Advances in Water Resources Volume 32, Issue 8, August 2009, Pages 1323-1335 doi:10.1016/j.advwatres.2009.05.008 | How to Cite or Link Using DOI Copyright © 2009 Elsevier Ltd All rights reserved. Cited By in Scopus (2) Permissions & Reprints Two-dimensional, high-resolution modeling of urban dam-break flooding: A case study of Baldwin Hills, California Humberto A. Gallegosa, Jochen E. Schubertb and Brett F. Sandersa, , , a Department of Civil and Environmental Engineering, University of California, Irvine, CA 92697-2175, USA b IESSG, The University of Nottingham, Nottingham NG7 2RD, UK Received 9 March 2009; revised 17 May 2009; accepted 19 May 2009. Available online 28 May 2009. Abstract Modeling of dam-break flooding in an urban residential area in southern California is presented. Modeling is performed using BreZo, an unstructured grid, Godunov-type, finite volume model that solves the shallow-water equations. The model uses terrain data from a 1.5 m Light Detection and Ranging (LiDAR) Digital Terrain Model (DTM) and contour data depicting the reservoir and breach geometry. A spatially distributed Manning coefficient based on a landcover classification derived from digital orthophotos and vector data (e.g., parcel outlines) is also used, and the interception of flow by storm drains is modeled with sink terms in the 2D continuity equation. The model is validated with flood extent and stream flow measurements, and a sensitivity analysis is completed to identify the necessary level of data and model complexity for accuracy purposes. Results show street depressions in the land surface should be resolved by the computational mesh for flood extent and stream flow accuracy. A ca. 5 m resolution mesh that spans streets by approximately 3 cells achieves a good balance between accuracy and computational effort. Results also show that heterogeneous resistance is important for stream flow accuracy, and the interception of overland flow by storm sewers is important for flood extent accuracy. The sensitivity of predictions to several additional factors such as the reservoir level, breach geometry and DTM source (LiDAR, National Elevation Data, Shuttle Radar Topography Mission Data) is also reported. Keywords: Urban hydrology; Flood inundation modeling; Shallow-water equations; Dam-break; Finite volume method; DTM; High resolution; LiDAR; National Elevation Data; Shuttle Radar Topography Mission Article Outline 1. Introduction 2. Materials and methods 2.1. Site description 2.2. Failure sequence 2.3. Data sources 2.4. Terrain modeling 2.5. Flood inundation modeling 2.6. Mesh generation and model parameterization 2.7. Initial conditions 3. Results 3.1. Validation of the flood prediction 3.2. Sensitivity analysis 4. Discussion 5. Conclusions Acknowledgements References 1. Introduction Urban flooding is becoming more frequent as a consequence of several factors including continued watershed development with impervious surfaces[15], population growth which places increasing pressure on communities to develop in flood prone areas [6] and [22], climate change which has magnified the intensity of rainfall [19], sea level rise which threatens coastal developments, and decaying or poorly engineered flood control infrastructure such as the levee system of California’s Sacramento-San Joaquin river delta [37]. Furthermore, the consequences of flooding are greater in urban versus rural sites due to the relative economic value and population density [22] and [33]. To manage the risk of flooding, damage assessments are needed and should consider not only economic but also social and environmental factors [41]. This can be accomplished by first applying hydraulic models to predict the depth and velocity distribution of probable floods, and then overlaying these data upon assets of an economic, social and environmental nature to quantify probable damages. Government at all levels is increasingly investing in Geographical Information Systems (GIS) to organize and efficiently utilize geospatial data for a diverse number of management and operational objectives. In Los Angeles County, for example, a consortium of public agencies known as the Los Angeles Area Imagery Acquisition Consortium (LAR-IAC) jointly funded the acquisition of several county-wide, high-resolution data sets including Light Detection and Ranging (LiDAR) terrain data, digital orthophotos, and oblique aerial photos. These data make it possible to resolve landscape geometry and surface features with a spatial resolution (ca. 1 m) and vertical accuracy (e.g., <10 cm RMSE) that is ideal for flood inundation modeling (FIM) [1], [2] and [29]. Damage estimates can subsequently be integrated in GIS. High-resolution modeling of urban flooding from a dam failure is the focus of this study. A twodimensional (2D) flood inundation model based on the shallow-water equations is applied and parameterized using LiDAR terrain data, digital orthophotos and other supporting data, and predictions of flood extent and stream flow are compared to observations for validation purposes. A number of 2D urban FIM studies have recently appeared in the literature. Researchers have addressed questions such as the necessary grid resolution [44], [6] and [33], resistance parameterization [26], [6] and [22], role of sub-surface storm drains [15], tradeoffs between shallowwater and diffusive wave routing schemes [17], and models for the impact of buildings on flood dynamics [44], [6], [20], [33] and [27] including use of porosity methods[44], [45], [12], [25], [34] and [31]. These studies have shown that urban flood flows manifest as a combination of sub- and super-critical flow along streets and between buildings, depending on street slopes and flood dynamics, and 2D models are poised to resolve these dynamics when important flow paths such as streets and gaps between buildings are resolved. This may require a grid resolution of 2 m or less [6] and [17] using structured grids or a variable resolution unstructured mesh that is constrained by building walls [33]. There remains a need for more urban FIM validation studies, to develop a sound understanding of good modeling practice (e.g., model formulation, data requirements, mesh resolution) and assess the overall predictability of urban flooding, particularly in the context of dam failures. Mignot et al. [26]simulated two flood events in Nimes, France where water is channeled along city streets. The model was found to accurately depict flood extent, as the site was bounded by steep topography, but relatively large root-mean-square (rms) errors in flood depth, ca. 50 cm or 50%, were reported despite efforts to calibrate model parameters. Neal et al. [27] modeled fluvial flooding of Carlisle, England using a considerably coarser resolution (25 m) than other urban flood modeling studies have suggested is necessary for resolving street flows and depicting building effects [6], [17] and [33]. After extensive calibration, model predictions of flood depth yielded smaller rms errors (ca. 30 cm) than the Mignot et al. study [26]. Valiani et al. [42] validated a 2D shallow-water model prediction of the Malpasset dam-break flood in France in 1959, but modeling was focused on the basin scale so smaller scale features germane to urban centers were not examined. Similarly, Begnudelli and Sanders [4] validated a 2D shallow-water model prediction of the St. Francis dam-break flood, but the flood zone was exclusively rural