You are on page 1of 22

Hydrological Sciences Journal

ISSN: 0262-6667 (Print) 2150-3435 (Online) Journal homepage: https://www.tandfonline.com/loi/thsj20

FluBiDi – a model to estimate flood based on


runoff: validation using extreme and natural basin
conditions

Faustino De Luna-Cruz, J.G. Ramos-Hernández, O.A. Fuentes-Mariles & J.


Gracia-Sánchez

To cite this article: Faustino De Luna-Cruz, J.G. Ramos-Hernández, O.A. Fuentes-Mariles


& J. Gracia-Sánchez (2019) FluBiDi – a model to estimate flood based on runoff: validation
using extreme and natural basin conditions, Hydrological Sciences Journal, 64:3, 297-317, DOI:
10.1080/02626667.2019.1578464

To link to this article: https://doi.org/10.1080/02626667.2019.1578464

Accepted author version posted online: 04


Feb 2019.
Published online: 27 Feb 2019.

Submit your article to this journal

Article views: 99

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=thsj20
HYDROLOGICAL SCIENCES JOURNAL
2019, VOL. 64, NO. 3, 297–317
https://doi.org/10.1080/02626667.2019.1578464

FluBiDi – a model to estimate flood based on runoff: validation using extreme


and natural basin conditions
Faustino De Luna-Cruza,b, J.G. Ramos-Hernándezb, O.A. Fuentes-Marilesb and J. Gracia-Sánchezb
a
Programa de Maestría y Doctorado en Ingeniería, Universidad Nacional Autónoma de México, UNAM-CU, Ciudad de México, México;
b
Instituto de Ingeniería, Coordinación de Hidráulica, Universidad Nacional Autónoma de México, UNAM-CU, Ciudad de México, México

ABSTRACT ARTICLE HISTORY


FluBiDi is a two-dimensional model created to simulate real events that can take days and Received 29 May 2018
months, as well as short events (minutes or hours) and inclusive laboratory tests. To verify the Accepted 22 November
robustness of FluBiDi, it was tested using a previous study with both designed and real digital 2018
elevation models. The results highlight good agreement between the models (i.e. Mike Flood, EDITOR
SOBEK, ISIS 2D, and others) tested and FluBiDi (around 90% for a specific instant and 95% for the A. Castellarin
complete time simulation). In the simulated hydrographs, the discharge peak value, time to peak,
and water level results were accurate, reproducing them with an error of less than 5%. The ASSOCIATE EDITOR
H. Kreibich
velocity differences observed in a couple of tests in FluBiDi were associated with very short
periods of time (seconds). However, FluBiDi is highly accurate for simulating floods under real KEYWORDS
topographical conditions with differences of around 2 cm when water depth is around 150 cm. FluBiDi; flood risk prediction;
The average water depth and velocities are precise, and the model describes with high accuracy shallow water equations;
the pattern and extent of floods. FluBiDi has the capability to be adjusted to different types of momentum and mass
events and only requires limited input data. conservation; flood
management

1 Introduction are no available records, in contrast to rainfall records,


which are accessible in several time series: daily, fortnightly
The main characteristics of basins exposed to floods are
or seasonal. Thus, the inter-annual variability in peak dis-
that they favour localized rainfall and rapid surface runoff,
charge needs to be efficiently correlated with the mean
and that tributary flood peaks reach the main stem in a
annual precipitation or the mean annual runoff (Wohl
time-synchronized manner (Robinson et al. 1995, Gupta
2007). Additionally, to determine how easy it is for rain-
and Waymire 1998, Kusumastuti et al. 2007). The flood
water falling on the land to reach a given stream, a fill–spill
impact can be local, affecting a community, or large, dis-
approach, the generation of preferential flows and surface
turbing the entire river basin, whether the flood is large
runoff connectivity thresholds, can be used (Beven and
(quite rare) or small (common). In particular, the recur-
Davies 2015). At this point, high connectivity represents a
rence of large floods in a river is a function of meteorolo-
major flood risk because larger volumes of water will reach
gical and physiographical conditions. Thus, a change in
streams and rivers rapidly, generating a high level of dis-
climate modifies flood potential in both magnitude and
charge. Also, the relationship between the shape of the
frequency, although Blum (2007) considered that it is not
hillslope and the stream, as well as steep slopes with no
directly proportional to the increase or decrease in precipi-
flat area next to the stream, represent a higher level of
tation, since basin physiography could accelerate flooding
connectivity compared with smooth slopes or flat lands.
as a result of alignment of highlands, drainage patterns and
Thus, there is a need to understand the hydrological con-
types of landforms, and the amount of moisture presented.
nectivity between a river and its floodplain, which requires
In general, river basins exposed to high rainfall events
knowledge of the interplay between the climate, ecological,
present an organized drainage net, increasing the water
geomorphological and hydrological processes in the eco-
flow at the mainstream according to the amount and
system (Hudson et al. 2013). Furthermore, runoff and
intensity of precipitation and the topography of the basin
floods are also nonlinear, threshold-driven processes as a
(Gupta 2010). Flow discharges also depend on the river
result of the interactions between climate and catchment
geometry and roughness of the channel, which are some of
characteristics, such as surface and subsurface pathways,
the nonlinear relationships present in a hydrological sys-
which are dynamic and heterogeneous and have highly
tem. However, even if flood discharges are common, there

CONTACT J.G. Ramos-Hernández jramosh@iingen.unam.mx


© 2019 IAHS
298 F. DE LUNA-CRUZ ET AL.

nonlinear relationships with other variables and para- known, and there are no losses by infiltration. Thus,
meters (Kusumastuti 2006). uncertainties could be related to the use of nonlinear
The nonlinearities of the system also imply, among aspects for both rainfall and runoff and in the flood
other characteristics, different volumes of rainfall boundary delineation based on the topography (slope,
inputs for different pristine conditions; thus, they mod- digital elevation model – DEM). In this way, a fine grid
ify the unit hydrograph of a river since there is not a enables accurate prediction of the pattern of flood
linear relationship transforming effective precipitation depth, whether or not a wide range of flood discharge
depths into the rainfall–runoff hydrograph (Sherman is being tested, without the need for recalibration of the
1932, 1942). However, runoff varies with the spatial roughness coefficient.
and temporal patterns of rainfall and the topographic Fiorentino and Iacobellis (2001) analysed the effect
and hydrological characteristics of a catchment, and of runoff thresholds that reinforce different generation
preferential and non-preferential flows are generated mechanisms on the resulting flood frequency distribu-
travelling at different distances and speeds (Robinson tions and established the concept of variable runoff
and Sivapalan 1997, Jothityangkoon et al. 2001, Spence contribution area. Nippgen et al. (2015) recognized
and Woo 2006, Kusumastuti et al. 2007). Thus, in that it is necessary to differentiate between active and
order to predict the rainfall–runoff hydrograph, it is contributing areas because they are not interchange-
necessary to recognize that different rainfall, drainage able. For Ambroise (2004), active areas have no
basin size and form, topography, vegetation and cli- hydraulic connection to the stream network, thus
mate generate a characteristic rainfall hydrograph of a they do not contribute to watershed runoff when the
given duration similar to the unit hydrograph, and downhill flow paths are discontinuous. In this way,
different flood responses are expected between seasons lateral connectivity of rivers and floodplains, as well
or over a period of many years in the same region. as peak discharges have been studied in more detail; for
Strahler (1964) pointed out that the drainage basin example, Middelkoop and Asselman (1998) examined
morphometry has a direct effect on the flood hydro- embankment floodplains. Hudson et al. (2013) pointed
graph. Also, Gupta (2010) mentioned other conditions out that using only a conventional cross-sectional
such as basin shape and bifurcation ratio. Additionally, channel approach to explain connectivity is limited.
it is important to consider the scale of temporal varia- They proposed a longitudinal perspective that reduces
bility – inside and between a storm, as well as seasonal the restriction between the river and floodplain, calcu-
(annual), inter-decadal and inter-annual – and to lating the area of floodplain inundation associated with
establish a hydrological connection from the main different discharge magnitudes and flow durations, and
channel at the basin to the drainage network. relating the pattern of inundation to the floodplain
Commonly, hydrograph users ignore the nonlinearities geomorphology. Also, authors have noticed that coarse
of rainfall, runoff and floods, although they could floodplain elevation data are sometimes used, which
introduce intermittency to the rainfall–runoff process. makes it difficult to estimate the exact threshold of
This is because natural or artificial reservoirs could connectivity accurately. Other studies have considered
attenuate or terminate the runoff response and produce differences in the floodplain geomorphology to under-
runoff via overflow (Kusumastuti 2006). stand the pattern and spatial extent of inundation. In
Moreover, Blöschl and Sivapalan (1997) showed that some cases, studies have used the same return period
the coefficient of variation of a flood frequency curve or flow duration without taking into account that the
increases by a factor of four when nonlinearity is not river stage varies longitudinally as a result of a non-
introduced into the rainfall–runoff relationship. linear downstream profile (Hudson et al. 2013).
According to Balthazar et al. (2009), nonlinear dyna- To simulate theoretical distributed floods favouring
mical systems are highly complex, since real systems the use of numerical simulation models, the variable
are imperfect and uncertainties are present in both runoff contribution area is fundamental (Fiorentino et
system parameters and modelling stage. Basically, al. 2006). Some of these models, from empirical to
uncertainties are related to the lack of precise knowl- physically-based, or with stochastic to distributed para-
edge of the system parameters, random or noisy exter- meters, have failed to reproduce the way in which a
nal loading, operating conditions, and variabilities in basin responds to precipitation (i.e. typical features of
the processes, among others. Also, Balthazar et al. the runoff process) because they consider the beha-
(2009) mentioned that these uncertainties are not viour to be linear. A stream hydrograph alone has
important and may be overlooked in the mathematical limited value in assessing whether the simulation
modelling. This is the case when rainfall is not forecast, obtained fits the water flow accurately within the
the basin size is small, the hydrogram or limnigram is basin in both space and time. To understand the
HYDROLOGICAL SCIENCES JOURNAL 299

