological changes on three profiles (24, 43 and 50 as indicated in
Dm – – 0.44 13.8

Fig. 15) sliced out of the 2D XBeach output. To the north, the dry

Fig. 12. Model vs. observed time series of wave parameters for February 4–12, 2015 at Varna Bay (yellow dot in Fig. 1B): (from top to bottom) significant wave height, peak wave period
and mean wave direction. Shaded areas represent transitions between identified events.

223
N. Valchev et al.
Fig. 13. Scatter plots of model and observed (from left to right) surge level (H), significant wave height (Hs), peak wave period (Tp). Thick line represents the best fit and dashed line –
linear fit. Estimated R2 are 0.64, 0.76 and 0.72, respectively.
beach is presented by a very narrow mildly-sloped (0.09) sand strip
backed by adjacent buildings. There, both calculated and observed
thicknesses of eroded layers are the least (profile 24). The difference
between them is mainly attributed to the fact that coarser sediments
(D50 ¼ 0.41 mm) were used as input parameter in the model, while in
reality they are finer (D50 ¼ 0.28 mm). Moreover, there is a shortage
of sediment supply at this location and the beach berm is not as thick
as assumed in the model. This can partially explain the low BSS value
of 0.15 (Table 3).
Profile 43 is representative for the central beach part, which is
exposed to wave action to a greater extent; therefore, the slopes are
steeper (0.16) and sediments – coarser (D50 ¼ 0.42 mm, equal to the one
used in the model). In addition, the berm is much thicker with respect to
profile 24. On that account, the difference between calculated and
measured eroded volumes is minor – evidence for very good model skills
– BSS ¼ 0.9 (Fig. 16, Table 3).
Generally, the increased accumulation that takes place in the south-
ern part of the beach during storms contributes to higher wave energy
dissipation and prevents excessive erosion. This pattern is observed in the
course of considered events as well and is illustrated by the morpho-
logical changes on Profile 50 (Fig. 16). The model performance is not as
good as in the previous case (BSS ¼ 0.47); nevertheless, there is a fair
resemblance between calculated and measured post-storm profiles. This
judgment is supported by the comparable eroded volumes (Table 3).
Additionally, due to the use of equilibrium profile formulation, the
presence of bars in this area is not reflected in the model bed that results
in excessive erosion of the beach face. As a whole, the measured post-
storm profiles are characterized with milder beach face slopes, which
can be attributed to the fact that the real wave field was longer-crested in
comparison to the modelled one.
As for the flooding hazard, XBeach underpredicts wave run-ups along
the entire beach as the difference between calculated and measured
positions ranges between 1.6 m (to the north) and 15 m (in the central
part) (Fig. 17). This problem might be connected with the ‘surfbeat’
mode that resolves only the long-wave run-up, in particular having in
mind that the wave field provided at the off-shore model boundary was
shorter-crested in comparison with the real one. This mode is more
suitable for dissipative slopes as those in the northern part of Varna
beach, where the difference between observed and calculated wave run-
up maxima is minor. However, on steeper slopes, as those in the central
and south beach, the contribution of the short waves run-up gains
importance resulting in underprediction of the run-ups. Similar problems
were reported in (Splinter et al., 2011; Stockdon et al., 2014). This might
224
Coastal Engineering 134 (2018) 212–228
be an argument for application of non-hydrostatic mode, which resolves
short wave run-up and overwash.
5. Obstacles and challenges
The obstacles on the way to successful implementation of an efficient
forecasting system can arise from uncertainties related to different
sources such as input data for the models, capabilities and limitations of
the models themselves, availability of data for estimation of the forecast
error, etc.
Previous studies, for instance (Jorda et al., 2012), have shown that
the uncertainties introduced by the ocean models are small compared to
the ones related to the accuracy and resolution of the atmospheric forc-
ing. Presented results also clearly show that even relatively small forcing
forecasting error could distort predictions along the model chain. In this
sense, although the GFS forecast is a plausible starting point for the
system testing, the lack of freely available high resolution regional
meteorological forecast is considered as one of the major constraints.
