Hydrol. Earth Syst. Sci.

, 15, 27472761, 2011

www.hydrol-earth-syst-sci.net/15/2747/2011/ Hydrology and
doi:10.5194/hess-15-2747-2011 Earth System
Author(s) 2011. CC Attribution 3.0 License. Sciences

A simple 2-D inundation model for incorporating flood damage in

urban drainage planning
A. Pathirana1 , S. Tsegaye2 , B. Gersonius1 , and K. Vairavamoorthy2
1 UNESCO-IHE Institute for Water Education, P.O. Box 3015, 2601 DA Delft, The Netherlands
2 School
of Global Sustainability, University of South Florida, 4202 E. Fowler Avenue, CGS101,
Tampa, Florida 33620-8100, USA
Received: 1 September 2008 Published in Hydrol. Earth Syst. Sci. Discuss.: 14 November 2008
Revised: 22 August 2011 Accepted: 22 August 2011 Published: 31 August 2011

Abstract. An urban inundation model was developed and 1 Introduction

coupled with 1-D drainage network model (EPA-SWMM5).
The objective was to achieve a 1-D/2-D coupled model that
is simple and fast enough to be consistently used in plan-
ning stages of urban drainage projects. The 2-D inundation
model is based on a non-standard simplification of the shal-
low water equation, lays between diffusion-wave and full dy-
namic models. Simplifications were made in the process rep-
resentation and numerical solving mechanisms and a depth
scaled Manning coefficient was introduced to achieve sta-
bility in the cell wetting-drying process. The 2-D model is
coupled with SWMM for simulation of both network flow
and surcharge induced inundation. The coupling is archived
by mass transfer from the network system to the 2-D system.
A damage calculation block is integrated within the model
code for assessing flood damage costs in optimal planning
of urban drainage networks. The model is stable in deal-
ing with complex flow conditions, and cell wetting/drying
processes, as demonstrated by a number of idealised exper-
iments. The model application is demonstrated by applying
to a case study in Brazil.
Correspondence to: A. Pathirana
(a.pathirana@unesco-ihe.org)
Considering the large spatial variability and dynamic nature
of urban floods due to sewer overflow, reliable assessment of
their impacts demand spatially distributed inundation mod-
els. There are numerous modelling products that can be used
for the purpose. However, most of the available models tend
to be complex both in terms of setting up and data prepa-
ration and are computationally expensive. Probably due to
these facts, employment of inundation models for optimal
planning of urban drainage systems by balancing costs and
benefits, is not widely practiced by the professional commu-
nity. For this purpose, arguably, a simple and quick predic-
tion of flood within an acceptable limit of accuracy is much
more important than going for an accurate but complex and
time consuming hydrodynamic simulation.
This work presents the development of a simple 2-D inun-
dation model for quick prediction of flood plain hydraulics
in urban area. The main focus in the model development was
to archive simplicity and speed, within an acceptable level
of accuracy. The inundation model was then coupled with
a drainage network model, US EPAs Storm Water Manage-
ment Model (SWMM 5.0) (Rossman, 2004). The coupled
model result can be used for the assessment of flood damage
costs in optimal planning of urban drainage networks.
The next section describes the 2-D model algorithm and
its numerical implementation. Then the coupling of the 2-D
model with SWMM is explained. The performance of the
coupled model is examined for several hypothetic conditions
and its application is demonstrated using a simplified case
study in Brazil.

Published by Copernicus Publications on behalf of the European Geosciences Union.

2748 A. Pathirana et al.: A simple inundation model for urban drainage

2 2-D inundation model code development

   
di1,j +di,j di,j +di+1,j
The inundation model is based on shallow water equations un+1
i1,j  21x un+1
(SWE) that describe depth averaged 2-D flow. The assump-
n+1
hi,j = 1t  21x  i,j (4)
di,j 1 +di,j n+1 di,j +di,j +1 n+1
tion of a 2-D flow over inundated plain as slow, shallow + 21y vi,j 1 21y vi,j
phenomena can reduce the reproduction of flood plain hy-
draulics to the minimum necessary level to achieve accept-
able predictions. This may lead to local inaccuracies but it +q1t + hni,j
has been shown that the uncertainties of data (e.g. the spec-
ification of topography and boundary roughness) are dom- where 1t is finite difference time step, 1x and 1y are regu-
inant and thus influence the model results to a greater ex- lar grid spacings in the x- and y-directions, respectively, n is
tent than those incurred through simplified (depth-averaged) time step counter and (i,j ) is grid counter.
mathematics (Hunter et al., 2007). Using Eqs. (2) and (3) rate of change of velocities in x-
and y-directions are written as follows:
2.1 Model formulation
 
ui,j hi,j hi+1,j hi,j
= g gSf x = g gSf x (5)
Simplified inundation models relay on the kinematics wave t x 1x
approximation which neglect the local acceleration, convec-
tive acceleration and pressure term in the shallow water equa- vi,j hi,j

hi,j +1 hi,j

tion. The non-inertia or diffusion wave model neglects the lo- = g gSf y = g gSf y (6)
t y 1y
cal and convective acceleration but considers the back water
effect. However in this paper, based on the assumption that
the velocity of the water flow in urban flood plains is small Further simplification of Eq. (5) and (6) is done by repre-
compared the other terms, the convective acceleration term senting g1tSf in term of the dependent variable- velocity
in the momentum equation is ignored from the full dynamic component (let g1tSf x = uSx ). The velocities at time step
wave equation. So the flood flow representation is somewhat n + 1 in x and y directions are then written as:
unique as it lays between diffusion wave and full dynamic
equation. The simplified shallow water flow governing equa- g1t  n+1 
un+1
i,j = hi+1,j hn+1
i,j uni,j (1 Sx ) (7)
tions shown below are derived from the original equations 1x
(Chow, 1964).
n+1 g1t  n+1 
  vi,j = hi,j +1 hn+1
i,j
n
vi,j (1 Sy ) (8)
Continuity equation............ h u v
t + d x + y = q (1) 1y

