Environmental Modelling & Software 90 (2017) 201e216

Contents lists available at ScienceDirect

Environmental Modelling & Software

journal homepage: www.elsevier.com/locate/envsoft

Flood inundation modelling: A review of methods, recent advances

and uncertainty analysis
J. Teng a, b, *, A.J. Jakeman b, J. Vaze a, B.F.W. Croke b, D. Dutta a, S. Kim a
a
CSIRO Land and Water, Canberra, ACT, Australia
b
The Fenner School of Environment and Society, The Australian National University, Canberra, ACT, Australia

a r t i c l e i n f o a b s t r a c t

Article history: This paper reviews state-of-the-art empirical, hydrodynamic and simple conceptual models for deter-
Received 24 November 2016 mining flood inundation. It explores their advantages and limitations, highlights the most recent ad-
Received in revised form vances and discusses future directions. It addresses how uncertainty is analysed in this field with the
6 January 2017
various approaches and identifies opportunities for handling it better. The aim is to inform scientists new
Accepted 9 January 2017
to the field, and help emergency response agencies, water resources managers, insurance companies and
other decision makers keep up-to-date with the latest developments. Guidance is provided for selecting
the most suitable method/model for solving practical flood related problems, taking into account the
Keywords:
Flooding
specific outputs required for the modelling purpose, the data available and computational demands.
Natural hazards Multi-model, multi-discipline approaches are recommended in order to further advance this research
Large floodplains field.
High resolution © 2017 Elsevier Ltd. All rights reserved.
1D, 2D, 3D hydrodynamic model
Remote sensing
Environmental flows

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
2. Overview of methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
2.1. Empirical methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
2.2. Hydrodynamic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
2.2.1. 1D models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
2.2.2. 2D models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
2.2.3. 3D models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
2.3. Simplified (non-physics-based) methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3. Advantages and limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3.1. Empirical methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3.2. Hydrodynamic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3.3. Simplified conceptual models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
4. Recent advances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
4.1. Empirical methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
4.2. Hydrodynamic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.2.1. 1D models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.2.2. 2D models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.2.3. 3D models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
4.3. Simplified conceptual models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
4.4. Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

* Corresponding author. CSIRO Land and Water, GPO Box 1700,Canberra, ACT,
2601, Australia.
E-mail address: Jin.Teng@csiro.au (J. Teng).

http://dx.doi.org/10.1016/j.envsoft.2017.01.006
1364-8152/© 2017 Elsevier Ltd. All rights reserved.
202 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216

4.4.1. Identifying sources of uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

4.4.2. Quantifying uncertainty from different sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.4.3. Representation of uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
5. Summary and next steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

1. Introduction the most attention and are the subject of ongoing research:
empirical methods such as measurements, surveys, remote sensing
Flooding is a global phenomenon that causes casualties and and statistical models evolved from these data-based methods (e.g.
property loss on every inhabited continent. It is probably the most Schumann et al., 2009; Smith, 1997); and hydrodynamic models.
devastating, widespread and frequent natural disaster for human The latter include one-dimensional (1D) (e.g. Brunner, 2016; DHI,
societies. On the other hand, the impact of flooding is not always 2003), two-dimensional (2D) (e.g. DHI, 2012; Moulinec et al.,
negative as it is a part of a natural cycle and can have great envi- 2011) and three-dimensional (3D) methodologies (e.g. Prakash
ronmental and social benefits, particularly in areas which have et al., 2014; Vacondio et al., 2011) that simulate water movement
suffered from prolonged drought. Periodic scouring floods are even by solving equations derived from applying physical laws to fluid
crucial for ecosystems in most riverine and coastal wetlands. motion with varying degrees of complexity. In recent years, a third
Throughout human history, there has been a constant group of approaches has been gaining increasing popularity for
endeavour to understand, assess and predict flood events and their modelling very large floodplains (such as for national scale flood
impact. Flood inundation models are therefore developed to serve risk assessment) and data sparse regions. These models can be
this purpose. As flooding accounts for a significant proportion of labelled as simplified conceptual models and are based on more
the total number of reported natural disasters occurring in the modest representations of physical processes and have run times
world, and over the last 30 years this proportion has been orders of magnitudes shorter than hydrodynamic models. They are
increasing (Freer et al., 2011), the development and application of particularly suitable for large study areas and/or stochastic
flood inundation models and relevant research have become a modelling for probabilistic flood risk assessment.
global endeavour. Despite active research in the field, rapid and accurate flood
Systematic efforts within the research community since the modelling at high spatio-temporal resolutions remains a significant
1970s have greatly improved the capability of flood inundation challenge in hydrologic and hydraulic studies. This is due to the
modelling. The models are widely used in flood risk mapping (Apel complex and chaotic nature of flooding and uncertainty currently
et al., 2006; Dutta et al., 2006), flood damage assessment (Bhuiyan enduring in flood inundation modelling (Freer et al., 2011; Merz
and Dutta, 2012; Merz et al., 2010), real-time flood forecasting and Thieken, 2005). Many new concepts, techniques and philo-
(Arduino et al., 2005), flood related engineering (Gallegos et al., sophical debates have detailed the difficulties of providing effective
2009), and water resources planning (Vaze et al., 2013), as well as guidance and an agreement on best practice. And there is a vast
having served as an important prerequisite for investigating river literature describing, comparing, and benchmarking various
bank erosion and floodplain sediment transport (Marriott, 1992; models and algorithms. For instance, Alcrudo (2004) provided a
Pizzuto, 1987), contaminant transport, floodplain ecology (Karim state-of-the-art review on mathematical modelling of flood prop-
et al., 2015), river system hydrology (Dutta et al., 2013) and agation for the Impact Project (www.impact-project.net). Pender
catchment hydrology (Abbott et al., 1986; Beven, 1989). Combined (2006) reviewed the hydraulic models that are used in flood risk
with climate models, hydrological models, and river models, the management research and classified the models based on the
application of flood modelling has been extended to modelling that maximum dimensionality of the flow processes represented.
aims to formulate climate adaptation and risk mitigation strategies. Woodhead et al., (2007) provide a comprehensive reference to
Reliable and robust simulation of inundation characteristics has flood inundation models for the flood manager as part of the
also made it possible to effectively plan for environmental flows in FLOODsite project consortium (ww.floodsite.net). From 2009, the
order to maintain healthy aquatic ecosystems. report series from DEFRA/Environment Agency, UK reviewed and
Typically, the application purpose of any modelling requires benchmarked the then latest 2D hydraulic modelling packages
contextual attention to the output variables of predictive interest (Ne
elz and Pender, 2009, 2010, 2013). The application of satellite
and their time and space scales, the level of accuracy required and remote sensing to river inundation was reviewed by Smith (1997)
computational efficiency demands. For flood forecasting, applica- and a more recent review by Schumann et al. (2009) report on
tions may require considerations of fast run time and real-time data progress in this field. However, most of these studies focus on a
assimilation. Flood risk assessments in urban areas rely on the ac- specific group of models or a particular aspect of modelling such as
curacy of supercritical flow representation that can be offered by a 2D hydrodynamic models, or shock capturing schemes. Moreover,
numerical model that simulates fluid dynamics. Velocity should be the technologies are constantly evolving and modelling packages
carefully modelled and reported for dam construction, flood dam- are always going through major changes, with some hugely
age assessment, or erosion studies, while maximum flood extent improved and some discontinued.
and water depth may be sufficient for hazard mapping, environ- This paper reviews the current state-of-the-art for the different
mental flow assessment, and water resources planning. All these modelling approaches, analyses their advantages and limitations,
considerations call for the end users to wisely select a model how they deal with uncertainty issues, highlights the latest de-
balancing their demands against model complexity and data velopments and discusses future directions. While informing the
requirements. novice scientist in this field, it also aims to help emergency
Over the past century, two groups of approaches have attracted response agencies, water resources managers, insurance
 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216 203

companies and other decision makers keep up-to-date with the
latest developments in flood inundation models. In particular it
provides guidance for choosing the most suitable method/model
for solving their practical problems. Nevertheless, it is impossible to
cover all published studies and all available models. Instead we aim
to concentrate on those studies that address new concepts or real
advances in flood inundation modelling, and those that compare
different modelling approaches.
2. Overview of methods
surface floodplain flow too, in which case floodplain flow is part of
the one-dimensional channel flow, is assumed to be in one direc-
tion parallel to the main channel, and one cross-section averaged
velocity is used to represent large variations in velocity across the
floodplain. Fig. 1 shows an example of a 1D model and its repre-
sentation of floodplain flow by a series of cross-sections.
Generally, 1D models solve equations derived by ensuring mass
and momentum conservation between two cross sections Dx apart,
which yields the well-known one-dimensional Saint-Venant
equations:

All the methods in this section are developed to: i) understand vQ vA

Conservation of mass þ ¼0 (1)
the processes involved in river and floodplain flow; and/or ii) pre- vx vt
dict instream and floodplain variables, such as volume, water level,
 
inundation extent and flow velocity, that are useful for risk Q2
v
assessment and floodplain management. There are many different 1 vQ 1 A vh
ways of grouping them but here we use the three main categories Conservation of momentum þ þg
A vt A vx vx
discussed below:
 
 g S0  Sf
2.1. Empirical methods ¼0 (2)

