Professional Documents
Culture Documents
Abstract: Recent developments in flood modeling have led to the concept of coupled 共sewer/surface兲 hydraulic models. In this paper two
coupled models are examined; a one-dimensional 共1D兲 sewer model coupled with a 1D surface network model 共1D/1D兲 and a 1D sewer
model coupled with a two-dimensional 共2D兲 surface flow model 共1D/2D兲. Flow over the terrain is better modeled by 2D models, whereas
Downloaded from ascelibrary.org by DAPS LIBRARY on 08/17/13. Copyright ASCE. For personal use only; all rights reserved.
in confined channels 1D models provide a good approximation with less computational effort. This paper presents a comparison of the
simulation results of 1D/1D model and a 1D/2D model. The methodology adopted for setting up the models is outlined and explained in
detail as well as the 1D/1D modeling techniques used for reproducing the 1D/2D model results. The surface flow comparison clarifies the
limitations of the 1D/1D model and indicates that the definition of the surface pathways, the linking elements sewer/surface, and inclusion
of virtual manholes are key factors for setting up a more accurate 1D/1D model.
DOI: 10.1061/共ASCE兲HY.1943-7900.0000037
CE Database subject headings: Floods; Hydraulic models; Two-dimensional; Overland flow; Sewers; Urban areas.
Rainfall events ⫻ ⫻
共5兲 Calibration DOF ⫻ —
冑
fusion wave兲 flow equations 共Chen et al. 2005兲. The UIM uses a
共hpkj − hskj − Hi − Hsh兲 two-step alternating direction explicit numerical scheme which is
Qcs5 = − K · A p · R2/3 共5兲
Lp conditionally stable, such that the time step is limited by the
Courant–Friedrichs–Lewy 共CFL兲 condition. Because the 1D/2D
where Qcsi = discharge flow through the control section i;
model is not fully coupled, a time synchronization technique is
hskj = water depth at the surface; hpkj = water depth at the sewer;
used for accounting for the different time steps in both models. In
Li = perimeter length of inlet box 共0.8 m兲; Hi = inlet box height
this way the exact variable values are exchanged at the same time.
共0.40 m兲; Hsh = height of the vertical shaft connecting the inlet
A weir or an orifice links the 1D sewer network to the 2D surface
box to the manhole 共0.60 m兲; L p = length of the horizontal shaft
system, depending on the flow direction. Chen et al. 共2007兲 define
共5 m兲, A p = area of the connecting pipe; R = hydraulic radius
the orifice/weir discharge as a function of five variables:
共80 mm of diameter兲; Cd = discharge coefficient 共0.5兲; and K
共HpK.D
j j
, Hsk.U j
, zlevel , Amj , w j兲, HpK.D
j j
and Hsk.U are the dependent
= Strickler roughness coefficient 共80 m1/3 / s兲. The values in paren-
variables, water level downstream, and upstream of the link j,
theses are used as defaults because that information is not avail-
respectively, at the sewer pipe and at the surface nodes, where Amj
able. The discharge Qkj of the SLE is determined by
is the manhole area at a given link j.
Qkj = min兵Qcs1,Qcs2,Qcs3其 共6兲 Table 2 summarizes the main characteristics of the two mod-
els. The two models have features common to available commer-
Qkj = max兵Qcs4,Qcs5其 共7兲 cial and academic software 共e.g., the models mentioned in the
“Introduction”兲 such that the qualitative conclusions of this paper
Eq. 共6兲 or Eq. 共7兲 is selected depending on the flow direction. The should also apply to those models.
first is used if the flow is from the surface to the sewer, and
the second is used if the flow is from the sewer to the surface. The
MLE is obtained as the product of Qkj and the number of connec- Methodology
tions 共SLEs兲 to each manhole. In sum, the MLE is a function In order to enable a meaningful comparison between the 1D/1D
of four main variables: 共Hskj , Hpkj , Neq j j
, Rcd 兲. Hskj and Hpkj and the 1D/2D, the models are built consistently. Table 1 summa-
= dependent variables, water levels at the surface nodes, and at the rizes the steps followed, where the cross indicates the need to use
j j
sewer pipe. Neq and Rcd = two parameters of the MLE, respec- the corresponding modeling element.
tively, the number of equivalent SLE and the coefficient to reduce 1. The first step deals with acquisition and handling of data.
instability 关the reader is referred to the work of Leandro et al. The same sewer network is used in both 1D/1D and 1D/2D
Results
Downloaded from ascelibrary.org by DAPS LIBRARY on 08/17/13. Copyright ASCE. For personal use only; all rights reserved.