Momentum along x-dir........ u h
t + g x + gSf x = 0 (2) 2 2 2 2 2 2
where Sx = g1tn d 4/3u +v ; Sy = g1tn d 4/3u +v .
Momentum along y-dir....... v h
t + g y + gSf y = 0 (3)
Equations (4), (7) and (8) can be combined to determine
the water stage above the datum.
where h is water stage over the datum, d is the depth
of flow, q is source or sink per unit area, t is flow g1t
di1,j +di,j hn+1

g1t
n+1
21x 2 i,j + 21x2 di1,j +di,j hi1,j
time, u and v are velocities along x- and y-direction, 
d +di,j

+ i1,j uni1,j (1Sx )

respectively, Sf x and Sf y are friction slopes along x-
21x
and y-direction, respectively. The friction terms are es-
g1t
n+1 g1t
n+1

+
21x2 di,j +di+1,j hi+1,j 21x 2 di,j +di+1,j hi,j


timated by using the empiricalresistance relationship in 
di,j +di+1,j

n

ui,j (1Sx )

Mannings equation: Sf x = (n2 u u2 + v 2 )/d 4/3 ; and Sf y = 21x

hn+1 =1t

i,j g1t

n+1 g1t
n+1
di,j +di,j 1 hi,j + 21y2 di,j +di,j 1 hi,j 1

(n2 v u2 + v 2 )/d 4/3 . where n is the Mannings roughness

21y2
 
di,j +di,j 1
coefficient. n


vi,j 1 1Sy

+ 21y


+ g1t di,j +1 +di,j hn+1


g1t
d +d

n+1
h
21y2 i,j +1 21y2 i,j +1 i,j i,j
2.1.1 Numerical model 
di,j +1 +di,j

n


21y vi,j 1Sy

Finite difference (FD) method is employed for solving the

flow equations. The following equation is obtained using
temporal and spatial discretisation of simplified continuity +hni,j +q1t
equation (Eq. 1) for depth averaged shallow water.
A simplified form of the above equation is as follows:

Hydrol. Earth Syst. Sci., 15, 27472761, 2011 www.hydrol-earth-syst-sci.net/15/2747/2011/

 Y-direction
-direction
A. Pathirana et al.: A simple inundation model for urban drainage 2749
routing
  
di1,j + 2di,j + di+1,j
g1t 2
21x2
  
di,j 1 + 2di,j + di,j +1
+ + 1 hn+1
i,j
21y2
  t- axis
g1t n+1 di1,j +di,j
21x 2 (di1,j + di,j )hi1,j + 21x uni1,j (1 Sx )
 
g1t di,j +di+1,j
+ 21x2 (di,j + di+1,j )hn+1
i+1,j  uni,j (1 Sx )

21x (9)
= 1t g1t

di,j +di,j 1

+ 21y2 (di,j + di,j 1 )hn+1 + v n (1 S )

i,j 1  21y  i,j 1
y
t n+1
g1t n+1 d i,j +1 +d i,j n
+ 21y2 (di,j +1 + di,j )hi,j +1 21y vi,j (1 Sy )

+hni,j + q1t
t n+1/2
This equation leads to a set of tridiagonal system of linear x-axis X-direction Routing
algebraic equations which can be solved by forward sweep
and a backward substitution. (See Sect. 2.1.2) Fig. 1. An alternating direction flood routing during each half time
step.
2.1.2 Alternating direction implicit (ADI) finite
difference method
where
Implicit solving of shallow water equations leads to solving
large sets of partial differential equations which is computa- g1t 2

aa = 21x 2 di1,j + di,j
tionally expensive. In the case of fluvial flooding, the com- 
d +2d +d

putational load of the implicit solutions is justified due to the bb = g1t 2 i1,j i,j
21x2
i+1,j

possible offsetting by longer time steps. However, in case d +2di,j +di,j +1
+ i,j 1 21y 2 +1
of small scale urban flooding (the subject of this paper), the g1t 2

modeller needs information at relatively finer time-resolution cc = 21x2 di,j + di+1,j
(due to the swift nature of urban floods), and therefore ex-
plicit approaches could often be more beneficial. In this pa- 
di1,j +di,j
 
d +d

per, an alternating direction implicit finite difference (ADI) 21x uni1,j (1 Sx ) i,j 21xi+1,j uni,j (1 Sx )
 

g1t
n di,j +di,j 1 n
procedure is used to solve the governing equations. In spite dd = 1t + 21y 2 di,j + di,j 1 hi,j 1 + vi,j 1 Sy

 21y  1
g1t di,j +1 +di,j

of the postfix implicit the ADI method behaves largely as

n n
+ 21y2 di,j +1 + di,j hi,j +1 21y vi,j 1 Sy
an explicit method (due to implicit formulation only along
one dimension at a time in the 2-D flood plain) and provide
all the said advantages of the explicit approach. Of course +hni,j + q1t
this comes with the usual drawback of explicit approach as
well, namely the limitation of time-step size imposed by a Along the y-direction
CFL condition (Caviglia and Dragani, 1996). According to
Peaceman and Rachford (1955) ADI method provides high From time step n+1/2 to n+1 the procedure is reversed so
computational efficiency which requires less computing time that the value of h along y-direction is determined implic-
because it involves a tridiagonal matrix. The method is based itly while the other terms are expressed explicitly and then
on splitting the differential equation in two parts and solves the water depths are updated from the calculated h values.
them sequentially in x- and y-directions within two half-time Analogues to the x-direction, in case of the y-direction the
steps t n+1/2 and t n+1 , respectively. Simple diagrammatical values for hn+1 n+1
i+1,j and hi1,j are approximated from the previ-
n+1/2 n+1/2
representation of ADI routing procedure for one full time ous half time step values hi+1,j and hi1,j , respectively and
step is shown in Fig. 1. Eq. (9) can be converted to linear algebraic equation shown
below:
Along the x-direction
aa hn+1 n+1 n+1
i,j 1 + bbhi,j + cchi,j +1 = dd (11)
From time step n to n + 1/2 the solution for water stage
hn+1/2 along the x-axis is determined implicitly within a row Where
while the value of other terms are expressed explicitly.
g1t 2
Thus, the values for hn+1 n+1

i,j +1 and hi,j 1 are approximated aa = 21y 2 di,j + di,j 1
from the previous time step values hni,j +1 and hni,j 1 respec-
 
di1,j +2di,j +di+1,j
bb = g1t 2 21x2
tively. Then Eq. (9) can be further simplified to a linear alge- 
di,j 1 +2di,j +di,j +1
braic equation shown below: + 21y2
+1
n+1/2 n+1/2 n+1/2 g1t 2

aa hi1,j + bbhi,j + cchi+1,j = dd (10) cc = 21y2 di,j +1 + di,j

www.hydrol-earth-syst-sci.net/15/2747/2011/ Hydrol. Earth Syst. Sci., 15, 27472761, 2011

