Professional Documents
Culture Documents
Main
Main
Benchmarking urban ood models of varying complexity and scale using high
resolution terrestrial LiDAR data
Timothy J. Fewtrell a,, Alastair Duncan b, Christopher C. Sampson c, Jeffrey C. Neal c, Paul D. Bates c
a
Willis Research Network, School of Geographical Sciences, University of Bristol, BS8 1SS, UK
Geomatics Group, Environment Agency, Phoenix House, Lower Bristol Road, Bath, BA2 9ES, UK
c
School of Geographical Sciences, University of Bristol, BS8 1SS, UK
b
a r t i c l e
i n f o
Article history:
Received 28 October 2010
Received in revised form 6 December 2010
Accepted 14 December 2010
Available online 21 December 2010
Keywords:
Hydraulic modelling
LYNX
Terrestrial LiDAR
Urban ooding
Surface water
Urban drainage
a b s t r a c t
This paper describes benchmark testing of a diffusive and an inertial formulation of the de St. Venant
equations implemented within the LISFLOOD-FP hydraulic model using high resolution terrestrial LiDAR
data. The models are applied to a hypothetical ooding scenario in a section of Alcester, UK which experienced signicant surface water ooding in the June and July oods of 2007 in the UK. The sensitivity of
water elevation and velocity simulations to model formulation and grid resolution are analyzed. The differences in depth and velocity estimates between the diffusive and inertial approximations are within
10% of the simulated value but inertial effects persist at the wetting front in steep catchments. Both models portray a similar scale dependency between 50 cm and 5 m resolution which reiterates previous ndings that errors in coarse scale topographic data sets are signicantly larger than differences between
numerical approximations. In particular, these results conrm the need to distinctly represent the camber and curbs of roads in the numerical grid when simulating surface water ooding events. Furthermore,
although water depth estimates at grid scales coarser than 1 m appear robust, velocity estimates at these
scales seem to be inconsistent compared to the 50 cm benchmark. The inertial formulation is shown to
reduce computational cost by up to three orders of magnitude at high resolutions thus making simulations at this scale viable in practice compared to diffusive models. For the rst time, this paper highlights
the utility of high resolution terrestrial LiDAR data to inform small-scale ood risk management studies.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
Recent studies on the effects of urbanization on the hydrologic
response of drainage networks have examined the impact of urban
drainage on ooding in terms of drainage network structure
(Meierdiercks et al., 2010), drainage network efciency (Aronica
and Lanza, 2005), drainage pathway distribution (Leito et al.,
2009) and model resolution (Schubert et al., 2008; Fewtrell et al.,
2008). The proliferation of recent modelling efforts is a direct consequence of large pluvial ood events over urban areas (e.g. Dead
Run in Baltimore, US in July 2004 (Ntelekos et al., 2008) or ooding
in Hull, UK in summer 2007) and the associated perceived increased risk from such high rainfall events. Indeed, the Environment Agency of England and Wales (EA) estimated that twothirds of the 57,000 homes affected in the June and July 2007
oods in the UK were ooded from surface water runoff exceeding
the capacity of the drainage system (DEFRA, 2008). In addition, the
Pitt Review (Pitt, 2008) noted that although the UKs understanding of ooding risks from coastal and uvial sources is well
Corresponding author.
E-mail address: t.fewtrell@bristol.ac.uk (T.J. Fewtrell).
1474-7065/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.pce.2010.12.011
282
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
served ood event in the area. Therefore, the inow boundary conditions for this test case were derived using the depthduration
frequency curves for estimating rainfall from the Flood Estimation
Handbook (FEH, Institute of Hydrology (1999)). The EA in England
and Wales is currently mapping surface water ooding risk using a
1-in-200 year return period 30-min rainfall storm. For this study,
we assume that the 200-year 30-min rainfall (47 mm) is collected
over a drainage area of 100 100 m upstream of the inow point
(see Fig. 1) to represent the ow coming from a blocked sewer
draining a small catchment (Fig. 2). The nal assumption in this
study is that the drainage system is operating at capacity such that
water on the surface does not interact with the drains at the road
side. The lack of observed data of the ooding at this test site
means that the ensuing modelling exercise takes the form of a sensitivity analysis.