Table 2. Summary Two Models Coupling, Linkage, Governing Equations, Numerical Scheme, and Stability at Surface
Model Governing equations Numerical scheme Stability Coupling Linkage
1D/1D Full St. Venant equations Implicit Preissmann Unconditionally stable Fully coupled MLE and weir
1D/2D Full St. Venant equations Implicit Preissmann Unconditionally stable Partially coupled Weir and orifice
共sewer兲 共sewer兲 共sewer兲
Noninertia flow equations Explicit two step level AD Conditionally stable CFL
共surface兲 共surface兲 共surface兲
Fig. 8. Discharges at three selected linking points, before 共top row兲 and after 共bottom row兲 calibration
damage curve 共Dawson et al. 2008兲. Velocities enable creation of Fig. 10 shows water levels at Ponds 8031, 8027, 8015, 8020,
“flood hazard” maps where flood hazard is rated as low/moderate/ and 8022. This figure shows that the peak levels were generally in
high as a function of unit discharge 共product of depth and veloc- better agreement than the times of peak. The selected area is
ity兲. Depths are readily available from results of both models, severely flooded during the simulations as can be seen in Fig.
hence these are directly usable for comparison. However, velocity 10共d兲. The flow spreads over the whole area 共i.e., it is distinctly
is much more difficult to use for comparison because in the two dimensional兲. The 1D/1D model fails to predict the correct
1D/2D model it is defined by two components and typically var- flow paths and, therefore, the filling sequence of ponds.
ies more distinctly 共in space兲 than the velocity in the 1D/1D For Ponds 8015 and 8020 共Fig. 7 shows the modeling details兲
model; therefore, the arrival times of flood waves will be used the time and peak level for the first pond are both well predicted.
instead to infer the flood flow dynamics. The terms “lagging” and The 1D/1D model in the second pond shows the water level
“overshooting” will be used, respectively, to refer to the delay or slightly underestimated and a lagged peak time when compared to
early arrival of the flood wave. the 1D/2D model. One of the possible explanations for the dis-
Fig. 9 shows the water levels at Ponds 8050 and 8043, and the crepancy in Pond 8020 is the overestimation of the stage-area
surcharged Manholes 3024, 3079, and 3085 共the first two man- curve 共during the aggregation of smaller ponds, done by the GIS
holes coincide with the UPsID’s of the links shown in Fig. 8兲. The procedure兲, which would cause a peak delay and consequently
time and peak levels are well captured in both ponds and good prevent one from obtaining the correct peak level.
agreement is observed in the general shape of the water level Pond 8022 in the 1D/1D model shows an overshoot of both
profile. The length of the channels modeled downstream of the peak time and water level. At first it may appear that an incorrect
surcharged manholes accurately predict the routing flood hy- modeling assumption was used for linking Ponds 8020 and 8022
drograph from the manhole to the pond. Therefore, both models 共Fig. 7兲. However, if a channel is used for connecting the two
reproduced the timing of flood propagation well. The locations of ponds, it would result in an overlap of flooding areas due to their
ponds are in agreement with the flooded areas 关Fig. 9共d兲兴. Focus- edge proximity. The schematization error in this case may have
ing on Pond 8050, the analysis shows that after surcharging and a resulted from failing to consider the delay effect of filling up the
maximum peak is attained, the sewer starts to drain the water smaller ponds. The error could also be exaggerated due to the
back. It also shows that the crest levels in both models are in good absence of multiple exit points at Ponds 8020 and 8022 in the
agreement. Pond 8043 similarly shows good agreement; however 1D/1D model. A closer look into the 1D/2D results supports this
in this case the water just remains on the surface and does not explanation, showing that the filling of several small ponds pre-
return to the sewer. Pond 8052 does not receive any water, be- cedes the filling of the lowest point in the 2D model, where Pond
cause neither manhole 3025 surcharges, nor the weir connection 8022 is located in the 1D/1D model. Since only one large pond
with the neighboring Pond 8043 overflows. This figure shows was identified in the 1D/1D model 共i.e., Pond 8022兲 it directly
that, for the selected ponds, the 1D schematization of the ponds starts filling without having the delay effect caused by the sur-
and channels, as well as the modeling discretization used 共Fig. 5兲 rounding smaller ponds.
were indeed effective in reproducing the 2D features of the The modeling details type used for Pond 8027 are similar to
1D/2D model. those shown in Fig. 5 共i.e., the first type兲. The figure shows the
Fig. 10. 1D/1D and 1D/2D model results in Flooded area 2 共north-
Fig. 9. 1D/1D and 1D/2D model results in Flooded area 1 共south-
east兲: 共a兲 1D/1D water levels; 共b兲 1D/2D water levels; 共c兲 difference
west兲: 共a兲 1D/1D water levels; 共b兲 1D/2D water levels; 共c兲 difference
in wave peaks time of occurrence; and 共d兲 location of nodes and
in wave peaks time of occurrence; and 共d兲 location of nodes and
flooded area 2
flooded area 1
time to peak and the peak levels reasonably well predicted, how-
ever, the instant when the pond starts to fill is overshot in the modeling details 共Fig. 6兲. Because of the short distance between
1D/1D model. In this case, it may be concluded that a schemati- the pond exit and the manhole, the VM connects to the pond
zation error similar to that made earlier for Pond 8022 has again using a weir. The good prediction of the exact moment when the
been made. Nevertheless, the modeling detail used here seems to flooding starts is an indicator that there was no need to consider a
allow a good prediction of peak time and the volume-depth curve channel for connecting the manhole and the pond. However, an
to enable a good agreement on the peak level. overshoot of the peak level in the 1D/1D model indicates either
Pond 8031 shows a good general agreement in the profile an underestimation of the volume-depth curve or a need to in-
shape for both models. This pond illustrates the second case in the clude an extra exit point in Pond 8031.