and the scale of predictions was relatively large compared to recent high-resolution studies of urban flooding (e.g., [26], [6], [20], [17] and [33]). Notwithstanding these differences in scale, the Malpasset and St. Francis applications show that Godunov-type finite volume shallow-water models perform well in practical applications, readily accommodating the challenge of transcritical over natural terrain with wetting and drying. This has motivated use of similar models in other FIM studies [43], [33] and [30]. Here, we present a high-resolution 2D FIM study of an urban dam-break flood that occurred in 1963 in the Baldwin Hills region of Los Angeles, California. Several key datasets have been obtained to support FIM including a LiDAR Digital Terrain Model (DTM), digital orthophotos of the study site, and post-disaster reports on the reservoir, its hydraulic infrastructure and the failure sequence [36] and [38]. Two types of field data have been obtained for validation purposes: (1) a survey of flood extent completed by the US Army Corps of Engineers (USACE) [38], and (2) stream flow data for the main channel below the flood zone, Ballona Creek [38]. To the knowledge of the authors, this represents the first attempt to validate a 2D urban dam-break flooding model that utilizes high-resolution data including LiDAR, aerial imagery and other miscellaneous vector data. The remainder of the paper is organized as follows, • Section 2 describes the hydrodynamic routing methodology including mesh generation, terrain and resistance parameterization, treatment of sub-surface storm drains, dam breach modeling, and model initialization. • Validation of the model is presented in Section 3.1, followed by a sensitivity analysis in Section 3.2 to identify the most important factors relative to model accuracy. • Section 4 provides a discussion of results, followed by conclusions in Section 5. 2. Materials and methods 2.1. Site description Baldwin Hills Reservoir was placed into service in 1951 by the Los Angeles Department of Water and Power (LADWP) for water supply purposes. The reservoir was situated on the north slopes of Baldwin Hills, approximately 3.2 km (2 miles) south of Interstate 10 and 4 km (2.5 miles) east of Interstate 405 as shown in Fig. 1. The topography of Baldwin Hills was ideal for water supply purposes, having sufficient elevation near the service area, though it was close to the Inglewood fault which at that time was one of the most active in California [36]. The reservoir capacity was 1,110,000 m3 (897 acre-ft) and the surface area was estimated at 79,200 m2 (20 acre) at the spillway crest, elevation 145.5 m (477.5 ft). The reservoir was rectangular in shape and encircled by an engineered embankment constructed of earthen materials and lined with asphaltic pavement, as shown in Fig. 2. The northern side of the embankment, or dam, rose 47.2 m (155.0 ft) above a ravine that would later become a channel of high velocity dam-break flood water. A spillway was located on the northeast corner of the reservoir and was designed with a drain pipe that would, if the reservoir was inadvertently over-filled, direct water to a catch basin just north of the reservoir. The reservoir was also engineered with an extensive drainage system designed to remove pore water which penetrated the asphaltic lining of the reservoir, e.g., through cracks resulting from differential settling. The drainage system directed water to the east side of the reservoir, where inlet and outlet works were also located (see Fig. 2) to support the water supply function of the reservoir, and to the north side of the reservoir. Hence, the dam and spillway were located on the northern side of the reservoir, while the water supply inlet and outlet works were on the eastern side of the reservoir. Full-size image (173K) Fig. 1. Baldwin Hills study area in Los Angeles, California including aerial imagery, observed flood extent, Ballona Creek gauging station location, and flood model boundary. Full-size image (161K) Fig. 2. Progression of the dam failure. Photographs reproduced with permission from Los Angeles Times. 2.2. Failure sequence December 14, 1963 began with approximately (790 acre-ft) in the reservoir and all systems operating normally. But at approximately 11:15 Pacific Standard Time (PST), the reservoir caretaker observed an unusual amount of drainage water coming from the northeast corner of the reservoir. By 12:15, the drainage had increased considerably and was observed to be muddy which indicated erosion of the dam. This water flowed east from the inlet/outlet works down a service road, then north along La Brea avenue. By 13:00, water was observed leaking from the east abutment of the dam, approximately 27 m (90 ft) below the spillway crest. And at 13:30, water was also leaking from a crack that had opened near the crest of the dam. These leaks would continue to grow over the next two hours before major breaching occurred. During this time, the region north of the dam was evacuated by emergency personnel, and LADWP personnel were taking all possible steps to reduce the volume of water in the reservoir. The inlet to the reservoir was closed, other reservoirs in the system were taken off-line to focus system demand on Baldwin Reservoir, and a number of “blow-off” valves in the water supply service area were opened to maximize outflow. LADWP estimates that of controlled flows were in place during this time. Efforts were also made to seal the crack in the dam (see 14:50 photo in Fig. 2), but it continued to widen. Between 15:00 and 15:15 the lower and upper leaks in the dam merged into one and formed an approximately 3 m (10 ft) wide breach. During this time, flows were contained in a catch basin just north of the dam. However, at 15:20 the flow through the breach increased considerably, the crack continued to widen, and by 15:30 the catch basin was overtopped. At 15:30, the final, major widening of the breach occurred as shown in Fig. 2. Video coverage of the event shows that the final breach caused a rarefaction wave in the reservoir, a signature feature of the so-called partial dam-break problem used for benchmark testing of hydrodynamic models. At 15:38, the roadway over the breach collapsed (Fig. 2) and failure was complete. LADWP estimates that approximately were in the reservoir (half of its capacity of 897 acre-ft) upon the second and final breach [8]. The flood impacted the area north of the dam that is bordered, roughly, by Santa Barbara Avenue to the East (since renamed Martin Luther King Blvd.), Jefferson Blvd. to the North, and Ballona Creek to the West as shown in Fig. 1. In addition, high velocities were reported on the ca. 7% slope below the dam, where homes were torn from their foundation and considerable erosion occurred. On more level ground further North, the flood fanned out and smaller velocities were reported. Five people died, the reservoir itself was lost, and flood damage was estimated at more than $15 million in 1964 dollars [36]. Structural damage included 41 homes destroyed and 986 houses, 100 apartment buildings, and 3000 automobiles damaged [40]. In addition, clean up and restoration efforts of streets, utilities, storm drains and repairs to the Ballona Creek Flood Control Channel were required. The cause of failure was investigated by California Department of Water Resources (CADWR) [36] who reported that earth movement under the reservoir cracked the asphaltic lining and subsequent leakage under pressure scoured the earthen fill within the embankment [36]. Today, the reservoir site has been transformed to a public park and one of the challenges addressed in this paper is the reconstruction of terrain as of 1963. 