P.

Willems – River Modelling – Hydrodynamic river modelling

Hydrodynamic river modelling

Reference: Timbe L., 2007, ‘River flooding analysis using quasi-2D hydraulic modelling and geospatial
data’, PhD thesis Faculty of Engineering K.U.Leuven
Hydrodynamic river modelling

Reference: Timbe L., 2007, ‘River flooding analysis using quasi-2D hydraulic
modelling and geospatial data’, PhD thesis Faculty of Engineering K.U.Leuven

2.1 Introduction
The determination of potential flooded (flood-prone) areas along the river floodplains has
become a key issue for the government and local water authorities due to its political and
economic implications. Hence, river flood analysis very important for the assessment of
flood risk evaluation of floodplains, flood damage evaluation for project planning, design
of hydraulic structures for flood control, and flood-runoff forecasting for project
operations [USACE, 1994].

This analysis is performed by river flood routing techniques, which can be classified as
hydrologic or hydraulic routing. The hydrologic routing approach is based on the storage
continuity equation and an equation that uses the storage volume as linear or nonlinear
function of inflow and outflow discharges. In this method the discharge downstream of
the reach can be calculated based only on the upstream discharge. On the other hand, the
hydraulic approach solves the Saint Venant equations which are nonlinear and have no
analytical solution. The complete Saint Venant equations can be solved using numerical
techniques such as the finite difference, finite element and finite volume methods.
Hydraulic routing requires a lot more of data compared with hydrologic routing, but
water levels and discharges can be obtained at discrete points along the river reach.

For flood analysis the water levels in the river and on the floodplain is most important
than the magnitude of the discharge. Therefore, hydraulic analysis is the more appropriate
technique to compute the water elevation. In general, hydrologic analysis is performed
first in order to get the runoff discharges, which are used as input date in the hydraulic
model. In turn, the water levels resulting from the hydraulic model as used to obtain flood
maps. Normally from the water levels to flood maps a Geographic Information System
(GIS) interface is employed. This general methodology is presented in figure 2.1, and the
different modelling procedures, namely, hydrological, hydraulics and flood modelling are
explained hereafter.

RAINFALL CATCHMENT CHANNEL

HYDROLOGIC RIVER/FLOOD GIS
MODEL PLAIN FLOOD
RUNOFF HYDRAULIC MODEL MAPPING
DISCHARGES/
WATER LEVELS
Figure 2.1: Schematic representation of river flood modelling.

2.2 Hydrologic modelling

Flood hydrology is the estimation of the magnitude of streamflow at various locations in
a watershed resulting from a given precipitation input. The hydrological cycle comprises
all of the physical processes that affect the movement of water in its various forms. The
main processes include interception, water storage in the subsurface, lakes, reservoirs,
infiltration, percolation, evaporation, transpiration, etc.

The precipitation is the main driven input of the hydrological system, the analysis can be
done using a historical event, a design storm or a stochastic generated. The storm type
can be convective, orographic or frontal, the first one is of special interest due to its
substantial spatial and temporal variability. The sampling of rainfall is normally done
using gauge networks which may result in precipitation estimates with a high uncertain.
The gauge measurements themselves may exhibit significant uncertainty, primarily due
to wind effects. Another source of uncertainty is the watershed characteristics due to its
heterogenety, such as the topography, soils, land use, geology, drainage density and river
characteristics [Wheater, 2002].

Because the complexity of the real hydrologic system, the analysis is performed using
hydrologic models, which are an approximation of the reality. The inputs and outputs are
measurable hydrological variables and its structure are a set of equations that relate the
inputs and outputs. The simplest hydrological model is the metric model which make use
of observed data (rainfall and streamflow) to characterise the response of a catchment.

2.2.1 Hydrologic models

In a first classification, hydrologic models can be physical and abstract. Physical models
can be scale models which represent the system in a reduced scale and analogue models
which make use of another physical system with similar properties to the prototype. In
contrast, abstract models represent the hydrological system in a mathematical form, the
system operation is described by means of a set of equations that relate the input and
output variables. These variables can be function of the space and time, also they can be
probabilistic or random variables.

If the randomness and the variation in space and time are considered, the models can be
classified as deterministic and stochastic. A deterministic model do not consider the
randomness in the variables, the same input produce always the same output [Chow,
1994]. This type of model can be classified according to the description of the physical
processes as empirical, lumped and distributed [Refsgaard and Kundsen, 1996]. The
variables in a empirical model are based in empirical relationships based on experience.
In a deterministic lumped model, the system is averaged in the space or considered as a
single point with no dimensions in the space. A deterministic distributed model considers
that the hydrological processes occurs in different points in the space and the model
variables are defined as a function of the space dimensions.
On the other hand, in a stochastic model the output variables are at least partially random.
This models can be independent of the space or correlated with it, depending how the
random variables are related between them in different points of the space.

According the hydrological processes modelled, hydrologic codes can be divided into
event-driven and continuous models [Cunderlik, 2003]. The former is designed to
simulate individual precipitation-runoff events, the emphasis is placed on infiltration and
surface runoff because the main objective is the evaluation of direct runoff. The main
constrain of event models is their lack for moisture content recovery between events,
which not allows the proper simulation of dry weather flows. In contrast, continuous-
process models take explicit account of all runoff components, including direct and
indirect runoff. These models focus on long-term hydrologic abstractions responsible for
the rate of moisture recovery during the periods of no precipitation and usually they can
simulate at various times step (hourly, daily, monthly or seasonal)

The fundamental criteria to consider for the selection of hydrological model are
[Cunderlik, 2003]:

• the required model outputs important to the project,

• the hydrologic processes that need to be modelled to estimate the desired output
adequately,
• the availability of input data, and ,
• the price

All the hydrological models are an approximation of the reality, the output of a real
system never can be predicted with certainty. Nowadays the most popular classification
of hydrological models in use are: lumped and distributed, and between them semi-
distributed models. In the next paragraphs these models are discussed in detail.