2750 A. Pathirana et al.: A simple inundation model for urban drainage
 
g1t
n+1/2 d +di,j
d + di,j hi1,j + i1,j uni1,j (1 Sx ) boundary condition, depth scaled factor for the Mannings
21x2 i1,j  21x 
g1t n+1/2 d +d
di,j + di+1,j hi+1,j i,j 21xi+1,j uni,j (1 Sx )


dd = 1t + 21x

coefficient, nmodified =n(dmin /d) is used to represent the real-
 2   
+
di,j +di,j 1
21y
n
vi,j


1 1 Sy
d i,j +1
21y
+d i,j n 1S
vi,j y


ity of drying process in flood flow by slowing sudden drying
of wet cells. The modification of Mannings coefficient is
n+1/2
done for wet-cells with water depth below dmin and suddenly
+hi,j + q1t dries within a time step. The proposed wetting/drying pro-
cess representations avoid frequent instability in model sim-
Once the value of h is determined, the velocity components
ulation.
along the x- and y-directions will be updated by implement-
ing Eqs. (7) and (8), respectively.
2.4 Process representation for flow barrier
2.2 Courant condition
Features like roads, buildings and dykes have great effect on
flow dynamics and flood propagation. These barriers cause
According to Caviglia and Dragani (1996) the ADI numeri-
numerical instability in most of the hydrodynamic models. In
cal scheme is not unconditionally stable when applied to the
this model algorithm, the flow interaction near topographic
complete set of governing equations in an area with real to-
obstacles (barriers) is treated in such a way that an obsta-
pography and irregular boundaries. The instability arises be-
cle cell will not contribute to the velocity calculation of the
cause the routing method uses implicit values along one di-
nearest lower cell till it acquires a threshold depth of water
rection while the values in the orthogonal direction are deter-
d=0.01 m.
mined explicitly from the previous time step. Thus courant
In addition, a realistic representation of flow near high bar-
limitation in time step 1t is implemented to ensure the nec-
riers is achieved by modifying coefficients of the linear al-
essary stability during simulation.
gebraic equations (Eqs. 10 and 11) for the cells under con-
 
|u| + gd |v| + gd sideration. The water stage of high barrier neighbouring is
+ 1t 1 (12) replaced with water stage value of itself ( cell under consid-
1x 1y
eration) to avoid the numerical instability of the model, and
where 1t is finite difference time step, 1x and 1y are regu- to avoid spurious flows from barriers cells to any neighbour-
lar grid spacings in the x- and y-directions, respectively, and ing cells.
d is the water depth.
In this inundation model the time step is defined as an in- 2.5 Flood damage analysis block
put parameter and it has to be less than the courant 1t for
each routing step. During each step of routing the minimum
Inundation models could help decision makers to have a good
courant 1t of all cells in the 2-D domain, is calculated using
perception about the flood extent and to identify the effec-
Eq. (12) and used for verifying the input 1t value.
tive flood mitigation measures through flood risk assessment.
Providing an elementary means of estimating the loss due to
2.3 Process representation for wetting/drying cells flood is important for this purpose. A simple damage calcu-
lation block is integrated within the inundation model code.
In this model the hydraulic parameters are represented for The maximum flood depth at each grid cell is used for the
the mid-point of each square cell. However partially wet- calculation of the damage in urban areas. The velocity of the
ted cells i.e. cells without enough water to submerge all cor- water flow in urban flood plains is very small and its contri-
ners of the cell, the average depth are badly represented by bution is neglected in the calculation of flood damages. A
depth at the centroid (Begnudelli and Sanders, 2006). Espe- regression equation for the possible flood damage curves is
cially for a cell having very small water depth, the incidence implemented in the model code. This equation is shown be-
of dry/wet condition will be very high and would frequently low.
lead to model instability. In order to avoid this problem, we
suggest a boundary depth value to classify wet and dry cells Damage = A + B ln(d) + Cd + Dd 2 + Ed 3
and depth factored Mannings coefficient for sudden drying
wet cells. where d is the flood depth and A to E are consecutive coef-
Due to the fact that friction terms are factored by water ficients for the possible flood damage curves. Typical stage-
depth, cells with very small depth value possibly create nu- damage curves can be represented by the input of suitable
merical instability of the model. Thus, threshold value of values for the coefficients. The damage analysis is performed
depth dmin =0.0001 m is proposed for classifying submerged if an input for the damage calculation block is provided, if
cells and updating hydraulic parameters. During each full not, the model skips the damage calculation part.
time step the water depth is checked and the velocity and
friction terms of those cells with depth less than dmin are set
to zero value to avoid instability. In addition to the above

Hydrol. Earth Syst. Sci., 15, 27472761, 2011 www.hydrol-earth-syst-sci.net/15/2747/2011/

 Duration of simulation
Overflow volume
Time
A. Pathirana et al.: A simple inundation model of occurrence
for urban drainage and duration of flooding 2751