3. Modelling framework
Hunter et al. (2008) demonstrated the need to benchmark
hydraulic models of varying complexity to fully understand the effect of process representations on the simulation of ood ows
through urban environments. Further evidence from the authors
suggests it proved difcult to ensure that each model interpreted
the model inputs and boundary conditions in the same way. Similarly, work on the EA benchmarking study (Pender and Nelz,
2010) suggests participants encountered similar problems of
ensuring consistency between models in the construction of each
test case. Incorporating different numerical solution schemes into
a single computational code reduces any such ambiguity in model
setup. Therefore, the LISFLOOD-FP model is used here as there are a
number solution schemes of varying complexity implemented
within the same numerical code (see Bates and De Roo, 2000; Hun-
283
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
Fig. 1. MasterMap of Alcester with 10 cm LYNX data overplotted in the model domain highlighting the location of the sewer surcharge inow point and the control points.
Q txi;j
hflow
i;j
hi1;j hi;j
Dx
!1=2
Dy
where Q is the discharge [L3 T1], hi,j is the depth of water in cell
(i, j), how is the depth of water available to ow [L] and Dt is the
model time step [T], Dx is the model grid resolution [L] and n is
the Mannings friction coefcient [L1/3 T]. Water depths are subsequently updated using the nite difference discretised form of the
continuity equation using discharge between cells, as:
tDt
hi;j
Fig. 2. Inow boundary conditions.
ter et al., 2005; Bates et al., 2010). The numerical schemes evaluated in this study are detailed below.
hi;j Dt
284
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
result, Hunter et al. (2005) proposed a stability criterion for the LISFLOOD-FP implementation of Mannings equation based on an analogue of the diffusion wave equation, as in Eq. (3):
Dt
1=2
1=2 !
Dx2
2n Dh
2n Dh
min 5=3 ; 5=3
4
hflow Dx
hflow Dy
qt ghflow Dt DhDxz
qtDt
10=3
t
t
1 ghflow Dtn2 jqt j= hflow
where q is the ow per unit width [L2 T1], g is the acceleration due
to gravity [L T2], how is the depth of water available to ow [L] and
Dt is the model time step [T], Dx is the model grid resolution [L], z is
the bed elevation [L], h is the water depth [L] and n is the Mannings
friction coefcient [L1/3 T]. Water depths are updated using Eq. (2)
above. The stability criterion for this numerical model is given by
the CourantFreidrichsLevy condition for shallow water ows
such that the stable model time step, Dt, is a function of the grid
resolution, Dx, and the maximum water depth, h, within the domain (Bates et al., 2010):
Dx
Dtmax a p
ght
where a varies between 0.2 and 0.7. Unlike previous diffusive storage cell models (e.g. Hunter et al., 2005), this stability constraint
yields time steps large enough that high resolution models of the
urban environment are computationally tractable (Bates et al.,
2010).
4. Results and discussion
Initially, the diffusive and inertial models were constructed at
Dx = 10 cm, 25 cm, 50 cm, 1 m, 2 m and 5 m resolutions. However,
due to computational expense of the diffusive formulation (Hunter
et al., 2005), the 10 and 25 cm models were excluded from the
analysis. Fig. 3 shows the propagation of the ood wave over the
50 cm terrestrial LiDAR DEM at four times (360, 900, 1800 and
7200 s) for the diffusive and inertial formulations of LISFLOOD-FP
and the difference between the two approximations. The water enters the domain at the simulated blocked drain in the north east
corner and ows down the main northsouth facing street. At
the outset, the ood is contained within the curbs; indeed the
water ows down one side of the road initially before spilling over
the raised central section and then owing down the other side of
the road. This ow pattern is due to the retention of the road cam-
ber within the DEM which has never before been captured by elevation products used in hydraulic modelling of oods. As the water
reaches the road junctions, it spreads out and ows down the roads
perpendicular to the main road before ponding in the low points
and at the extremities of the DEM. The ponding at the DEM edges
occurs because the boundaries of this model are specied as closed
and therein lies the inherent disadvantage of using terrestrial LiDAR products derived from a system mounted on a vehicle as data
only exists in line of sight from areas accessible by road. As such,
there is clear scope to fuse terrestrial LiDAR products with other
sources of digital elevation data such as airborne LiDAR or IfSAR
products to resolve topographic features outside the street
network.