water flow effect in a flood, McDonnell and Beven and determines the contribution volume of this site to the
(2014) mentioned that it is necessary to test models total basin runoff (including local rain). To do this, the
to settle multiple working hypotheses for different analysis performed with the model uses rainfall data to
catchment responses, as was done by Néelz and develop a rainfall–runoff hydraulic model. Second, as a
Pender (2013). distributed, physically-based model, FluBidDi provides
Mathematical models, as distinct from hydrological an interpretation closer to reality, since it incorporates
ones, can represent a flood by using only cross-sec- several variables and parameters of the hydrological cycle
tions, or by including the development of the mixing and basin characteristics. This is possible because the
layer between the main channel and floodplain in two- mathematical model is founded on physical principles
dimensional (2D) models. Such 2D models also include that consider equations at a small scale, assuming that
features of the components of the drainage net. Bates scale changes are possible using parametric values.
and De Roo (2000) indicated that topography needs to The FluBiDi model is a hydraulic model; it includes a
be considered in particular detail to model the water subroutine that incorporates complex analysis involving
levels achieved, in order to avoid significant errors in temporal and spatial variations of input (precipitation)
the estimation of flooding when using 2D models. For and output parameters (runoff, evapotranspiration and
instance, the use of DEMs or remote sensing data such infiltration) for long-term events (months). For the out-
as LiDAR helps in understanding the floodplain pro- put parameters, the US SCS (Soil Conservation Service)
cesses and allowing utilization of 2D models, such as curve number (CN) method is used with the premise that
finite elements, due to the accuracy of the topography the soil dries out after each rain event. Basically, the
data (Marks and Bates 2000). model involves the momentum and mass conservation
The main purpose of this research is to propose a equations; thus, to calculate water flow of the drainage
distributed, physically-based model called FluBiDi network and floodplain in a basin, it is necessary to solve
(“Bidimensional Distributed Flow”), as an alternative a differential equation system under certain initial and
to the commercial models already used worldwide As boundary conditions. For this, the model uses the finite
with other 2D codes that predict runoff and flooding in difference method because there is no analytical method
large river basins, FluBiDi simulates both historical and for the solution of the equations mentioned (De Luna
statistical flood events from several recurrence times. 2015). Therefore, to model the process of transformation
The time step is determined as function of the pixel of rainfall into runoff, it is assumed that (1) part of the
size: hour or minutes for 1 km to 1 second for 10 m to rain does not become runoff (loss) while the remainder
achieve accurate rainfall–runoff hydrographs. To com- does (differences) (production module), and (2) runoff
pare the accuracy and robustness of the model, several movement is observed through the basin until it reaches
tests were replicated from the benchmark report of the site of interest (transfer module) Additionally,
Néelz and Pender (2013) to simulate (1) water flow FluBiDi simulates runoff in the surrounding plains and
movement under extreme conditions (initial and at hydraulic structures (i.e. bridges) in order to obtain the
boundary, and designed topography), and (2) quasi- extent of flooded area, magnitude of water levels and
natural flood events considering hydraulic input water velocity reached during the discharge. The dynamic
(boundary conditions and initial condition) and a com- character of floods and the influence of water displace-
plex topography (i.e. a real ecosystem). In Section 2, ment downstream requires the use of mathematical mod-
the FluBIDi model derivation and numerical integra- els that, at least, include flow equations in two horizontal
tion are presented together with the test of its applica- dimensions, in which water velocity corresponds to the
tion, while the tests of accuracy and robustness are average vertical value. This allows FluBiDi to define dif-
described in Sections 3 and 4, with the results of an ferent types of risk associated with either the velocity of
existing DEM being provided in Section 4; concluding the stream or the level of water surface reached.
remarks are presented in Section 5. The first FluBiDi version was coded in VisualBasic
(Fuentes and Franco 1997), so the user could follow the
model in a graphical way as the simulation process
2 FluBiDi model
proceeded. This version was tested on the low basin
The FluBiDi model is a fourth-generation modelling sys- area of the La Sierra River, Tabasco State, Mexico, and
tem for forecasting runoff developed by Fuentes and in this case “n” hydrographs were used as input data
Franco (1997) and complemented by De Luna (2015) at (Fuentes-Mariles et al. 2000). This means that hydro-
the Institute of Engineering of the Universidad Nacional grams from upstream input basins are considered, even
Autónoma de México (UNAM). First, FluBiDi seeks to if they are not modelled. In 2009, “generalized rainfall”
establish runoff for a given site within a basin under study was included; this means a single precipitation record
300 F. DE LUNA-CRUZ ET AL.

was considered for the whole basin (González et al. Fsx and Fsy are force by mass unit associated with the
2009). After some field measurements, the model was movement resistance in the x and y directions (m s−2),
shown to have the ability to process surface water levels u and v are the flow velocities in the x and y directions,
as boundary conditions. Thus, the model uses known respectively (m s−2), and x and y are the horizontal and
values, taking records from flood limnigrams from vertical directions in the Cartesian system. Fsx and Fsy
hydrometric stations when the discharge is into the can be written in terms of friction slopes Sfx and Sfy:
continental land, or mareograms if the discharge goes
1 @p du
to the sea (Fuentes et al. 2010). Once the reliability of  þ X  gSfx ¼ (3)
ρ @x dt
the model was tested, precipitation records from sev-
eral climatological weather stations were considered in 1 @p dv
 þ Y  gSfy ¼ (4)
2011. Currently, the FluBiDi model uses hydrographs, ρ @y dt
but in 2013 mathematical modelling was performed
using only surface water levels as a variable for the Also, it is necessary to consider that velocity (u and v)
boundary conditions. Two reservoirs connected by a changes in time and space; thus Equations (3) and (4)
single channel were considered in order to compute the are re-written as:
transference discharge (Domínguez-Mora et al. 2014). 1 @p @u @u @u
This was an innovative contribution, since very few  þ X  gSfx ¼ þu þv (5)
ρ @x @t @x @y
models have the ability to establish boundary condi-
tions and most of them require as input both the 1 @p @v @v @v
 þ Y  gSfy ¼ þu þv (6)
discharge and water level. FluBiDi can input two ρ @y @t @x @y
water levels to compute peak discharges since it uses
Due to gravitational acceleration, X and Y are equal to the
an intrinsic variable of the differential equations. The
gravitational potential derived on x and y, respectively. If
2013 version of FluBiDi was used to map the risk level
the gravitational potential is defined as the gravity accel-
for housing (severity as depth per velocity) during a
eration on the z axis (perpendicular to x and y):
flood, and a validation of the flood surface water levels
was performed using remote sensing (Fuentes-Mariles @
X¼ ðgzÞ ¼ gð@=@xÞðzÞ (7)
et al. 2013). FluBiDi is a promising flood simulation @x
model that has already been calibrated and validated on @
real events in several river basins in Mexico and could Y¼ ðgzÞ ¼ gð@=@yÞðzÞ (8)
@y
be useful in other regions.
It is considered that, for water flows on broad terrain
with little gradient background, the convective accel-
2.1 Model theoretical derivation eration is negligible. Thus, this model is applicable in
As already stated, for the mathematical models, the situations where the Froude number is small (Ogden et
flow equations need to be included in two horizontal al. 2015).
dimensions. Here, the depth-averaged 2D shallow Equations (7) and (8) are similar if one considers the
water equations (SWE) are derived by integrating the bed slope (S0x and S0y). p = ρ g h (fluid density, gravity
Navier-Stokes equations (based on the study of the acceleration and water depth) substituted in Equations
motion of viscous fluids involving parametrization at (3) and (4) provides:
a macroscale from the basic microscale equation) 1 @u @h  
assuming that hydrostatic pressure distribution and þ ¼ S0x  Sfx (9)
g @t @x
uniform velocity profiles take place in the vertical
direction. As the model involves the momentum and 1 @v @h  
þ ¼ S0y  Sfy (10)
mass conservation equations, the governing equations g @t @y
for momentum are (Mahmood and Yevjevich 1975):
For friction slopes, it is important to introduce the
1 @p du Manning-Strickler equation, where n is Manning’s
 þ X  Fsx ¼ (1)
ρ @x dt roughness coefficient:
1 @p dv n2 juju
 þ Y  Fsy ¼ (2) Sfx ¼ (11)
ρ @y dt h4=3
where p is hydrostatic pressure (kg m−2), X and Y are n2 jvjv
Sfy ¼ (12)
forces by mass unit in the x and y directions (m s−2), h4=3
HYDROLOGICAL SCIENCES JOURNAL 301