2.2.1.1 Lumped models

The parameters of lumped hydrologic models have no spatial variability in the catchment,
therefore, the response is evaluated only at the basin outlet. Normally, in this type of
models the parameters do not represent the physical features of hydrologic processes and
frequently involve some degree of empiricism. The spatial variability of the model
parameters can not be evaluated directly, instead effective values area calculated for the
entire basin (area-weighted average). Lumped models are not usually applicable to event-
scale processes. They have a good performance compared with more complex model
such as distributed models if we the main interest is only the calculation of discharge
[Beven, 2000]

The main advantages of lumped models are their simple structure, the minimum
requirements of data, and fast to setup and calibrate the model. These characteristics lead
to satisfactory model results if the discharge is the primary concern.
Normally the only data input in these models is the precipitation, evapotranspiration and
the basin area, the discharge at the outlet of the catchment is also needed for calibration
purposes. Additionally air temperature and radiation is required if snowmelt is modelled.
Geographical data such as topography, vegetation and soils are not considered. Typical
examples of lumped hydrological models are the NAM and the IHACRES models.

NAM model: it is a commercial software, part of the DHI’s MIKE 11 generalized river
modelling package, which simulates the rainfall-runoff processes at catchment scale. The
model can be characterised as a deterministic, lumped, conceptual model. The
conceptualization of the hydrological system is performed by continuously accounting for
the moisture content in four mutually interrelated storages. The different physical
elements of the catchment is represented in each storage. The model is able to carry out
both, continous longterm simulation and single events with any time step selected by the
user. The model results are the outlet discharge and its different component (base, quick
and slow flow), effective rainfall, evaporation and transpiration [DHI, 2002].

IHACRES model: it is a hybrid conceptual-metric model, the simplicity of the metric

model is used for reduce the parameter uncertainty inherent in hydrological models while
at the same time attempting to represent more detail of the internal processes than is
typical for a metric model. IHACRES simulates continuous or individual events using
time steps equal or greater than 1 minute. The model uses a non-linear loss module that
converts rainfall into effective rainfall and a linear module which transfer effective
rainfall to streamflow. The inclusion of a range of non-linear loss modules can be used
for assessing the effects of climate and land use change. The linear module can routes the
effective rainfall to stream making use of any configuration of stores in parallel and/or in
series. The model outputs are time series of streamflow and the basin wetness index, unit
hydrographs, hydrograph separation (quick and slow flow) and indicative uncertainties
associated with the unit hydrograph parameters [Croke at al., 2005].

2.2.1.2 Distributed models

Distributed models considered that the parameters fully vary in space at a resolution
usually selected by the uses. Compared with lumped models, distributed codes require a
large amount of data for the parameterization in each grid cell increasing computational
requirements. Because the physical processes are modelled in detail, and if properly
applied, they can provide the highest degree of accuracy.

MIKE SHE model: Developed by Danish Hydraulic Institute (DHI) is a further

development of the European Hydrological System – Systéme Hydrologique Européen,
SHE (Abbot et al., 1986). It is a deterministic, fully-distributed and physically-based
modelling system for describing the mayor processes of the entire land phase of the
hydrological cycle: precipitation (rain or snow), evapotranspiration, interception,
overland sheet flow, channel flow, unsaturated subsurface flow and saturated ground
water flow. The evapotranspiration is calculated using the Kristensen and Jensen method.
Overland flow analysis includes a two-dimensional (2D) finite difference diffusive wave
approach using the same 2D mesh as the ground water component. The groundwater
model is computed using a 2D or 3D finite difference method. Vertical flow in the
unsaturated can be calculated using the full Richard’s equation, a simplified gravity flow
procedure, or a single two-layer water balance method for shallow water tables. MIKE
SHE solves the partial diferential equations for the processes of overland and channel
flow and unsaturated and saturated subsurface flow. The flow equations are solved
numerically using finite difference methods [DHI, 2004].

2.2.1.3 Semi-distributed models

In this type of models the parameters are partially allowed to vary in space by dividing
the basin into a number of smaller subbasins. They can be further classified into two main
types: kinematic wave theory models and probability distributed models.

The benefit of semi-distributed models is that their structure is more physically-based

than the structure of lumped models, and they are less demanding on input data and
computational requirements than fully distributed models. Examples of semi-distributed
are the HEC-HMS, SWAT and TOPMODEL, which are public domain codes.

HEC-HMS model: developed by the US Army Corps of Engineers to simulate both event
and continuous periods. Distributed runoff is computed using a grid-cell approach in
depiction of the watershed. The infiltration processes can be simulated for event
modelling by initial and constant, SCS curve, gridded SCS curve number, and Green &
Ampt methods. Five layer soil moisture accounting model can be used for continuous
simulation of complex infiltration and evapotranspiration. A number of hydrologic
routing methods such as lag, Muskingum, Muskingum-Cunge, modified puls, and
kinematic are available for simulating flow in open channels [US-ACE 2001].

SWAT (Soil and Water Assessment Tool) model: developed by the USDA-ARS has a
comprehensive structure that models all the processes in the watershed. The basins can be
subdivided into subbasins with the purpose of consider the differences in soils, land use,
crops, topography and weather. In addition, a snow model allows the subbasin to be split
in a set of elevation bands. The potential evapotranspiration can be computed using the
Hargreaves, Priestley-Taylor, or the Penman Monteith methods. The surface runoff
volume is calculated using a modification of the SCS curve number, or the Green &
Ampt infiltration methods. Flow in open channels is routed using a variable storage
coefficient method and the Muskingum method [Neitsch et al., 2002]. Finally, a user
friendly graphic user interface developed in ArcView32.

TOPMODEL: The distinctiveness of this model is the simulation of single or multiple

basins in a semi-distributed way using gridded elevation data for the catchment area
founded on the topographic index ln(a/tan b), where a is the drainage area to this point
and b is the land slope. TOPMODEL is a variable contributing area conceptual model,
where the catchment topography and the soil transmissivity that disminishes with the
depth are the major factors affecting the runoff generation. The estimation of surface
runoff production is computed based on variable saturated areas using infiltration excess
or saturation excess, while the subsurface flow is calculated using an exponential
function of water content in the saturated zone. The Penman-Monteith method is used to
calculate the potential evapotranspiration. Channel routing and infiltration excess area
computed using the Beven & Kirby method [Beven, 1996, 2000].