A mega flood of the Quaternary Period (which extends from Here Q is the flow discharge (Q ¼ uA, where u is the cross-
about 1.8 million years ago to the present) had a peak discharge of sectional averaged velocity and A is the flow cross-section area), t
nearly 20 million m3/s but could have remained unknown had represents time, h is the water depth, g is the gravitational accel-
there not been geologic evidence left behind. Many ancient cata- eration, Sf is the friction slope and So is the channel bed slope. One-
strophic floods in the world's largest river basins were learnt about dimensional Saint-Venant equations are a simplification of the
in this way by geologists and hydrologists, or from historical two-dimensional shallow water equations which are also known as
measurements/descriptions discovered by historians (O'Connor the two-dimensional Saint-Venant equations. Equations (1) and (2)
and Costa, 2004). Many more floods of this magnitude may have have no analytical solution, but can be solved using numerical
occurred in the past without leaving any trace in the recorded techniques. The solution of these equations comprises estimates of
history. Fortunately, since then, human societies have developed Q and h for every cross-section at each time step.
sophisticated technologies to collect, process, integrate, and anal-
yse flood related data and, most of the time, we call these data 2.2.2. 2D models
“observations.”. Nevertheless, these data, including on-ground The 2D models represent floodplain flow as a two-dimensional
measurements, surveys, interviews, aerial photographs and satel- field with the assumption that the third dimension e water depth -
lite imageries, are indeed a limited representation of the reality. is shallow in comparison to the other two dimensions (DHI, 2012;
They are derived with underlying assumptions and inevitable un- Roberts et al., 2015). Most approaches solve the two-dimensional
certainties, and therefore should be treated as model results. This shallow water equations, which represent mass and momentum
type of modelling is a reflection of the past and generally consid- conservation in a plane, and can be obtained by depth-averaging
ered as robust and accurate. The models have received the longest the Navier-Stokes equations:
research attention and are fast evolving and undergoing continual
development. The results of these models are widely used to sup- vh vðhuÞ vðhvÞ
Conservation of mass þ þ ¼0 (3)
port decision making and have served as inputs to other types of vt vx vy
methods; for example, remotely sensed flood extent has been
 
widely used in flood monitoring, as well as a benchmark for hy- vðhuÞ v 1
Conservation of momentum þ hu2 þ gh2
drodynamic model calibration and validation (Smith, 1997). vt vx 2
vðhuvÞ
2.2. Hydrodynamic models þ
vy
Hydrodynamic models are mathematical models that attempt to ¼0 (4)
replicate fluid motion and typically require solving computation-
 
ally. These models simulate water movement by solving equations vðhvÞ vðhuvÞ v 1
þ þ hv2 þ gh2 ¼ 0 (5)
formulated by applying laws of physics. Depending on their spatial vt vx vy 2
representation of the floodplain flow, the models can be dimen-
sionally grouped into 1D, 2D and 3D models. Table 1 lists some where x and y are the two spatial dimensions, and the 2D vector
typical hydrodynamic models and their developers. ðu; vÞ is the horizontal velocity averaged across the vertical column.
The solution of these equations comprises estimates of u, v, and h
2.2.1. 1D models over space and time. As with the one-dimensional Saint-Venant
The simplest representation of floodplain flow is to treat the equations, the two-dimensional shallow water equations have no
flow as one-dimensional along the centre line of the river channel analytical solutions. Many numerical schemes are therefore
(Brunner, 2016; DHI, 2003). Many hydraulic situations can make developed for algebraic approximation.
the 1D assumption, either because a more detailed solution is un- Depending on numerical discretisation strategies, the models
necessary (e.g. the purpose does not require knowledge in other can be classified into finite element, finite difference, and finite
dimensions) or because the flow is markedly 1D, such as in a volume methods. According to discretisation in time, the models
confined channel or in a pipe. 1D models can be used to model open can be divided into implicit (solver cannot proceed to next time
204 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216

Table 1
Some well-known software/models that are capable of modelling flood inundation.

Developer 1D 2D 1D þ 2D 3D Simplified Status Note

Ambiental Flowroute-i Commercial

Australian National University ANUGA Hydro Open https://anuga.anu.edu.au/
& Geoscience Australia source
BMT WBM TUFLOW Classic TUFLOW Classic TUFLOW TUFLOW Commercial
1D 2D TUFLOW GPU Classic FV
TUFLOW FV
Cardiff University FASTER DIVAST Research Is the foundation of ISIS 2D.
DIVAST-TVD
CH2M Hill (formerly Halcrow Flood Modeller Flood Modeller Pro Flood Flood Modeller Pro Commercial Flood Modeller Pro is the successor of
Group) Pro 1D solvers 2D solvers Modeller Pro 2D FAST solver the ISIS suite.
CSIRO SPM TVD Research
DELTARES SOBEK Suite SOBEK Suite SOBEK Suite DELFT3D Commercial DELFT3D is open source http://oss.
DELFT3D deltares.nl/web/delft3d/get-started.
DHI MIKE11/MIKE MIKE21 MIKE Flood MIKE 3 Commercial
HYDRO River


Electricit
de France
e MASCARET TELEMAC 2D TELEMAC Open
3D source
HR-Wallingford RFSM Commercial
Innovyze InfoWorks RS InfoWorks 2D InfoWorks Commercial Use ISIS flow simulation engine.
ICM
JBA JFLOW Commercial
Nottingham University TRENT TRENT Research
Svasek Hydraulics FINEL 2D FINEL 3D Commercial
Tokyo University CaMa-Flood Research Global-scale model.
University of Bristol LISFLOOD-FP LISFLOOD-FP Research
University of California BreZo Research
University of Exeter SIPSON UIM UIM þ SIPSON Research
US Army Corps of Engineers HEC-RAS HEC-RAS 2D Free http://www.hec.usace.army.mil/
software/hec-ras
XP Solutions XP2D XPSWMM Commercial
XPSTORM
ANSYS CFX, Commercial
Fluent
Autodesk €st
Bifro Commercial As part of Maya
Blender Blender Open https://www.blender.org
source
Chaos Group PHEONIX Commercial
FD
DPIT Navie
Commercial
Effex
Kyushu University Flip3D Open https://code.google.com/p/flip3d/
source
Maxon Cinema 4D Commercial
Next Limit RealFlow Commercial
Red Giant Psunami Commercial
Side Effects Software Houdini Commercial
TU München & ETH Zürich Mantaflow Open http://mantaflow.com
source

step until the whole domain is solved) and explicit (solving of the may be adequate, especially given the type and quality of data
current unit independent of solving the rest of the domain for any typically available for model construction and validation (Alcrudo,
given time step) models. In terms of spatial representation, the 2004). However, modelling of vertical turbulence, vortices, and
models can use structured mesh (rectangular grids), unstructured
mesh (triangular grids), and most recently, flexible mesh as shown
in Fig. 2.
2D hydrodynamic models are perhaps the most widely used
models in flood extent mapping and flood risk estimation studies.
There has now been much assessment of the capability of the 2D
models. For example, Ne
elz and Pender (2009) have given a
comprehensive review of 2D hydraulic modelling packages and,
based on the review, a benchmarking study has been carried out to
compare the performance of some commonly used 2D models
(Ne
elz and Pender, 2010). More recently a later set of 2D models
was reviewed and compared once again (Ne
elz and Pender, 2013).

2.2.3. 3D models
For many scales of floodplain flow, complex three-dimensional
representation of flow dynamics has tended to be regarded as
unnecessary, as a two-dimensional shallow water approximation Fig. 1. The one-dimensional representation of flow using a series of cross-sections.
 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216 205

Fig. 2. Different meshes used in 2D modelling: a) structured mesh; b) unstructured mesh; and c) flexible mesh (TUFLOW FV, reproduced from http://www.tuflow.com/tuflow%20fv.
aspx with permission).

spiral flow at bends is important during catastrophic floods, such as Here u is the velocity; r is the fluid density; p is pressure; g is
those occurring due to dam breaks, tsunamis, flash floods or gravitational acceleration; m is kinematic viscosity. Equation (6)
embankment and levee breaches. 3D models were therefore arises from applying Newton's equation F ¼ ma to fluid motion.
developed to allow representation of vertical features (Monaghan, Equation (7) is based on the assumption that the material density is
1994; Ye and McCorquodale, 1998). Some of these models solve the constant within a fluid parcel. Depending on the representation of
horizontal flow with 2D shallow water equations and include a the processes, these models may be divided into two broad cate-
quasi-3D extension to model velocity in vertical layers (Casulli and gories: Eulerian (grid-based) and Lagrangian (particle-based).
Stelling, 1998). Other 3D models are derived from the three- Application of 3D models in flood inundation modelling was
dimensional Navier-Stokes equations, which describe the motion once considered not viable at the reach-scale (>1 km), being
of fluid substances and are usually written as: limited by computational feasibility and the problems of accurately
representing free surface flows, high-order turbulence and a tran-
vu 1 sient flood shoreline. This limitation has eased in recent years due
Conservation of momentum þ u$Vu þ Vp ¼ g þ mV$Vu
vt r to fast development of particle-based models and vast improve-
ments in computing techniques. Particle-based approaches are
(6)
now favoured owing to a number of advantages over grid-based
approaches: capable of representing small-scale features (of size
Incompressibility condition V$u ¼ 0 (7) smaller than a grid cell); no need for spatial discretisation; have
206 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
little or no problem with mass preservation or spatial diffusion.
Nevertheless, the use of particle-based 3D models in flood
modelling that incorporates real topography is fairly recent (Cleary
and Prakash, 2004), and such studies are rare in the literature in
stark contrast to 2D models.
2.3. Simplified (non-physics-based) methods
There is another type of model that can be used in a predictive
way but that does not fit the description of a hydrodynamic model.
These models have sometimes been referred to as 0D models
(Pender, 2006) but should be classified as 2D models due to their
dimensional representation of flow. They were also sometimes
considered as 0-term models (Ne
elz and Pender, 2013) in contrast
with the full-term models that solve full 2D shallow water equa-
tions, and 3-term or 2-term models that solve simplified shallow
water equations omitting one or two acceleration terms. This
classification, however, could be considered as conforming to a very
shallow water equation-centric view. In this paper, we separate
them from the hydrodynamic models and put them in their own
category e simplified conceptual models e as they do not involve
any simulation of the physical process of inundation and are based
on simplified hydraulic concepts.
One of these models is the Rapid Flood Spreading Method
(RFSM) (L'homme et al., 2008) which divides the floodplain into
elementary areas that represent topographic depressions in the
pre-processing. It then spreads the flood volumes by filling these
areas using a filling/spilling process (Flood Modeller Pro 2D Fast
Solver also uses this technique). Fig. 3 captures the function of this
type of model. Benchmarking studies (Ne
elz and Pender, 2010,
2013) show that this type of model can produce approximate
predictions of final inundation distributions, with clear benefits in
terms of computational cost compared to hydrodynamic models
(up to 1000 times faster as claimed by Flood Modeller Pro 2D Fast
Solver).
Another simplified conceptual model is based on the so-called
theory of “planar method” or “bathtub method”. It derives the
flood extent by intersecting a series of planes at fine intervals with a
high resolution DEM and instantly links the water stage/volume
with the flood extent (Teng et al., 2015) as depicted in Fig. 4.
Operational at continental scale, the Height Above the Nearest
Drainage (HAND) model normalizes topography according to the
local relative heights found along the drainage network and de-
termines nearby flood inundation extent by selecting the sur-
rounding cells whose HAND values are less than the known water
depth in the stream (Nobre et al., 2011).
These models are orders of magnitude faster to run than 2D or
even 1D hydrodynamic models, making them useful tools for large-
scale applications where only final/maximum flood extent and
water levels are required, dynamic effects being insignificant.
3. Advantages and limitations
The main relative advantages and limitations of these different
modelling techniques, as well as their suitability to modelling
purpose, are summarised in Table 3 and discussed below.
3.1. Empirical methods
Empirical methods are the most intuitive and straightforward
approach to retrieve useful flood information from observations.
They are commonly believed to be accurate, although the accuracy
of different methods depends on the acquisition and processing
techniques adopted for each individual method. Higher accuracy is
usually achieved by higher cost of acquisition and more complex
processing of the data. Wisdom of hindsight is probably the most
obvious limitation of this type of methods. They are snapshots of
the past and cannot directly predict responses to future or changed
conditions.
The data derived by these methods used to be of coarse spatial
and temporal resolution. But this situation is fast improving
because of advances in remote sensing. Other restrictions of these
methods include: engineering limitations, such as those associated
with the design of sensors, carriers and transmission devices;
environmental limitations, including clouds, wind, vegetation
cover, weather conditions, and constraints from other natural fac-
tors; processing limitations, for instance, algorithms used in data
mining can play a major part in the value and accuracy of the in-
formation; and artificial errors can also be introduced during any
part of the whole process.
3.2. Hydrodynamic models
Hydrodynamic models are the most widely used tools to
simulate detailed flood dynamics. They can be directly linked to
hydrological models and river models to provide flood risk map-
ping, flood forecasting and scenario analysis. Unlike empirical
models, the input of hydrodynamic models can be manipulated/
perturbed to investigate the impact of changes in initial conditions,
boundary conditions, or topographic input to account for
augmentation/destruction of hydraulic features/structures. The 1D
versions are computationally efficient, but they suffer from a
number of drawbacks including the inability to simulate lateral
diffusion of the flood wave, the discretisation of topography as cross
sections rather than as a continuous surface and the subjectivity of
cross-section location and orientation. The 2D versions that solve
full shallow water equations have been reported to be able to
simulate timing and duration of inundation with high accuracy.
Nevertheless they are computationally intensive. The 2D and 3D
models are generally considered as unviable for areas larger than
1000 km2 when the resolution required is less than 10 m and/or in a
probabilistic approach requiring multiple simulations as the time
taken to run simulations may be prohibitively long.
Benchmarking studies (Ne
elz and Pender, 2010, 2013) have
confirmed that 2D models are capable of adequately predicting
those variables, including velocity, flood extent and water level,
upon which flood risk management decisions are based. But
studies (Ne
elz and Pender, 2013) have also pointed out strong
differences in modelling results, especially for predictions made
using 1D river to 2D floodplain linkage, and for velocity pre-
dictions in urban areas. This has brought the uncertainty of the
modelling results into focus.
3.3. Simplified conceptual models
These models require significantly less computer effort than the
hydrodynamic models. They are fast and robust, most desirable for
applications that do not require velocity output and have low de-
mands on the representation and accuracy of flow dynamics.
Runtime savings portend that this type of model is suitable for large
floodplains (>2000 km2 in Teng et al. (2015), while HAND is used at
continental scale) and for probabilistic risk assessment requiring
large numbers of simulations. Where clear flow paths across the
floodplain exist, these models produce predictions of inundation
extent, overbank volume and water depth that compare well with
shallow water equation 2D models. However, the comparisons
diverge for more complex topographies, particularly in areas where
momentum conservation is important e such as the prediction of
water levels and velocities in the complex flow field close to a dam
failure and where the spreading flood encounters an adverse slope
 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216 207