Regrouping, considering Z as surface water level vectors, which link microscopic to macroscopic cir-
related to the land topography, these equations are culation via the Navier-Stokes theorem. These surface
written as: integrals represent the conversion of the lineal inte-
gral into area integrals, this implies changing the
1 @u n2 juju @h @Z
þ 4=3 ¼   (13) boundary conditions of the region (Green´s theo-
g @t h @x @x rem). Some schemes consider the homogeneous, con-
1 @v n2 jvjv @h @Z servative part of the system, and a discretization of
þ 4=3 ¼   (14) the non-conservative term with a “lateralization”
g @t h @y @y
(Costabile et al. 2012). Thus, velocities are located
Finally, the governing mass continuity equation needs in half of the cell sides and the surface water and
to consider rain contributions and infiltration losses; topography are at the centre of the cells (Fuentes and
thus: Franco 1997). In this case, the flood basin area (in
@h @hu @hv horizontal projection mesh) is divided into rectangu-
þu þv ¼ rf (15) lar cells, with length Δx and width Δy, and equations
@t @x @y
are solved for each variable at the centre of the model
where r is the rain intensity and f is the infiltration domain (Fig. 1) (Fuentes and Franco 1997, Costabile
losses. et al. 2012). Under this scheme, the momentum
equations may be written as:
 
1 @u @h @Z
2.2 Numerical integration scheme juju þ α ¼ α þ (16)
g @t @x @x
The finite volume method was used for the numerical  
integration of the governing equations. It considers 1 @v @h @Z
jvjv þ α ¼ α þ (17)
the integral form of the SWE to define schemes on g @t @y @y
different mesh types. The system of equations is
4=3
integrated over an arbitrary control volume to obtain with α ¼ hn2 . Thus, Equations (1) and (2) can be
surface integrals for each component of the flux expressed in terms of finite differences as:

Figure 1. Zone in which the flood area is allocated in the model.


302 F. DE LUNA-CRUZ ET AL.

  pþ1 p
uiþ1=2;j  uiþ1=2;j As the solution of Equation (23) considers two values for
 pþ1  pþ1 p
uiþ1=2;j uiþ1=2;j þ αiþ1=2;j the numerical fluxes, it is necessary to consider initial and
gΔt
" p p
# boundary conditions. Initial boundaries consider an initial
p hiþ1;j  hi;j ziþ1;j  zi;j time, t0, in order to assign values to the variables u, v and h;
¼ αiþ1;j þ (18)
2 Δx Δx in this case, the floodplain is dry, so these variables are zero.
If there is a water body (source/sink, q), depth values (h) in
  pþ1 p
vi;jþ1=2  vi;jþ1=2
 pþ1  pþ1 p some cells will be zero, but others will correspond to a
vi;jþ1=2 vi;jþ1=2 þ βi;jþ1=2
gΔt known surface water level. Thus, it is necessary to take
p p
!
h  h ziþ1;j  zi;j into account that the mesh could have known depth and
p iþ1;j i;j
¼ βi;jþ1 þ (19) velocity values (Fig. 2).
2 Δy Δy
The border conditions consider four points from the
centre of the cell: for the right side (R), for the left side
where
(L), and for above and below. The first assumption is for
!4=3
p
hiþ1;j þ hi;j
p
1 null velocity. Thus, looking at Figure 2, if i is equal to 1,
p
αiþ1=2;j ¼ (20) the velocity is ui-1/2,j = u1/2,j = 0; otherwise, for a max-
2 n2iþ1=2;j
imum value of i (i = M), the velocity is ui-1/2,j = uM-1/2,j.
p p
!4=3 Above the cell, if j = N, the velocity is vi,j-1/2 = vi,N-1/2 = 0,
hi;jþ1 þ hi;j 1
p
βi;jþ1=2 ¼ (21) and below the cell, for j = 1, the velocity is vi,j−1/2 = vi,1/
2 n2i;jþ1=2 2 = 0. The final equations for the left (i = 1), right (i = M),
above and below are, respectively, Equations (24), (25),
ni;j þ niþ1;j
niþ1=2;j ¼ (22) (26) and (27):
2
p pþ1
hpþ1 ¼ hi;j þ qi;j ΔxΔyΔt
Subscripts i and j are used to geolocate the variables and p i;j

represents the moment in which these variables are con- Δt h pþ1  p p



pþ1

p p
i
 uiþ1;j hiþ1;j þ hi;j  ui1;j hi;j þ hi1;j
sidered. Also, some considerations, such as whether the 2Δx 2 2

Δt h pþ1  p p

pþ1

p p
i
surface water level is positive or null on the basis of the  vi;jþ1 hi;jþ1 þ hi;j  vi;j1;j hi;j þ hi;j1
2Δy 2 2
mass equation, provide Equation (23) as the final equation:
(24)
pþ1 p pþ1
hi;j ¼ hi;j þ qi;j ΔxΔyΔt pþ1 p pþ1
hM;j ¼ hM;j þ qM;j ΔxΔyΔt
Δt h pþ1  p   i
Δt h pþ1  p i
p pþ1 p p
 uiþ1;j hiþ1;j þ hi;j  ui1;j hi;j þ hi1;j p
2Δx 2 2  uM1;j hM;j þ hM1;j
Δt h pþ1  p   i 2Δx
Δt h pþ1  p   i
2
p pþ1 p p
 vi;jþ1 hi;jþ1 þ hi;j  vi;j1;j hi;j þ hi;j1 
p pþ1 p p
vM;jþ1 hM;jþ1 þ hM;j  vM;j1 hM;j þ hM;j1
2Δy 2 2
2Δy 2 2

(23) (25)

Figure 2. Frontier boundaries considered in the mathematical modelling.


HYDROLOGICAL SCIENCES JOURNAL 303

pþ1 p pþ1
hi;N ¼ hi;N þ qi;N ΔxΔyΔt the latest generation of 2D hydraulic modelling
Δt h pþ1  p p

pþ1

p p
i packages” developed by Néelz and Pender (2013). In
 uiþ1;N hiþ1;N þ hi;N  ui1;N hi;N þ hi1;N
2Δx this report, several 2D simulation models were selected
Δt h pþ1  p i
2 2


p
vi;N1 hi;N þ hi;N1 to represent different theoretical 2D models for pre-
2Δy 2
dicting real or hypothetical flood events, some consid-
(26) ering a flash response and others a slow response. The
pþ1 p pþ1
hi;1 ¼ hi;1 þ qi;j ΔxΔyΔt main benefit of performing this benchmark compari-
     son was the accessibility of data for all cases of flood
Δt pþ1 p p pþ1 p p
 u 1 h þ hi;j  ui 1 hi;1 þ hi1;1 situations. The simplified characteristics of the 2D
2Δx iþ2;1 iþ1;1 2;1
h
Δt pþ1 p  i models maybe found revised in Neal et al. (2012).
p
 v 3 hi;2 þ hi;j
2Δy i;2 The SWE model involves terms for convective accel-
(27) eration, pressure, bottom slope, friction slope and
As noted by Costabile et al. (2012), some controls are boundary conditions. As a full SWE model, FluBiDi
necessary in order to prevent instability problems con- has the following characteristics:
sidering fully dynamic equations, particularly when
small depths are computed over complex topography ● The first version, which was created for research
and at wet–dry interfaces. They identified three possi- purposes, corresponds to a 1.0 Beta version in
ble sources of problems: the interface between dry and Visual Basic, and the hardware platform is an
wet cells, steep bed levels and friction slope. The first Intel Core i7 at 3.60 GHz, with a 32 GB RAM
problem requires a temporary modification of the right desktop system and 64 bit Windows 8.1
cell bed level, and the wet cell velocity component Enterprise operating system.
normal to the wet–dry cell interface is set to zero to ● The minimum recommended hardware for run-
ensure a no-flow condition. This temporary modifica- ning a FluBiDi simulation is a desktop with Intel
tion does not consider that a wet cell next to a dry cell Core i5 at 3 GHz and 16 GB RAM, with a 64 bit
on the right (R) is defined by hL > 0, hR = 0, so the Windows 8 operating system.
surface water level is other than hR > hL. For the second ● For distribution and use by the international com-
problem, it is necessary to avoid negative water depths munity FluBiDi will be presented in Fortran C
because more water than that actually contained in the and will be available from the website of the
wet cell may be computed as flowing into the dry cell. UNAM Engineering Institute (www.iingen.
In this case, a minimum water depth hs has to be unam.mx).
considered and, when a lower value is computed, the ● FluBidi is different from other models because it
cell is considered dry with the velocity components set does not use sub-menus to input data; it only
to zero and the cell itself is filled up to hs. Then, to requires four files in txt format: a DEM, rainfall
ensure the mass conservation, water is subtracted from or input runoff to the basin, limnigrams and sur-
the adjacent cell containing most water and, in the face roughness (Manning’s n). Also, the initial
same cell, the variables hu and hv are also modified in hydrodynamic conditions (velocity and water
such a way to maintain u and v the same as before. depth) need to be introduced. Another character-
Finally, the friction slopes effect can be removed using istic is that the user can monitor any stage of the
an implicit or semi-implicit treatment of the friction process, in parallel with the calculations, through
source term (bed friction), trying to achieve a discreti- raster and vector files that provide limnigrams or
zation of the independent term of the equation. This hydrograms, respectively, for each cell or section
guarantees that the bed friction does not result in of interest. The advantage in this is that the user
errors when simulating runoff if the water depth does not need to wait until the simulation finishes
becomes very small. For small water depths, the bed to obtain useful information. The grid resolution
friction term dominates over other terms in the (or number of nodes in the area being modelled)
momentum equation as the term h4/3 appears in the and total simulation time are specified for each
denominator. test.