2.3. Data sources Several sets of data were obtained to support model parameterization and validation. Items (1)–(5) below are used for model parameterization purposes, while (6) and (7) allow for validation: (1) A 1.5 m (5 ft) resolution bare-earth Digital Terrain Model (DTM) from the 2006 LAR-IAC survey, as shown in Fig. 3. This was provided by the Los Angeles County Department of Public Works (LACDPW). The DTM exceeds National Standard for Spatial Data Accuracy (NSSDA) and Federal Emergency Management Agency (FEMA) standards for vertical accuracy, with a RMSE of 8.5 cm [23]. Full-size image (158K) Fig. 3. (a) LiDAR DTM (Raster), (b) reservoir and breach geometry (contours), and (c) merged DTM (TIN). (2) A set of 10 cm (4 in.) resolution digital orthophotos from the 2006 LAR-IAC survey, as shown in Fig. 1. These data, provided by LACDPW were obtained for resistance parameter estimation and georeferencing purposes. The orthophotos exceed NSSDA standards for horizontal accuracy with a radial RMSE of 26 cm [24]. (3) Parcel outline data were obtained from LACDPW to support the landcover classification and resistance parameterization shown in Fig. 4. Full-size image (106K) Fig. 4. Landcover classification and Manning n value, and location of catch basins and storm drain outlets. (4) Contour maps depicting the reservoir and dam breach geometry, as shown in Fig. 3, were obtained from the CADWR report [36]. The contour intervals were 6 m (20 ft) for the reservoir geometry and 3 m (10 ft) for the breach geometry. These data were scanned, geo-referenced, and digitized using polylines in ArcGIS 9.2 (ESRI, Redlands, CA). (5) Catch-basin locations in the study area were obtained from the City of Los Angeles Bureau of Engineering and are shown in Fig. 4. Catch basins collect water from street gutters and divert it to subsurface pipes that transfer flow to Ballona Creek. A field survey by UC Irvine personnel was completed to verify the existence of these basins as of 1963 and to characterize the type and size. Catch basins were largely of the curb-inlet type with a 20 cm (8 in.) height and a 2.1 m (7 ft) length. A small number of grate inlets were also found and noted. (6) Flood extent data, shown in Fig. 1, were obtained from a USACE report [38]. This consisted of a map with hand-drawn markings of flooding. This was scanned, geo-referenced, and digitized using polylines in ArcGIS 9.2 (ESRI, Redlands, CA). (7) Ballona Creek stream flow data at the gauging station shown in Fig. 1 were obtained from a USACE report [38]. This is limited to the following information: drainage water first arrived at 14:10 PDT, a peak flow of approximately exceeded to occurred at 16:40, flow between 16:10 and 17:00 and decreased by 19:10. 2.4. Terrain modeling ArcGIS 9.2 (ERSI, Redlands, CA) was used to merge the LiDAR DTM and reservoir and breach contour data into a Triangular Irregular Network (TIN) reflective of 1963 conditions, as shown in Fig. 3. Contours were available only as printed drawings or plates, so each image was scanned and georeferenced. Contours were then manually digitized as polylines and their nodes converted to x, y, z points. The LiDAR DTM was also converted from a raster format to x, y, z points, and a combined set of x, y, z points was obtained by filtering LiDAR points in areas of overlap. Finally, the merged set of points was converted to a TIN DTM as shown in Fig. 3. Street flows are important in urban flood hydrology [17] and [33], and we are assuming that 2006 terrain data provide a good description of 1963 terrain heights in the vast majority of the flood zone. This appears justified based on a comparison of modern digital orthophotos and historical aerial photos (not shown), which shows that the street layout has not changed. The photographic comparison also shows differences in the size and configuration of buildings, which is expected considering the damage of the flood. However, it should be stressed that the DTM is designed to capture bare-earth heights. 2.5. Flood inundation modeling To predict dam-break flood inundation, the 2D shallow-water equations were solved using BreZo, a Godunov-based finite volume code that runs on an unstructured mesh of triangular cells similar to a TIN [5]. The TIN computational mesh is different from the TIN DTM shown in Fig. 3, as it is configured for model efficiency, accuracy, and stability purposes. However, elevation data for the former is extracted from the latter. BreZo uses an approximate Riemann solver to estimate mass and momentum fluxes. This accommodates mixed flow regimes common to dam-break floods and handles wetting and drying problems without loss of stability, accuracy or conservation [3] and [5]. BreZo has been previously validated in a rural dam-break study [4], and applied to simulate urban flooding caused by overtopping of a culvert [33], but has not previously been applied to an urban dam-break application such as Baldwin Hills. The TIN computational mesh used by BreZo defines ground height at vertices and assumes that ground height varies linearly within each triangle; this achieves second-order accuracy relative to terrain height truncation errors. To minimize numerical dissipation, BreZo switches locally between two first-order accurate methods of variable reconstruction [5]. In practical test cases, the combination of a second-order accurate terrain model and a first-order accurate flow solver has been found to strike the best balance between numerical error and computational effort. In contrast to earlier versions of BreZo which used a global time step [5], here a three-level local time stepping (LTS) scheme is used to reduce run times [30] as in the study by Schubert et al.[33]. Cells are assigned a time step of either , or 4Δt; the largest that satisfies the Courant, Friedrichs, Lewy (CFL) condition is used. To maintain conservation with LTS, flux calculations and solution updates must be carefully sequenced but otherwise there is no loss of accuracy compared to global time stepping schemes. Curb inlets in the study area, shown in Fig. 4, divert surface water through sub-surface pipes to Ballona Creek. To account for this, the continuity equation solved by BreZo was modified with a set of point sink and source terms corresponding to curb inlets and sub-surface pipe outlets, respectively. Each time step, the volumetric flow rate into each curb inlet was computed with a modified weir equation as follows, (1) where g is the gravitational constant, h is the local depth of flow, ho is the height of the curb, L is the length of the inlet measured along the curb, and CD is a dimensionless discharge coefficient set by trial and error to 0.5. This selection was motivated by a local experimental study by the City of Los Angeles Bureau of Engineering [7], of catch-basin inflows, which indicates that CD falls between 0.1 and 0.5. Values of ho and L were measured for several catch basins by UCI personnel as described in Section 2.3. Time integration of the sink/source terms was implemented with an explicit, fractional step method. In the first step, the continuity equation was updated to account for all fluxes of surface water. In the second step, the