Input topography data of 1 m horizontal and 10 cm vertical resolution
are of high accuracy and cannot be considered a major source of uncer-
tainty. However, the situation is completely different when it comes to
shallow bathymetry and beach face topography data. Even if those in use
are considered of reasonable accuracy, they should be regularly updated,
in particular during seasons marking distinct changes in the bed profile.
The sediment data also could affect the accuracy of simulation since
homogeneous sediments were used as model input, while in reality
sediment sizes vary along the coast being coarser in the southern and
finer in the northern section of the Varna beach. The importance of issues
related to accuracy of the coastal geomorphic data underpins the need for
constant and more thorough coastal monitoring.
Furthermore, the limitations of forecasting and diagnostic capacity of
the system can be connected to any of the models in the chain. Models'
accuracy is related to reliable downscaling of hydrodynamic and wave
conditions from basin-wide to near-shore scale and further onshore to the
complex setting of the hotspot. Authors assume that the models' perfor-
mance does not constitute a major constraint. The presented results
affirmed that Delft3D FLOW-WAVE model reproduces well the water
level fluctuations. A possible limitation in the flow model set-up is that
flow conditions of the large scale models are not properly resolved since
they do not account for the effect of density differences due to a non-
uniform temperature and salinity distribution. The model error related
to this issue is still to be quantified but it is not expected to be an
important source of inaccuracy due to time scale of the event driven
N. Valchev et al.
Fig. 14. Calculated from field data (top) versus model (bottom) 2D wave spectrum at
ADCP location on February 6, 2015 15:00 GMT.
processes and absence of strong near-shore density gradient. Moreover,
as it can be seen in Fig. 7B, the large scale flow conditions during the
important storms hitting from NE quarter does not differ from the typical
circulation pattern.
225
Coastal Engineering 134 (2018) 212–228
Fig. 15. Sedimentation-erosion pattern: blue area indicates erosion, red area – sedimen-
tation. Position of profiles 24, 43 and 50 is indicated. (For interpretation of the references
to colour in this figure legend, the reader is referred to the web version of this article.)
Evidently, the wave model predictions are very sensitive to accuracy
of the available forcing data. Although the models coped to predict the
overall spatial variability of the wave field, it tended to depict it as
shorter crested, reaching its fully developed stage with a certain delay.
This is deemed to be related to the capability of correct reproduction of
wave direction, which brings on other issues such as underestimation of
wave energy. However, an account should be rendered to the duration of
validation period, which is too short for any definitive conclusion to be
drawn. The models' forecasting error should be estimated on the basis of
longer term functioning of the system and providing that there exists
continuous flow of observation data.
As for the hazard modeling, XBeach tends to overpredict the erosion
and to underestimate the wave run-up extent. The latter limitation is due
partly to the application of ‘surfbeat’ mode.
Finally, problems with model performance are related to the grid
design since it is quite difficult to represent the complex coastal
configuration within the hotspot in equal details, especially since, the
grid resolution should be consistent with the forecasting time
limitations.
Thus, a further improvement of the forecasting system is possible by
provision of a more accurate regional atmospheric forcing input, up-to-
date beach face topography and shallow bathymetry data and by in-
crease of grid resolution, including the use of alternative approaches to
compose the model grid such as domain decomposition or flexible mesh.
N. Valchev et al. Coastal Engineering 134 (2018) 212–228

Fig. 16. Comparison between calculated and measured morphological changes on profiles 24, 43 and 50.

sedimentation-erosion pattern, among others, that can be further used for

Table 3
impact assessment in the Bayesian Decision Support System and even-
Calculated and measured eroded volumes and BSS for profiles 24, 43 and 50.
tually can result in warning release.