Néelz and Pender (2013) carried out 10 hydraulic tests


using 2D models with designed and existing DEMs
2.3 Model application test
(Crowder et al. 2004, Néelz and Pender 2010), but here
To test the reliability of the FluBiDi model, its perfor- only seven tests were replicated. Test 1 and Test 6 (A and B)
mance was checked against the report “Benchmarking of Néelz and Pender (2013) were removed based on the
304 F. DE LUNA-CRUZ ET AL.

conclusion of the report: the first because it was a trivial test detail, the reader is referred to Néelz and Pender
and covers the capability tested in Test 2, and the second (2013).
because of the difficulty to replicate real data (6A) and the
alteration to the data required to accomplish the boundary
conditions (6B). Test 7 “river to floodplain linking” was not 3.1 Filling of floodplain depressions (Test 2)
used as it will be further studied considering floods of
longer than 48 h duration. The FluBiDi model was set up The objective of this test is “to assess the ability to
for the seven tests as a hydraulic model without the appli- handle disconnected water bodies, wetting and drying
cation of any hydrological subroutine, since the initial of floodplains, and to predict the inundation extent due
conditions correspond to events of short duration (sec- to low momentum flooding on a complex topography,
onds, minutes, hours and some days) and only consecutive with an emphasis on the final distribution of flood water
rainfall events were considered. rather than peak levels” (Néelz and Pender 2013).
Smooth topographic transitions in a hydrograph with
a peak flow of 20 m3 s−1 and time base of approx.
85 min were applied at the top left corner of the
3 Results for a designed DEM
domain where the bed is dry. The total processing
In order to replicate accurately the benchmark of 2D time was 48 h, offering the opportunity to allow
hydraulic modelling packages reported by Néelz and water levels to become stabilized in the 16 output
Pender (2013), FluBiDi was initially set up to use a points (holes). The boundary conditions implied a
similar DEM resolution in each test. Then the point closed system with uniform cover (Manning’s
source hydrograph was established. This results section n = 0.03).
briefly presents the objective of each test, the histogram Figure 3(b) presents the fan distribution obtained
and the initial and boundary conditions. For more using FluBiDi with a preferential flow direction

Figure 3. (a) Inflow hydrograph and map of the area showing the 16 output point locations with the dashed (red) line indicating
the upstream boundary condition, and (b) water level results as a function of the flood characteristics, obtained using FluBiDi.
HYDROLOGICAL SCIENCES JOURNAL 305

towards the east by conservation of momentum. In initial dry bed. Moreover, upstream boundary condi-
general, FluBiDi predicts the same behaviour as almost tions consider a varying inflow discharge causing a
all the SWE models for Points 2, 3, 4, 7 and 8, where flood wave to travel down along a 1:200 slope. As a
the same water levels were achieved. This is expected, function of the conservative momentum models, codes
because all the points present a maximum value com- were expected to predict a water level rise at Point 2
pared with later decreases, as a response to the due to the acceleration terms. Manning’s coefficient n
entrance hydrogram at Point 4. As the water flow is uniform at 0.01 and the grid resolution is 5 m.
through the cells loses energy, it is distributed to the It can be seen from Figure 4 that FluBiDi predicted a
lower levels, which absorb the wave, including Point 2 less rapidly moving wave front at Point 1, starting with
where the increase in flow is no longer reflected. In an initial velocity of 0 m s−1 at t ≈ 80 s, whereas for the
fact, one characteristic of the 2D flows is the wave other models velocity started at an average of t ≈ 60 s.
regulation, as was observed for Points 1, 5 and 6 The generation of this slow moving wave means that
(Fig. 3). These points are located at the greatest dis- for FluBiDi the water level of 9.75 m is reached from
tance from the origin; thus, less water arrives in t ≈ 90 s, with a maximum value of 9.95–9.97 m at
response to, as a function of, the roughness in the t ≈ 200 s. This differs from the other models, which
topography. In the case of Point 5, FluBiDi starts to reached the same initial water level at t ≈ 60 s and the
fill the hole at 5 h, in contrast to the other points where maximum water levels of 9.98–9.99 m at t ≈ 120 s. In
it took up to 13 h. For this reason, water reaches the all cases there is some oscillatory behaviour, which was
predicted water level faster and the low is towards related to two probable causes:
Point 9. At Points 12, 11, 10 and 9 we observed a
large amount of water arriving in the holes, in parti- a. the wave energy is not completely absorbed in
cular in the first three, due to the fan distribution in the channel walls; for FluBiDi this implies a small
FluBiDi overestimating the amount with respect to the water level height in comparison with the other
other models. Point 9 received almost no water in the models of almost 0.025–0.030 m, which is insig-
other models, although, thanks to the water available in nificant, so there is not a major risk of flood; and
Point 5, there is more water at Point 9 in the FluBiDi b. as the initial varying inflow is dependent on the
model. In general, in the FluBiDi model, the holes slope, when the slope increases a supercritical
could be filled in a few hours, but if the purpose is to flow is created, which is difficult to reproduce
delay filling, Manning’s n needs to be modified, as this by FluBiDi.
is the roughness in the topography and is an important
aspect to consider in this test. At Point 2, a contrasting rapid rise in the level was
The FluBiDi modelling parameters were: ΔT = 0.5 s, seen after 140 s for FluBiDi compared to 180 s for the
with a grid resolution of 20 m and a run time of 1210 s; other models, except for ISIS 2D, where the height rise
the total water volume was 97 238 m3 and the mean started at t ≈ 90 s. The ISIS 2D and FluBiDi models
water volume (for all the models including FluBiDi) achieved quite similar responses, obtaining water levels
was 97 085.4 m3, with a standard deviation (SD) of of 9.89 and 9.88 m, respectively, whereas for the other
4713.5 m3. For two of the 20 models (Flowroute-iTM models the level was in the range 9.79–9.83 m. The
and ISIS 2D) at least one value was outside the range of overall model response (referring to those tested by
mean ±2SD. Thus, the mean tendency followed by the Néelz and Pender 2013) could be divided into three
models without the outliers was 97 200 m3. Since groups:
FluBiDi was 0.04% higher, it demonstrates that this
model can achieve reliable values (practically no water (i) Overestimating: in the ISIS 2D and FluBiDi
loss or gain was observed), improving the process models, the main reason is because the accel-
knowledge. eration term is not included in either model due
to the flash flow generated, although the total
volume is at the initial condition. The ISIS 2D
3.2 Momentum conservation over a small
model was improved to reduce the flash flow
obstruction (Test 3)
effect by introducing the MacCormack-TVD
The objective of this test is “to evaluate the channel scheme. In the case of FluBiDi, the oscillation
conservation of momentum considering an obstruction could be reduced by decreasing DT.
of 100 m width in the sloping bed topography between (ii) Underestimating: e.g. in Flowroute-iTM, where
two depressions (point 1 at 150 m and point 2 at 250m)” the lowest Manning’s n value was 0.03 instead
(Néelz and Pender 2013). The boundaries include an of 0.01, due to stability of the model, and
306 F. DE LUNA-CRUZ ET AL.

Figure 4. (a) Profile of the DEM and inflow hydrograph used as upstream boundary conditions, and results of (b) velocity (Point 1)
and (c) water level (Points 2 and 1) obtained by FluBiDi compared with the models tested by Néelz and Pender (2013).