2.3 River hydraulics

Rivers are complex and dynamic, they always changing in response to human
intervention and changing climatic, geologic, and hydrological regimen. Effective
analysis of river problems requires recognition and understanding of the governing
processes in the river system. In hydraulic analysis must be considered:

• The characteristic of the flow in the river

• The geomorphologic behaviour of the river channel

These two components are sometimes treated separately; however, in alluvial channels
(channels with movable boundaries) the flow and the shape of the boundary are
interrelated.

One-dimensional, steady state, fixed bed water surface profiles are often computed as
part of traditional river hydraulics studies. However some floodplain management, flood
control, or navigation studies may require consideration of unsteady (time dependent)
flow, mobile boundaries, or multidimensional flow characteristics (flow with nonuniform
velocities distributions) to properly perform the required studies. There does not yet
exists definitive criteria which can be routinely applied to yield a clear choice of method.

2.3.1 Considerations for river hydraulics studies

To follow some topics to be considered for the proper application of river hydraulic
studies [US-ACE, 1993].

2.3.1.1 Flow dimensionality considerations

• Realms of one-dimensionality: to decide if a multidimensional study is needed, or a

one-dimensional approach is sufficient, a number of questions must be answered. If
only one principal direction can be identified, there is a good possibility that a one-
dimensional study will suffice.
• Limitations of one dimensionality. One-dimensional analysis implies that the
variation of relevant quantities in directions perpendicular to the main axis is either
assumed or neglected, not computed. Common assumptions are the hydrostatic
pressure distribution, well-mixed fluid properties in the vertical, uniform velocity
distribution in a cross section, zero velocity components transverse to the main axis.
• Two-dimensional flow: It is possible that actual transverse variations will differ so
greatly from the assumed variation that streamwise values, determined from a one-
dimensional study will be in significant error. If flow velocities in floodplains are
much less than that in the main channel, actual depths everywhere will be greater than
those computed on the basis of uniform velocity distribution in the entire cross
 section. Also, flow in a bend causes super-elevation of the water surface on the
outside of the bend which may be a significant.
• Determination of flow dimensionality: It is not possible to state with theoretical
certainty that a given reach can be assumed one-dimensional unless multi-
dimensional studies on the reach have been carried out and compared to the results of
a one dimensional approach. A practical rule of thumb, however, if the reach length is
more than twenty times the reach width, and if transverse flow and stage variations
are not specifically of interest, the assumption of one dimensionality will prove
adequate.
• Composite Channels: The concept of a composite channel is typically used to account
for retardation of flow by very rough floodplain in a one-dimensional analysis. It is
assumed that, with a horizontal water surface and energy slope common to main
channel and overland flows, the total discharge can be distributed among the main
channel and overbanks in proportion to their individual conveyances.
• Floodplains: A river raising rapidly and going overbank may take significant time to
inundate the floodplain. The transverse water surface will then not be horizontal and
will slope downward (laterally outward from the main channel) to provide the force
for the flood proceeding up the floodplain.
• Networks: while a network of interconnected streams is surely two-dimensional, the
individual channels comprising each reach of the network can usually be treated as
one-dimensional.

2.3.1.2 Water waves

Water flowing (or standing) with a free surface open to the atmosphere is always
susceptible to wave motion. The essence of wave motion exists in the concept of
propagation of disturbances. If a given flow is perturbed by something somewhere within
its boundaries, some manifestation of that perturbation is transmitted at some velocity of
propagation to other portions of the water body.

The essence of flood prediction is the forecasting of maximum stages in bodies of water
subject to phenomena such as precipitation runoff, tidal influences (including those from
storm tides), dam operations and possible dam failures. Also of interest are discharge and
stage hydrographs, velocities of anticipated currents, and duration of flooding.
Deterministic methods for making such predictions, typically called flood routing, relate
the response of the water to a particular flow sequence.

Flood routing

Many flood routing techniques were developed in the late nineteenth and early twentieth
centuries. Various simplifying assumptions were introduced to permit solutions with a
reasonable amount of computational effort. While analytical techniques for solving linear
wave equations were known, those solutions could not, in general, be applied to real
floods in real bodies of water because of the nonlinearity of the governing equations and
the complexity of the boundaries and boundary conditions. The advent and proliferation
of high-speed electronic computers in the second half of the twentieth century
revolutionized the computation of flood flows and their impacts. Numerical solutions of
the governing partial differential equations can now be accomplished with reasonable
effort.

Data for flood routing

Solution of the partial differential equations of river flow requires prescription of

boundary and initial conditions. In particular, the geometry of the watercourse and its
roughness must be known, as well as the hydraulic conditions at the upstream and
downstream ends of the reach and at all lateral inflows and outflows (tributaries,
diversions) along the reach. Due to the extreme irregularity of a natural watercourse, the
channel geometry and hydraulic properties (such as roughness and infiltration) cannot be
specified exactly.

Water motion

The motion of water particles at a cross section during a flood is nearly uniform, top to
bottom. The celerity of a flood wave is dependent, in a fundamental way, on the water
depth. In a flood wave, the pressure distribution is nearly hydrostatic; i.e., it increases
uniformly with depth below the surface. They are so-colled “long waves” that are, in fact,
gradually varied unsteady flows in open channels. The term unsteady implies that the
measurements of water velocity at one point in such a channel will show time variance at
a scale larger than turbulent fluctuations. “Varied” means that, at any instant, velocities at
different points along the channel are different. “Gradually varied” means that the
pressure distribution in a cross section is hydrostatic.

Wave speed

The analyst must be cognizant of the fact that the response of water in a river to a flood
or other disturbance is a wave which propagates at some speed and influences water
levels consecutively, not simultaneously.

The kinematic wave speed, that is, the speed of propagation of the main body of the
flood, is strongly dependent on the channel slope and roughness and must be considered
(Ponce 1989).