Profile Eroded volume [m3] BSS
Considering the validation, it can be asserted that Delft3D FLOW-
Calculated Measured WAVE model accuracy, although arguable, is satisfactory and results
24 3.7 1.6 0.15 obtained give a sound basis for estimation of onshore hazards. XBeach
43 8.3 6.7 0.90 tends to overpredict the erosion and to underestimate the flooding extent
50 5.3 6.1 0.47 but in reasonable limits, thus enabling us to claim that the model forecast
skill is good enough for the greater part of Varna beach and can suc-
Regarding the XBeach model, the application of non-hydrostatic mode cessfully serve for evaluation of onshore impact.
may lead to the correct reproduction of the flooding extent by taking into A strong side of the proposed coastal hazard forecasting system is that
account the short wave run-up and overwash, which are important on it is based on Delft-FEWS, which has proved to be a reliable and effective
steeper slopes. However, these improvements should be pursued in bal- operational platform allowing for flexible integration of data and models.
ance with overall system computational time requirements. Another advantage of the system is the use of state-of-the-art models
capable to downscale the large-scale hydrodynamic conditions to the
complex setting of the hotspot with reasonable accuracy. Moreover, the
6. Conclusions
way the system advanced within RISC-KIT project made possible its
evolution into an impact-oriented prediction tool additionally able to
The presented paper describes the process of set-up and validation of
provide rapid estimation of coastal vulnerability as well as to test effec-
a multi-domain coastal forecasting system aimed to predict intensities
tiveness of various DRR measures. Besides, the system could easily be
and extents of flooding and erosion hazards at the study site of Varna
upgraded to an early warning system.
(Bulgaria). Conditions at the hotspot boundary are provided by a chain of
Further improvements could be sought by provision of a high-
coupled Delft3D FLOW-WAVE models. To this end, Delft3D models are
resolution regional atmospheric forcing input, up-to-date beach face
applied consecutively on three domains: Black Sea, Bulgarian shelf and
topography and shallow bathymetry data achieved through a regular
Varna Bay. As a final step, the XBeach model generates forecast for the
coastal monitoring, and increase of model grid resolution, which
hotspot consisting of time series of maximum inundation depth and

226
N. Valchev et al.
Fig. 17. Comparison between measured and calculated positions of maximum wave run-
up on Varna beach.
should be in concurrence with the operational requirements of
the system.
Acknowledgement
This work was supported by the European Commission 7th Frame-
work Programme through the grant to RISC-KIT (Resilience-increasing
Strategies for Coasts – Toolkit), contract no. 603458, and contributions
by the partner institutes.
References
Akpinar, A., Ponce de Le on, S., 2016. An assessment of the wind re-analyses in the
modelling of an extreme sea state in the Black Sea. Dyn. Atm. Oceans 73, 61–75.
Andreeva, N., Valchev, N., Trifonova, E., Eftimova, P., Kirilova, D., Georgieva, M., 2011.
Literary review of historical storm events in the western Black Sea. In: Proc.. of Union
of Scientists – Varna, “Marine Sciences”, pp. 105–112 (In Bulgarian).
Baart, F., van der Kaaij, T., van Ormondt, M., van Dongeren, A., van Koningsveld, M.,
Roelvink, J.A., 2009. Real-time forecasting of morphological storm impacts: a case
study in The Netherlands. J. Coast. Res. SI 56, 1617–1621.
Berens, P., 2009. CircStat: a MATLAB Toolbox for circular statistics. J. Stat. Softw. 31,
1–21.
Bogaard, T., De Kleermaeker, S., Jaeger, W.S., van Dongeren, A., 2016. Development of
generic tools for coastal early warning and decision support, 3rd European
Conference on Flood Risk Management FLOODrisk 2016. E3S Web Conf. 7, 18017.
Bogaard, T., De Kleermaeker, S., J€ager, W.S., 2017. Development of Generic tools for
Coastal Forecasting and Warning.
227
Coastal Engineering 134 (2018) 212–228
Burnash, R., 1995. The NWS river forecasting system catchment modelling. In: Singh, V.
(Ed.), Computer Models of Watershed Hydrology. Water Resources Publications, New
York, USA, pp. 311–366.
Cherneva, Z., Andreeva, N., Pilar, P., Valchev, N., Petrova, P., Guedes Soares, C., 2008.