(iii) Average: the remainder of the models is in this The FluBiDi model parameters in Test 3 were a grid
group. resolution of 1 m with ΔT = 0.01 s and a run time of
2100 s. A large-resolution grid provides a better under-
In general, the results presented in Test 3 in Néelz standing of the wave movement in the channel. In
and Pender (2013) are pretty poor, considering that the addition, it is highly probable that, if Manning’s n
test run time is a matter of seconds. However, the aim were changed to represent more surface roughness,
was to show how FluBiDi behaves as it was created to the ISIS 2D and FluBiDi models could provide more
simulate the rainfall–runoff phenomenon in a real similar values to the majority of the other models.
basin; this means the run time may be minutes, hours
or days. Based on this premise, the FluBiDi results are
3.3 Speed of flood propagation over an extended
not as poor as they seem. This is because FluBiDi
floodplain (Test 4)
simplifies equations and, for this test in particular, the
model’s equations do not need to be simplified. In This test assessed “the code’s ability to simulate the
conclusion, even in this case, the FluBiDi model may celerity of propagation of a flood wave and predict
provide a close idea of the flood, which is very impor- transient velocities and depths in a flat horizontal flood-
tant in flood risk analysis. The miscalculations were plain” (Néelz and Pender 2013). Celerity implies the
due to the original condition, where the energy wave speed with which perturbations are transmitted, con-
is absorbed at the walls of the channel, as this was not trolled by the hydrograph. The inflow boundary con-
truly replicated. Also, the impact of the grid size and dition simulates the failure of an embankment by
Manning’s n coefficient are determinant aspects to be breaching or overtopping, with a peak flow of 20 m3 s−1
considered. and the model run time is t = 5 h (Fig. 5). The
HYDROLOGICAL SCIENCES JOURNAL 307

Figure 5. (a) Modelled domain, showing the location of the 20-m inflow and six output points, and (b) the hydrograph applied as
inflow boundary conditions. Source: Néelz and Pender (2013).

boundary condition is applied along a 20-m line in the FluBiDi slightly higher than the other codes except
middle of the western side of the floodplain. Manning’s for SOBEK. One probable reason for the differences
n is uniform at 0.05 and the grid resolution is 5 m. in wave arrival time and in velocity is the accuracy of
Water levels and velocity as tested by Néelz and the simplification (adjustment) carried out for each
Pender (2013) are presented in Figure 6 showing the model.
FluBiDi response to each point superimposed. All the As the benchmark allows one to test how water
full SWE models, including FluBiDi, predict water level levels and velocity are affected by the distance from
with a small percentage difference, and the difference the boundary to the distance reached in 1 h of run
in run time between the models is also small. This time, the FluBiDi model was tested using cross-sec-
implies the need to identify the parameter model tions across the domain (1000 m) (Fig. 7). The
against values of stream discharge, in order to optimize findings show that FluBiDi presents similar beha-
the simulation. However, a stream hydrograph alone viour to all the SWE models tested by Néelz and
has limited value in assessing whether the simulation Pender (2013), whose modelling parameters were
obtained accurately fits the water flow within the basin ΔT = 0.5 s, with a grid resolution of 5 m and a
in both space and time (McDonnell and Beven 2014). run time of 3800 s. The only difference was that, in
Thus, Points 1 and 2 receive almost the same amount the case of both depth and velocity, FluBiDi started
of energy. At Point 5, the FluBiDi wave arrives late at a high value of 0.5 m depth and 1.6 m s−1 velo-
(after 1 h) compared with the other models and, as a city, and reached the lower heights close to 360 and
consequence, the water level reached is low but still 380 m next to Points 4 and 5, respectively. This
within the expected interval. confirms the findings of Néelz and Pender (2013),
In terms of velocity, FluBiDi presented a mean value who concluded that “advective acceleration (the
of 0.38 m s−1, which is comparable with the other momentum equation of the hydraulic theory of free
models. Differences observed at Point 1 were explained surface flow) is not the dominant process since the
by Néelz and Pender (2013) as the ability of all the slope angle is not small enough to maintain the
models to implement the boundary condition. This is condition of flood spreading arising from a rupture”.
because the velocity is a fast phenomenon and, to On the contrary, both the laminar and transitional
represent this, the model input parameters need to be flows under the boundary conditions are the ones
adjusted. The advantage of the FluBiDi model is that that define the dominant process since small differ-
there are few parameters to adjust, in contrast to the ences in the floodplain result in a smooth surface
other models which require several parameters to run where u  at z = h is not small. Results agree very well
the simulation. As a result of these adjustments, the with the Néelz and Pender (2013) obtained values,
flood wave arrives at the same time for all the models, and they provide an opportunity to understand the
but there is a variation in the velocities reached, ran- system and to identify the model parameters when
ging from 0.3 to 0.51 m s−1. As the energy wave the exit condition at the basin is unknown. Thus, as
advances, it was observed for FluBiDi that it arrives the event does not reach supercritical flows in less
delayed by 10–15 min, although its prediction is within than 1 s, then FluBiDi’s simulation provides optimal
the interval for all the codes; only at Point 6 was results.
308 F. DE LUNA-CRUZ ET AL.

Figure 6. Water levels (depths) and velocities showing the results of FluBiDi compared with those of the models tested by Néelz
and Pender (2013).

4 Results for an existing DEM low resolution are used. However, use of such DEMs
may result in limitations of omitting or generalizing
Commonly, hydrodynamic models are successful in some of the important features affecting flow dynamics
topographically homogeneous areas, where DEMs of in heterogeneous zones (Erpicum et al. 2010). The use
HYDROLOGICAL SCIENCES JOURNAL 309

Figure 7. Cross-section of depths and velocities at 0.15-depth contour lines (t = 1 h) showing the results of FluBiDi compared with
those of the models tested by Néelz and Pender (2013).

of a fine grid enables the accurate prediction of the a unidimensional model and that the downstream border
pattern of flood depth, especially where a wide range of condition determines the value. Thus, FluBiDi matches
flood discharges is being tested, without the need for very well with the proposed water level of between 148.7
recalibration of the roughness coefficient (Haile 2005, and 148.9 m for the simulation time. A difference in water
Erpicum et al. 2010, Brandt 2016). Some examples are depth was obtained at Point 7, where it was 0.3 m lower
presented below. For more detail, the reader is referred than the predicted value. This is because FluBiDi does not
to Néelz and Pender (2013). consider the exact position of Point 7; for this reason it
starts at 153 m, whereas the initial values for the models
tested by Néelz and Pender (2013) were between 152.7 and
4.1 Valley flooding (Test 5) 153.2 m. Also, the final water level is slightly higher than for
To simulate a major flood inundation and predict the the other models, in which it agrees more with the pre-
flood hazard arising from a dam failure (peak levels, dicted values, because for them there is still water in the
velocities and travel times), a valley was defined with a depression, whereas in FluBiDi it is already dry. It was
designed skewed trapezoidal inflow hydrograph and a interesting to note that, even though FluBiDi has some
short early peak at 3000 m3 s−1 and a length of ~260 m limitations when representing supercritical flows in a
(Fig. 8). The upstream end was designed to account for a short run time, the response at Point 1 seems not to be
typical failure of a small embankment dam and to ensure affected because the run time is more than a few seconds.
that both supercritical and subcritical flows will occur in The velocities shown in Figure 9 present a significant
different parts of the flow field. Other boundaries are variation over time for all models, being more similar at
closed and the initial condition is dry. The DEM of the Point 1. The FluBiDi model presented similar behaviour to
valley is approx. 0.8 km × 17 km with a slope of approx. TUFLOW FV1, rising at 0.38 h, and then to TUFLOW
0.01 in its upper region to approx. 0.001 in its lower FV2, reaching almost 2.4 m s−1. It is important to mention
region. Manning’s coefficient, n is 0.04 (uniform). The that the TUFLOW FV model was enhanced in 2015 in
model is run until time t = 30 h to allow the flood to settle order to cope with supercritical flows, and the number of
in the lower parts of the valley. The NB Direct RFSM and iterations was increased to guarantee convergence (Néelz
ISIS Fast models were used to fix the “final water level” and Pender 2013).
(horizontal lines), as shown in Figure 8. At Points 3, 4 and 7, there are important variations
In general, the differences in peak water levels for almost in velocity values, but not in shape, it being difficult to
all codes were around 0.4 m, which represents approx. 10% establish an optimum velocity. In all test cases, FluBiDi
of the predicted peak depth in the deeper areas of the discharged earlier than the other models, the wave
channel. As may be observed, FluBiDi had a similar arriving sooner. These differences could be associated
response in predicting the peak depth level at Points 1, 3 with the location of each site.
and 5 compared with the other models predicting flood However, the most important consideration is that
arrival times (Fig. 8). Points 1, 3 and 5 correspond to some models were applied as 1D, but in Point 4, given
depressions, so some water could be retained. As Néelz the geometrical characteristics of the river bed, there is a 2D
and Pender (2013) pointed out, the differences in water flow with components in ʋ (x) and v (y), which cannot be
levels at Point 5 correspond to variances further upstream represented by 1D models. FluBiDi as a 2D model could be
associated with depressions and the ability of the DEM to adjusted to work as 1D as well as RFSM, a quasi-2D model
represent this. Our modelling agrees with this, but it is also version, the only difference between these codes is that for
noted that the flow at Points 1, 3 and 5 could correspond to RFSM the wave arrives faster than for FluBiDi because
310 F. DE LUNA-CRUZ ET AL.