Fig. 3. The principal of RFSM as described in L'homme et al. (2008): a) defining basic storage elements on floodplains called Impact Zones; b) key features of the Impact Zones; and
c) different spilling/merging steps (Figures 1, 2 and 3 in L'homme et al., 2008 ©2009 Taylor & Francis Group, London, UK. Used with permission).

Fig. 4. The principal of CSIRO TVD model: a) intersecting the DEM with multiple planes (profile view); b) derive flood extent based on the connectivity of the depressions with the
river (plan view).

on the floodplain. This limits the application of simplified concep-
tual models to applications where dynamic effects are less signifi-
cant in determining the direction of water flow. They also have no
representation of flow dynamics, nor force field, and most of them
do not include prediction of velocities, which makes them unsuit-
able for modelling tsunamis, flash floods, dam breaking types of
flooding and bank erosion studies.
4. Recent advances
4.1. Empirical methods
Traditionally, water levels or discharge from streamflow gauging
stations are the means most relied upon for flood observation.
While this type of empirical method tends to shrink over time
(gauging stations closing down, surveys rare, and lack of manpower
208 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
to do on-ground measurements), the observing of floods from air or
space is thriving. The most recent advances in this group are mainly
related to remote sensing. Over the last decade radar remote
sensing has emerged as a powerful tool to support and advance
flood modelling. Synthetic aperture radar (SAR) was found to be
useful for flood inundation maps as the radar return signals are
normally low for smooth open water bodies without vegetation
due to reflection from the water surface. This characteristic permits
flood boundaries to be determined with a good level of accuracy
under many conditions. However, turbulence, wind, emergent
vegetation and trees can all cause significant increases in radar
back-scatter, making inundation extent difficult or impossible to
determine.
New satellite launches (e.g. SWOT, ALOS, RADARSAT-2, Terra-
SAR-X, COSMO-SkyMed, and Sentinel-1), better sensors, shorter
return periods, faster image acquisition and processing have
resulted in higher quality and finer resolution remotely sensed data
becoming more readily available. Algorithms, data-mining tech-
niques and decision trees have been developed to extract more
accurate and larger amounts of information from data. For instance,
the Normalised Difference Water Index (NDWI) was developed by
(Gao, 1996) to delineate the open water bodies from satellite-
derived Near-Infrared (NIR) and Short Wave Infrared (SWIR)
channels. Since then it has been widely used as well as improved by
(Xu, 2006) and integrated into other water body detection methods
such as Open Water Likelihood (OWL) (Ticehurst et al., 2014) and
Enhanced Water Index (EWI) (Wang et al., 2015). There is always a
trade-off, however, between accuracy of the technique and
complexity of data processing. Some of the remote sensing prod-
ucts (such as MODIS, LandSat TM, SPOT 5) have been around long
enough for third party agencies to provide value-added products or
services (for example, the Water Observations from Space website
http://www.ga.gov.au/scientific-topics/hazards/flood/wofs).
Remote sensing data are not only used to simply derive flood
hydrology information (e.g. discharge, flood extent and area, and
water stage) for flood assessment, but also are incorporated into
other modelling approaches. Two remote sensing products, Shuttle
Topography Radar Mission (SRTM) and Light Detection and Ranging
(LiDAR) Digital Elevation Models (DEM), are routinely used as
topographic input for both hydrodynamic models and simplified
conceptual models owing to their wide availability and high qual-
ity. The integration of remote sensing with flood modelling has also
progressed from using the data to calibrate and validate modelling
results to using remote sensing to understand and improve model
structures, and facilitate data assimilation with models (Bates et al.,
1997; Martinis et al., 2009; Matgen et al., 2011; Mure-Ravaud et al.,
2016; Pulvirenti et al., 2011; Sanyal and Lu, 2004; Schumann et al.,
2009; Smith, 1997; Ticehurst et al., 2015). Nevertheless, the un-
certainty in remotely sensed flood extent and water stages can be
significant. This has been acknowledged and quantification of un-
certainty in remote sensing data has become a major aspect of
research in recent years (Stephens et al., 2012).
4.2. Hydrodynamic models
Hydrodynamic models are undergoing fast improvements both
in accuracy and computational efficiency. Apart from advances in
individual models deliberated below, the combining of models is
also receiving wider recognition as it allows almost limitless pos-
sibilities for maximising the benefits of 1D, 2D and 3D modelling
approaches.
4.2.1. 1D models
The most significant advance in 1D models is their ability to be
coupled with 2D models (Liu et al., 2015). There are a few different
ways to forge this linkage, such as: modelling a channel partly in
1D, and partly in 2D where more processes are required to be
represented; modelling culverts in 1D within a mainly 2D model;
linking a 1D drainage network model and a 2D surface flood model;
and linking a 1D river model and a 2D floodplain model.
Many techniques have been developed to link 1D and 2D
models. These include lateral link (an example is shown in Fig. 5),
longitudinal link, and vertical link, all used by different developers
(Ne
elz and Pender, 2009). They model the exchange flows between
1D and 2D models differently and are bounded by limitations such
as: omitting momentum exchange between channel and flood-
plain; neglecting return flow from floodplain back to channel; or
assuming flow is purely one-dimensional until the river reaches
bankfull stage and then switching to 2D modelling. The difference
in the linkage techniques results in significantly different pre-
dictions of volume of water exchanged between the river and the
floodplain, and flow-on differences in the prediction of other flood
properties from various modelling packages/suites. The signifi-
cance of this difference and what it means for future applications is
subject to further investigation (Ne
elz and Pender, 2013).
4.2.2. 2D models
The 2D models have received the most research attention in
recent decades. A large amount of work has been devoted to
developing two-dimensional numerical models for unsteady
shallow flows, and various computational techniques using finite
difference, finite element and finite volume schemes have been
reported (see Table 2). Among them, finite volume schemes are
increasingly popular and have become the most widely used
method in the 2D models that solve shallow water equations
(Caleffi et al., 2003; DHI, 2012; Sleigh et al., 1998). This can be
explained by their advantages in terms of conservativeness, geo-
metric flexibility and conceptual simplicity (Alcrudo, 2004).
Research focus has been put on the accurate representation of
flow over topographic discontinuities (shocks or hydraulic jumps in
shallow flows). As the 2D shallow water equations are typically of
hyperbolic type, the numerical solution of the dry front is usually
characterised by unphysical oscillations and positional error of the
front, which grows as a function of time. Many methods have been
developed to address this issue. One distinctive group of these
methods relies on essentially non-oscillatory reconstructions of the
data and the solution of a generalized Riemann problem (governing
equations subject to special initial conditions consisting of two
constant states separated by a discontinuity) (Toro and Garcia-
Navarro, 2007). provide a detailed review of this group of
methods. It is possible to find analytical solution of the Riemann
problem, but in most practical situations approximate solutions
provides simplicity and efficiency. Popular linearized Riemann
solvers include Roe's Riemann solver (Roe, 1981), and Harten, Lax
and van Leer's (HLL) Riemann solver (Harten et al., 1983). A class of
non-linear second-order Godunov type schemes that have been
very successful are the so called Total Variation Diminishing (TVD)
methods. These methods are able to capture large gradients of the
solution, or even discontinuities, without (or with much reduced)
spurious oscillations. Second-order accuracy is achieved both in
time and space through the use of the MacCormack scheme, in
conjunction with a total variation diminishing modification. The
model is able to predict accurately the shock fronts, even over
initially dry beds with friction and abrupt slope changes (Liang
et al., 2006).
Modelling of wetting and drying is another major area of
ongoing research in numerical modelling of the shallow water
equations. Medeiros and Hagen (2013) reviewed four groups of
wetting and drying algorithms used by contemporary models and
discussed their benefits and drawbacks. They concluded that
 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216 209