2.3.1.3 Flow classification

To determine which principles apply to a particular situation in river mechanics, it is

necessary to properly classify the flow. Various categories of flow are amenable to
different simplifying assumptions, data requirements, and methods of analysis. The first
step in the analysis of river hydraulics situations is classification of the state, type and
characteristics of the flow. Once the presumed flow characteristics have been
categorized, the engineer can identify the data, boundary conditions, and simulation
techniques appropriate for the situation.
Table 1 presents the flow classification according the different factors involved. Firstly
the water flow can be conveyed in pressure conduits and open channels (free water
surface), which can have movable or rigid boundaries; most natural rivers have movable
boundaries also know as alluvial channels. Considering the effects of viscosity the flow is
classified as laminar, transitional and turbulent depending on the ratio of viscous to
inertial forces defined by the Reynolds number. In natural river channels the flow is
always turbulent, while laminar flow is very rare. Taking into account the gravity force
the flow can be subcritical, critical and supercritical, depending of the Froude number
which is defined as the ratio of inertial to gravitational forces. If the flow is subcritical the
water surface profile is controlled by channel characteristics at the downstream end of the
river reach, and the calculation is done from downstream to upstream (backwater
calculation). In contrast if the flow is supercritical, the water surface profile is calculated
from upstream to downstream.

Table 1: Flow classification

Classification Parameter
Channel boundaries Pressure conduits
Open channels
Viscosity Laminar Re < 500
Transitional 500 ≤ Re ≤ 2000
Turbulent Re > 2000
Gravity Sub-critical Fr < 1
Critical Fr = 1
Supercritical Fr > 1

2.3.1.4 Regimes of flow

In natural rivers the flow is always fully turbulent, therefore the determination of flow
regime for open channels and river hydraulics is mostly verified if the state of the flow is
subcritical (Fr<1) or supercritical (Fr>1).

2.3.1.5 Types of flow

Considering the velocity, the flow is classified as steady flow if the velocity at a point
does not change in magnitude and direction with time, and unsteady flow if the velocity
at a location changes with time. Also the flow can be classified as uniform and
nonuniform flow. If the depth, water area, velocity and discharge do no change with the
distance along the channel the flow is uniform. These conditions are rarely met in natural
river channels, uniform flow can be only steady flow. On the contrary, most flow in
natural rivers are nonuniform or spatially varied flow, hydraulic variables vary only along
the length of the river reach.

Furthermore, nonuniform flow can be classified as rapidly varied flow if the spatial
changes in depth and/or velocity occur abruptly and the pressure distribution is not
hydrostatic. If the stream lines are practically parallel, e.g. the hydrostatic pressure
distribution exists throughout the channel section the nonuniform flow is classified as
gradually varied flow.

2.3.2 Flow routing

Flow routing is a mathematical method to predict the changing magnitude, speed, and
shape of a flood wave at one or more locations along a river, channel, reservoirs, etc. The
flood wave can originate from precipitation runoff (rainfall or snowmelt), reservoir
releases (spill flows or dam-failures), and tides (astronomical and/or wind-generated).

Flow routing is the procedure to compute the flow hydrograph at a point on a watercourse
from a known hydrograph(s) upstream. This is done in order to account for changes in
flow hydrograph shape as a flood wave passes downstream (figure 2.2).

Figure 2.2: Flood routing through a river reach.

There are two types of flow routing:

- Lumped or hydrologic: the flow is calculated as a function of time, the spatial

variability is not taken into account. Hydrologic routing is governed by continuity
equation and flow/storage relationships.

- Distributed or hydraulic: the flow is calculated as a function of space and time. It is

governed by the continuity and momentum equations.

2.3.3 One dimensional flow

The basic governing equations for one-dimensional unsteady flow were derived by Barré
de Saint-Venant, commonly know as the Saint-Venant equations represented by the
conservation of mass and conservation of momentum equations in a volume control (see
fig. 2.3):
Assumptions:

- the flow is one-dimensional: the water-lengths are large compared to water depths.
- hydrostatic pressure prevails and vertical accelerations are negligible
- streamline curvature is small
- bottom slope of the channels is small
- Manning’s equation is used to describe resistance effects
- the water is incompressible and homogeneous (no significant variation in density)

Equation of continuity

∂Q ∂A
+ =q Eq. (1)
∂x ∂t

Equation of momentum

1 ∂Q 1 ∂  Q 2  ∂y
+   + g − g (So − S f ) = 0 Eq. (2)
A ∂t A ∂x  A  ∂x

where
Q = flow
A = active flow area
S = storage area
q = lateral flow per unit flow distance
V = Q/A = average flow velocity
g = acceleration of gravity
z = water surface elevation
Sf = friction slope
VL = average velocity of the lateral inflow
x = flow distance
t = time
Figure 2.3: Control volume for the definition of continuity and momentum equations.

The different terms in the momentum equation represent the local acceleration, the
convective acceleration, the pressure force, the gravity force and the friction force
respectively. Eq. 1 and Eq. 2 can not be solved analytically (only for very simplified
cases), therefore numerical solutions are used for practical applications.

The momentum equation can be rewritten as:

∂V ∂V ∂y
+V +g − g (So − S f ) = 0 Eq. (3)
∂t ∂x ∂x
Kinematic wave

Diffusion wave

Dynamic wave

According the terms considered in the momentum equation (Eq 3) for the analysis the
Saint Venant equations can be simplified in the following models:

Kinematic wave: when gravity forces and friction forces balance each other (steep slope
channels with no back water effects).
Diffusion wave: when the pressure forces are important in addition to gravity and
frictional forces.

Dynamic wave: when both inertial and pressure forces are important and backwater
effects are not negligible (mild slope channels with downstream control).

Solution of the Saint-Venant equations

(a )Analytical

The Saint-Venant equations are solved by integrating partial differential equations, but
this procedure is applicable to only a few special simple cases of kinematic wave (due to
the presence of nonlinear terms).