3 Model coupling

Coupling which supports bi-directional interactions between

sewer flow and 2-D inundation models is optimal to simulate
sewer flooding. However, we assumed that the implementa-
tion of back flow into the sewer system is not essential, par- D ra in a g e n e tw o rk
ticularly for urban flood damage estimation work, where the
prime concern is the maximum flood depth (not the duration
as in the case, for example agricultural land). In swift urban
floods, the maximum flood height is not heavily influenced
by the ability of sewer network to receive return flow. Thus,
coupling is achieved by mass transfer from the drainage net- F lo o d e d A re a
work system to 2-D system.
There are different frameworks that can be used for modelFig. 2. Interacting physical systems whichsystems
can be simulated by 1-D2-D coupled
Fig. 2. Interacting physical which can be simulated byflood
1- model.
coupling. According to Bulatewicz Figure 3:1 Interacting
Jr. (2006), they can physical systems
D2-D coupled flood model.
which can be simulated by 1D-2D coup
be organised into four different categories of coupling ap-
model
proaches: monolithic, scheduled, communication, and com-
ponent. In this paper, a unidirectional communication is used
for coupling the 2-D-Inundation model Thewithfollowing
1-D-SWMM.functions are Onedevised
of the coupling functions receives
for developing overflow infor-between v
a relationship
mation and transforms it into a flow rate per unit grid
The unique features of this approach are: 3084
incorporated in the main simulating cell area withinengine of the 2D model.
the computational domain.
Model codes remain independently executing programs Because of the coarser time step in SWMM, the inunda-
that interact only by exchanging data via message pass- tion model steps a number of times for a single SWMM
ing during execution. report time step.
It allows communication libraries that support direct 3.2 Coupled model algorithm
model-to-model communication as well as model-to-
coupler communication. The model algorithm, which is used for surface flow routing
and flood damage calculation, is written in C/C++ program-
It supports existing models to be coupled with minimal ming language, the choice of language largely influenced by
changes to the model source codes. the fact that SWMM code is available in C language. The
simplified algorithm is shown in Fig. 3.
3.1 Creating a coupled model

The sewer overflow is the main interacting physical process
between the two selected models. This physical process usu-
ally occurs at the point where outfall and surcharged man-
holes are located. The features which are integrated in the in-
teracting physical process between 1-D and 2-D system are:
Routing and reporting time step length
3.3 Initial and boundary conditions
The 2-D model allows setting up of several different bound-
ary conditions (e.g. Free flow, wall, specified velocities) and
initial condition (initially dry, specified flood level). How-
ever, most common initial condition for sewer overflow situ-
ation is that of initially dry condition.

Duration of simulation
Overflow volume at each time step
Time of occurrence and duration of flooding
The following functions are devised for developing a rela-
tionship between variables and incorporated in the main sim-
ulating engine of the 2-D model.
The initialisation of both models takes place at the
start of simulation. The inundation model starts rout-
ing when the manholes begin overflowing. Delaying
the flood routing process increases the efficiency of this
coupled model.
4 Idealised experiments
The model was subjected to a number of tests using idealized
situations. The flow process representations and simplifica-
tions used in this model are checked against mass conserva-
tion, stability and stationary of the result.
Several idealised test problems were utilised to examine
the hydrodynamic behaviour, such as boundary conditions,
2-D flood wave diffusion, cell drying and wetting, and flow
interaction with topographic obstacles. Hypothetical exam-
ples are used for quick visual inspection of the models sen-
sitivity and its performance against the simplifications made
in the model code.

www.hydrol-earth-syst-sci.net/15/2747/2011/ Hydrol. Earth Syst. Sci., 15, 27472761, 2011

2752
Several hypothetical tests were conducted for checking the
boundary condition and the model result confirms that the
numerical representation works properly by preserving the
continuity and stability at boundaries.
4.1 Model performance against mass balance error
The presence of mass-balance errors in flood flow is a known
problem in non-conservative numerical schemes (as opposed
to conservative methods like Finite Volume approach) and
improving the accuracy and mass-balance properties of fluid
flow in general is an active area of research (Kippe et al.,
2007). In this model the percentage of mass balance error is
checked at different instants of simulation.