Figure 8. (a) Scheme of the valley used for the valley flooding test (Test 5): DEM with cross-section along the centre line and
location of the output points (top); and inflow hydrographs (below). (b) Results (at Points 1, 3, 5 and 7) of FluBiDi water level time
series compared with the models tested by Néelz and Pender (2013).

RFSM code corresponds to a simplified flood routing situations after intense rainfall: (A) a precise location
hydraulic model. (release of water after its use) and (B) overwhelming of
At Point 7, FluBiDi discharge arrived earlier than the the drainage system. In some cases, the second
other codes, with a velocity of 1.7 m s−1, while the max- approach requires the linking of a 1D pipe model to a
imum values reported for the models tested by Néelz and 2D overland flow model.
Pender (2013) were >2, 1.95 and 1.82 m s−1 (ISIS 2D GPU,
LISFLOOD FP and UIM, respectively). As occurred at 4.2.1 Rainfall and point source surface flow in urban
Point 4, the change of direction led to some differences areas (Test 8A)
associated with the cross-sections of Points 3 and 7; this was This test simulated a shallow inundation originating from a
the major change at Point 3 for the depression. Some water point source and from rainfall applied directly to the model
remained at Point 7 because it was not completely dry. grid, at relatively high resolution. The flood came from two
Additionally, a comparison test was carried out to ana- sources: a uniformly distributed rainfall event with a peak
lyse water level and velocity cross-sections in the valley at 400 mm h−1 over a time base of 3 min, and a point source
centre line 0–15 km down from the location of the dam at the location occurring over a time base of ~15 min, with a
break. Here, FluBiDi demonstrated that water level and peak at 5 m s−1 occurring ~35 min after the rainfall event
velocity results could be reproducible with high accuracy (see Fig. 10). In this case, a DTM (no vegetation or build-
as well as the talweg obtained from the DEM This supplies ings) was used with a resolution of 0.5 m created from
a model with high capability to represent different events LiDAR data (grid resolution of 2 m or ~97 000 nodes). A
and its limitations (i.e. fast floods) could be solved since, as land-cover-dependent roughness value was applied, with
it uses a real topography, the model could cope with the two categories: roads and pavements (Manning’s n = 0.02)
hydraulic boundary conditions improving the simulation. and any other land-cover type (Manning’s n = 0.05). The
The FluBiDi modelling parameters were ΔT = 5 s, with a model was run until t = 5 h to allow the flood to settle in the
grid resolution of 50 m and a run time of 1500 s. lower parts of the modelled domain. The “final water level”
(Fig. 10, horizontal dashed lines) was predicted by NB
Direct RFSM and ISIS Fast (Néelz and Pender 2013).
4.2 Flooding in urban areas (Test 8)
In general, the FluBiDi model behaved similarly to
In Test 8, Néelz and Pender (2013) present the condi- the other models, predicting water levels similar to ISIS
tion for a surface water flood that arises from two Fast. In the second peak at Point 1, after rainfall
HYDROLOGICAL SCIENCES JOURNAL 311

Figure 9. Velocity time series (at Points 1, 3, 4 and 7) showing the results of the FluBiDi model compared with the models tested by
Néelz and Pender (2013).

(~35 min) when the inflow is presented from the point the dry areas, affecting the beginning of each graph.
source, surface roughness conditions change due to the However, it is interesting that, at all points, the SOBEK
moisture associated with the rain, thus for FluBiDi model seemed to provide a curve with fewer variations
water arrives 4 min earlier than for the other codes, (Fig. 10) (Néelz and Pender 2013). Looking at the
presenting a 2 cm difference as result of the different characteristics of the SOBEK model, it is possible that
Manning’s n values. The friction coefficient and the it makes different simplifications and adjustments to
topography of the street are sufficient to lose energy. different factors, since it uses a square grid without
The FluBiDi model exhibited interesting behaviour at considering the detail, in contrast to the other models,
Point 2 because the flow remained in a depressed area including FluBiDi. Other models, such as TUFLOW,
just before Point 2, so the second peak was almost 2 cm were improved to cope with flood events, but still
less than that for the other models. After this point, showed some restrictions in running the test using
FluBiDi showed similar behaviour to the other models. finite numbers in the initial conditions. The FluBiDi
However, some oscillations were observed at Point 6 model has the ability to introduce an initial water level,
for all models, which corresponds to certain numerical achieving quick stabilization afterwards. This allows
instability, since flow is forced to go within the con- the FluBiDi performance to be quite similar to that of
fines of the street. the SOBEK model.
Velocity in this test was addressed as modules, At Point 2, the velocity increased 3 or 4 min after
where the component y is so small that in practice reaching the first peak, thus the local rainfall was
only velocity in the x direction is significant. All mod- enough to generate a velocity close to 0.8 m s−1. This
els followed the same behaviour and the oscillations was followed by a dry period until the point source
observed were significant (see Fig. 10(b). According to surface flow started to increase to reach the second
Néelz and Pender (2013), the grid resolution is a peak at 44 min after the rainfall was presented; finally,
restriction since it introduces a local variability in the the water was lost as drainage. The topography chan-
topography and to the surface roughness that modifies ged from 26 m at Point 6 to 29 m at Point 7, thus, at
312 F. DE LUNA-CRUZ ET AL.

Figure 10. (a) DTM used, with the location of the point source, the hyetograph and the inflow hydrograph at point location. (b)
Water level time series at Points 1, 2, 3 and 6 (left) and velocity time series at Points 2 and 6 (right). Results of FluBiDi compared
with the models tested by Néelz and Pender (2013).

Point 6, SOBEK and FluBiDi respond to the topogra- rainfall and the point source there was more water
phy variation. Similarly to the SOBEK model (Néelz storage in the FluBiDi model. It is important to note
and Pender 2013), the FluBiDi model responded to the that FluBiDi predicted that Point 1 has a high water
variation in topography. As the velocity reduced from level contributing with the flow towards Point 6 and
1.15 to 0.42 m s−1 during the dry period between the Point 7. As there is water available at the start of the
HYDROLOGICAL SCIENCES JOURNAL 313