As the ood wave propagates through the street network, there
are differences, albeit small, in the simulated water depths between the diffusive and inertial model. At each stage of the simulation, it is clear that the diffusive approximation is deeper at the
wetting front than the inertial model. The inertia that the body
of water needs to be overcome in the inertial formulation, not present in the diffusive form, to activate ow thus leading to the shallower water depths observed in Fig. 3. This physical attribute also
accounts for the deeper water at the top of the slope in the inertial
model after 900 s of simulation time. Once the inertia has been
overcome, water ows more readily down the street network creating shallower water depths at the northern extent of the domain
and deeper water on the roads after 1800 s. At the end of the simulation, the diffusive formulation is deeper by up to 0.015 m (2%
of the depth) throughout most of the domain but the inertial model
is up to 0.041 m deeper (6% of the depth) in the pond at the lowest point in the domain. The increased momentum caused by the
gravity terms in the inertial model (see Eq. (4)) is likely to reduce
the diffusive behaviour of the model and reduce ow perpendicular to the main slope and thus causing the deeper ponding at the
southern edge of the domain. The differences between the diffusive
and inertial model formulations throughout the entire simulation
are less than the errors associated with the underlying LiDAR data
from which the 50 cm DEM is derived. Indeed, the differences are
of similar relative magnitude to those found in previous studies
of these two formulations (Bates et al., 2010), but are signicantly
smaller than the differences found between other simplied models and models based on the full shallow water equations (Hunter
et al., 2008). Nevertheless, evidence from the latter study suggests
that working with models within the same numerical code will
minimise any differences in data interpretation and solely highlight differences between the numerical approximations, thus
allowing more concrete conclusions to be formed regarding model
capabilities and deciencies.
In order to analyze the global effect of model resolution on simulation results, the root mean squared difference (RMSD) between
the coarse models (1 m, 2 m and 5 m) and the benchmark, high resolution (50 cm) model for the diffusive and inertial models is computed (Fig. 4). As noted in previous studies on the inuence of
model resolution (Horritt and Bates, 2001; Yu and Lane, 2006a;
Fewtrell et al., 2008), there is noticeable reduction in model performance at coarse resolutions up to 0.065 m RMSD over the entire
domain in this case. The 1 and 2 m resolution models are within
the typical vertical error of terrestrial LiDAR data (5 cm RMSE)
whereas the 5 m model produces differences greater than this LiDAR error. Indeed, at 5 m resolution, the global RMSD is greater
than 10% of the maximum simulated water depth and of similar
magnitude to the errors found in Bates et al. (2010) for a wave
propagating over a planar surface and up a planar beach. As in
Fewtrell et al. (2008), the RMSD is actually describing differences
in relative submergence as the slopes in the topography are significantly greater than the gradient of the water surface. Furthermore,
the differences between the diffusive and inertial model formula-
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
285
Fig. 3. Progression of ooding predicted by (a) the diffusive model and (b) the inertial model at various times throughout the simulation using the 50 cm DEM. The difference
between the ood depths from the two formulations is shown in panel (c).
286
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
(a)
(b)
Fig. 4. Evolution of the root mean squared difference between the benchmark Dx = 50 cm model and the coarser 1 m, 2 m and 5 m models throughout the simulation for the
(a) diffusive and (b) inertial model formulations.
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
287
Fig. 5. (Top row) plots of simulated water depth over time at the four control points at Dx = 50 cm, 1 m, 2 m and 5 m using the diffusive formulation. Other plots (rows 24)
show simulated water depth from the plots on the top row minus simulated water levels from the other resolutions.
Fig. 6. Plots of simulated water elevation over time at the four control points at the four grid resolutions. The diffusive approximation is represented by a solid line and the
inertial approximation is represented by a dashed line. Note that when the dotted line is not visible, it is plotting directly below the solid line.