massbal= Twdepth Tsource /Tsource 100
where massbal is the mass balance error in percentage ,
Tsource is the total input volume and Twdepth is the total vol-
ume on 2-D grids at same time of simulation.
The mass-balance check for different hypothetical terrains
is done at different flood routing times and model outputs
for flat, sloping and complex topographies are illustrated in
Fig. 4, 5 and 6. The result shows the error increase with
increasing complexity of the terrain (range up to 3.35 % for
a very irregular topography). Inaccuracies associated with
mass balance error are inevitable and the model accuracy is
expected to decrease with increasing simplification in model
formulation and numerical solving mechanism, with increas-
ing time step of model integration and with increasing grid
size.
4.2 Model performance in wetting and drying problems
The model was examined over hypothetical topography with
the various flow conditions that may occur in actual flood-
plains and shown to be capable to simulate wetting and dry-
ing processes that will occur as the flood flows over an ur-
ban area. This is achieved by using a threshold value for
water depth and depth scaled Mannings coefficient for wet-
ting/drying process representation (see Sect. 2.3 ). This rep-
resentation has the following advantages:
Avoids possible numerical instabilities due to very small
water depth values.
An implicit representation of the reality of high fric-
tion at very small water depths (boundary layer, laminar
flow).
Avoid sudden drying of wet cells.
4.3 Model performance in uneven topography
including flow barriers
Flow barriers are often challenges for numerical stability.
The inundation maps for a hypothetical terrain involving flow
Hydrol. Earth Syst. Sci., 15, 27472761, 2011
A. Pathirana et al.: A simple inundation model for urban drainage
barriers are presented in Fig. 8 and 9. The model perfor-
mance is checked on hypothetical irregular topographies. For
problems involving large flow barriers, the flood wave diffu-
sion is visually inspected for synthetic terrains (e.g. the one
shown in Fig. 7) and the probable mass balanced errors were
examined at cell level. Close attention is paid for the be-
haviour of flood wave near barriers and other sudden changes
of topography (e.g. falls, walls) and hydraulics (e.g. wet-
ting/drying). The model works properly with global mass
balance error for irregular topography (Fig. 6) of less than
3.4 percent. The result shows that the model performs well
on complex topographies and frequent wetting and drying.
Frequent numerical instabilities are avoided by the simplified
flood flow representations discussed in Sects. 2.3 and 2.4.
5 Case Study in Porto Alegre, Brazil
Porto Alegre is the capital city of Brazils southernmost state,
Rio Grande do Sul. Porto Alegre itself has a population of
1.3 million inhabitants. It covers an area of 470 square kilo-
metres, 40 per cent of which is urban and 60 per cent rural.
The Areia basin is located in the north of the city. It covers
an area of 21 km2 , of which approximately half corresponds
to the basin of Arroio da Areia and the rest belongs to the
Airport polder.
The main drainage network runs below the street grid,
except for a small part that intersects a housing block and
passes under the foundations. The drainage system in the
basin of Arroio da Areia can be divided into two distinct sys-
tems: one drained by gravity and the other by the pumping
station Silvio Brum. The areas with a level above 8,13 m are
drained by closed conduits, while the pumping station drains
an area of 139,2 ha that is below the level of 8,13 m. In the
end the drainage from the upstream basin flows inside a pres-
sured pipe.
The actual capacity of the urban drainage in some parts
of the Areia basin is not enough to discharge the upstream
increase in flood peak and volume as a result of the urban-
isation process. The heaviest inundations happen on the in-
tersection of the roads Nilo Peanha and Texeira. On this
point that is the lowest of the region the drainage system,
which transports the water to the Arroio the Areia, over-
flows and inundation levels can reach one metre. Previous
floods had resulted in damage to property and even in the loss
of life. It is this general area that was used for the demonstra-
tion here. (Fig. 10)
5.1 Input data for the coupled model
Delineation of the model domain for 2-D computation can be
challenging because the modeller doesnt know the region of
inundation before execution (Begnudelli and Sanders, 2007).
The approach adopted here was to delineate a boundary sig-
nificantly far from the flooded manholes. Standard Shuttle
www.hydrol-earth-syst-sci.net/15/2747/2011/
 Coupled Model Algorithm
e modelA.algorithm, which is used for surface flow routing and flood damage 2753
Pathirana et al.: A simple inundation model for urban drainage
calculation
written in C++ computer programming language. The simplified algorithm is shown i
ure 3:2:
Input Data Boundary & Initial
(2D) Condition (2D)

Input Data
(SWMM) Open SWMM

Start SWMM

Step SWMM
t s = t sn + 1

No ts tr
Yes

No Overflow>0

Yes

X-Flood Routing

t 2 D = t 2n D+ 1
Y-Flood Routing

No
t2D tr
Yes

No SWMM last Stop Damage No

time step ? SWMM Curve ?
Yes
Yes
Calculate
Damage Cost

Print values of d, u, v and

flood extent, damage cost(?)

Stop
2D

Fig. 3. Algorithm showing computation sequences of the 1-D/2-D coupled model (where t2D is inundation model time step, tr and ts are the
Algorithm showing computation sequences of the 1-D2-D coupled model (Wher
SWMM reporting and routing step, respectively).

dation model time step, tr and ts are the SWMM reporting and routing step).

www.hydrol-earth-syst-sci.net/15/2747/2011/ Hydrol. Earth Syst. Sci., 15, 27472761, 2011

1
3085
 2754 A. Pathirana et al.: A simple inundation model for urban drainage

Legend
Depth (m)
0
0.05
0.1
0.15

Time step (t) 0.8 sec Time step (t) 0.8 sec
x & y 30m x & y 30m
Time elapsed 60 min Time elapsed 70 min
Mass bal. error -0.3098% Mass bal. error -0.6605%

Time step (t) 0.8 sec Time step (t) 0.8 sec
x & y 30m x & y 30m
Time elapsed 80 min Time elapsed 90 min
Mass bal. error -0.9153% Mass bal. error -1.0464%

Time step (t) 0.8 sec Time step (t) 0.8 sec
x & y 30m x & y 30m
Time elapsed 100 min Time elapsed 110 min
Mass bal. error -1.1323% Mass bal. error -1.2154%
Figure 4:1 Illustration of mass balance error for flat terrain at different time of simulation.
Fig. 4. Idealized case of flat terrain with single source. The mass balance error is shown below each figure.
4. Illustration of mass balance error for flat terrain at different time of simulation.
14
3086
Hydrol. Earth Syst. Sci., 15, 27472761, 2011 www.hydrol-earth-syst-sci.net/15/2747/2011/
A. Pathirana et al.: A simple inundation model for urban drainage 2755

Legend
Depth (m)
0
0.05
0.1 A
0.15
Legend
Depth (m)
0
0.05
0.1
0.15

Time step (t) 0.8 sec Time step (t) 0.8 sec
x & y 30m x & y 30m
Time elapsed 80 min Time elapsed 100 min
Time
Mass step (t)
bal. error 0.8 sec
-1.0311% Mass
Timebal.
steperror
(t) -1.19776%
0.8 sec
x & y 30m x & y 30m
Time elapsed 80 min Legend Time elapsed 100 min
C
Mass bal. error -1.0311%
Depth (m)
0
Mass bal. error -1.19776%
0.05
0.1
0.15
Legend
Depth (m)
0
0.05
0.1
0.15

Time step (t) 0.8 sec Time step (t) 0.8 sec
x & y 30m x & y 30m
Time elapsed 120 min Time elapsed 150 min
Mass bal. error -1.28478% Mass bal. error -1.36459%
Time step (t) 0.8 sec Time step (t) 0.8 sec
x
Figure & y
4:2wit
Illustration 30m x &
of mass balance error for sloping y
terrain 30m
Fig.Fig.
5. A 5.
sloping terrain aofsingle source.
Time elapsed
Illustration mass Time elapsed
120 min error for sloping terrain.
balance 150 min
Mass bal. error -1.28478% Mass bal. error -1.36459%
Legend
Depth (m)
Figure 00.05
4:2 Illustration of mass balance error for sloping terrain
0.1
0.15 3087
Legend
Depth (m)
0
0.05
0.1
0.15