Validation of the WAMC4 wave model for the Black sea. Coast. Eng. 55, 881–893.
Cumiskey, L., Priest, S., Valchev, N., Viavattene, C., Costas, S., Clarke, J., 2017. A
Framework for Considering the (Inter)dependencies between Disaster Risk Reduction
Measures in Coastal Risk Assessment.
Davidan, I., Valchev, N., Belberov, Z., Valcheva, N., 2006. Assessment of the reanalyzed
wind field accuracy for wave modelling purposes in the Black Sea region. In: Proc. of
4th Int. Conf. on EuroGOOS “European Operational Oceanography: Present and
Future”, pp. 801–805.
De Kleermaeker, S., Verlaan, M., Kroos, J., Zijl, F., 2012. A new coastal flood forecasting
system for The Netherlands. Proc.. Hydro12 135–140.
Dean, R.G., 1977. Equilibrium Beach Profiles: U.S. Atlantic and Gulf Coasts, Department
of Civil Engineering. Ocean Engineering Report No. 12. University of Delaware,
Newark, DE.
Deltares, Delft-FEWS Documentation, 2015. URL: https://publicwiki.deltares.nl/display/
FEWSDOC/Home.
Deltares, Delft3D-FLOW, User Manual, 2014, p. 710.
Deltares, Delft3D-WAVE, User Manual, 2016, p. 220.
Galabov, V., Kortcheva, A., Bogatchev, A., Tsenova, B., 2015. Investigation of the hydro-
meteorological hazards along the Bulgarian coast of the Black Sea by reconstructions
of historical storms. J. Env. Protec. Eco 16, 1005–1015.
Harley, M.D., Ciavola, P., 2013. Managing local coastal inundation risk using real-time
forecasts and artificial dune placements. Coast. Eng. 77, 77–90.
Inghilesi, R., Catini, F., Bellotti, G., Franco, L., Orasi, A., Corsini, S., 2012. Implementation
and validation of a coastal forecasting system for wind waves in the Mediterranean
Sea. Nat. Hazards Earth Syst. Sci. 12, 485–494.
J€ager, W.S., Christie, E.K., Hanea, A.M., den Heijer, C., Spencer, T., 2017. A Bayesian
network approach for coastal risk analysis and decision making. Coast. Eng. http://
dx.doi.org/10.1016/j.coastaleng.2017.05.004 (in press).

Jorda, G., Gomis, D., Alvarez-Fanjul, E., Somot, S., 2012. Atmospheric contribution to
Mediterranean and nearby Atlantic sea level variability under different climate
change scenario. Global Planet. Chang 80–81, 198–214.
Kubryakov, A.I., Korotaev, G.K., Dorofeyev, V.L., Ratner, Yu. B., Palazov, A., Valchev, N.,
Malciu, V., Matescu, R., Oguz, T., 2012. Black Sea coastal forecasting systems. Ocean.
Sci. 8, 183–196.
Matthies, M., Giupponi, C., Ostendorf, B., 2007. Environmental decision support systems:
current issues, methods and tools (Preface). Env. Model. Softw. 22, 123–127.
National Statistical Institute, 2011. 2011 Population Census in the Republic of Bulgaria
(Final Data), 2011, p. 51. URL: www.nsi.bg/census2011/PDOCS2/
Census2011finalen.pdf.
EMODnet Bathymetry Consortium, 2016. EMODnet Digital Bathymetry (DTM). EMODnet
Bathymetry. http://dx.doi.org/10.12770/c7b53704-999d-4721-b1a3-
04ec60c87238.
Ranasinghe, R., Swinkels, C., Luijendijk, A., Roelvink, D., Bosboom, J., Stive, M.,
Walstra, D., 2011. Morphodynamic upscaling with the MORFAC approach:
dependencies and sensitivities. Coast. Eng. 58, 806–811.
Roelvink, D., Reniers, A., van Dongeren, A., van Thiel de Vries, J., McCall, R.,
Lescinski, J., 2009. Modelling storm impacts on beaches, dunes and barrier islands.