point source surface flow (at 35 min), the FluBiDi Pender (2013) was TUFLOW. In part, this good per-
model increases the flow velocity to reach a maximum formance was due to the enhancements realized for the
of 1.22 m s−1, on average, in the second peak, finally 2012.05 version, providing it with better velocity out-
reducing to 0.36 m s−1, with some inundated area puts, increasing the running time and, as a conse-
remaining after 80-min run time. quence, improving stability. These enhancements are
shown in the velocity graph in Figure 11, in which the
4.2.2 Surface flow from a surcharging sewer in FluBiDi model is compared with the models tested by
urban areas (Test 8B) Néelz and Pender (2013) using different approaches to
This test simulates “a shallow inundation originating represent the initial conditions after the dry surface.
from a surcharging underground pipe with a single man- The FluBiDi model started to inundate 4 min later than
hole, without effect in the manhole outflow. The inflow TUFLOW, which implies that the surface is drier, so it
boundary condition is applied at the upstream end of the has the energy to achieve an instantaneous velocity
pipe, with a surcharge expected to occur at the manhole peak of 0.95 m s−1, which dropped to 0.55 m s−1 as a
that spread water across the whole surface defined by a maximum value; and it was dry again after 190 min of
DTM at 2 m resolution” (Néelz and Pender 2013). The run time. The pipe changed direction twice, with build-
other conditions are similar to Test 8A, but, as an ings transverse to the street where Point 3 is located.
urban area is going to be considered in this case, the Introducing a different value of Manning’s coefficient
introduction of the biophysical and human factors that could probably reduce the variation at the moment of
modify the water flow, such as buildings, streets (i.e. discharge, although this could also generate a slow
Point 1) and low topographical elevation locations (i.e. response of the flow which could demand more run-
Point 3), is significant. Thus, a land-cover roughness ning time to start increasing the water level.
value for each cover needs to be applied (Néelz and The manhole flow predictions were generally similar
Pender 2013, Jenkins et al. 2016, Courty et al. 2018). As between codes, with a difference of around 15% range
a consequence of there being paved areas, infiltration is including FluBiDi with a total volume of 5393 m3
not possible due to the low permeability of the different (Néelz and Pender 2013). In general, FluBiDi provides
areas, favouring surface runoff. This extra runoff satu- a value 2% (110 m3) higher than the mean total volume
rates the drainage system network that conveys water of 5283 m3, very close to the ISIS Fast Dynamic pre-
out of the cities and changes the distribution of the diction of 5234 m3. These small differences are asso-
runoff. Only nine models were tested, as in the original ciated with the fact that FluBiDi, as well as ISIS FAST
report, and the “final water level” was predicted by ISIS Dynamic, as a 2D model did not carried out a 1D–2D
Fast (horizontal dashed lines), as shown in Figure 11, link-up simulation. Thus, the hydrogram is FluBiDi’s
in which the water levels at Points 1, 2 and 3, and the input volume value, which corresponds to 5393m3.
velocity at Point 3 are plotted (Néelz and Pender 2013). A final comparison of the results obtained using
After leaving the pipe, flow was favoured moving FluBiDi with those of Néelz and Pender (2013) is pre-
towards Point 1, which was immediately inundated. In sented in Figure 12.
the FluBiDi model, the wave of water moved faster, For Test 8A, local rainfall and a DTM without
reaching a peak 20 min before the majority of models; buildings provides a better idea of the inundated sites
although after 200 min all the models followed the according to the street network showing the intercon-
water level predicted. This is because the topography necting lines and points for the given area. This net-
of the streets marks a lower point than the surface work analysis allows one to identify preferential flows
water level presented. Close to Point 2, FluBiDi pre- that follow the topography according to the surface
sents a low inundated area, which makes the shallow roughness proposed for the type of land cover. In
flow slow down, and results in final peak values at Test 8B, the preferential flow (Points 1, 6, 3) is more
Point 2 of 0.03 m (lower than that of the other models) evident due to the buildings allocated. Water level
to finally dry the surface after 180 min. Less variation values were more constant for the flow at Points 7, 2,
was observed at Point 3, which is related to the change 8 and 4, since the street geometry is quasi-regular. For
in topography and roughness following the trajectory the velocity, this does not apply since velocity depends
of the pipe. In this case, Point 3 corresponds to a on the drainage (pluvial and sewage) network to effi-
location of lower topographical elevation, so there is a ciently release runoff coming from several sources. In
major impact of the inundation water remaining for this case, peak velocity magnitudes change rapidly
more than 300 min, with a difference in the water level (within seconds), so methods that take infrequent
of the FluBiDi model being 0.3% lower than in the snapshots of model state variables may not capture
other models. The best model observed by Néelz and maximum velocities. This is an advantage of FluBiDi,
314 F. DE LUNA-CRUZ ET AL.

Figure 11. (a) The DTM used, showing the location of the manhole, and the inflow hydrograph at the upstream end of the culvert.
(b) Water level time series (results at Points 1, 2 and 3) and velocity time series at Point 3 showing the results of the FluBiDi model
compared with the models tested by Néelz and Pender (2013).

since it integrates the computation with the post-pro- runoff and the impact of the urbanized area down-
cessing, providing graphic output that allows the flood stream, but also there is a need to consider the changes
phenomenon to be followed step by step. Although it generated by the heterogeneous urban catchment with
demands greater computation resources, making the impermeable, semi-impermeable or infiltration zones.
whole process slow, the run time for FluBiDi is only As FluBiDi takes into account all the aspects of runoff
1.5 times greater than that of the other codes. contributions or losses, and these can be followed step-
The final FluBiDi modelling parameters were by-step due to the facility to generate outputs at any
ΔT = 0.01 s, with a grid resolution of 1 m (instead moment in the simulation, the capability of the model
the 2 m proposed originally) and a run time of 4200 s. is highly improved.
The change in the grid resolution in the simulation of
an urban flood (Tests 8A and 8B) indicates the need to
know the effect of buildings, streets and low topogra- 5 Conclusions
phical elevation locations within an urban area. The Today, there are different systems that provide high-
increment of surface runoff reduces the basin charac- resolution topographic data which could be used in
teristics, such as retention time, along with the flow flood simulation models. The accuracy in representing
concentration time in a drain section. Thus, it is neces- an event from these models defines their success,
sary to take into account the upstream contribution to allowing to study them and understand how to achieve
HYDROLOGICAL SCIENCES JOURNAL 315

Figure 12. Flood water level and travel time. The dashed rectangle shows the similar output contours of peak depth between
FluBiDi and the real flood for Test 8A at 10, 50 and 120 min (2 h) and Test 8B at 50, 120 and 200 min (3.3 h).

integrated river management. The software used in the cases they are in the average with the other models tested by
FluBiDi model is user friendly, and has been improved Néelz and Pender (2013), and the model describes with
over the years since it was first designed. high accuracy (2%) the pattern and extent of floods. There
Also, FluBiDi is a reliable system that has the ability is less than 10% difference for a specific instant and less
to simulate real events lasting days and months, as well than 5% for the complete time simulation, with respect to
as short events (of minutes or hours) and laboratory the average values of the other models. This is possible
tests (of seconds). In the laboratory tests, the capacity since FluBiDi is a model that takes into account several
of FluBiDi to cope with flash flows is low, but this has aspects, such as topography, geomorphological and geo-
been solved by reducing the time step (ΔT) and the metrical characteristics, climate, channel and floodplain
grid size. conditions and hydraulic structures. Thus FluBiDi is a
The numerical results obtained from each ideal DEM good model to choose for use in flood management.
test show considerable agreement between FluBiDi and In addition to the robustness of the model, in all
other models in terms of discharge peak value, time to cases the quality of the input data and DEM are deci-
peak and water level. In a couple of tests, the FluBiDi sive. For that it is important to have a suitable grid that
velocities did not compare adequately with the other mod- represents the conditions and provides a water level
els, which was, in part, associated with the fast flood show- and velocity that fit each pixel; it also relies on the
ing in Test 6 and with the fact that FluBiDi did not consider experience of the modeller to represent conditions
a perfect wave (where the energy of displacement is that have a physical meaning without forcing the mod-
absorbed at the walls of the channel, as shown in Test 3). els to solve situations that do not exist.
The effect of the back-wave changes some of the initial
conditions, and, as a consequence, generates oscillation, as
was observed in FluBiDi. However, the results demonstrate Acknowledgments
that FluBiDi is a reliable model for the simulation of flood
The authors would like to thank Dr Ismene Libertad America
events using real topography and run times of more than Rosales-Plascencia (PhD), CEO of Proyectos Multidisciplinarios
seconds. There are peak velocity oscillations but the average del Centro, S.A. de C.V., for the continued discussions helping to
values of water depth and velocities are accurate, since in all improve this paper.
316 F. DE LUNA-CRUZ ET AL.