Time step (t) 0.8 sec Time step (t) 0.8 sec
x & y 30m x & y 30m
Time elapsed 150min Time elapsed 300 min
Mass bal. err 2.2719% Mass bal. err 3.3551%
Time step (t) 0.8 sec Time step (t) 0.8 sec
x &4:3
y Illustration of mass x & y
30mbalance error for complex 30m
Figure Timetopography
elapsed 300 min
Time elapsed 150min
Mass bal. err 2.2719% Mass bal. err 3.3551%

Figure
Fig. 6. Inundation 4:3 Illustration
performance of mass
in a complex balance
error for complex
topography. topography 15
Fig. 6. Illustration of mass balance error for complex topography.

www.hydrol-earth-syst-sci.net/15/2747/2011/ 15
Hydrol. Earth Syst. Sci., 15, 27472761, 2011
2756 A. Pathirana et al.: A simple inundation model for urban drainage

Table 1. Location of nodes in 2-D grid domain.

Node ID 2 3 4 6 7 8 9 10 11 12 13 14 48
i 47 47 45 42 33 34 41 41 42 58 55 45 41
Location
j 69 64 60 65 64 58 58 50 58 55 26 29 30

Source
Notch
4
3 4
2 3.5
1 3
0 2.5
2
-30 1.5
1
-20 0.5
0
-10
0
10
20 30
10 20
30 -30 -10 0
-20

Fig. 7. A The synthetic terrain with a cylindrical notched flow barrier used to test the mode.

Radar Topographic Mission (SRTM) elevation data (Farr and
Kobrick, 2000) is used as the primary digital elevation model
(DEM) for the area. The google-earth aerial photography
was used together with site-visit experience to digitize the
buildings and roads. These were superimposed on the DEM
(buildings as flow barriers, roads as having a slightly lower
elevation than the surrounding topography). Therefore, the
buildings and roads were explicitly considered (Fig. 11) in
the 2-D flow space to represent urban topography appropri-
ately (An important requirement for small-scale urban flood
modeling). Addition of buildings and roads to the eleva-
tion model introduced flow barriers and sharply bending flow
paths in the topography which could lead to instabilities in
the model.
Thirteen nodes (physically these are man-holes) in the
drainage network were connected with the inundation model.
The locations based on grid counters (i,j ) along x- and y-
direction are shown in Table 1. (See also Fig. 14) The model
was simulated with 50 yr2 h design rainfall event (Fig. 13).
5.1.1 Flood damage data
Damage data, which relate flood damage to flood inundation
parameters, are different for different classes of land use and
properties. This is an essential component is flood damage
estimation. In most cities the central agencies have data and
experience in making damage estimates but often no compre-
hensive guides are available at the local level. In this study
residential class damage curve developed by Nascimento et
al. (2006) using a survey data in the city of Itajuba, a town lo-
cated in the South-eastern region of Brazil, during the year of
2002 was used as an input for completing the damage analy-
sis (see Fig. 12). D = 130.9 + 56.3ln(d) where D is damage
cost in Brazilian Real currency per sq. m and d is depth of
inundation in m.
5.2 Simulation results
The 1-D pipe flow simulation results of the existing urban
drainage network, show that flooding occurs at a number of
nodes (shown in Fig. 14). The flood hydrographs for some
of the manholes is shown in Fig. 13. There was no data for
proper calibration and validation of the inundation mode, so
this part was not done. Reliable inundation data are excep-
tionally scarce for most cities in the world, in terms of both
events and spatial and temporal coverage during an event.
The coupled model simulation is performed to see the spa-
tial and temporal variation of flood flows. Two hours of
1-D/2-D coupled simulation was done at 2-D model time step
of 1 s and grid size of 20 m. The inundation model generates
flood map based on maximum water depth (Fig. 15).
In addition to the flood depth, the model estimates mone-
tary value of the damage associated with flooding, which is
one of the fundamental pieces of information upon which
expenditure decisions are based. For this case study the
stage-damage curve developed in 2002 was used and the

Hydrol. Earth Syst. Sci., 15, 27472761, 2011 www.hydrol-earth-syst-sci.net/15/2747/2011/

 7
4
61 6
3
41 5
2
4
1
A. Pathirana et al.: A simple inundation
61 model for urban drainage 2757
21 30
41 2
S1
1 S21 1
21 S41 0
S1
Figure 4:4 A 3D image of the hypothetical
S21
terrain data
1
S41
Legend
Depth (m) Figure 4:4 A 3D image of the hypothetical terrain data
0
0.05
0.1
Legend
0.15
Depth (m)
0
0.05
0.1
0.15

(a) Flood Map (simulation time 65min) (b) Flood Map (simulation time 80min)

Figure 4:5 Inundation maps at different time of simulation

(a) Flood Map (simulation time 65min) (b) Flood Map (simulation time 80min)
Legend Figure 4:5 Inundation maps at different time of simulation
Depth (m)
0
0.05
Legend
0.1(m)
Depth
00.15
0.05
0.1
0.15

(c)Flood Map (simulation time 95min) (d) Flood Map (simulation time 100min)
Legend
(c)Flood
Depth (m) Map (simulation time 95min) (d) Flood Map (simulation time 100min)
0
0.05
17
0.1
0.15
17

(e) Flood Map (simulation time 120min) (f) Flood Map (simulation time 140min)

Figure 4:6 Inundation maps at different time of simulation

Fig. 8. Inundation maps at different
In the time of simulation.
above hypothetic test, flood wave diffusion is visually inspected for irregular terrain
ig. 8. Inundation maps
shown at different
in Figure 4:4. Thetime of simulation.
flood maps at different times of simulation demonstrate how the
flood propagates over the chosen hypothetical topography. The other hypothetical terrains
which are used for testing the model are shown in fig below.

www.hydrol-earth-syst-sci.net/15/2747/2011/ 3090 Hydrol. Earth Syst. Sci., 15, 27472761, 2011