Coast. Eng. 56, 1133–1152.
Roelvink, D., van Dongeren, A., McCall, R., Hoonhout, B., van Rooijen, A., van Geer, P., de
Vet, L., Nederhoff, K., Quataert, E., 2015. XBeach Technical Reference: Kingsday
Release, p. 141. Deltares.
Sembiring, L., van Ormondt, M., van Dongeren, A., Roelvink, D., 2015. A validation of an
operational wave and surge prediction system for the Dutch coast. Nat. Hazards Earth
Syst. Sci. 15, 1231–1242.
Splinter, K., Palmsten, M., Holman, R., Tomlinson, R., 2011. Comparison of measured and
modeled run-up and resulting dune erosion during a lab experiment. Proc.. of the
Coastal Sediments 2011. World Sci. 1, 782–795.
Stakev, M., 1980. About the break-down state of some coastal structures after the storm in
February 1979. Shipbuil. Navig 10, 26–30 (in Bulgarian).
Stanev, E., Schulz-Stellenfleth, J., Staneva, J., Grayek, S., Seemann, J., Petersen, W., 2011.
Coastal observing and forecasting system for the German Bight – estimates of
hydrophysical states. Ocean. Sci. 7, 569–583.
Stockdon, H.F., Thompson, D.M., Plant, N.G., Long, J.W., 2014. Evaluation of wave run-
up predictions from numerical and parametric models. Coast. Eng. 92, 1–11.
Sutherland, J., Peet, A.H., Soulsby, R.L., 2004. Evaluating the performance of
morphological models. Coast. Eng. 51, 917–939.
Trifonova, E., Valchev, N., Andreeva, N., Eftimova, P., 2012. Critical storm thresholds for
morphological changes in the western Black Sea coastal zone. Geomorphology
143–144, 81–94.
Valchev, N., Trifonova, E., Andreeva, N., 2012. Past and recent trends in the western
Black Sea storminess. Nat. Hazards Earth Syst. Sci. 12, 1–17.
Valchev, N., Andreeva, N., Eftimova, P., Trifonova, E., 2014. Prototype of early warning
system for coastal storm hazard (Bulgarian Black Sea coast). Compt. Rend. Acad.
Bulg. Sci. 67–7, 971–978.
Valchev, N., Andreeva, N., Eftimova, P., Prodanov, B., Kotsev, I., 2016. Assessment of
vulnerability to storm induced flood hazard along diverse coastline settings, 7 10002.
In: 3rd European Conf. on Flood Risk Manag. FLOODrisk 2016, E3S Web of
Conferences. http://dx.doi.org/10.1051/e3sconf/20160710002.
Van Ormondt, M., Van Dongeren, A., Briere, C., Sembiring, L., Winter, G., Lescinski, J.,
Swinkels, C., 2012. Simulating storm impacts and coastal flooding along The
Netherlands Coast. In: 2nd Europ. Conf. on Flood Risk Manag. FLOODrisk 2012,
pp. 28–29.
N. Valchev et al.
Verlaan, M., Zijderveld, A., de Vries, H., Kroos, J., 2005. Operational storm surge
forecasting in The Netherlands: developments in the last decade. Phil. Trans.. R. Soc.
A Math. Phys. Eng. Sci. 363, 1441–1453.
Werner, M., Kwadijk, J., Schellekens, J., 2005. Flood early warning systems for
hydrological (Sub) catchments. In: Encyclopaedia of Hydro. Sci, vol. 1. John Wiley &
Sons, pp. 349–364.
228
Coastal Engineering 134 (2018) 212–228
Werner, M., Schellekens, J., Gijsbers, P., van Dijk, M., van den Akker, O., Heynert, K.,
2013. The Delft-FEWS flow forecasting system. Env. Model. Softw. 40, 65–77.
Yang, F., Pan, H., Krueger, S.K., Moorthi, S., Lord, S.J., 2006. Evaluation of the NCEP
global forecast system at the arm SGP site. Mon. Weather Rev. 134, 3668–3690.