Fig. 5. An example given by SOBEK on coupling 1D and 2D models using lateral link: a) combined 1D/2D staggered grid; b) combined continuity equation for 1D and 2D com-
putations (Figure 2.1 in Woodhead et al., 2007. Used with permission).

wetting and drying is still a nontrivial challenge to numerical
modellers, as in most cases stable, easy solutions are inaccurate and
accurate solutions are computationally costly and frequently
unstable.
To reduce the complexity and runtime, simplification of the
two-dimensional shallow water equations can be made to omit
certain terms, resulting in kinematic or diffusive wave
approximations. These include so called ‘3-term’ models that omit
convective acceleration but preserve pressure, bottom slope and
friction slope (LISFLOOD-FP); and ‘2-term’ models that omit pres-
sure as well (JFLOW, UIM, Cama-Flood). These types of simplifica-
tions were reviewed and discussed in Hunter et al. (2007). Results
from models solving simplified equations are considered to be
comparable with full shallow water equations models in situations

Table 2
The 2D hydrodynamic models listed in Table 1 and their technical features.

Model name Solving equation Numerical scheme Discretisation Discretisation of Shock capturing Parallelisable Note
of time space algorithm

Flowroute-i Shallow water equations Finite volume Explicit Structured No Yes

mesh
ANUGA Hydro Shallow water equations Finite volume Explicit Flexible mesh Riemann solver Yes
TUFLOW Shallow water equations Finite difference Implicit Structured No No
Classic 2D mesh
TUFLOW GPU Shallow water equations Finite volume Explicit Structured No No
mesh
TUFLOW FV Shallow water equations Finite volume Explicit Flexible mesh Yes Yes
DIVAST Shallow water equations Finite difference Implicit Structured No No
mesh
DIVAST-TVD Shallow water equations Finite difference Explicit Structured TVD No
mesh
Flood Shallow water equations Finite difference Implicit Structured No No Based on DIVAST, successor
Modeller mesh of ISIS 2D
Pro
2D ADI solver
Flood Shallow water equations Finite difference Explicit Structured TVD No Based on DIVAST-TVD
Modeller mesh
Pro
2D TVD solver
SOBEK Suite Shallow water equations Finite difference Implicit Structured Yes No
mesh
DELFT3D Navier Stokes equations Finite difference Implicit Flexible mesh Yes Yes
MIKE21 Shallow water equations Finite volume Explicit Flexible mesh Yes Yes
TELEMAC 2D Shallow water equations Finite element/Finite Implicit/ Unstructured Yes Yes
volume Explicit mesh
InfoWorks 2D Shallow water equations Finite volume Explicit Flexible mesh Riemann solver No Based on ISIS 2D
JFLOW Diffusive wave Finite volume Explicit Structured Riemann solver Yes No inertia terms
mesh
TRENT Shallow water equations Finite volume Explicit Structured Riemann solver Yes
mesh
FINEL 2D Shallow water equations Finite element Explicit Flexible mesh Riemann solver Yes
CaMa-Flood Diffusive wave Finite difference Explicit Structured No Yes
mesh
LISFLOOD-FPa Dynamic wave models that Finite difference Explicit Structured No Yes No inertia terms
neglect advection mesh
BreZo Shallow water equations Finite volume Explicit Unstructured Riemann solver Yes
mesh
UIM Diffusive wave Finite difference Explicit Structured No No
mesh
XP2D Shallow water equations Finite difference Implicit Structured No No
mesh
a
The latest LISFLOOD-FP include several different versions of code including a version that solves full shallow water equations. Here we based our knowledge on the most
widely recognised version that was described in (Neal et al., 2012).
210 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216

Table 3
Comparative summary of the relative merits and weaknesses of different modelling approaches.

Method Strength Limitation Suitability

Empirical Relatively quick and easy to Non-predictive Flood monitoring

methods implement No/indirect linkage to hydrology (difficult to use in Flood damage assessment
Based on observation scenario modelling) Serve as observations to support calibration, validation
Derived inundation estimate is Coarse spatial and temporal resolution (although and data assimilation for other methods
independent improving)
Technology is rapidly improving Engineering limitations (sensors, carriers, transmission
devices)
Environmental limitations (clouds, wind, damaging
weather conditions, other natural constrains)
Processing limitations (algorithm, artificial errors …)
Hydrodynamic Direct linkage to hydrology High data requirements Flood risk assessment
models Detailed flood risk mapping Computationally intensive Flood damage assessment
Can account for hydraulic features/ Input errors can propagate in time Real-time flood forecasting
structures Flood related engineering
Quantify timing and duration of Water resources planning
inundation with high accuracy River bank erosion
Floodplain sediment transport
Contaminant transport
Floodplain ecology
River system hydrology
Catchment hydrology
Simplified Computationally efficient No inertia terms (not suitable for rapid varying flow) Flood risk assessment
conceptual No/little flow dynamics representation Water resources planning
models Floodplain ecology
River system hydrology
Catchment hydrology
Scenario modelling