2758 A. Pathirana et al.: A simple inundation model for urban drainage

HE
5, 3061

A simple
model
dra

(a) (b) A. Path

Time(t) 1sec Time(t) 1 sec
x & y 30m x & y 20m
Time elapsed 100 min Time elapsed 120 min Titl
Mass bal. err 1.295% Mass bal. err 2.081%
Abstract

Conclusions

Tables

(c) (d) Back

Time(t) 1 sec Time(t) 1 sec
x & y 30m x & y 30m Full Sc

Time elapsed 100 min Time elapsed 120 min

Mass bal. err -0.748% Mass bal. err -1.1281% Printer-fri

Fig. 9. Inundation Figure

Fig. 9.map for a 4:7
singleInundation
Inundation map forcell
source map
a single
(a) for a single
source
surrounded bycell
smallsource
(a) cell (a)
surrounded
flow obstacle surrounded
i.e.by small
map afterflow by fillsmall
up andflow
obstacle
water i.e. obstacle
map
overtop the obstacle
Interactiv

(b) in a narrow cannels; (c) bounded by an ideal straight barrier; (d) near large flow barrier.
after water fill up and overtop the obstacle (b) in a narrow cannels; (c) bounded by an ideal
i.e. map after water fill up and overtop the obstacle (b) in a narrow cannels
straight barrier; (d) near large flow barrier.
(c) Bounded by an ideal straight barrier
3091(d) near large flow barrier.
model output showed the extent of flooding is 19 ha and well as being able to simulate wetting and drying processes
For isproblems
the associated damage about 297involving large flow
000 BRL (Brazilian barriers,
Real thatthe
willflood
occur wave diffusion
as the flood flows is visually
over an urbaninspected
area.
currency). for several synthetic terrains and the numerical
The model works
model was properly.
applied However,
for a case study toanalogous
determine the
detail inundation zones, depths and velocities due to sur-
to wetting/drying process, the flow barriercharged
representations implemented
water. It was in our
also examined formodel incur
some of the im-
6 Conclusions some mass balance error as shown inportant Figurehydrodynamic
4:7. For the features,
samesuch as handling
reason mentionof various
in
boundary conditions, 2-D flood wave propagation, cell dry-
A new 2-D inundationwetting/drying process it and
model was developed is reasonable
was then to accept
ing and the inaccuracies.
wetting, and flow interaction with topographic obsta-
coupled with 1-D SWMM model to simulate surcharge in- cles. Since the inundation model is based on detailed spatial
duced inundation in urban areas. Due to the high computa- information i.e. land use and topography, its output will be
tional efficiency of ADI numerical scheme, the model is fast, more realistic to determine flood damage. A flood damage
while maintaining stability and an acceptable level of numer- calculation block is integrated within the inundation model
ical accuracy. code. This tool can help the planner to make quick, informed
The 1-D/2-D coupled model was examined with a number decisions on flood damages control measures on cost-benefit
of numerical experiments involving idealised topography, in- basis.
cluding large flow barriers, sloping terrains and irregular to- In planning of urban drainage a simple and fast tool for
pography. It was found that the model performs well against prediction of flood within an acceptable level of accuracy
stability. It has also been shown to be capable of dealing 19
is often useful for the practitioner as an gentle introduction
with the various flow conditions that may occur in actual ur- to the (currently uncommon) use of inundation models in
ban floodplains (e.g. flowing over barriers, sharp turns), as optimal planning of drainage infrastructure. In the current

Hydrol. Earth Syst. Sci., 15, 27472761, 2011 www.hydrol-earth-syst-sci.net/15/2747/2011/

pproach adopted
A. Pathirana here model
et al.: A simple inundation was todrainage
for urban delineate a boundary significan
2759

oles. i

Figure 5:1 selected flood plain area

Fig. 10. Selected flood plain area.
Fig. 10. Selected flood plain area.
HESSD
5, 30613097, 2008
250
raphic datei for the selected flood plain was downloaded fr A simple inundationStage-Damage curve
200 model for urban
Legend
drainage

Topographic Mission (SRTM)

3092 and converted to Raster and
Damage (R$/m2)

j Boundary cells A. Pathirana et al.

Buildings 150
Streets
Green Area Title Page

reet, building and other flow barrier informations are added

Terrain
100 Abstract Introduction

Conclusions References

50 Tables

an topography precisely.
Figures

J I
0
J I
0 0.5 1 1.5 2 2.5 3 3.5 4
Back Close

Full Screen / Esc Depth (m)

Figure 5:211.Digitized
Fig. 11. Digitized
Fig. buildings
buildings and street
Digitised buildings for part of and street
selected
and street for area. for part of selected area
part of selected area. Printer-friendly Version
Fig. 12. Flood damage (stage damage) curve ( adapted from Nasci-
ations with in 2D calculation domain mento et al. (2006).
Interactive Discussion

only 13 nodes with in the domain

3093of the selected flood plain area. The locations
www.hydrol-earth-syst-sci.net/15/2747/2011/ Hydrol. Earth Syst. Sci., 15, 27472761, 2011
grid counters (i, j) along x and y direction are shown below.
Table 5:1. Location of nodes in 2D grid domain
 manholes. The distribution of flood volume in some of the manholes is shown in Figure 5, 3

5:4.
2760 A. Pathirana et al.: A simple inundation model for urban drainage A sim
0 mo
13 0
12 1 1
11
10 2 2 A.
Flooding(CMS)
9 3 3

Depth(mm)
Depth(mm)
8
7 4 4
6 5
5 5
4 6 Abs
3 6
7
2 7 Concl
1 8
0 8 9
Tab
0 15 36 57 78 99 120 127 134 9 141 148 155
0 15 36 57 78 99 120 127 134 141Time (min)
148 155
(min) The coupled
Rainfall
Time model simulation
Manhole_2 is performed
Manhole_3 to see the spatial and temporal v
Manhole_4
Manhole_5 flood flows.
Manhole_6 J
Rainfall Manhole_2 Manhole_3 A time stepManhole_12
of 1 sec and grid size of 20 m are used for this case
Manhole_4
Ba
Manhole_5 Manhole_6 inundation
Manhole_12 model generates flood maponbased
these onon
thewater
network.depth. The flood map at
Fig. 13.Fig.
Flood13.
hydrographs
Showing of some
Figure 5:4 Showing flood distribution in some of the manholes
flood1-D model nodesin
distribution (Manholes).
some ofSee
theFig. 14 for the
manholes. locations

simulation is shown in Figure 5:6.