continuity equation was updated to account for each source/sink term using the result of the first step to evaluate the right hand side of Eq. (1). Note that many of the curb inlets share the same outlet along Ballona Creek, as a result, a few cells are updated multiple times each time step. For stability purposes and to avoid negative depth predictions, the volumetric flow rate from each storm drain was limited so no more than 50% of the available volume could be withdrawn in a single time step. The preceding approach is proposed as a simple alternative to address the problem of catch-basin diversions in 2D dam-break FIM, compared to a fully coupled 1D/2D solver. There are clearly limitations to the method, for example there is no restriction to flow through the network (only flow into the network), the model cannot predict sewer surcharging which is often a driver of urban flooding, and the model assumes that flow is instantaneously routed from catch basins to the storm drain outlet. Further, this type of approach should not be used to design sub-surface storm drains. However, dam-break studies have rarely considered sub-surface storm drains on the grounds that sub-surface flow is a small fraction of overland flow. Hence, we utilize this relatively simple approach as a first step to judge the importance of sub-surface flows in an urban flooding scenario. The overall complexity of the storm drain flows, including clogging by sediment and debris, provides further motivation for simplicity as a first step. 2.6. Mesh generation and model parameterization A mesh of triangular cells was generated using Triangle, a flexible and powerful open source tool for 2D constrained Delaunay mesh generation [32]. The input to Triangle is an ASCII file that defines the boundary of the domain with a list of vertices and line segments, what graph theorists call a Planar Straight Line Graph (PSLG). Triangle enforces user supplied angle and cell area constraints for mesh quality purposes. Area constraints control the resolution of the mesh, and variable resolution meshes can be easily created. Further, meshes can be customized to study sites by aligning edges with building walls or street curbs, which improves model accuracy with relatively coarse meshes [33]. However, nearly uniform meshes were utilized in this study for simplicity and we do not attempt to resolve buildings with the mesh. This would require too fine a resolution for the available computational resources. Instead, the mesh was designed to resolve street depressions in the land surface, which are thought to act as channels during urban flooding, and to resolve heterogeneity in landcover (roads, developed parcels, vegetated open space, etc.) which affects flow resistance. Once a few preliminary runs had been completed to identify the impacted region, a PSLG was circumscribed around the area of interest to define the boundary (Fig. 1) and support mesh generation. The boundary was set back sufficiently far from the flood zone that boundary conditions became irrelevant, with one exception. Ballona Creek directs water south from the study area towards the Pacific Ocean. Here, a non-reflecting boundary condition was used so water can freely exit [28]. The boundary was placed downstream of the gauging station, where data are available, to facilitate comparisons between predictions and observations of stream flow. Meshes were generated using a 30° minimum angle constraint and a dual-zone maximum area constraint. An area constraint of was used in a region surrounding the breach, due to its narrow cross-section, and an area constraint of either 9.3, 37.2, or was used everywhere else. This created a set of three meshes, respectively: Mesh A with 1,337,155 triangles or cells, Mesh B with 336,681 cells and Mesh C with 86,835 cells. The resolution of these meshes, taken as the square root of the average cell area, corresponds to 2.5, 4.9, and 9.6 m, respectively. Typical street widths in Los Angeles County are 18 m (60 ft). Hence, Mesh A, B, and C resolve streets with at least 7, 3, and 1 cell, respectively. In addition, Ballona Creek is ca. 60 m wide so it is spanned by ca. 24, 12, and 6 cells with Mesh A, B, and C, respectively. Most predictions in the study utilize Mesh B, because it is fine enough to resolve street depressions but coarse enough (336,681 cells) for execution on a desktop computer in a few hours. Mesh A allows us to report the convergence error of Mesh B, and Mesh C allows us to report the consequences of an overly coarse mesh that does not accurately depict street depressions. Following mesh generation, ground elevation at mesh vertices was interpolated from the merged TIN DTM. Note that TINs utilize a linear reconstruction of terrain height which is the basis for interpolating ground elevation at mesh vertices. A Manning n was assigned to each cell in accordance with a simple landcover classification that was manually created from parcel outlines and digital orthophotos supplied by LACDPW. Parcel outlines were used as a mask to define the road network, and digital orthophotos were used to manually outline apartment building footprints, asphalt parking lots and vegetated open space areas which, based on the review of a historical orthophoto [36], were confirmed to exist at the time of the flood. As shown in Fig. 4, Manning n values of 0.014, 0.016, 0.013, 0.30, and were assigned to roads, channels, reservoir, developed parcels with buildings, and vegetated open space, respectively. A value of is typical of asphalt pavement, is typical of concrete channels with gravel and sediment along the channel bottom, is typical of smooth concrete surfaces, and corresponds to pasture with high grass [9]. This was chosen because the vegetated areas included many shrubs and small trees. Further, a value of has been recommended for developed parcels with buildings [39], but this would likely depend on the flow obstruction. Both historical and modern photos show that at least 50% of parcel footprints are occupied by buildings. 2.7. Initial conditions The failure sequence described in Section 2.2 and shown in Fig. 2 indicates that the breaching process began gradually before 15:00 and effectively ended at 15:30 with a major widening. Further, LADWP officials estimated that storage in the reservoir was approximately half its capacity at 15:30, ca. (ca. 449 acre-ft). However, the volume at the beginning of the major breaching processes, around 15:20, is not clear. The volume at this time is important as it represents what flooded north into the study area. Approximately (790 acre-ft) were stored at the beginning of the day, but LADWP took a number of steps to lower the level prior to catastrophic dam failure. Using design drawings of the reservoir which include the slope and height of the dam face, and photogrammetric scaling techniques, the height of the reservoir at 15:20 and 15:30 was estimated from the photographs shown in Fig. 2. Results suggest that the reservoir elevation was between 140.9 and 141.5 m (462 and 464 ft) at 15:20 and between 138.7 and 139.3 (455 and 457 ft) at 15:30. Based on the geometry of the reservoir, this corresponds to at 15:20 and at 15:30. Note that the volume at 15:30 is consistent with the LADWP [8] report that the reservoir was “half full” at 15:30. Further, this analysis suggests that the combination of controlled and uncontrolled flows sent approximately (186 acre-ft) from the reservoir before