involving very low momentum flow. But the benefit of computa-
tional cost-saving is unclear with some benefits reported in terms
of reduced computational effort in applying LISFLOOD-FP
compared with the full shallow water equation models, but
significantly longer run time than other models recorded for UIM
due to very short time steps required by the model formulation
(Ne
elz and Pender, 2013). A number of studies have recently
focused on developing simple 2D flood models using the cellular
automata (CA) approach (Dottori and Todini, 2011; Guidolin et al.,
2016). A CA model usually consists of a set of cells that represent
discretised space, each of which has a state, a distribution of
neighbouring cells, a discrete time step, and a set of transition rules.
The transition rules are composed of simple operations that govern
the evolution of each cell's state, which make use of the previous
state of the cell itself and of those in its neighbourhood. Since
computing the new state of a cell depends only on the state of the
neighbouring cells at the previous time step, CA algorithms are well
suited to parallel computation. Guidolin et al. (2016) has reported
that CA model can be 8 times faster than the traditional model,
when InfoWorks ICM 3.0 were used as a benchmark model.
The discretisation of time is an important consideration in nu-
merical methods. The explicit schemes (which are adopted by the
vast majority of newly developed models) are prevailing over im-
plicit schemes as they are simpler to implement and easier to
parallelise. However, they are subject to some form of time-step
limitation (for stability) analogous to the Courant Friedrichs-Lewy
condition (u.Dt/Dx < 1).
The latest advance in model discretisation of space is the
development of flexible meshes. More and more developers have
started to incorporate this type of mesh in their model. To name a
few, there are MIKE FM, TUFLOW FV, ANUGA Hydro, DELFT3D
Flexible Mesh Suite, and FINEL 2D. A computational mesh is care-
fully constructed to have finer resolution around complex areas to
reflect subtle changes in the topography, and coarser resolution
over vast areas with little spatial variance to save computational
effort. The design of the flexible mesh tends to have great influence
on model performance. Thus, although more time and effort are
spent preparing the model geometry, over the life cycle of a
modelling project a well assembled mesh will save time for both
the modellers and the computers.
With advances in computing science, the performance of hy-
drodynamic models has improved exponentially. Parallel
computing and distributed workflows are frequently used to meet
the high computational demand. The contemporary models have
become very computationally cautious by incorporating various
techniques to parallelise and speed up the model runs. These
include: i) programming using OpenMP, an application program-
ming interface (API) that supports multi-platform shared memory
multiprocessing programming (e.g. adopted by Brezov, DELFT3D);
ii) utilising graphics processing units (GPU) to achieve 10 to 100
time speed increase compared to CPU (e.g. TUFLOW, MIKE FLOOD,
MIKE 3 and JFLOW); and iii) partitioning the model domain and
spreading the computation on distributed-memory high-perfor-
mance computing clusters or cloud computing devices using MPI
standards (e.g. FINEL2D, DELFT3D). Models that use explicit
schemes are particularly suitable for parallelising.
Over recent decades, the accessibility of the model codes has
been broadened from research to commercial areas and much is
now open source. The main benefit of open source is that the model
becomes a collaborative effort where developers improve upon the
source code and share the changes within the community, and
users download, modify and provide their versions back to the
community. Although the open source models still cannot compete
overall with the commercial models in terms of documentation
(user manuals, technical references), Graphic User Interface (GUI),
technical support, visualisation tools and other desirable “acces-
sories” such as data format converting tools, they are gradually
improving in these areas. A new commercial mode has also
emerged with some model codes moved to open source and
companies providing technical support at a cost (e.g. DELFT3D).
4.2.3. 3D models
Possibly the most ground-breaking prospects in flood inunda-
tion modelling come from 3D models, especially the particle-based
 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
Fig. 6. Progression of flood inundation for different dam collapse scenarios modelled by SPH coupled with a discrete element model (Figure10 in Prakash et al., 2014 ©2014 Elsevier
Inc. Used with permission).
3D models. The concept of smoothed particle hydrodynamics (SPH)
was invented to simulate non-axisymmetric phenomena in astro-
physics (Gingold and Monaghan, 1977; Lucy, 1977). Monaghan
(1994) was the first to use an SPH model to simulate free surface
fluid flows. Further background information on the SPH model can
be found in a review by Monaghan (1992). Cleary and Prakash
(2004) pointed out SPH's potential in environmental science.
Prakash et al. (2014) used an SPH numerical model (coupled with a
discrete element model) to simulate different possible scenarios in
a dam failure flood event (Fig. 6). They proved that SPH is capable of
modelling flood inundation characterised by rapidly varying flow,
such as dam breaks, tsunamis, and levee breaches. Particle-based
modelling methods have specific advantages over traditional grid
or mesh-based methods for these types of problems. These include
highly accurate and non-diffusive prediction of convection and the
straightforward representation of small features.
A special mention should be given to the development of 3D
fluid simulators in the computer graphics research community,
which contributes widely in the entertainment industry such as to
animations, movie special effects, and interactive video games.
Many recent advances in 3D fluid solvers were achieved there,
probably owing to the nature of the industry and its huge profits
that have attracted some of the most talented minds. Before the
computer graphics industry became involved, however, fluid
simulation was being actively modelled mathematically in the
1950's and 60's. The techniques invented then formed the foun-
dation of what is used in the industry today (Bridson, 2008; Stam,
1999). These techniques include the staggered Marker-And-Cell
(MAC) (Harlow and Welch, 1965) grid structure, and Particle-In-
Cell (PIC) (Harlow, 1962) method, which is the precursor to the
modern day Fluid-Implicit-Particle (FLIP) (Brackbill and Ruppel,
1986) method, the material point method (MPM) (Sulsky et al.,
1994) and other Eulerian/Lagrangian hybrid methods.
There are great commercial fluid simulation tools such as those
in Maya, Houdini (specialist animation software) and those from
the open source family such as Blender and Mentaflow.
211
DreamWorks has recently released an open source volumetric data
structure and toolkit OpenVDB (www.openvdb.org) which can
significantly improve the performance of fluid solvers. Although
these techniques were developed to achieve a high degree of visual
realism instead of physical accuracy of flow dynamics, and inevi-
tably are subject to simplifications and unrealistic assumptions,
they do constitute real progress in the fluid simulation field and are
attracting active research interest. Other active areas of fluid
simulation research in the graphics community include smoke
simulation, fire simulation, simulation of highly viscous fluids, and
coupled simulations (Bridson, 2008). Coupled simulations combine
two or more fluid solvers which interact plausibly to improve
computational efficiency and/or to allow simulation of both dense
and diffuse water volumes (Losasso et al., 2008). It is definitely an
area worth watching closely and the ideas such as efficient data
storage and manipulation can surely be shared with real world
flood inundation modelling.
4.3. Simplified conceptual models
The main development in simplified conceptual models has
been to address their deficiency of lacking mass conservation and
representation of velocity. RFSM Dynamic uses continuity to
distribute flood volume between storage areas and then computes
the flow rates between these using Manning's equation. RFSM EDA
employs a similar approach but is based on a diffusive approxi-
mation to shallow water equations. Attempts have been made to
limit the volume on the floodplain using weightings derived from
flow accumulation areas, as in the CSIRO TVD model (coincidently
having the same name as the shock capturing scheme, but they
represent entirely different modelling approaches). The enhance-
ments of HAND model reportedly include enabling non-uniform
inundation within a catchment, accounting for backwater effects
in catchments with precipitation and flooding downstream of the
catchment, and bypassing hard flow paths that are shorted during
large flooding events.
212 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
As with the other model types, model assimilation using remote
sensing data is another research hotspot that can improve the
predictions from simplified conceptual models, largely by integra-
tion of remotely sensed flood extent and water depth during a
model simulation. These methods can also benefit from incorpo-
ration of explicit schemes and parallel computing to further
improve their efficiency. Due to their relatively low computational
cost, these methods have great potential in modelling large flood-
plains, facilitating uncertainty analysis, scenario modelling, prob-
abilistic flood risk assessment, and environmental impact studies.
4.4. Uncertainty
Uncertainty analysis is an indispensable part of model predic-
tion for non-idealised environmental systems. It has received
increasing attention by the flood inundation modelling community
and has become a crucial part of modelling practice in some pro-
jects partly due to legislative requirements (Freer et al., 2011). To
address uncertainty issues systematically, the following steps
should be considered: firstly, the sources of uncertainty are to be
identified; secondly, uncertainty from different sources need to be
quantified (or at least ranked qualitatively) and then prioritised;
and uncertainty should be properly communicated (Beven et al.,
2015; Dottori et al., 2013).
4.4.1. Identifying sources of uncertainty
Although there are many sources of uncertainty in flood inun-
dation modelling, Merz and Thieken (2005) state that ‘it is impor-
tant to recognise two basic kinds of uncertainty that are
fundamentally different from each other: natural and epistemic
uncertainty. Natural uncertainty stems from variability of the un-
derlying stochastic processes. Epistemic uncertainty results from
incomplete knowledge about the process under study.’ Most of the
natural uncertainties can be treated formally by statistical proba-
bilistic methods, although it can be difficult to identify an appro-
priate statistical model. Epistemic uncertainties, including
subjective uncertainty, lack-of-knowledge uncertainty, and errors,
imply that the nature of the uncertainty may not be consistent in