288
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
(a)
(b)
(c)
(d)
(e)
Fig. 7. Detailed view of road junction in the four different resolution DEMs (ad) with the OS MasterMap road network vector layer overplotted and an indicative
cross-section of the road and curbs in panel (e). Note the generalization of the road network in coarse resolution DEMs.
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
289
Fig. 8. (Top 2 rows) Simulated velocity over time at the four control points at the four grid resolutions where the top row is the diffusive approximation and the second row is
the inertial approximation. The third row plots the velocity from the inertial approximation minus the diffusive approximation. Note the difference in scale on panel (c) of
row 3 compared to the other panels of row 3.
the maximum scalar velocity. Immediately noticeable is the similarity in predicted velocities between the diffusive and inertial formulations across all model resolutions. Indeed, with the exception
of points 2 and 4 at 2 m resolution, the difference in velocity between the two approximations is less than 10% of the predicted
velocity although this is not altogether unsurprisingly given the
similarity in simulated water elevations. The difference in velocity
is most apparent during the wetting phase where the diffusive
wave approximation predicts lower velocities which are commensurate with the deeper water depths at the wetting front highlighted in Fig. 3. At point 1, there is also a clear scale dependence
of velocity calculations whereby the 1, 2 and 5 m signicantly
over-estimate velocities compared to the 50 cm model in both
the diffusive and inertial codes. Considering the shape and scale
dependence of the road cross-sections (see for example Fig. 7e),
the ow area around point 1 is considerably constrained in coarse
resolution models causing a higher velocity through the road section. However, the opposite occurs at point 3 where the 1, 2 and
5 m models progressively under-estimate the velocity. The aggregation of topography to these coarser resolutions creates a smoother DEM surface in this region (Fig. 7e) which reduces the inuence
of the road camber in constraining water ow. This creates a greater ow area which reduces the velocity and increases water depths
at coarse scales. Therefore, there is clearly a spatially varying inuence of model resolution that will be site and DEM specic and unknown a priori. Although the 1 and 2 m resolution models appear
to provide robust estimates of water elevation, risk estimates considering some function of velocity and depth may be inconsistent
at such scales; a result also noted in a study of a larger suite of
numerical models (Pender and Nelz, 2010).
The evidence base developed in this study, allied with experience from other similar modelling work (e.g. Yu and Lane, 2006a;
Schubert et al., 2008; Hunter et al., 2008; Pender and Nelz,
2010), suggests that although there are localized differences and
scale dependencies, simplied numerical models provide a viable
alternative to solutions of the 2D de St. Venant equations in terms
of simulating ood depths and velocities. Therefore, in terms of
practical use, the key aspect lies in computational speed to deliver
simulation results in a timely fashion for industrial applications.
Fig. 9a shows the evolution of stable model time step throughout
the simulation and Fig. 9b shows the computation time required
for this test case for each model resolution. As noted in Bates
et al. (2010), the stable time step in the diffusive formulation is
several orders of magnitude smaller than at the same point in
the simulation in the inertial formulation. This translates into a
three orders of magnitude increase in computation time at the
highest resolution and a 10 increase in computation time even
at the 5 m scale. Indeed, the 50 cm inertial model completes in
9 min on eight processors compared to the 8400 min required
for the 50 cm diffusive model using the OpenMP version of the
code (Neal et al., 2009). At this site, the 1 m inertial model completes in 1.5 min compared to the 1 min for the 5 m diffusive
290
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
(a)
(b)
Fig. 9. Panel (a) shows the variation in stable time step (Dt) over time for the diffusive approximation (solid line) and the inertial approximation (dotted line). Panel (b) plots
the computation time for the various resolution models for the diffusive and inertial approximations. Note the log y-axis on both panels.
model indicating that the inertial formulation provides manageable runtimes at the scales required for detailed ood risk assessment which the diffusive formulation does not.