re 5:4 Showing flood distribution in some of the manholes i
Legend Pri
Depth (m) HESSD
0 Int
0.05 5, 30613097, 2008
0.1
0.15 A simple inundation
3095
model for urban
drainage

A. Pathirana et al.
j

Title Page

Abstract Introduction

Conclusions References

Figure
Fig. 15. Flood Fig.5:6.
inundationFlood
map atInundation
15. Maximum flood
the enddepth at Map at the
each Figures
grid
of simulation.
Tables end of
cell during thesimulation
2h sim-
ulation.
In addition to the flood map, the coupled model prints the velocities in each dir
J I

flood depths for each grid cells, the floodJ extent and
I
monetary value of the floo
state-of-the-art of urban storm drainage planning, there is a
Figure 5:5 Location on a text file. Frominthe
bottlenecks
largeoutput, it isofdrainage
polarisation
the selected observed
models
Back
in that the
termsClose
network model performs
of complexity and pre- well on a
cision. On one extreme there3097are simple 1-D models that do
topography containing large flow
not provide barriers
any inundation i.e. buildings
information
Full Screen andother
/ Esc and on the it is also examin
there
are sophisticated 1-D/2-D (often commercial) models that
performs well againstsacrifice
stability.
speed/simplicity to precision.Version There is not much
Printer-friendly
middle ground! Our objective was to contribute filling this
22
g. 14. Location bottlenecks in the selected drainage network.
Fig. 14. Bottlenecks in the drainage network indicated by nodal Interactive Discussion
void. Sacrificing some of the precision for accessibility for a
flooding during 1-D analysis. Time: 1 h after6start Conclusions
of simulation.
e 5:5 Location bottlenecks in the selected drainage network wider audience (in terms of simplicity, speed) can indeed im-
prove current
3096 In this study, development of asituation
new 2D of under-employing
inundation model inundation
codemod-is carried o
elling in urban drainage planning among practitioners, par-
computer programming language.
ticularly The model
in the developing world. is then coupled with 1D-SWMM t
surcharge induced inundation in urban areas and22to relate the spatial inform
Hydrol. Earth Syst. Sci., 15, 27472761, 2011 www.hydrol-earth-syst-sci.net/15/2747/2011/
design of urban drainage network. Due to the high computational efficienc
numerical scheme, the model also achieves an advantage in required hyd
A. Pathirana et al.: A simple inundation model for urban drainage
Cost benefit analysis and optimisation of drainage in-
frastructure design could make significant savings and
increase the reliability of drainage projects. However,
optimisation problems demand large number (thousands) of
simulations runs and many existing inundation models are
too numerically heavy to employ for that purpose. There-
fore, often these studies are done based on the volumetric
flooding (as overflow from sewers), without considering the
inundation effects explicitly. In the process the large errors
are introduced in damage estimation. However, the current
model is a good candidate for optimisation problems due to
its numerically light implementation.
Edited by: N. van de Giesen
References
Begnudelli, L. and Sanders, B. F.: Unstructured Grid Finite-Volume
Algorithm for Shallow-Water Flow and Scalar Transport with
Wetting and Drying, J. Hydraul. Eng., 132(4), 371384, 2006.
Begnudelli, L. and Sanders, B. F.: Simulation of the St. Francis
Dam-Break Flood, J. Eng. Mech., 133(11), 12001212, 2007.
Bulatewicz Jr., T. F.: Support for model coupling: An interface-
based approach, Ph.D. dissertation, University of Oregon, pp216,
2006.
Caviglia, F. J. and Dragani, W. C.: An Improved 2-D Finite-
Difference Circulation Model For Tide- And Wind-Induced
Flows, International Association for Mathematical Geology, 22,
10831096, 1996.
www.hydrol-earth-syst-sci.net/15/2747/2011/
2761
Chow, V. T.: Handbook of applied hydrology, McGraw-Hill, New
York, 1964.
Farr, T. G., Rosen, P. A., Caro, E., Crippen, R., Duren, R., Hens-
ley, S., Kobrick, M., Paller, M., Rodriguez, E., Roth, L., Seal, D.,
Shaffer, S., Shimada, J., Umland, J., Werner, M., Oskin, M., Bur-
bank, D., and Alsdorf, D.: The Shuttle Radar Topography Mis-
sion, Rev. Geophys., 45, RG2004, doi:10.1029/2005RG000183,
2007.
Hunter, N. M., Bates, P. D., Horritt, M. S., and Wilson, M.
D.: Simple spatially-distributed models for predicting flood
inundation: A review, Geomorphology, 90(34) p208-225,
doi:10.1016/j.geomorph2006.10.021, 2007.
Kippe, V., Haegland, H., and Lie, K. A.: A Method to Improve the
Mass Balance in Streamline Methods, SPE Reservoir Simulation
Symposium, Houston, Texas, USA, 2007.
Nascimento, N, Baptista, M., Silva, A., Lea Machado, M.,
Costa de Lima, J., Goncalvesm, M. A., Silva Dias, R., and
Machado, E.: Flood-damage curve: Methdological develop-
ment for the Brazilian context, Water Practice Technol., 1(1),
doi:10.2166/wpt.2006.022, 2006.
Peaceman, D. W. and Rachford Jr., H. H.: The Numerical Solu-
tion of Parabolic and Elliptic Differential Equations, J. Soc. Ind.
Appl. Math., 3(1), pp. 2841, 1955.
Rossman, L. A.: Storm water management model Users manual
version 5.0. Water Supply and Water Resources Division Na-
tional Risk Management Research Laboratory Cincinnati, 2004.
Hydrol. Earth Syst. Sci., 15, 27472761, 2011