time or space, and may therefore be difficult to treat formally by
probabilistic methods (Freer et al., 2011). Another practical way to
identify the sources of uncertainty is to focus on model processes
(Beven et al., 2015; Dottori et al., 2013). In that connection, the
following sources of uncertainty are considered the most relevant
and widely studied: choice of model structures (Apel et al., 2009),
model parameters (e.g. friction, conveyance parameters etc.) (Bates
et al., 2004; Pappenberger et al., 2005; Romanowicz and Beven,
2003), model inputs (floodplain and channel geometry, initial and
boundary conditions such as inflow hydrograph and lateral inflows)
(Abily et al., 2016; Pappenberger et al., 2008; Savage et al., 2016),
so-called validation data (Stephens et al., 2012; Werner et al., 2005),
change in floodplain landscape over time (e.g. land use change)
(Beven, 2010), and change in climate conditions (Neal et al., 2013,
Vaze et al., 2011).
4.4.2. Quantifying uncertainty from different sources
Purpose-designed methods are available to efficiently quantify
uncertain drivers in order to find those that have the greatest in-
fluence, and/or to rank or weight them based on their importance.
These techniques are commonly referred to as ‘sensitivity analysis’
(Hall et al., 2005; Saltelli et al., 2000). These can be local sensitivity
analysis (SA) where variables or parameters are perturbed one at a
time by a small amount around some fixed point and the effect on
the model output of interest is calculated (Tsubaki and Kawahara,
2013); and global methods where all variables or parameters are
varied simultaneously over their entire feasible space, typically
using a sampling based approach, and the effects on the output of
both individual variables/parameters and interactions between
them are assessed (such as Sobol, KullbackeLeibler entropy, Morris,
regionalised sensitivity analysis and regression) (Abily et al., 2016;
Le Cozannet et al., 2015; Mara et al., 2015; Norton, 2015;
Pappenberger et al., 2008; Sarrazin et al., 2016).
Published sensitivity studies have mainly focused on the un-
certainties sourced from model parameters and inputs. Although
friction parameters are usually considered the most influential and
are therefore given first consideration in model calibration (Hall
et al., 2005; Pappenberger et al., 2005), Pappenberger et al.
(2008) demonstrated that different methods can lead to
completely different ranking of importance of the parameter fac-
tors and that it is impossible to draw firm conclusions about the
relative sensitivity of different factors. Shin et al. (2013) show, for a
wide class of conceptual rainfall-runoff models, that climate input
period, quantity of predictive interest and objective function affect
parameter sensitivity rankings. However the major sensitivities can
be identified, it is the lesser ones that may change with circum-
stance, making the use of a sensitivity analysis and precise speci-
fication of the quantity of interest, we would argue, still important
to understand the relative and crucial sources of uncertainty.
Model emulation which involves the production of a surrogate
model, usually being an abstract formalism not a physical model,
can also be utilized to provide global sensitivities. The surrogates
are derived by sampling the factor space (inputs and parameters) of
the original model, producing a response surface for the quantity of
predictive interest by simulation through the original model, and
then fitting the surrogate's parameters to that response surface. The
most common surrogates are based on Gaussian Processes (GP)
(O'Hagan and Kingman, 1978) or Polynomial Chaos Emulation (PCE)
(Sudret, 2008). The advantage of an emulation exercise is that one
achieves a much computationally faster model that can be used for
subsequent uncertainty analysis (see Laloy et al., 2013 for a hy-
drological application of this technique).
4.4.3. Representation of uncertainty
An important attitude in taking account of, or managing, un-
certainty in flood inundation mapping is obtaining an appreciation
of uncertainty as far as possible from multiple viewpoints,
including methods and assumptions, the latter therefore including
multiple models. An aspect not often recognised as an uncertainty
in itself is how uncertainty is communicated. A key part of this is
the communication of the meaning of the outputs of predictive
interest from the uncertainty analysis (Beven et al., 2015). Typically,
uncertainty is expressed by a combination, or ensemble, of model
realisations after eliminating the obviously impossible outcomes.
Such model realisations are conventionally produced from a Monte
Carlo-based sampling of different parameter values, inputs and/or
(less often) model structures. However we return to alternate ways
of characterising uncertainty later in the section.
Broadly speaking, two types of uncertainty estimation have
been invoked in flood inundation models. Both types involve
realisations of a model by sampling of prior parameter and input
functions to produce a space of parameters that capture their un-
certainty, and hence the output uncertainty for the quantity of
predictive interest. Bayesian approaches use formal probability
distributions for the priors and formal likelihood measures, and
under certain conditions, if satisfied, yield statistically rigorous
confidence intervals (Hall et al., 2011). Convergence issues to obtain
the posterior parameter estimates, however, may need strong
attention (Sarrazin et al., 2016). The Generalized Likelihood Un-
certainty Estimation (GLUE) (Beven and Binley, 1992) methodology,
however, is less formally based and invokes, perhaps more flexible
or pseudo-likelihood (Stephens and Bates, 2015), measures that
 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
partition the parameter space into acceptable and non-acceptable
model behaviours. It yields confidence intervals that should not
pretend to be statistically rigorous but which at least give an
appreciation of uncertainties.
The GLUE-based methods have proven to be useful in producing
(pseudo) probabilistic flood inundation predictions (Aronica et al.,
2002; Romanowicz and Beven, 2003; Romanowicz et al., 2006).
They have gained vast popularity and are subject to on-going use
and extension. For example, Bates et al. (2004) combined likelihood
values from different events using Bayes equation. Romanowicz
and Beven (2003) and Pappenberger et al. (2005) applied fuzzy
based measures in evaluating a flood inundation model to allow for
measurements with high uncertainties. The main result constituted
a map showing the distributed probabilities of inundation for a
given flow event, thereby providing an easy way to evaluate the risk
of inundation in future specified events.
However, application of the GLUE methodology, as do Bayesian
methods, require a number of subjective decisions, such as deciding
which parameters of the model will be fixed and which will be
varied, the parameter ranges, the ranges of acceptable effective
observational error, the likelihood criteria, the treatment of error in
input and boundary conditions. Previous studies suggest that this
choice leads to clear differences in inundation probabilities
(Blazkova and Beven, 2009; Stephens and Bates, 2015).
Another constraint on the application of both methods is the
computing limitations on running sufficient model realisations,
particularly when models with a high number of parameter di-
mensions, or long run times, or multiple model structures are being
considered. This can be overcome by i) improving model efficiency,
including use of more advanced computing facilities; ii) using
simplified conceptual models; or iii) using a surrogate model or
model emulation (e.g. Keating et al., 2010; Doherty and
Christensen, 2011; Asher et al., 2015). The authors are not aware
of emulation using GP or PCE in a hydrodynamic context but it
appeals as a way forward especially as it also can provide parameter
sensitivities for free. Emulation needs to be appropriately validated
against the original model, however, so that it can be used instead
when undertaking sampling of the factor space for either type of
uncertainty analysis.
As expressed earlier there are other ways to capture and
communicate uncertainties (e.g. Jakeman and Jakeman, 2016). One
is scenario modelling with and without formal probabilities: In its
broadest sense, this involves exploring multiple, plausible sce-
narios about future conditions, model structure and parameter
values. It can be used for many purposes, such as to promote dis-
cussion and sharing of knowledge and perspectives and/or to
search for those scenarios that lead to good, intermediate and poor
outcomes.
Another approach is to crash test or torture the model. This can
be similar to scenario modelling in that attempts are made to see
what model parameter sets, observation periods and other condi-
tions establish limitations or invalidate the model. This can involve
examining the performance of the model through time and/or
space to assess inadequate performance.
Comparing outputs of predictive interest from multiple models
is another approach. Different model candidates or perspectives
can be used with the other tools mentioned above and below to
understand sources of uncertainty. Various techniques such as
Bayesian model selection can then be used to assess the strengths
and weaknesses of each, and under which conditions each model is
more suitable.
One should also not neglect seeking qualitative information
about uncertainty. A qualitative approach may be sufficient or at
least a useful adjunct in some cases. The emphasis in this case may
be on how the prediction was produced, and the limitations
213
involved in doing so. As covered in Jakeman et al. (2016), one way of
approaching this is through quality assurance of the modelling
process (Refsgaard et al., 2007) and its constituent assumptions
whilst another is to include qualitative judgements about the in-
formation and how it is produced (Van Der Sluijs et al., 2005).
As crucial as any step is understanding one's data, including data
errors, both type and magnitude, by analysis prior to modelling. But
this should also be applied to model residuals to check primarily for
bias, and cross-correlation with model inputs and with model
outputs. Simple text-book analysis of data to reveal their un-
certainties before modelling is under-practised or at least under-
reported. There is a wealth of tools available to detect outliers,
trends, implausible correlations, timing errors in model response,
and generally extract information from data. The value of simple
plotting and visualisation should not be ignored.
5. Summary and next steps
Despite the wide utility of flood inundation modelling as out-
lined in the Introduction, there are still unanswered research
questions that warrant addressing in the near future. Faster and
more accurate fluid solvers are always desirable and research in this
area is likely to continue in any case for the years to come. On the
other hand, some processes, in particular those associated with
antecedent soil moisture conditions and surface water-