(b) Numerical

Nowadays, the availability of high speed computers, it is possible to solve the complete
dynamic wave equations through numerical methods, which starts with initial and
boundary conditions at time t=0 (uniform steady flow is specified at all locations) and at
distance x=0 the flood hydrograph is known. The most common method used is the
finite-difference approximation. In this method the calculations are performed on a grid
placed over the (x,t) plane. The flow and water surface elevation are obtained for
incremental time and distances along the channel (see figure 2.4).

Figure 2.4: Finite-difference approximation, finite-difference grid.

The finite-difference approximation can be solved using an explicit or implicit scheme.
An explicit finite-difference scheme is referred when the spatial partial derivatives are
replaced in terms of the variables at the known time level, while in the implicit scheme
the spatial partial derivatives are expressed in terms of the values at the unknown time
level. The unknown variables, therefore, appear implicitly in the algebraic equations and
the entire system have to be solved simultaneously.

Table 2 presents the backward, forward or central finite-difference approximation for a

function f(x) with one independent variable. The three methods differs in how the tangent
at point B is represented (see figure 2.4). The backward and forward approximation
replace the tangent by the slope of line AB and BC respectively, while in the central
approximation the tangent is replaced by the slope of line AC. Obviously the central
finite-difference approximation is better (second order accurate) compared with the
backward and forward approximation (first order accurate).

Table 2: Finite-difference approximation

Backward Forward Central
df f ( x0 ) − f ( x0 − ∆x) df f ( x0 + ∆x) − f ( x0 ) df f ( x0 + ∆x) − f ( x0 − ∆x)
= = =
dx x = x0 ∆x dx x = x0 ∆x dx x = x0 ∆x

Figure 2.5: Finite-difference grid for one dimensional flow.

For river flow routing the finite-difference calculations are performed on a grid in the
plane x-t, the flow and water surface elevations are function of distance and time f(x,t)
With a grid interval of ∆x for the distance and ∆t for the time (see figure 2.5). Table 3
presents the spatial partial derivatives for the explicit and implicit finite-differences
schemes.
Table 3: Finite-difference approximation
Explicit finite-difference Implicit finite-difference
Backward ∂f f k − f i −1
k
∂f f k +1 − f i −k1+1
= i = i
∂x ∆x ∂x ∆x
Forward ∂f f − fik
k
∂f f − f i k +1
k +1
= i +1 = i +1
∂x ∆x ∂x ∆x
Central ∂f f − f i −k1
k
∂f f − f i −k1+1
k +1
= i +1 = i +1
∂x 2∆x ∂x 2 ∆x

One-dimensional hydrodynamic models

Examples of 1D hydrodynamic software models are the Mike 11 (DHI 2002), HEC-RAS
(HEC 2002), InfoWorks RS (ISIS), Sobek (Delf Hydraulics), TUFLOW, FLDWAV,
which can compute one-dimensional unsteady flow. All these models solve the Saint-
Venant equations for shallow water waves in open channels using a finite-difference
scheme (see table 4). These models require the river geometry (cross-sections
perpendicular to the flow), stream bed resistance factors (Manning coefficient) and time
series of flow discharge and stage height boundary conditions. For each grid point (cross-
section) an average water depth and velocity is calculated using a finite-difference
approximation.

Table 4: One-dimensional hydrodynamic models

Model Owner Commercial Numerical solution scheme

HEC-RAS U.S. Army Corps of No 4-point Box scheme

Engineers (Preissmann implicit scheme)
InfoWorks RS Yes 4-point Box scheme
(Preissmann implicit scheme)
Mike 11 DHI Yes 6-point Abbott-scheme
SobeK Delft Hydraulics Yes Delft-scheme
LISFLOOD-FP No explicit finite-difference
FLDWAV U.S. National No
Weather Service

A special 1D model is the raster-based storage cell codes, such as the LISFLOOD-FP
(Bates and De Roo, 2000). This kind of models take advantage of high resolution Digital
Elevation Models (DEM), which are easily available nowadays. The basic component is a
raster Digital Elevation Model with enough resolution and accuracy to identify river
channel (location and slope) and key elements of the floodplain topography (dykes,
embankments, depressions, roads). Channel flow is computed using a 1D approach, the
continuity equation relating flow into a cell and its change in volume and a momentum
equation for each direction where flow between cells is calculated according to
Manning’s law. The equations are solved using an explicit finite-difference discretisation.
The LISFLOOD-FP uses Manning's equation with explicit time stepping to calculate
flows between square cells, producing the simplest numerical representation of floodplain
flow. Channel flows are dealt with using a 1-D kinematic or diffusive wave
approximation coupled with the floodplain flows. An adaptive time step algorithm has
now been developed to ensure stability of the explicit scheme without use of flow
limiters.

2.3.4 Two dimensional flow

In recent years there has been a move from 1D to 2D models due to the increasing power
of computers, and the improved understanding of hydraulics process and their
representation in numerical models.

The shallow-water equations describe two-dimensional, depth-averaged unsteady flow

and are based on the conservation of mass and momentum. The equations assume that the
water is incompressible and that pressure distribution is hydrostatic. Density is constant
throughout the water column and vertical velocities are considered small relative to the
horizontal velocities.

The depth-averaged continuity and momentum and equations derived from the Navier-
Stokes equations (for an incompressible flow) are [M. Hanif Chaudhry 1993]:

Continuity equation

∂Z ∂ (u d ) ∂ (v d )
+ + =0 Eq. (4)
∂t ∂x ∂y

Momentum equation

∂ ∂ ∂  ∂d  g
(u d ) + (u 2 d ) + (u v d ) =  g x + g z d − 2 u u + v
2 2
Eq. (5)
∂t ∂x ∂t  ∂x  C

∂ ∂ ∂  ∂d  g
(v d ) + (u v d ) + (v 2 d ) =  g y + g z d − 2 v u 2 + v 2 Eq. (6)
∂t ∂x ∂t  ∂y  C

The equations represent the momentum with respect to a coordinate system x-y parallel
to the bottom of the channel.

u, v are the component of the velocity along the x and y directions,

u and v are the mean values of the u and v over the depth of the channel,
g= (gx, gy, gz)T is the gravitational force per unit mass,
C is the Chezy coefficient,
d is the water depth measured perpendicular to the bottom of the channel,
x, y
In very flat floodplains with complex topographic features due to the presence of
infrastructure, flood wave propagation is not an one-dimensional phenomenon. To
accurately capture the lateral flows, a two-dimensional modelling approach is required.