5. Conclusions
This paper presents a benchmarking and sensitivity analysis of a
two simplied formulations of the de St. Venant equations in the
LISFLOOD-FP hydraulic model on high resolution terrestrial LiDAR
data applied to a surface water ood event in Alcester, UK. To the
authors knowledge, this represents the rst attempts at conducting hydraulic modelling using sub-metre scale elevation data derived from terrestrial LiDAR systems collected at the 410 cm
scale. The diffusive and inertial forms of the de St. Venant equations deliver similar estimates of ood elevations throughout the
simulation, regardless of model resolution, such that the differences in water elevation are less than the errors associated with
the terrestrial LiDAR data. Furthermore, these differences in water
elevation are similar in relative magnitude to those found in previous analysis of diffusive and inertial approximations (Bates et al.,
2010) and signicantly smaller than those found when comparing
diffusive and dynamic hydraulic models (Hunter et al., 2008).
Evaluating the scale dependence of the two LISFLOOD-FP formulations point to a step change in model performance at between
2 and 5 m grid resolution which is conrmed as a consequence of
the degradation of the road network and camber representation in
coarse grids. This result is consistent with results from previous
studies which suggest three computational cells are required
across a street in order to provide accurate simulation results in
a timely fashion (Gallegos et al., 2009). As highlighted in Bates
et al. (2010), the computational cost of high resolution model simulations is considerably reduced in the inertial formulation compared to the diffusive approximation making possible the use of
terrestrial LiDAR data in practical applications. A comparison of
velocity simulations in this study shows that the difference between the diffusive and inertial approximations is generally less
than 10% of the predicted velocity but that this difference is most
T.J. Fewtrell et al. / Physics and Chemistry of the Earth 36 (2011) 281291
Bates, P.D., Horritt, M.S., Fewtrell, T.J., 2010. A simple inertial formulation of the
shallow water equations for efcient two dimensional ood inundation
modelling. Journal of Hydrology 387, 3345.
Bradbrook, K.F., Lane, S.N., Waller, S.G., Bates, P.D., 2004. Two dimensional diffusion
wave modelling of ood inundation using a simplied channel representation.
International Journal of River Basin Management 2, 113.
Cobby, D.M., Mason, D.C., Horritt, M.S., Bates, P.D., 2003. Two-dimensional hydraulic
ood modelling using a nite-element mesh decomposed according to
vegetation and topographic features derived from airborne scanning laser
altimetry. Hydrological Processes 17, 19792000.
DEFRA, 2008. Future Water: The Governments Water Strategy for England. CM7319
London, p. 98.
Estrela, T., Quintas, L., 1994. Use of a GIS in the modelling of ows on oodplains. In:
White, W.R., Watts, J. (Eds.), 2nd International Conference on River Flood
Hydraulics. John Wiley and Sons Ltd., Chichester, pp. 177190.
Fewtrell, T.J., Bates, P.D., Horritt, M., Hunter, N.M., 2008. Evaluating the effect of
scale in ood inundation modelling in urban environments. Hydrological
Processes 22, 51075118.
Fewtrell, T.J., Bates, P.D., de Wit, A., Asselman, N., Sayers, P., 2009. Comparison of
varying complexity numerical models for the prediction of ood inundation in
Greenwich, UK. In: Samuels, P., Huntington, S., Allsop, W., Harrop, J. (Eds.),
FLOODrisk2008 Flood Risk Management: Research and Practice. Taylor &
Francis Group, Keble College, Oxford.
Gallegos, H.A., Schubert, J.E., Sanders, B.F., 2009. Two-dimensional, high-resolution
modeling of urban dam-break ooding: a case study of Baldwin Hills, California.
Advances in Water Resources 32, 13231335.
Horritt, M.S., Bates, P.D., 2001. Effects of spatial resolution on a raster based model
of ood ow. Journal of Hydrology 253, 239249.
Horritt, M.S., Bates, P.D., 2002. Evaluation of 1D and 2D numerical models for
predicting river ood inundation. Journal of Hydrology 268, 8799.
Hsu, M.H., Chen, S.H., Chang, T.J., 2000. Inundation simulation for urban drainage
basin with storm sewer system. Journal of Hydrology 234, 2137.