groundwater interaction, are important on the floodplain during
a flood event but are currently under-represented in models re-
ported in the literature.
Differences in the predictions derived from different modelling
approaches highlight the uncertainty associated with the model-
ling practices. Conversely, there is some virtue in having an eclectic
range of assumptions and approaches to appreciate the range of
possible uncertainties; for example so that a precautionary
approach can be adopted to understand hazards and meet envi-
ronmentally beneficial needs associated with floods.
Quantifying and reducing uncertainty will consequently
continue to be a research need and emphasis. Using surrogate
models can potentially ease the constraint of the heavy computa-
tional demand from uncertainty analysis and should be explored
further. Methodologies acknowledging the uncertainty inherent in
the inundation modelling process, such as probabilistic flood risk
maps have proven to be preferable to single deterministic maps of
inundation extent and are likely to be more in demand in the near
future.
As remote sensing techniques continue to improve, and the data
become more readily available, the integration of remote sensing
data with flood inundation modelling will prevail. The remotely
sensed imageries will be more routinely used in model calibration,
validation and assimilation. A data-rich environment also encour-
ages development of models with various fidelity that can produce
the predictions to fit the available data within error tolerance. This
in turn allows robust flood inundation modelling to be extended to
greater scale, and more difficult-to-model terrain (e.g. urban sim-
ulations). Yamazaki et al., 2011, 2013 provided an example on global
flood prediction systems. Another good example is given by the
recent work being done in the US at the NOAA National Water
Centre. Beginning in 2015 with National Flood Interoperability
Experiment (NFIE), they developed and launched the National
Water Model (NWM) in August 2016 (http://water.noaa.gov/about/
nwm). It integrated weather forecasting model and hydrological
model to generate real-time water prediction and flood forecasting
for the continental United States at 2.7 million river channels.
HAND approach was used to establish a foundation for locally
informative, real-time flood inundation mapping continuously
across the continental United States (http://www.caee.utexas.edu/
214 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
prof/maidment/NFIE/SI2016/SI2016Draft.pdf). Being an operational
system, United States Geological Survey (USGS) streamflow obser-
vations are assimilated into the analysis and assimilation configu-
ration benefits from the inclusion of 1260 reservoirs.
From a technology point of view, parallel computing has become
a trend for reducing a model's computational cost. Any use of
parallel and supercomputing technologies will require that model
developers have the appropriate hardware in mind when writing
model codes. Open source availability of software is another di-
rection for the future as its elevated accessibility encourages high
and diverse usage which in turn assists in model development
improvements.
Last but not the least, inter-model, inter-discipline approaches
can take advantage of the merits of various approaches while
avoiding shortcomings. The astute combining of models has been
achieved and implemented in numerous ways and can be expected
to continue to succeed.
6. Conclusions
In conclusion, over the past century, and particularly since the
1970s, there have been vast improvements in flood inundation
modelling. The review of the state-of-the-art methods in this paper
represents our knowledge at a snap-shot in time. New de-
velopments are currently being undertaken in nearly all aspects
discussed above, especially in the functions of related software
packages. We recommend that managers and decision makers refer
to the developers' technical reports for latest updates.
It cannot be over-emphasised that there is no such thing as a
“perfect model” and the aim of developing and using models that
are ‘as realistic as possible’ must be balanced against the compu-
tational demand, investment in data collection and model set-up,
and the requirements of the end user. When it comes to choosing
the right model for the right application, parsimony is an overriding
principle: avoid more complication than is necessary to fulfil the
objectives, and be precise and realistic about the objectives
(Jakeman et al., 2006).
While empirical methods could be quite sufficient for flood
monitoring and post-disaster assessment, hydrodynamic models
are required to represent detailed flow dynamics to investigate
impacts of dam break, tsunamis or flash floods. Simplified con-
ceptual models can be good candidates for probabilistic flood risk
assessment, and multi-scenario modelling on a large floodplain
with well-defined channels.
There seems almost unlimited potential for improving,
combining, and applying flood inundation models to understand,
assess and predict the impact of floods on the environment and
human society. Flood inundation modellers should avoid the
temptation of being too obsessed with their own models and have
the courage and wisdom to be always open to the new ideas, new
technologies, and new opportunities that may come from even a
totally alien field.
Acknowledgements
This work was carried out as part of a PhD research. The authors
thank the CSIRO Land and Water (the Water Resource Management
Program) and the Murray-Darling Basin Authority for funding and
supporting this research.
References
Abbott, M.B., Bathurst, J.C., Cunge, J.A., O'Connell, P.E., Rasmussen, J., 1986. An
introduction to the European Hydrological System e syste me Hydrologique
Europe
en, ‘‘SHE’’, 1: history and Philosophy of a physically-based distributed
modelling system. J. Hydrol. 87, 45e59.
Abily, M., Bertrand, N., Delestre, O., Gourbesville, P., Duluc, C.-M., 2016. Spatial
Global Sensitivity Analysis of High Resolution classified topographic data use in
2D urban flood modelling. Environ. Model. Softw. 77, 183e195.
Alcrudo, F., 2004. A State of the Art Review on Mathematical Modelling of Flood
Propagation, IMPACT Project, p. 22. http://www.impact-project.net/cd/papers/
print/008_pr_02-05-16_IMPACT_Alcrudo.pdf.
Apel, H., Aronica, G.T., Kreibich, H., Thieken, A.H., 2009. Flood risk analysesdhow
detailed do we need to be? Nat. Hazards 49 (1), 79e98.
Apel, H., Thieken, A., Merz, B., Blo€schl, G., 2006. A probabilistic modelling system for
assessing flood risks. Nat. Hazards 38 (1e2), 79e100.
Arduino, G., Reggiani, P., Todini, E., 2005. Recent advances in flood forecasting and
flood risk assessment. Hydrol. Earth Syst. Sci. 9 (4), 280e284.
Aronica, G., Bates, P.D., Horritt, M.S., 2002. Assessing the uncertainty in distributed
model predictions using observed binary pattern information within GLUE.
Hydrol. Process. 16 (10), 2001e2016.
Asher, M.J., Croke, B.F.W., Jakeman, A.J., Peeters, L.J.M., 2015. A review of surrogate
models and their application to groundwater modelling. Water Resour. Res. 51
(8), 5957e5973.
Bates, P.D., Horritt, M.S., Aronica, G., Beven, K., 2004. Bayesian updating of flood
inundation likelihoods conditioned on flood extent data. Hydrol. Process. 18
(17), 3347e3370.
Bates, P.D., Horritt, M.S., Smith, C.N., Mason, D., 1997. Integrating remote sensing
observations of flood hydrology and hydraulic modelling. Hydrol. Process. 11
(14), 1777e1795.
Beven, K., 1989. Changing ideas in hydrology - the case of physically-based models.
J. Hydrol. 105 (1e2), 157e172.
Beven, K., 2010. Distributed Models and Uncertainty in Flood Risk Management.
Flood Risk Science and Management, pp. 289e312.
Beven, K., Binley, A., 1992. The future of distributed models: model calibration and
uncertainty prediction. Hydrol. Process. 6 (3), 279e298.
Beven, K., Lamb, R., Leedal, D., Hunter, N., 2015. Communicating uncertainty in flood
inundation mapping: a case study. Int. J. River Basin Manag. 13 (3), 285e295.
Bhuiyan, M., Dutta, D., 2012. Analysis of flood vulnerability and assessment of the
impacts in coastal zones of Bangladesh due to potential sea-level rise. Nat.
Hazards 61 (2), 729e743.
Blazkova, S., Beven, K., 2009. Uncertainty in flood estimation. Struct. Infrastruct.
Eng. 5 (4), 325e332.
Brackbill, J.U., Ruppel, H.M., 1986. FLIP: a method for adaptively zoned, particle-in-
cell calculations of fluid flows in two dimensions. J. Comput. Phys. 65 (2),
314e343.
Bridson, R., 2008. Fluid Simulation for Computer Graphics. A K Peters/CRC Press.
Brunner, G.W., 2016. HEC-RES River Analysis System - User's Manual Version 5.0. US
Army Corps of Engineers. Institute for Water Resources, Hydrologic Engineering
Center (HEC), p. 962.
Caleffi, V., Valiani, A., Zanni, A., 2003. Finite volume method for simulating extreme
flood events in natural channels. J. Hydraul. Res. 41 (2), 167e177.
Casulli, V., Stelling, G.S., 1998. Numerical simulation of 3D quasi-hydrostatic, free-
surface flows. J. Hydraul. Eng. 124 (7), 678e686.
Cleary, P.W., Prakash, M., 2004. Discrete-element modelling and smoothed particle
hydrodynamics: potential in the environmental sciences. Philos. Trans. R. Soc. A
Math. Phys. Eng. Sci. 362 (1822), 2003e2030.
DHI, 2003. MIKE 11-A Modelling System for Rivers and Channels - User Guide. DHI,
p. 430.
DHI, 2012. MIKE 21-2D Modelling of Coast and Sea. DHI Water & Environment Pty
Ltd.
Doherty, J., Christensen, S., 2011. Use of paired simple and complex models to
reduce predictive bias and quantify uncertainty. Water Resour. Res. 47 (12),
W12534.
Dottori, F., Di Baldassarre, G., Todini, E., 2013. Detailed data is welcome, but with a
pinch of salt : accuracy, precision, and uncertainty in flood inundation
modeling. Water Resour. Res. 49 (9), 6079e6085.
Dottori, F., Todini, E., 2011. Developments of a flood inundation model based on the
cellular automata approach: testing different methods to improve model per-
formance. Physics and Chemistry of the Earth. Parts A/B/C 36 (7e8), 266e280.
Dutta, D., Herath, S., Musiake, K., 2006. An application of a flood risk analysis system
for impact analysis of a flood control plan in a river basin. Hydrol. Process. 20
(6), 1365e1384.
Dutta, D., Teng, J., Vaze, J., Lerat, J., Hughes, J., Marvanek, S., 2013. Storage-based
approaches to build floodplain inundation modelling capability in river system
models for water resources planning and accounting. J. Hydrol. 504 (0), 12e28.
Freer, J., Beven, K.J., Neal, J., Schumann, G., Hall, J., Bates, P., 2011. Flood Risk and
Uncertainty. Risk and Uncertainty Assessment for Natural Hazards, Cambridge,
UK, pp. 190e233.
Gallegos, H.A., Schubert, J.E., Sanders, B.F., 2009. Two-dimensional, high-resolution
modeling of urban dam-break flooding: a case study of Baldwin Hills, California.
Adv. Water Resour. 32 (8), 1323e1335.
Gao, B.C., 1996. NDWI - a normalized difference water index for remote sensing of
vegetation liquid water from space. Remote Sens. Environ. 58 (3), 257e266.
Gingold, R.A., Monaghan, J.J., 1977. Smoothed particle hydrodynamics - theory and
application to non-spherical stars. Mon. Not. R. Astron. Soc. 181 (2), 375e389.
Guidolin, M., Chen, A.S., Ghimire, B., Keedwell, E.C., Djordjevi
c, S., Savi