In two-dimensional modelling equations of continuity and momentum are written in two

dimensions and results are calculated at each grid point in the solution domain. Thus only
a finite spatial resolution dx can be used that makes computing slow and requires a lot of
computer memory.

Contrary to 1-D Modelling approach where results (h and Q) can only be obtained at
points where cross sections information is available, 2-D modelling results are available
at every grid point in the solution domain. Moreover discharges and velocities area
available in two dimensions (along the flow and in lateral direction).

The main advantage of using the 2D approach is that it provides information on variable
discharges and velocities in both x and y coordinates at each point at each computational
interval. The computation of velocity profiles in two dimensions allows the accurate
representation of flood wave propagation and better prediction of the effects of river
training, scouring and sediment transport processes.

The shallow-water equations are not amenable to direct solution and require that the
continuous derivatives in the equations be approximated by finite-difference techniques.
Finite-difference techniques solve the governing equations for a finite number of
locations and time. The techniques require the subdivision of the application domain into
a mesh or grid with a finite number of node points (see figure 2.6). The two-dimensional
partial differential equations can be solve using explicit and implicit finite-difference
methods as for one-dimensional problems.

Figure 2.6: Finite-difference grid for two dimensional flow.

The finite-difference methods can be used to solve one and two dimensional flow
problems.

The finite element method is an alternative method for solving the same differential
equations. For one-dimensional problem it does not have a significant advantage over the
finite-difference method. The main advantage of the finite-element method is its ability to
handle irregular boundaries and grid refinement (fig. 2.7).

Figure 2.7: Finite-element grid with one-dimensional elements.

Two-dimensional Hydrodynamic models

2D hydrodynamic models solve the Saint-Venant equations in two dimensions applying

finite-difference like Mike 21, or a finite element such as Telemac 2D. These models
need a continuous representation of the terrain topography and bathymetry of the riverbed
and floodplains, bed resistance and wind friction factors, which demands a high
computational time and cost. Table 4 presents some popular 2D hydrodynamic models.

Table 4: Two-dimensional hydrodynamic models

Model Owner Commercial Numerical solution scheme

Mike 21 DHI No Implicit finite-difference scheme

Telemac LNHE No Triangular finite-element
2D
TRIM2RD USG Yes Semi-implicit finite-difference
scheme
Mike Flood DHI No Implicit finite-difference scheme
FLO-2D No
TUFLOW Yes
Sobek Delf Hydraulics No

The hydrodynamic module of Mike 21 makes use of a so-called Alternating Direction

Implicit (ADI) technique to integrate the equations for mass and momentum conservation
in the space-time domain. The model uses a finite-difference scheme with a constant grid
spacings in the x and y directions (therefore the model are has to be rectangular). A
Double-Sweep (DS) algorithm is used to solve the equation matrices for each direction
and each individual grid line.

Telemac-2D was developed by the National Hydraulics and Environmental Laboratory

(Laboratoire National d’Hydraulique et Environnement - LNHE) of the Research and
Development Directorate of the French Electricity Board (EDF-DRD). Telemac-2D can
be used for free-surface maritime (the sizing of port structures, the study of the effects of
building submersible dikes or dredging, the impact of wasted discharged from a coastal
outfall) or river hydraulics (the impact of construction works: bridges, weirs, groynes;
dam breaks, flooding, the transport of decaying or non-decaying tracers). It solves the 2D
shallow water (St. Venant or depth averaged) equations of free surface flow using the
Galerkin’s method of weighted residuals over an unstructured mesh of triangular finite
elements. The main results are the depth of water and the depth averaged velocity
components.

TRIM2RD (Transient Inundation Model for Rivers – 2 Dimensional) solves the two-
dimensional depth-averaged flow equations. The model uses a semi-implicit, semi-
Lagrangian finite-difference method. The equations are spatially discretized as a
staggered rectangular mesh that consists of square computational cells with length x and
width y (x=y). The model can start computations from a “dry” bed and converge to
accurate solutions. Inflows are expressed as source terms, which limits the use of the
model to sufficiently long reaches where the flow reaches equilibrium with the channel.

Some models like FLO-2D, Mike Flood [DHI,2003], TUFLOW and Sobek (with the
Overland Flow module) integrate a 1D and a 2D approach. The 1D model can accurately
simulate water levels and discharge along the riverbed channel. In contrast, the 2D model
is used when overland flow occurs onto the floodplain.

2.4 Floodplain modelling

Floodplain modelling can be performed using a 1D or 2D approach or a coupling system
that integrates both techniques. The first, 1D simplified method for floodplain modelling
is considering the floodplain as part of the river cross section (figure 2.8a). In this case
there is the assumption that the water level is the same in the river and floodplain. As a
recommendation this approach should be used when the floodplain and the river are not
separated with dikes or embankments, and the floodplains cover a limited transverse
dimension, (floodplain width not larger than 3 times the width of the main river) Pender
G. et. al 2005. Under these conditions the floodplain flow is supposed to be one direction
parallel to the main river channel.

A second simplified approach to model floodplains is to consider the floodplain as a

storage reservoir (figure 2.8b). A relationship elevation-volume (or area) can be derived
using the topographical data (DEM) for each floodplain, and all these storage elements
can be connected to the 1D model of the main river. In this method the water surface
profile in the floodplain is no modelled. 1D hydronamic models like HEC-RAS,
InfoWorks and Mike 11 can implement storage elements.

Finally, a quasi or pseudo two-dimensional floodplain modelling can be performed, using

a 1D hydrodynamic model. Here, the floodplains are modelled as a network of river
branches and spills (figure 2.8c). The floodplains are connected to the main river by link
channels (Mike 11) or spill units (InfoWorks), and the discharge through these spills is
calculated using a weir equation.