catastrophic failure. After careful analysis of all available information, breaching of the dam was modeled as a two-stage process. In the first stage, the breach was assumed to instantaneously open at 15:20 to a trapezoidal shape approximately 21.3 m (70.0 ft) wide at the crest of the dam and 7.6 m (25.0 ft) at the base of the dam. This roughly approximates the breach geometry around 15:20. In the second stage, which was assumed to instantaneously occur at 15:30, the breach was assumed to take on the final geometry reported by CADWR and shown in Fig. 2 and Fig. 3. The reservoir elevation at 15:20 was taken as 141.1 m (463.0 ft) based on the 15:20 reservoir photo; this represents the initial condition. BreZo was run for 10 min using the first breach geometry, and restarted using the second breach geometry at 15:30. BreZo predicted a reservoir volume of at 15:30; this is consistent with the 15:30 reservoir photo and the LADWP assessment of a half full reservoir. BreZo was integrated for a total period of 3 h to simulate the flood. The solution was saved at 4–8 min intervals for analysis purposes, the maximum depth and velocity was saved in each computational cell, and the discharge in Ballona Creek and through the storm drain system was also saved. 3. Results 3.1. Validation of the flood prediction The progression of dam-break flooding predicted by the model is shown in Fig. 5 from 15:20 onward. This shows the flood quickly funnelled north through the steep canyon below the dam, reaching the relatively flat terrain north of Coliseum St. by 15:25. Over the next 5 min, the flood pushed further north to Rodeo Rd. and spread laterally. By 15:30, the flood is shown to be fingering west along Rodeo Rd. and flooding those streets perpendicular and parallel to it. By 15:54, it appears that flood water reached Ballona Creek, which then directed water south towards the Pacific Ocean. Fig. 5 shows water in Ballona Creek prior to the arrival of surface flows along Rodeo Rd. which is due to routing through storm drains. USACE [38] reported a small baseflow in Ballona Creek prior to the dambreak flood, and attributed this to storm drain routing of the initial dam leakage because Ballona Creek is typically dry in the absence of rainfall. Full-size image (172K) Fig. 5. Progression of flooding predicted by model (Run 1). Red outline represents observed flood extent. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) Eastward flooding is also evident in Fig. 5, for example between 15:38 and 15:54, the model shows that flood water subsequently spread north into the junction of Jefferson and Exposition Blvd., and southeast along Santa Barbara Ave. Fig. 4 shows a number of catch basins along Jefferson and Exposition Blvd., and as water moved relatively slowly into this area compared to the westward flooding, these helped to prevent further flooding northward. There is evidence of flood recession by 16:18 as terrain in the eastern portion of the flood zone is shown to be drying. Recession of the flood becomes clearly evident by 17:02 and nearly 90 min later at 18:30, the situation changes very little which reflects a relatively slow recession of the flood compared to the initial surge. As described in Section 2.6, Mesh B was designed to validate the model. To quantify the accuracy of the flood extent prediction, a fit measure FE=0.76 was computed, which compares favorably with other FIM studies [14]. The fit was computed as follows [14], (2) where E indicates flood extent (m2) and P and M correspond to prediction and measurement, respectively. The symbol ∩ indicates the intersection of two domains, and represents the union of two domains. Fig. 7a presents predicted (labeled “Run 1”) and observed stream flow in Ballona Creek. Error bars on the observed stream flow data correspond to a 10% level of uncertainty, which is a rough estimate typical of stage-discharge errors. A fit measure for the peak stream flow FQ=1.08 was computed as follows, (3) where Q indicates peak stream flow (m3/s) at the gauging station and P and M are the same as in Eq. (2). This is within the assumed level of uncertainty (ca. 10%). A fit measure for the prediction of travel time FT=1.06 was also computed as follows, (4) where T represents the time elapsed from 15:20 until peak flow at the gauging station. Note that for all three fit measures, F=1 corresponds to perfect agreement. Furthermore, for FQ and FT, a value greater than one indicates an over-prediction and a value less than one indicates an underprediction. Hence, Run 1 slightly overpredicts the peak stream flow and travel time. Further, the model appears to more accurately predict the rising limb of the hydrograph than the falling limb. The model over-predicts stream flow at 17:00, although between 18:00 and 19:00 the prediction is again consistent with observations. The flood extent predictions shown in Fig. 6 and the stream flow predictions shown in Fig. 7a clearly validates the 2D model formulation, configuration, and parameterization. Run 1 does benefit from calibration; CD values of 0.1, 0.3, and 0.5 were tested to arrive at 0.5. However, the notion of calibrating and validating model parameters should not be confused with the validation of a modeling approach. Full-size image (190K) Fig. 6. Flood extent predictions for Runs 1–12. White outline represents observed flood extent. Full-size image (71K) Fig. 7. Ballona Creek hydrograph predictions for Runs 1–12 at gauging station shown in Fig. 1, compared to observations. Model predictions also highlight the complexity of urban dam-break flood flows: flow is highly unsteady and transported along preferential flow paths (streets) where terrain is depressed like a river thalweg and resistance is minimal due to a relatively smooth surface (concrete or asphalt) compared to natural surfaces. Previous studies have also emphasized the importance of street flows, and the need to accurately depict street geometry and resistance within the model framework [33]. These results suggest that a rich set of urban geospatial data is needed to accurately depict urban flooding, including high-resolution terrain data, spatially distributed resistance parameters, storm drain network data, knowledge of the reservoir level at the time of failure, as well as the breach geometry. Advances in remote sensing and information technologies will undoubtedly make some of these data more readily accessible in the future, while other factors such as the breach geometry will rarely be known a priori for predictive modeling purposes. Next in Section 3.2, several additional model simulations are presented to examine the relative importance of these data sources and modeling techniques. The sensitivity of model predictions to these factors are measured, and the most critical aspects are identified along with modeling guidelines for future studies. 