Hunter, N.M., Horritt, M.S., Bates, P.D., Wilson, M.D., Werner, M.G.F., 2005. An
adaptive time step solution for raster-based storage cell modelling of oodplain
inundation. Advances in Water Resources 28, 975991.
Hunter, N.M., Bates, P.D., Nelz, S., Pender, G., Villanueva, I., Wright, N.G., Liang, D.,
Falconer, R.A., Lin, B., Waller, S., Crossley, A.J., Mason, D.C., 2008. Benchmarking
2D hydraulic models for urban ooding. Proceedings of the Institution of Civil
Engineers Water Management 161, 1330.
Institute of Hydrology, 1999. Flood Estimation Handbook. Institute of Hydrology,
Wallingford.
Leito, J.P., Boonya-aroonnet, S., Prodanovic, D., Maksimovic, C., 2009. The inuence
of digital elevation model resolution on overland ow networks for modelling
urban pluvial ooding. Water Science & Technology 60, 31373149.
291
Lichti, D., Pfeifer, N., Maas, H.-G., 2008. ISPRS journal of photogrammetry and
remote sensing theme issue Terrestrial Laser Scanning. ISPRS Journal of
Photogrammetry and Remote Sensing 63, 13.
Mason, D.C., Cobby, D.M., Horritt, M.S., Bates, P.D., 2003. Floodplain friction
parameterization in two-dimensional river ood models using vegetation
heights derived from airborne scanning laser altimetry. Hydrological
Processes 17, 17111732.
Mason, D.C., Horritt, M.S., Bates, P.D., Hunter, N.M., 2007. Use of fused airborne
scanning laser altimetry and digital map data for urban ood modelling.
Hydrological Processes 21, 14361447.
Meierdiercks, K.L., Smith, J.A., Baeck, M.L., Miller, A.J., 2010. Analysis of urban
drainage structure and its impact on hydrologic response. Journal of the
American Water Resources Association 46, 932943.
Merz, B., Kreibich, H., Thieken, A., Schmidtke, R., 2004. Estimation uncertainty of
direct monetary ood damage to buildings. Natural Hazards and Earth System
Sciences 4, 153163.
Neal, J., Fewtrell, T., Trigg, M., 2009. Parallelisation of storage cell ood models using
OpenMP. Environmental Modelling & Software 24, 872877.
Nelz, S., Pender, G., 2006. The Inuence of Errors in Digital Terrain Models on Flood
Flow Routes. In: River Flow 2006. IAHR, 19551962.
Ntelekos, A.A., Smith, J.A., Baeck, M.L., Krajewski, W.F., Miller, A.J., Goska, R., 2008.
Extreme hydrometeorological events and the urban environment: dissecting
the 7 July 2004 thunderstorm over the Baltimore MD Metropolitan Region.
Water Resources Research 44, 119.
Pender, G., Nelz, S., 2010. Benchmarking of 2D Hydraulic Modelling Packages.
SC080035/R2 Environment Agency, Bristol, p. 169.
Pitt, M., 2008. Learning Lessons from the 2007 Floods. The Pitt Review Cabinet
Ofce London, p. 505.
Schubert, J.E., Sanders, B.F., Smith, M.J., Wright, N.G., 2008. Unstructured mesh
generation and landcover-based resistance for hydrodynamic modeling of
urban ooding. Advances in Water Resources 31, 16031621.
Yu, D., Lane, S.N., 2006a. Urban uvial ood modelling using a two-dimensional
diffusion-wave treatment, part 1: mesh resolution effects. Hydrological
Processes 20, 15411565.
Yu, D., Lane, S.N., 2006b. Urban uvial ood modelling using a two-dimensional
diffusion-wave treatment, part 2: development of a sub-grid-scale treatment.
Hydrological Processes 20, 15671583.
Zanobetti, D., Longer, H., Preissmann, A., Cunge, J.A., 1968. Le modele
mathmatique du delta du Mekong. La Houille Blanche, p. 23.
Zanobetti, D., Longer, H., Preissmann, A., Cunge, J.A., 1970. Mekong delta
mathematical model program construction. Journal of the Waterways and
Harbors Division 96 (WW2), 181199.