Figure 2.8: One-dimensional floodplain modelling

Raster-based storage cell codes, such as the LISFLOOD-FP (Bates and De Roo, 2000)
and Delft FLS take advantage of high resolution Digital Elevation Models (DEM), which
are easily available nowadays. Channel flow is computed using a 1D approach, the
continuity equation relating flow into a cell and its change in volume and a momentum
equation for each direction where flow between cells is calculated according to
Manning’s law.

2D hydrodynamic models described in section 2.7 can be used for floodplain modelling.
The main constraints of 2D models are the high requirements of hardware, data and
computational time. However these models are more accurate for floodplain modelling,
because they conserve the momentum when the water overflows from the main river onto
the floodplain.

Since 1D models are accurate to simulate water level in the main river channel, some
packages are coupling 1D-2D hydrodynamic models. The 1D model is used to simulated
the water flow in the river, while the 2D model simulates the floodplain flow when the
water exceeds the dike or embankment. Using this method very complex floodplains can
be modelled, the model is more stable compared with the 2D model, and the
computational time is lower.

2.5 Flood mapping

Flood mapping has become a hot subject due this information is valuable to identify
highly risk areas for flood control planning, flood damage evaluation and alleviation,
floodplain management and development, and flood warning systems.

Basically there are two types of flood maps, based on historical information where
relevant information on floods is collected and plotted on maps, the experience shows
that historical flood data does not provide a full coverage of the flood limits and
sometime the data collection can be difficult to interpret. The second type of flood maps
area based on a consistent return period (synthetic events). Historical flood maps have the
advantage that flood envelopes are difficult to dispute and they are relatively easy to
prepare. On the other hand the probability of floods is not consistent (i.e. return period
varies), the coverage is incomplete, future floods can exceed limits by unknown amounts
and finally, they do not take into account physical changes after a flood event. In contrast,
flood maps based on a return period have a consistent approach with complete coverage
and the flood maps can be updated, nevertheless, they are more difficult to produce, the
results could be disputed, and flood limits could be exceeded [Ramsbottom, 2000].

Currently hydrological and hydraulic models are commonly used to simulate historical
(when data is available) or synthetic flood events. Model results, such as simulated water
levels together with a DEM can be used to delineate flood maps or potential flood risk
zones. This methodology can overcome the problem flood maps based on historical
information, since the historical flood extent can be validated with the flood extend based
on the model results. Once the river flood model has been calibrated and validated, it can
be used to produce flood maps based on synthetic events for different return periods.

In the following sections some approaches to integrate models results and GIS,
specifically for flood mapping applications are presented.
2.6 GIS integration with hydraulic/hydrological models
The arrival of GIS technology in the late 1960’s had allowed the access to large data sets
as well as a consistent visualization environment for displaying model data and results.
Complex models where data manage and preparation can be cumbersome could take
advantage of GIS to overcome this limitation and concentrate only in the modelling effort
[Karimi et al., 1996]. At the beginning, GIS and environmental modelling (e.g.
hydraulic/hydrological) developed in parallel. Environmental modelling is intrinsically
spatial, hence, spatial variability has to be taken into account. To achieve a efficient
integration the most important factor is to understand the goals of the environmental
model.

The main constrain to using GIS in modelling is the difficulty of integrating the model
with GIS. As stated by Bian (2004) it should not be only a link between a GIS package
and a model, or the extraction of values from a GIS layer to feed them into a model. But
think of environment in terms of its perception and representation, the representation of
both its form and function, and the representation of both spatial variation and temporal
change of these forms and functions.

Nowadays it is a useful tool for hydraulic/hydrological engineering, allowing for

example, to determine flooded areas or areas at risk of been flooded along the river
floodplains.

In a broad sense integration can be implemented for preprocessing data into a specific
format for analysis, direct support for modelling directly within the GIS package, and as a
post-processing data tool for reformatting, reporting and mapping [Zerger and Wealands,
2004]

Different approaches of integration between GIS and environmental modelling,

specifically for hydrological and hydraulic modelling, have been developed [Sui and
Maggio, 1999], [Zerger and Wealands, 2004], [Nunes Correia et al., 1998] and they will
described in the following sections.

2.6.1 Embedding GIS in hydraulic/hydrological models

In this type of integration GIS functionalities are embed in hydrological/hydraulic

modelling packages and in essence the GIS is used as a mapping tool. The advantage of
this implementation that is not constrained by any existing GIS data structures. On the
other hand, the data management and visulization capabilities of these software packages
are in no way comparable to those available in commercial GIS software packages.
Examples of this approach are River CAD, HEC-RAS, and Modflow.

The 1D model InfoWorks is an example of embedding GIS into a hydraulic model where
all the functions of a GIS environ are implemented (preprocessing, modelling and post-
processing). The hydraulic modelling is performed simultaneously with the flood
mapping, due to the direct implementation of the DEM in the model. The main benefit is
that all the tasks are done in one package, in contrast, it is more time demanding, because
flood maps are generated for every time step in the simulation.

Because of their complexity 2D models, such as MIKE 21 and Telemac-2D, also prefer
to embed a GIS interface into the hydrodynamic model. A loose coupling approach
would become too tedious and awkward in this case to manage large amounts of data
between the model and the GIS. These models employ their own grid formats for the
DEM and flood maps.

2.6.2 Embedding hydraulic/hydrological models into GIS

Because the popularity and worldwide application of hydraulic/hydrological models

several developers of GIS packages have included stand-alone GIS modules with
functions that can be used for a variety of hydraulic/hydrological needs. For instance
certain hydrological modelling functions have been embedded in ESRI’s ArcStorm and
ArcGrid and Intergraph’s in Roads. This approach takes full advantage of built-in GIS
functionalities, but modelling capabilities are usually simplistic and calibration must take
place outside of the package.