3.2. Sensitivity analysis A total of 12 model runs are presented, including the base case (Run 1) shown in Section 3.1. Table 1 presents the attributes of the 12 runs, labeled Runs 1–12. Each run differs from Run 1 in only one respect as follows: • Runs 2 and 3 utilize a twice finer (Mesh A) and twice coarser (Mesh C) computational mesh versus the base case (Mesh B), respectively. • Runs 4–6 utilize different sources and resolution of terrain data. Run 4 uses a DTM that was coarsened to 9.1 m (30 ft) by window averaging the 1.5 m (5 ft) LiDAR DTM, Run 5 uses 1/3 arc-s (10 m or 33 ft) National Elevation Data (NED), and Run 6 uses of 3 arc-s (30 m or 99 ft) Shuttle Radar Topography Mission (SRTM) data. • Run 7 uses a spatially uniform Manning . This value (coincidentally, perhaps) represents: (a) the spatial average of the distributed Manning n shown in Fig. 4 and (b) the effective value of Manning n resulting from the application of Hejl’s method [13], which considers the fraction of the floodplain available for conveyance (i.e., not blocked by buildings). • Run 8 uses a smaller catch-basin inflow coefficient, CD=0.3. • Run 9 uses a higher initial reservoir level, 143.9 m (472 ft), which represents a typical operating level. • Runs 10–12 use a single-stage trapezoidal breach approximation with a bottom width B=H, 2H, and 3H, respectively, where H is the height of the dam, and a 1:1 side slope. Table 1. Attributes of model runs and performance metrics: presented in Section 3.1. , and FT. Run 1 represents the base case Wat er lev DTM el (m) Catc h basi n CD Comment FE FQ FT Me sh Manni ng n Brea ch proc ess 1 B Distrib uted 2 stag e 141 .20 1.5 m LiDAR 0.50 Base case 0. 76 1. 08 1. 06 2 A Distrib uted 2 stag e 141 .20 1.5 m LiDAR 0.50 Finer mesh 0. 79 1. 15 0. 94 3 C Distrib uted 2 stag e 141 .20 1.5 m LiDAR 0.50 Coarser mesh 0. 63 0. 72 1. 33 4 B Distrib uted 2 stag e 141 .20 9.1 m LiDAR 0.50 Coarsened DTM 0. 71 1. 11 1. 02 5 B Distrib uted 2 stag e 141 .20 10 m NED 0.50 National DEM for USA 0. 47 1. 07 1. 05 6 B Distrib uted 2 stag e 141 .20 30 m SRTM 0.50 Global DEM 0. 31 0. 00 N A 7 B Unifor m 2 stag e 141 .20 1.5 m LiDAR 0.50 Uniform n 0. 73 0. 59 1. 86 8 B Distrib uted 2 stag e 141 .20 1.5 m LiDAR 0.30 Less flow to catch basins 0. 69 0. 98 1. 09 9 B Distrib uted 2 stag e 143 .90 1.5 m LiDAR 0.50 Higher reservoir level 0. 68 1. 37 0. 95 10 B Distrib uted 1 stag e 141 .20 1.5 m LiDAR 0.50 Breach 0. width = dam height 73 1. 05 0. 80 11 B Distrib 1 141 1.5 m 0.50 Breach 1. 0. R u n 0. R u n Me sh Manni ng n uted 12 B Distrib uted Brea ch proc ess stag e 1 stag e Wat er lev DTM el (m) .20 LiDAR 141 .20 1.5 m LiDAR Catc h basi n CD 0.50 Comment FE FQ FT width = 2 × dam height 68 14 84 0. 65 1. 16 0. 82 Breach width = 3 × dam height Fig. 6 shows flood extent predictions corresponding to Runs 1–12, Fig. 7 shows hydrographs for Ballona Creek, and Table 1 shows , and FT. First consider the impact of mesh resolution. Fig. 6 shows that an increase in mesh resolution (Run 2, FE=0.79) improves the flood extent prediction very slightly compared to Run 1 (FE=0.76). For example, south of Coliseum street, just to the east of the primary flood path below the dam, there is a case of street flow that is more accurately depicted by Run 2 than Run 1. Also, Run 2 more accurately depicts flooding at the junction of Jefferson and Exposition Blvds. At the gauging station, Fig. 7a and Table 1 show that Run 2 leads to a 7% greater peak stream flow, and 12% shorter travel time compared to Run 1. These differences are small compared to the consequences of using a coarser mesh (Run 3). Run 3 shows a significant overprediction of flood extent in the northeast corner of the flood zone which leads to FE=0.63. Further, peak stream flow is significantly under-predicted FQ=0.72 and the travel time is significantly overpredicted FT=1.33. Previous dam-break modeling studies with BreZo have shown predictions are overly dissipative when the mesh is too coarse, causing an underprediction of peak stream flow and flood propagation speed [4]. Consider now the source of terrain data. Fig. 6 and Table 1 show that coarsened LiDAR (Run 4), NED (Run 5), and SRTM degrade the accuracy of flood extent predictions to different extents: , and 0.31 for Runs 4–6, respectively. Neither NED nor SRTM resolve street-scale topographic variations, but NED resolves the larger scale features that appear to bound the primary flood path north from the dam and west toward Ballona Creek. Where NED performs poorly is in the northeast corner of the study site, where NED does not resolve street scale variations in the terrain that appear to bound the flood zone. SRTM depicts relatively flat terrain with a nonphysical waviness that has been termed “radar speckle” [11] and [18], and predictions of flooding at similar spatial scales have shown non-physical pools of water corresponding to local minima in the DTM [29]. This pooling effect is evident in Run 6 shown in Fig. 6, and the flood extent prediction is notably crude. However, in the canyon North of the dam, flood extent based on SRTM is little different from Run 1 based on LiDAR. These results show that the DTM can have a significant effect on flood extent accuracy, but interestingly, the impact on stream flow predictions at the gauging station are relatively small with the exception of SRTM (Run 6), which does not resolve the geometry of the flood control channel and therefore cannot support modeling of flow through it. Fig. 7b shows predicted and measured hydrograph data, and Table 1 shows thatFQ and FT values across Run 1, Run 4, and Run 5 differ by at most 4%. Fig. 6 also shows that a spatially uniform Manning n has relatively little impact on flood extent (Run 7), particularly when compared to a smaller catch-basin discharge coefficient CD=0.3 (Run 8) and a higher initial water level (Run 9) which both significantly increase flood extent in the northeast corner of the study site. Stream flow at the gauging station is also sensitive to these factors. For example, Fig. 7d shows that a spatially uniform Manning n leads to a significant under-prediction of peak stream flow FQ=0.59 and over-prediction of travel time FT=1.86. Fig. 7d shows that an increase in the initial reservoir height leads to an over-prediction of stream flow FQ=1.37 and a slight under-prediction of travel time FT=0.95. Finally, Fig. 7d shows that a smaller CD has relatively little impact on the stream flow hydrograph, compared to the base case. Runs 10–12 examine the effect of an increasing breach width. Fig. 6 shows that flood extent increases with breach width, and all three single-stage breach scenarios show a greater flood extent than the two-stage breach scenario used in Run 1. However, Fig. 7c shows that breach width has relatively little impact on the stream flow at the gauging station even compared to the base case (Run 1). 4. Discussion Every perturbation of the model set-up, with the exception of Run 2 (finer mesh) resulted in a larger flood extent compared to Run 1. This was most notable in the northeast corner of the site where flooding was incorrectly predicted north of Jefferson Blvd. Terrain is gently sloped to the northwest here, so gravitational effects tend to stretch out the slightest over-prediction of flooding. It appears that no aspect of the model set-up described here can be simplified without sacrificing either flood extent or stream flow accuracy. The source and resolution of terrain data, reservoir volume, breach configuration, computational mesh resolution, and sub-surface storm drains all affected flood extent predictions by a similar amount. Resistance parameters and the reservoir volume affected stream flow more than other factors. Previous studies of dam-break flood modeling have noted an insensitivity of flood extent predictions to resistance parameters [4], but stream flow predictions here show that distributed resistance parameters are essential for accurately routing the flood across the street network and along the flood control channel. Similar findings