Disclosure statement distribution of floods. Water Resources Research, 37 (3),


721–730. doi:10.1029/2000WR900315
No potential conflict of interest was reported by the authors. Fuentes, O.A., De Luna, F., and Cruz, J.A., 2010. Revisión
Hidráulica Mediante Simulación Matemática del Flujo en
los ríos Zanapa, Blasillo y Naranjeño: capítulo 2. México
References City (in Spanish): Instituto de Ingenieria Project for the
Comisión Nacional del Agua.
Ambroise, B., 2004. Variable ‘active’ versus ‘contributing’
Fuentes, O.A. and Franco, L.E., 1997. Modelo Matemático de
areas or periods: a necessary distinction. Hydrological
Áreas de Inundación. Cuadernos de Investigación. No. 41.
Processes, 18, 1149–1155. doi:10.1002/hyp.5536/abstract
Mexico City: Mexico Centro Nacional de Prevención de
Balthazar, J.M., Batista Goncalves, P., and Brasil, R.M.R.L.F.,
Desastres (in Spanish).
2009. Uncertainties in nonlinear structural dynamics.
Fuentes-Mariles, O.A., et al., 2000. Simulación numérica del
Mathematical Problems in Engineering, 2008, 1–4.
flujo en cauces y en las planicies que se podrían inundar de
doi:10.1155/2008/538725
la cuenca baja del río Grijalva, Estado de Tabasco. Mexico
Bates, P.D. and De Roo, A.P.J., 2000. A simple raster-based model
City (in Spanish): Instituto de Ingenieria Project for the
for flood inundation simulation. Journal of Hydrology, 236, 54–
Comisión Nacional del Agua.
77. doi:10.1016/S0022-1694(00)00278-X
Fuentes-Mariles, O.A., et al., 2013. Estudio de Inundaciones
Beven, K. and Davies, J., 2015. Velocities, celerities and the
Fluviales y Mapas de Peligro para el Atlas Nacional de Riesgo
basin of attraction in catchment Response. Hydrological
por Inundaciones Instituto de Ingenieria Project for the
Processes, 29, 5214–5226. doi:10.1002/hyp.10699
Subdirección General Técnica. México City (in Spanish):
Blöschl, G. and Sivapalan, M., 1997. Process controls on
Comisión Nacional del Agua.
regional flood frequency: coefficient of variation and
González, F.J., et al., 2009. Plan Hídrico Integral de Tabasco dentro
basin scale. Water Resources Research, 33 (12), 2967–
del cual se inscribe la ejecución del Plan de Acción Urgente
2980. doi:10.1029/97WR00568
(PAU) y la formulación del Plan de Acción Inmediata (PAI),
Blum, M.D., 2007. Large river systems and climate change.
para la rehabilitación de la infraestructura dañada por las
In: A. Gupta, ed. Large rivers: geomorphology and
lluvias atípicas. Mexico City SGIH-GDTT-OP-IIUNAM-
Management. Chichester: Wiley and Sons, Ltd, 627–656.
20.07.09-04 (in Spanish): Instituto de Ingenieria Project for
Brandt, S.A., 2016. Modeling and visualizing uncertainties of
the Comisión Nacional del Agua.
flood boundary delineation: algorithm for slope and DEM
Gupta, A., 2010. The hazardouness of high-magnitude floods. In:
resolution dependencies of 1D hydraulic models.
I. Alcántara-Ayala and A.S. Goudie, eds. Geomorphological
Stochastic Environmental Research and Risk Assessment,
Hazards and disaster prevention. Cambridge: Cambridge
30, 1677–1690. doi:10.1007/s00477-016-1212-z
University Press, 97–107.
Costabile, P., et al., 2012. Two-dimensional model for over-
Gupta, V. and Waymire, E., 1998. Some mathematical aspects of
land flow simulations: A case study. European Water, 38,
rainfall landforms and floods. In: O.E. Barndorff-Nielsen, et al.,
13–23, ISSN (electronic version) 1792-085X.
eds. Stochastic methods in hydrology, rainfall, landforms and
Courty, L.G., Rico-Ramirez, M.A., and Pedrozo-Acuña, A.,
flood. Advanced series on statistical science & applied probabil-
2018. The significance of the spatial variability of rainfall
ity. 7. Singapore: World Scientific, 129–171.
on the numerical simulation of urban floods. Water, 10,
207. doi:10.3390/w10020207 Haile, A.T., 2005. Integrating hydrodynamic models and high
Crowder, R.A., et al., 2004. Benchmarking hydraulic river resolution DEM (LIDAR) for flood modelling, master of science
modelling software packages. Results – test C (triangular in geo-information science and earth observation, specialisation:
channels). Defra/Environment Agency Flood and Coastal watershed management, conservation and river basin planning
Defence R&D Programme. R&D Technical Report: W5- March. Enschede: International Institute For Geo-Information
105/TR2C, Product Code SCHO0305BIXR-E-P, Bristol. Science And Earth Observation.
De Luna, F., 2015. Modelo hidráulico de flujo bidimensional Hudson, P.F., Sounny-Slitine, M.A., and LaFevor, M., 2013.
para estimar el escurrimiento a partir de la precipitación. A new longitudinal approach to assess hydrologic connec-
Thesis (PhD Candidate). Universidad Nacional Autónoma tivity: embanked floodplain inundation along the lower
de México, Mexico (in Spanish). Mississippi River. Hydrological Processes, 27, 2187–2196.
Domínguez-Mora, R., De Luna, F., and Gómez-Aguilar, E., doi:10.1002/hyp.9838
2014. Simulación Numérica del Comportamiento Jenkins, K., et al, 2016. Assessing surface water flood risk and
Hidráulico del canal Pescaditos. Mexico (in Spanish): management strategies under future climate change: an
XXIII Congreso Nacional de Hidráulica Puerto Vallarta. agent-based model approach [online]. Centre for Climate
Erpicum, S., et al., 2010. Detailed inundation modelling Change Economics and Policy (Working Paper No. 252)
using high resolution DEMs. Engineering Applications of and Grantham Research Institute on Climate Change and
Computational Fluid Mechanics, 4 (2), 196–208. the Environment (Working Paper No. 223). Available
doi:10.1080/19942060.2010.11015310 from: http://www.lse.ac.uk/GranthamInstitute/publica
Fiorentino, M., et al., 2006. Analysis on flood generation tion/assessing-surface-water-flood-risk-and-management-
processes by means of a continuous simulation model. strategies-under-future-climate-change-an-agent-based-
Advances in Geosciences, 7, 231–236. doi:10.5194/adgeo- model-approach/ [Accessed 24 November 2016].
7-231-2006 Jothityangkoon, C., Sivapalan, M., and Farmer, D.L., 2001.
Fiorentino, M. and Iacobellis, V., 2001. New insights about Process controls of water balance variability in large semi-
the climatic and geologic control on the probability arid catchment: downward approach to hydrological model
HYDROLOGICAL SCIENCES JOURNAL 317

development. Journal of Hydrology, 254 (1–4), 174–198.d. Néelz, S. and Pender, G., 2013. Delivering benefits through
doi:10.1016/S0022-1694(01)00496-611fc9b06b923&acd- evidence. Benchmarking the latest generation of 2D hydrau-
nat=1540575040_0dd58e8c3b70c6cd48f1e6827c5aeb20 lic modelling packages. Environment Agency/Defra Flood
Kusumastuti, D.I., 2006. Effects of threshold nonlinearities on and Coastal Erosion Risk Management Research and
the transformation of rainfall to runoff to flood in a lake Development Programme, Department for Environment
dominated catchment system. Thesis (PhD), University of Food and Affairs. Bristol, UK: Environment Agency,
Western Australia. Report SC120002.
Kusumastuti, D.I., et al., 2007. Threshold effects in catchment Nippgen, F., McGlynn, B.L., and Emanuel, R.E., 2015. The
storm response and the occurrence and magnitude of flood spatial and temporal evolution of contributing areas.
events: implications for flood frequency. Hydrology and Earth Water Resources Research, 51, 4550–4573. doi:10.1002/
System Sciences, 11, 1515–1528. doi:10.5194/hess-11-1515- 2014WR016719
2007 Ogden, F.L., 2015. A new general 1-D vadose zone flow
Mahmood, K. and Yevjevich, V., 1975. Unsteady flow in open solution method. Water Resources Research, 51.
channels, Vol 1. Fort Collins, CO: Water Resources doi:10.1002/2015WR017126
Publications. Robinson, J.S. and Sivapalan, M., 1997. Temporal scales and
Marks, K. and Bates, P.D., 2000. Integration of high-resolution hydrological regimes: implications for flood frequency
topographic data with floodplain flow models. Hydrological scaling. Water Resources Research, 33 (12), 2981–2999.
Processes, 14, 2109–2122. doi:10.1002/1099-1085(20000815/ doi:10.1029/97WR01964
30)14:11/12<2109::AID-HYP58>3.0.CO;2-1 Robinson, J.S., Sivapalan, M., and Snell, J.D., 1995. On the
McDonnell, J.J. and Beven, K., 2014. Debates – the future of relative roles of hillslope processes, channel routing, and
hydrological sciences: A (common) path forward? A call to network geomorphology in the hydrologic response of
action aimed at understanding velocities, celerities, and natural catchments. Water Resources Research, 31 (12),
residence time distributions of the headwater hydrograph. 3089–3101. doi:10.1029/95WR01948
Water Resources Research, 50, 342–5350. doi:10.1002/ Sherman, L.K., 1932. Streamflow from rainfall by unit-graph
2013WR015141 method. Engineering News-Record, 108, 501–505.
Middelkoop, H. and Asselman, N.E.M., 1998. Spatial varia- Sherman, L.K., 1942. Hydrograph of runoff. In: O.E.
bility of floodplain sedimentation at the event scale in the Meinzer, ed. Physics of the Earth, IX, Hydrology. New
rhine–meuse Delta, The Netherlands. Earth Surface York: McGraw-Hill.
Processes and Landforms, 23, 561–573. doi:10.1002/(SICI) Spence, C. and Woo, M.K., 2006. Hydrology of subarctic
1096-9837(199806)23:6<561::AID-ESP870>3.0.CO;2-5 Canadian Shield: heterogeneous headwater basins.
Neal, J., et al., 2012. How much physical complexity is needed to Journal of Hydrology, 317, 138–154. doi:10.1016/j.
model flood inundation? Hydrological Processes, 26 (15), 2264– jhydrol.2005.05.014
2282. doi:10.1002/hyp.8339 Strahler, A.N., 1964. Quantitative geomorphology of drai-
Néelz, S. and Pender, G., 2010. Benchmarking of 2D hydraulic nage basins and cannel networks. In: V.T. Chow, ed.
modelling packages. Environment Agency/Defra Flood and Handboodk of applied hydrology. New York: McGraw
Coastal Erosion Risk Management Research and Hill, 4.40–4.74.
Development Programme, Department for Environment Wohl, E.E., 2007. Hydrology and discharge. In: A. Gupta, ed.
Food and Affairs. Bristol, UK: Environment Agency, Large rivers: geomorphology and management. Chichester:
Report SC080035/R2. Wiley and Sons, Ltd., 27–44.

You might also like