2.6.3 Loose coupling

Loose coupling involves a standard GIS package and hydraulic/hydrological modelling
programs or a statistical package. The integration between the different software is
carried out via data exchange using either ASCII or binary data format, consequently they
do not have a common interface. This methodology allows minimal computer
programming, however data conversion between different packages can be tedious and
error prone. Loose coupling has been the basis for building GIS-based decision support
systems, providing integrated access to otherwise incompatible analytical tools. In
practice the approach may be the most realistic method for most GIS and
hydraulic/hydrologic engineers.

Usually a loose coupling technique is used when the floodplains are modelled using a 1D
hydrodynamic approach. Examples of this approach are the packages HEC-GeoRAS
[US-ACE, 2002] and MIKE/GIS [DHI, 2001], both models are implemented in ArcView
3.2 and they are a common interface to link model results and the GIS software using
ASCII or binary files. The model results (discharges and water levels) of the hydraulics
models HEC-RAS and MIKE 11 respectively are imported into the GIS where the flood
maps are determined by the interception between interpolated water levels with the DEM
for specific time moments selected by the user. For the analysis HEC-RAS make use of a
TIN derived-DEM, while MIKE/GIS uses a gridded elevation data, however, flood maps
are derived in grid format, allowing the calculation the area .and volume flooded.

2.6.4 Tightly coupling

Tightly coupled models are built-up completely within a GIS package and are useful of
modellers depending on the analytical functions of the GIS. The coupling is via either
GIS macro or conventional programming. GIS software vendors are providing macro and
script programming capabilities so that users can develop customized applications. Such
languages are seldom powerful enough to implement sophisticated models. They provide
direct access to spatial databases, fully integration of model data with other project data
and the model results can be visualised immediately. Nevertheless, this approach requires
a well defined interface to the data structure held by the GIS.

References

Abbot M.B., Bathurst J.C., Cunge J.A., O’Connel P.E. and Rasmussen J., 1986, An
introduction to the European Hydrological System - Systéme Hydrologique Européen.
SHE. II. Structure of a physically-based, distributed modelling system, Journal of
Hydrology, vol. 87, pp.45-5.

Bates, P. D. and De Roo, A. P. J., 2000, A simple raster-based model for floodplain
inundation, Journal of Hydrology, Vol. 236, pp. 54-77.

Beven K.J. , 1996, Toward a generalization of the TOPMODEL concepts: Topographic

indices of hydrological similarity, Water Resources Research, Vol.32, pp. 2135-2145.

Beven K.J., 2000, Rainfall-runoff modelling, The Primer, Wiley, 360 p.

Bian L., 2004, GIS for environmental modeling: an introduction, editorial, Computers,
Environment and Urban Systems, Vol. 28, pp. 171-173.

Chow V.T., Maidment D.R., Mays L.W., 1994, Applied Hydrology, McGraw-Hill, New
York, 584 p.

Croke G.B.F:, Andrews W., Jakeman A.J., Cuddy S. and Luddy A., 2005, Redesign of
the IHACRES Rainfall-Runoff model, 29th Hydrology and Water Resources
Symposium, 21-23 February, Camberra, 7p.

Cunderlik J.M., 2003. Hydrologic model selection for the CFCAS project: Assessment of
Water Resources Risk and Vulnerability to Changing Climate Conditions, Project
Report, pp.

DHI, 2001, MIKE 11 GIS: Floodplain mapping and analysis, User guide, Danish
Hydraulic Institute, Hørsholm, Denmark, 147 p.

DHI, 2002, MIKE 11: A modelling system for rivers and channels, User manual, Danish
Hydraulic Institute, Hørsholm, Denmark, 396 p.

DHI, 2003, Mike Flood 1D-2D Modelling, User manual, Danish Hydraulic Institute,
Hørsholm, Denmark, 16p.
DHI, 2004, MIKE SHE, User guide, Danish Hydraulic Institute, Hørsholm, Denmark,
XXX p.

Karimi H.A. , 1996, Evaluating strategies for integrating environmental model with GIS,
Computers, Environment and Urban Systems, Vol. 20, No. 6, pp. 413-425.

Neitsch S.L., Arnold J.G., Kiniry J.R., Srinivasan R. and Williams J.R., 2002, Soil and
Water Assessment Tool, User’s Manual, Texas Water Resources Institute, College
Station, TWRI report TR-191, 506 p.

Nunes Correia F., Castro Rego F., Da Graça Saraiva M. and Ramos I., 1998, Coupling
GIS with Hydrologic and Hydraulic Flood Modelling, Water Resources Management,
Vol. 12, pp. 229-249

Pender G., Bates P. B., Wright N. G., Hall J. and Mason D. 2004, Next Generation
Computer Modelling for the Prediction of Flood Level and Inundation Extent Part 1:
Technical Specification of Research Needs.

Refsgaard J.C. and Kundsen J., 1996, Operational validation and intercomparison of
different types of hydrological model, Water Resources Research, Vol. 31, p. 2189-
2202.

Ramsbottom D., 2000, A Suggested Approach to Flood Mapping (based on experience in

England and Wales), Proceedings of the National Hydrology Seminar, England, pp. 60-
70.

Sui D.Z. and Maggio R.CV., 1999, GIS integration with hydrological modelling,
Computers, Environment and Urban Systems, pp:

US-ACE, 1993, River hydraulics, Engineer Manual, U.S. Army Corps of Engineers, pp.
XXX.

US-ACE, 1994, Flood-Runoff Analysis, Engineer Manual, U.S. Army Corps of

Engineers, pp. XXX

US-ACE, 2001, Hydrologic Modeling System HEC-HMS, User’s Manual, US Army

Crops of Engineers, Hydrologic Engineer Center, 178p.

US-ACE, 2002, HEC-GeoRAS 3.1, User Manual, U.S. Army Corps of Engineers, pp.
XXX.

Zerger A. and Wealands S., 2004, Beyond Modelling: Linking Models with GIS for
Flood Risk Management, Natural Hazards, Vol. 33, pp. 191-208.
Wheater H.S., 2002, Progress in and prospects for fluvial flood modelling, Phil. Trans. R.
Soc. London, Vol. 360, pp. 1409-1431.