In order to impose consistent boundary conditions for the sur- f kj ⫽ objective function;
face flow domain in both 1D/1D and 1D/2D approaches through- g ⫽ gravity force;
out the entire simulation, the flows discharged from the manholes Hi ⫽ inlet box height;
in both models are adjusted in a calibration procedure that makes Hpkj ⫽ water level at subsurface network nodes;
use of the MLEs potential to control and regulate the flows dis- j
HpK.D ⫽ water level downstream of link at sewer pipe;
charged from the sewer network. Due to the absence of dynami- Hskj ⫽ water level at surface network nodes;
cally measured flood data, we assumed that the use of a 2 m Hsk.Dj
⫽ water level downstream of link at surface;
⫻ 2 m grid in the DEM is sufficient to enable accurate modeling Hsk.Uj
⫽ water level upstream of link at surface;
using the 1D/2D model. Thereby the results of the 1D/2D could Hsh ⫽ height of vertical shaft connecting inlet box to
be used as data for calibrating the 1D/1D model. manhole;
The comparison results presented strong evidence for what the hpkj ⫽ water depth at subsurface network nodes;
key factors and recommendations in setting up an accurate 1D/1D hskj ⫽ water depth at surface network nodes;
model are. These can be summarized as follows: i ⫽ link ID;
1. The definition of a 1D surface network of pathways and K ⫽ Strickler roughness coefficient;
ponds based on a DEM: k ⫽ iteration number of DOF;
a. Ponds defined by stage-area curves 共by aggregation of Li ⫽ perimeter length of inlet box;
ponds within ponds if below a specified threshold兲, bot- Lp ⫽ length of vertical shaft 共of pipe connecting inlet
tom levels 共lowest point of the pond兲, and crest levels box to manhole兲;
共the first exit point of a pond兲; and MS,TS, ⫽ control parameters of DOF;
b. Flow pathways defined by lengths, paths, and slopes, Neq j
⫽ number of equivalent single-linking-elements;
based on analyzing the elevations of the neighboring Qcsi ⫽ discharge for given control section i;
cells 共from highest to lowest兲, and cross-sectional profile Qkj ⫽ discharge for given multiple-linking element;
defined by extending it perpendicular to the paths up to R ⫽ hydraulic radius of pipe connecting inlet box to
form a confined cross section 共according with the DEM兲; manhole;
2. To apply sensible elements for linking the sewer with surface Rcd j ⫽ coefficient to reduce instability;
共MLE兲, and that connection identified based on the proximity w j ⫽ weir crest width;
between manholes and pathways, and linking ponds to ponds zlevel ⫽ weir crest level;
j
and to pathways 共weirs兲 based on the shortest distances; and ␣kj ⫽ calibration parameter of DOF;
3. The inclusion of virtual manholes to bring the topographic  ⫽ number of previous iterations of DOF;
features in the 1D/2D model that control the sewer dis- ⌬x ⫽ grid size; and
charges into the 1D/1D model. ¯ j ⫽ grid length of DOF.
k
The outlined methodology has been successfully tested on a
detailed area of the Keighley sewer network. In this study case,
the results obtained by the 1D/1D model have shown consistency
with the 1D/2D model. However, they have also highlighted po- References
tential weaknesses suggesting questions that should be considered
in future research, such as the cost-benefit of lowering the DEM Aronica, G. T., and Lanza, L. G. 共2005兲. “Drainage efficiency in urban
resolution while keeping some of the main topographic features areas: A case study.” Hydrolog. Process., 19共5兲, 1105–1119.
and maintaining an acceptable computational requirement, and Bazaraa, M. S., Sherali, H. D., and Shetty, C. M. 共1993兲. Nonlinear
how to develop better visualization tools for representing the programming, 2nd Ed., Wiley, New York.
1D/1D results. Bolle, A., et al. 共2006兲. “Hydraulic modelling of the two-directional in-
teraction between sewer and river systems.” Proc., Urban Drainage
Modelling and Water Sensitive Urban Design, Monash Univ., Mel-
bourne, Australia.
Acknowledgments Boonya-Aroonnet, S., Maksimović, Č., Prodanović, D., and Djordjević,
S. 共2007兲. “Urban pluvial flooding: Development of GIS based path-
The research presented in this paper was funded by the UK ways model for surface flooding and interface with surcharged sewer
Engineering and Physical Sciences Research Council project model.” Proc., 6th NOVATECH International Conf., Workshop I,
Flood Risk Management Research Consortium 共FRMRC兲 Work Graie, Lyon, France, 481–488.
Package 6.1 共Grant No. GR/S76304/01兲. The writers are grateful Carr, R. S., and Smith, G. P. 共2006兲. “Linking of 2D and pipe hydraulic