c, D.A., 2016.
A weighted cellular automata 2D inundation model for rapid flood analysis.
Environ. Model. Softw. 84, 378e394.
Hall, J.W., Manning, L.J., Hankin, R.K.S., 2011. Bayesian calibration of a flood
 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
inundation model using spatial data. Water Resour. Res. 47 (5), W05529.
Hall, J.W., Tarantola, S., Bates, P.D., Horritt, M.S., 2005. Distributed sensitivity anal-
ysis of flood inundation model calibration. J. Hydraul. Eng. 131 (2), 117e126.
Harlow, F.H., 1962. The Particle-in-cell Method for Numerical Solution of Problems
in Fluid Dynamics. United States report LA-DC-5288, 39, http://www.osti.gov/
scitech/servlets/purl/4769185.
Harlow, F.H., Welch, J.E., 1965. Numerical calculation of time-dependent viscous
incompressible flow of fluid with free surface. Phys. Fluids 8 (12), 2182.
Harten, A., Lax, P.D., Vanleer, B., 1983. On upstream differencing and Godunov-type
schemes for hyperbolic conservation-laws. Siam Rev. 25 (1), 35e61.
Hunter, N.M., Bates, P.D., Horritt, M.S., Wilson, M.D., 2007. Simple spatially-
distributed models for predicting flood inundation: a review. Geomorphology
90 (3e4), 208e225.
Jakeman, A.J., Barreteau, O., Hunt, R.J., Rinaudo, J.-D., Ross, A., 2016. Integrated-
Groundwater Management; Concepts, Approaches and Challenges. Springer,
762 pp.
Jakeman, A.J., Jakeman, J.D., 2016. An Overview of Methods to Identify and Manage
Uncertainty for Modelling Problems in the Water-agriculture-environment
Sector. Forum “Math for Industry” 2016, pp.24e30. MI lecture Note volume
70. Kyushu University, Institute Mathematics for Industry. ISSN 2188e1200.
http://www.imi.kyushu-u.ac.jp/.
Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development
and evaluation of environmental models. Environ. Model. Softw. 21 (5),
602e614.
Karim, F., Dutta, D., Marvanek, S., Petheram, C., Ticehurst, C., Lerata, J., Kim, S.,
Yang, A., 2015. Assessing the impacts of climate change and dams on floodplain
inundation and wetland connectivity in the wet-dry tropics of northern
Australia. J. Hydrol. 522, 80e94.
Keating, E.H., Doherty, J., Vrugt, J.A., Kang, Q., 2010. Optimization and uncertainty
assessment of strongly nonlinear groundwater models with high parameter
dimensionality. Water Resour. Res. 46 (10), W10517.
Laloy, E., Rogiers, B., Vrugt, J.A., Mallants, D., Jacques, D., 2013. Efficient posterior
exploration of a high-dimensional groundwater model from two-stage Markov
chain Monte Carlo simulation and polynomial chaos expansion. Water Resour.
Res. 49 (5), 2664e2682.
Le Cozannet, G., Rohmer, J., Cazenave, A., Idier, D., van de Wal, R., de Winter, R.,
Pedreros, R., Balouin, Y., Vinchon, C., Oliveros, C., 2015. Evaluating uncertainties
of future marine flooding occurrence as sea-level rises. Environ. Model. Softw.
73, 44e56.
L'homme, J., Sayers, P., Gouldby, B., Samuels, P., Wills, M., Mulet-Marti, J., 2008.
Recent development and application of a rapid flood spreading method. In:
Samuels, P., Huntington, S., Allsop, W., Harrop, J. (Eds.), Flood Risk Management:
Research and Practice. Taylor & Francis Group, London, UK.
Liang, D., Falconer, R.A., Lin, B., 2006. Comparison between TVD-MacCormack and
ADI-type solvers of the shallow water equations. Adv. Water Resour. 29 (12),
1833e1845.
Liu, Q., Qin, Y., Zhang, Y., Li, Z.W., 2015. A coupled 1D-2D hydrodynamic model for
flood simulation in flood detention basin. Nat. Hazards 75 (2), 1303e1325.
Losasso, F., Talton, J., Kwatra, N., Fedkiw, R., 2008. Two-way coupled SPH and par-
ticle level set fluid simulation. IEEE Trans. Vis. Comput. Graph. 14 (4), 797e804.
Lucy, L.B., 1977. Numerical Approach to testing of fission hypothesis. Astron. J. 82
(12), 1013e1024.
Mara, T.A., Tarantola, S., Annoni, P., 2015. Non-parametric methods for global
sensitivity analysis of model output with dependent inputs. Environ. Model.
Softw. 72, 173e183.
Marriott, S., 1992. Textural analysis and modeling of a flood deposit - river Severn.
UK. Earth Surf. Process. Landf. 17 (7), 687e697.
Martinis, S., Twele, A., Voigt, S., 2009. Towards operational near real-time flood
detection using a split-based automatic thresholding procedure on high reso-
lution TerraSAR-X data. Nat. Hazards Earth Syst. Sci. 9 (2), 303e314.
Matgen, P., Hostache, R., Schumann, G., Pfister, L., Hoffmann, L., Savenije, H.H.G.,
2011. Towards an automated SAR-based flood monitoring system: lessons
learned from two case studies. Phys. Chem. Earth 36 (7e8), 241e252.
Medeiros, S.C., Hagen, S.C., 2013. Review of wetting and drying algorithms for nu-
merical tidal flow models. Int. J. Numer. methods fluids 71 (4), 473e487.
Merz, B., Kreibich, H., Schwarze, R., Thieken, A., 2010. Review article 'Assessment of
economic flood damage'. Nat. Hazards Earth Syst. Sci. 10 (8), 1697e1724.
Merz, B., Thieken, A.H., 2005. Separating natural and epistemic uncertainty in flood
frequency analysis. J. Hydrol. 309 (1e4), 114e132.
Monaghan, J.J., 1992. Smoothed particle hydrodynamics. Annu. Rev. Astron. Astro-
phys. 30, 543e574.
Monaghan, J.J., 1994. Simulating free surface flows with SPH. J. Comput. Phys. 110
(2), 399e406.
Moulinec, C., Denis, C., Pham, C.T., Rouge, D., Hervouet, J.M., 2011. TELEMAC: an
efficient hydrodynamics suite for massively parallel architectures. Comput.
Fluids 51 (1), 30e34.
Mure-Ravaud, M., Binet, G., Bracq, M., Perarnaud, J.-J., Fradin, A., Litrico, X., 2016.
A web based tool for operational real-time flood forecasting using data
assimilation to update hydraulic states. Environ. Model. Softw. 84, 35e49.
Neal, J., Keef, C., Bates, P., Beven, K., Leedal, D., 2013. Probabilistic flood risk mapping
including spatial dependence. Hydrol. Process. 27 (9), 1349e1363.
Neal, J., Schumann, G., Bates, P., 2012. A subgrid channel model for simulating river
hydraulics and floodplain inundation over large and data sparse areas. Water
Resour. Res. 48.
Ne
elz, S., Pender, G., 2009. Desktop Review of 2D Hydraulic Modelling Packages.
215
DEFRA/Environment Agency, UK. http://evidence.environment-agency.gov.uk/
FCERM/Libraries/FCERM_Project_Documents/SC080035_Desktop_review_of_
2D_hydraulic_packages_Phase_1_Report.sflb.ashx.
Ne
elz, S., Pender, G., 2010. Benchmarking of 2D Dyfraulic Modelling Packages.
DEFRA/Environment Agency, UK. https://www.gov.uk/government/uploads/
system/uploads/attachment_data/file/290884/scho0510bsno-e-e.pdf.
Ne
elz, S., Pender, G., 2013. Benchmarking the Latest Generation of 2D Hydraulic
Modelling Packages. DEFRA/Environment Agency, UK. http://evidence.
environment-agency.gov.uk/FCERM/Libraries/FCERM_Project_Documents/
SC120002_Benchmarking_2D_hydraulic_models_Report.sflb.ashx.
Nobre, A.D., Cuartas, L.A., Hodnett, M., Renno, C.D., Rodrigues, G., Silveira, A.,
Waterloo, M., Saleska, S., 2011. Height above the Nearest Drainage - a hydro-
logically relevant new terrain model. J. Hydrol. 404 (1e2), 13e29.
Norton, J., 2015. An introduction to sensitivity assessment of simulation models.
Environ. Model. Softw. 69, 166e174.
O'Connor, J.E., Costa, J.E., 2004. The World's Largest Floods, Past and PresentdTheir
Causes and Magnitudes: U.S. Geological Survey Circular, vol. 1254, p. 13.
O'Hagan, A., Kingman, J.F.C., 1978. Curve fitting and optimal design for prediction.
J. R. Stat. Soc. Ser. B Methodol. 40 (1), 1e42.
Pappenberger, F., Beven, K., Horritt, M., Blazkova, S., 2005. Uncertainty in the cali-
bration of effective roughness parameters in HEC-RAS using inundation and
downstream level observations. J. Hydrol. 302 (1e4), 46e69.
Pappenberger, F., Beven, K.J., Ratto, M., Matgen, P., 2008. Multi-method global
sensitivity analysis of flood inundation models. Adv. Water Resour. 31 (1), 1e14.
Pender, G., 2006. Briefing: introducing the flood risk management research con-
sortium. Proc. Inst. Civ. Eng. - Water Manag. 159 (1), 3e8.
Pizzuto, J.E., 1987. Sediment diffusion during overbank flows. Sedimentology 34 (2),
301e317.
Prakash, M., Rothauge, K., Cleary, P.W., 2014. Modelling the impact of dam failure
scenarios on flood inundation using SPH. Appl. Math. Model. 38 (23),
5515e5534.
Pulvirenti, L., Chini, M., Pierdicca, N., Guerriero, L., Ferrazzoli, P., 2011. Flood
monitoring using multi-temporal COSMO-SkyMed data: image segmentation
and signature interpretation. Remote Sens. Environ. 115 (4), 990e1002.
Refsgaard, J.C., van der Sluijs, J.P., Højberg, A.L., Vanrolleghem, P.A., 2007. Uncer-
tainty in the environmental modelling process e a framework and guidance.
Environ. Model. Softw. 22 (11), 1543e1556.
Roberts, S., Nielsen, O., Gray, D., Sexton, J., Davies, G., 2015. ANUGA User Manual.
Commonwealth of Australia (Geoscience Australia) and the Australian National
University, p. 127.
Roe, P.L., 1981. Approximate Riemann solvers, parameter vectors, and difference-
schemes. J. Comput. Phys. 43 (2), 357e372.
Romanowicz, R., Beven, K., 2003. Estimation of flood inundation probabilities as
conditioned on event inundation maps. Water Resour. Res. 39 (3),
SWC41eSWC412.
Romanowicz, R.J., Beven, K.J., Young, P.C., 2006. Uncertainty Propagation in a
Sequential Model for Flood Forecasting, 303 ed. IAHS-AISH publication,
pp. 177e184.
Saltelli, A., Chan, K., Scott, E.M., 2000. Sensitivity Analysis. John Wiley & Sons, LTD.,,
494pp.
Sanyal, J., Lu, X.X., 2004. Application of remote sensing in flood management with
special reference to monsoon Asia: a review. Nat. Hazards 33 (2), 283e301.
Sarrazin, F., Pianosi, F., Wagener, T., 2016. Global sensitivity analysis of environ-
mental models: convergence and validation. Environ. Model. Softw. 79,
135e152.
Savage, J.T.S., Bates, P., Freer, J., Neal, J., Aronica, G., 2016. When does spatial reso-
lution become spurious in probabilistic flood inundation predictions? Hydrol.
Process. 30 (13), 2014e2032.
Schumann, G., Bates, P.D., Horritt, M.S., Matgen, P., Pappenberger, F., 2009. Progress
in integration of remote sensingederived flood extent and stage data and hy-
draulic models. Rev. Geophys. 47 (4), RG4001.
Shin, M.-J., Guillaume, J.H.A., Croke, B.F.W., Jakeman, A.J., 2013. Addressing ten
questions about conceptual rainfallerunoff models with global sensitivity an-
alyses in R. J. Hydrol. 503, 135e152.
Sleigh, P.A., Gaskell, P.H., Berzins, M., Wright, N.G., 1998. An unstructured finite-
volume algorithm for predicting flow in rivers and estuaries. Comput. Fluids
27 (4), 479e508.
Smith, L.C., 1997. Satellite remote sensing of river inundation area, stage, and
discharge: a review. Hydrol. Process. 11 (10), 1427e1439.
Stam, J., 1999. Stable Fluids, Proceedings of the 26th Annual Conference on Com-
puter Graphics and Interactive Techniques. ACM Press/Addison-Wesley Pub-
lishing Co.,, pp. 121e128.
Stephens, E., Bates, P., 2015. Assessing the reliability of probabilistic flood inunda-
tion model predictions. Hydrol. Process. 29 (19), 4264e4283.
Stephens, E.M., Bates, P.D., Freer, J.E., Mason, D.C., 2012. The impact of uncertainty in
satellite data on the assessment of flood inundation models. J. Hydrol. 414e415,
162e173.
Sudret, B., 2008. Global sensitivity analysis using polynomial chaos expansions.
Reliab. Eng. Syst. Saf. 93 (7), 964e979.
Sulsky, D., Chen, Z., Schreyer, H.L., 1994. A particle method for history-dependent
materials. Comput. Methods Appl. Mech. Eng. 118 (1), 179e196.
Teng, J., Vaze, J., Dutta, D., Marvanek, S., 2015. Rapid inundation modelling in large
floodplains using LiDAR DEM. Water Resour. Manag. 29 (8), 2619e2636.
Ticehurst, C., Dutta, D., Karim, F., Petheram, C., Guerschman, J., 2015. Flood mapping
at a daily scale using MODIS imagery in large floodplains and reduction of
216 J. Teng et al. / Environmental Modelling & Software 90 (2017) 201e216
uncertainty with the aid of hydrodynamic modelling. Nat. Hazards Earth Syst.
Sci. 78, 803e820.
Ticehurst, C., Guerschman, J.P., Chen, Y., 2014. The strengths and limitations in using
the daily MODIS open water likelihood algorithm for identifying flood events.
Remote Sens. 6 (12), 11791e11809.
Toro, E.F., Garcia-Navarro, P., 2007. Godunov-type methods for free-surface shallow
flows: a review. J. Hydraul. Res. 45 (6), 736e751.
Tsubaki, R., Kawahara, Y., 2013. The uncertainty of local flow parameters during
inundation flow over complex topographies with elevation errors. J. Hydrol.
486, 71e87.
Vacondio, R., Rogers, B., Stansby, P., Mignosa, P., 2011. SPH modeling of shallow flow
with open boundaries for practical flood simulation. J. Hydraul. Eng. 138 (6),
530e541.
Van Der Sluijs, J.P., Craye, M., Funtowicz, S., Kloprogge, P., Ravetz, J., Risbey, J., 2005.
Combining quantitative and qualitative measures of uncertainty in model-
based environmental assessment: the NUSAP system. Risk Anal. 25 (2),
481e492.
Vaze, J., Teng, J., Chiew, F.H.S., 2011. Assessment of GCM simulations of annual and
seasonal rainfall and daily rainfall distribution across south-east Australia.
Hydrol. Process 25 (9), 1486e1497.
Vaze, J., Viney, N., Stenson, M., Renzullo, L., Van Dijk, A., Dutta, D., Crosbie, R.,
Lerat, J., Penton, D., Vleeshouwer, J., Peeters, L., Teng, J., Kim, S., Hughes, J.,
Dawes, W., Zhang, Y., Leighton, B., Perraud, J.-M., Joehnk, K., Yang, A., Wang, B.,
Frost, A., Elmahdi, A., Smith, A., Daamen, C., 2013. The Australian Water
Resource Assessment System (AWRA). In: Proceedings of the 20th International
Congress on Modelling and Simulation (MODSIM2013). Adelaide, Australia.
Wang, S., Baig, M.H.A., Zhang, L., Jiang, H., Ji, Y., Zhao, H., Tian, J., 2015. A simple
enhanced water index (EWI) for percent surface water estimation using landsat
data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8 (1), 90e97.
Werner, M., Blazkova, S., Petr, J., 2005. Spatially distributed observations in con-
straining inundation modelling uncertainties. Hydrol. Process. 19 (16),
3081e3096.
Woodhead, S., Asselman, N., Zech, Y., Soares-Fraza ~o, S., Bates, P., Kortenhaus, A.,
2007. Evaluation of Inundation Models. FLOODsite report T08-07-01, p. 28.
http://www.floodsite.net/html/partner_area/project_docs/t08_07_01_
inundation_model_evaluation_m8_1_v1_7_p15.pdf.
Xu, H., 2006. Modification of normalised difference water index (NDWI) to enhance
open water features in remotely sensed imagery. Int. J. Remote Sens. 27 (14),
3025e3033.
Yamazaki, D., de Almeida, G.A.M., Bates, P.D., 2013. Improving computational effi-
ciency in global river models by implementing the local inertial flow equation
and a vector-based river network map. Water Resour. Res. 49 (11), 7221e7235.
Yamazaki, D., Kanae, S., Kim, H., Oki, T., 2011. A physically based description of
floodplain inundation dynamics in a global river routing model. Water Resour.
Res. 47.
Ye, J., McCorquodale, J.A., 1998. Simulation of curved open channel flows by 3D
hydrodynamic model. J. Hydraul. Eng. 124 (7), 687e698.