have resulted from modeling studies of rural flooding [16]. Further, distributed resistance parameters are needed for local predictions of velocity [2] and [14] which may be needed for damage assessments or predictions of sediment erosion and deposition. Street widths appear to be a useful guide for selecting a mesh resolution. In a county where street widths of 18 m are typical, Mesh B with a resolution of 4.9 m (3 cells across street) gave good predictions (FE=0.76) but Mesh C with a resolution of 9.6 m (1 cell across street) significantly degraded flood extent accuracy (FE=0.63). Similarly, when LiDAR terrain resolution was coarsened to 9.1 m (Run 4), a loss of accuracy was also observed (FE=0.71) compared to the base case. These results show that flow along street depressions in the land surface should be resolved to accurately depict flood extent in urban settings. The meshing requirements may depend slightly on the mesh type (e.g., structured versus unstructured) and whether streets are aligned with the grid, so the common practice of convergence testing is recommended to ensure that the mesh is sufficiently resolved. What do these results say about good modeling practice for urban dam-break studies? Essentially, high-resolution terrain data, aerial imagery and catch-basin data can and should be obtained to support flood modeling because the potential exists for a high degree of accuracy (FE 0.8). NED is attractive because it can be obtained without charge from the USGS. However, results here show that flood extent is overpredicted using NED compared to LiDAR. Secondly, urban flooding is characterized by preferential flow along streets; thus heterogeneity in flow resistance parameters should be resolved to accurate depict overland flow. Third, flow through sub-surface storm drains can be important but it may be possible to use a relatively simplistic modeling approach that essentially transfers water from catch basins to the storm drain outlet. Otherwise, flood extent is likely to be overpredicted. Lastly, modelers should strive to resolve streets with at least three computational cells. Otherwise, models are likely to over-predict flood extent, under-predict peak flows downstream, and over-predict travel time. What about the predictability of dam-break flood inundation? What appears most challenging is the reservoir level at the time of failure, and its volume. Water level sensors, either in situ or remote (e.g, satellite altimetry), stand to enable the real-time monitoring that could support real-time dam-break emergency management efforts with flood forecasts. In case of real-time monitoring through satellite sensors, spatial resolution and re-visit time are important factors and the Surface Water Ocean Topography (SWOT) mission planned by NASA for the coming decade may provide critical information [35]. The breaching process appears less important than the reservoir volume, but more research should be done to evaluate the best strategies to couple modern breaching models (e.g., [10] and [21]) with dam-break flood models. 5. Conclusions Urban dam-break flood modeling demands a rich set of high-resolution geospatial data for accuracy purposes, based on the results of this study. High-resolution terrain data such as LiDAR are needed to depict street depressions in the land surface, and an unstructured mesh similar to the one used here should be refined with at least three cells across each street. Landcover heterogeneity should be resolved to guide the spatial distribution of resistance parameters, and the location of catch basins and storm drain outlets should be considered. Efforts to simplify model formulation or coarsen the resolution generally cause an over-prediction of flood extent or inaccurate stream flow predictions. In addition, poor flood extent accuracy was achieved with NED and SRTM terrain data. Further, a spatially uniform resistance parameter lead to poor stream flow accuracy compared to a spatially distributed parameter with the same mean value. A simple method of routing flow through storm drains was introduced here and successfully implemented. This involved pairs of sink and source terms in the continuity equation co-located with catch basins and storm drain outlets, respectively. A modified weir equation was used to scale flow into each catch basin, based on its height and curb length, and the model was validated with a dimensionless discharge coefficient CD=0.5. This falls at the upper end of what is expected (0.1–0.5) based on a laboratory study of catch-basin inflows by the City of Los Angeles. Water volume in the reservoir at the time of failure is a critical factor for accurate flood predictions, and over- or under-estimate of water levels will lead to an over- or under-prediction, respectively of flood extent and stream flow, all else being equal. A less critical but still important factor is the breach geometry. Use of a trapezoidal breach with a 1:1 side slope and a bottom width equal to the dam height reproduced flood extent better than breaches with wider bottom widths. Given the sensitivity of flood dynamics to the reservoir level, these results suggest that more detailed modeling of the breach geometry can be justified if the reservoir level is known with a high degree of certainty. Finally, dam safety programs should monitor water levels in real-time to support simulation based emergency management of dam-break flooding. Acknowledgements This work was supported by a grants from the UC Water Resources Center (WR-1016), the National Science Foundation (CMMI-0825165), and the UC Irvine Urban Water Research Center (Contribution # 39) whose support is gratefully acknowledged. The authors also thank LADWP, LACDPW, LAR-IAC, City of Los Angeles Bureau of Engineering, and USACE (Los Angeles District) for their cooperation. Lastly, the authors thank the reviewers for constructive comments that improved the paper. References [1] Archambeau P, Dewala B, Erpicum S, Detrembleur S, Pirotton M. New trends in flood risk analysis: working with 2D flow models, laser DEM and a GIS environment. River Flow 2004, Naples, Italy; 2004. [2] P.D. Bates, M.S. Horritt, C.N. Smith and D. Mason, Integrating remote sensing observations of flood hydrology and hydraulic modelling, Hydrol Process11 (14) (1997), pp. 1777–1795. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (49) [3] L. Begnudelli and B.F. Sanders, Unstructured grid finite volume algorithm for shallow-water flow and transport with wetting and drying, J Hydraul Eng132 (4) (2006), pp. 371–384. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (28) [4] L. Begnudelli and B.F. Sanders, Simulation of the St. Francis dam-break flood, J Eng Mech 133 (11) (2007), pp. 1200–1212. Full Text via CrossRef |View Record in Scopus | Cited By in Scopus (7) [5] L. Begnudelli, B.F. Sanders and S.F. Bradford, An adaptive Godunov-based model for flood simulation, J Hydraul Eng 134 (6) (2008), pp. 714–725. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (6) [6] J.D. Brown, T. Spencer and I. Moeller, Modeling storm surge flooding of an urban area with particular reference to modeling uncertainties: a case study of Canvey Island, United Kingdom, Water Resour Res 43 (2007), p. W06402 10.1029/2005WR004597. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (6) [7] City of Los Angeles, Department of Public Works. Hydraulic model study, Catch Basin Inlet Capacity. Los Angeles, Hydraulic Research Laboratory, HRL 10-77; 1977. [8] City of Los Angeles, Department of Water and Power. Facts about the Baldwin Hills Reservoir. Los Angeles; 1964. [9] V.T. Chow, Open-channel hydraulics, McGraw-Hill, Boston (1959). [10] Fread DL. BREACH: an erosion model for earthen dam failures. US National Weather Service Report, Silver Spring, MD; 1988. [11] G. Falorni, V. Teles, E.R. Vivoni, R.L. Bras and K.S. Amaratunga, Analysis and characterization of the vertical accuracy of digital elevation models from Shuttle Radar Topography Mission, J Geophys Res 110 (2005), p. F02005 10.1029/2003JF000113. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (32) [12] V. Guinot and S. Soares-Frazão, Flux and source term discretization in two-dimensional shallow water models with porosity on unstructured grids, Int J Numer Method Fluid 50 (2006), pp. 309– 345. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (16) [13] H.R. Hejl, A method for adjusting values of Manning’s roughness coefficient for flooded urban areas, J Res US Geol Surv 5 (5) (1977), pp. 541–545.View Record in Scopus | Cited By in Scopus (2) [14] M.J. Horritt and P.D. Bates, Evaluation of 1D and 2D numerical models for predicting river flood inundation, J Hydrol 268 (2002), pp. 87–99. Article | PDF (1163 K) | View Record in Scopus | Cited By in Scopus (100) [15] M.H. Hsu, S.H. Chen and T.J. Chang, Inundation simulation for urban drainage basin with storm sewer system, J Hydrol 234 (2002), pp. 21–37. [16] N.M. Hunter, P.D. Bates, M.S. Horritt and M.D. Wilson, Improved simulation of flood flows using storage cell models, ICE J Water Manage 159 (1) (2006), pp. 9–18. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (5) [17] N.M. Hunter, P.D. Bates, S. Neelz, G. Pender, I. Villanueva and N.G. Wright et al., Benchmarking 2D hydraulic models for urban flood simulations,ICE J Water Manage 161 (1) (2008), pp. 13– 30. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (23) [18] Intermap Corp. Product handbook and quick start guide. Version 3.3. June 15; 2004. [19] Kawahara Y, Uchida T. Integrated modeling for inundation flows in urban areas. In: World environment and water resources congress Ahupua’a, ASCE; 2008. [20] D. Liang, R.A. Falconer and B. Lin, Coupling surface and subsurface flows in a depth averaged flood wave model, J Hydrol 337 (1–2) (2007), pp. 147–158. Article | PDF (1221 K) | View Record in Scopus | Cited By in Scopus (8) [21] F. Macchione, Model for prediction floods due to earthen dam breaching. I: Formulation and Evaluation, J Hydraul Eng 134 (12) (2008), pp. 1688–1696.Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (3) [22] D.C. Mason, M.S. Horritt, N.M. Hunter and P.D. Bates, Use of fused airborne scanning laser altimetry and digital map data for urban flood modelling,Hydrol Process 21 (2007), pp. 1436– 1447. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (14) [23] Maune DF. Vertical accuracy assessment report, 2006 LiDAR Bare-Earth Dataset for Los Angeles Region Imagery Acquisition Consortium (LAR-IAC), Dewberry, Fairfax, VA; 2006. [24] Maune DF. Report of horizontal accuracy testing of 4 digital orthophotos for Los Angeles County Imagery Acquisition Consortium (LAR-IAC), Dewberry, Fairfax, VA; 2006. [25] H.K. McMillan and J. Brasington, Reduced complexity strategies for modelling urban floodplain inundation, Geomorphology 90 (2007), pp. 226–243.Article | PDF (2743 K) | View Record in Scopus | Cited By in Scopus (17) [26] E. Mignot, A. Paquier and S. Haider, Modeling floods in a dense urban area using 2D shallow water equations, J Hydrol 327 (1–2) (2006), pp. 186–199. Article | PDF (1504 K) | View Record in Scopus | Cited By in Scopus (24) [27] J.C. Neal, P.D. Bates, T.J. Fewtrell, N.M. Hunter, M.D. Wilson and M.S. Horritt, Distributed whole city water level measurements from the Carlisle 2005 urban flood event and comparison with hydraulic model simulations, J Hydrol 368 (2009), pp. 42–55. Article | PDF (1754 K) | View Record in Scopus |Cited By in Scopus (4) [28] B.F. Sanders, Non-reflecting boundary flux function for finite volume shallow-water models, Adv Water Resour 25 (2002), pp. 195–202. Article | PDF (1281 K) | View Record in Scopus | Cited By in Scopus (14) [29] B.F. Sanders, Evaluation of on-line DEMs for flood inundation modeling, Adv Water Resour 30 (8) (2007), pp. 1831–1843. Article | PDF (4533 K) |View Record in Scopus | Cited By in Scopus (17) [30] B.F. Sanders, Integration of a shallow-water model with a local time step, J Hydraul Res 46 (8) (2008), pp. 466–475. View Record in Scopus | Cited By in Scopus (5) [31] B.F. Sanders, J.E. Schubert and H.A. Gallegos, Integral formulation of shallow-water equations with anisotropic porosity for urban flood modeling, J Hydrol 362 (2008), pp. 19– 38. Article | PDF (1670 K) | View Record in Scopus | Cited By in Scopus (3) [32] Shewchuk JR. Triangle: engineering a 2D quality mesh generator and Delaunay triangulator. In: Lin MC, Manocha D, editors. Applied computational geometry: towards geometric engineering. Lecture Notes in Computer Science, vol. 1148. Springer-Verlag; 1996. p. 203–22. Software may be obtained at<http://www-2.cs.cmu.edu/~quake/triangle.html>. [33] J.E. Schubert, B.F. Sanders, M.J. Smith and N.G. Wright, Unstructured mesh generation and landcover-based resistance for hydrodynamic modeling of urban flooding, Adv Water Resour 31 (2008), pp. 1603–1621. Article | PDF (4816 K) | View Record in Scopus | Cited By in Scopus (7) [34] S. Soares-Frazão, J. Lhomme, V. Guinot and Y. Zech, Two-dimensional shallow-water model with porosity for urban flood modelling, J Hydraul Res46 (1) (2008), pp. 45–64. View Record in Scopus | Cited By in Scopus (6) [35] Surface Water Ocean Topography (SWOT) Mission. <http://bprc.osu.edu/water/> [accessed 13.05.2009]. [36] State of California, Department of Water Resources. Investigation of failure. Baldwin Hills Reservoir; 1964. [37] State of California, Department of Water Resources. Sacramento Delta, San Joaquin Atlas; 1995. [38] United States Army Corps of Engineers, Los Angeles District. Report on flood damage and disaster assistance; 1964. [39] United States Army Corps of Engineers, Los Angeles District. Interim guidelines for estimating n values on flood plains; 1981. [40] United States Department of Agriculture. Issues, resolutions, and research needs related to embankment dam failure analysis. In: FEMA workshop proceedings; 2001. [41] United States National Research Council. Risk analysis and uncertainty in flood damage reduction studies. Report by the committee on risk-based analysis for flood damage reduction. National Academy Press; 2000. [42] A. Valiani, V. Caleffi and A. Zanni, Case study: Malpasset dam-break simulation using a twodimensional finite volume method, J Hydraul Eng 128 (5) (2002), pp. 460–472. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (43) [43] I. Villanueva and N.G. Wright, Linking Riemann and storage cell methods for flood prediction, Proc Inst Civ Eng – Water Manage 159 (WM1) (2006), pp. 27–33. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (12) [44] D. Yu and S.N. Lane, Urban fluvial flood modelling using a two-dimensional diffusion-wave treatment. Part 1: Mesh resolution effects, J Hydrol Process 20 (7) (2005), pp. 1541–1565. [45] D. Yu and S.N. Lane, Urban fluvial flood modelling using a two-dimensional diffusion-wave treatment. Part 2: Development of a sub-grid-scale treatment, Hydrol Process 20 (7) (2005), pp. 1567– 1583. Corresponding author. Tel.: +1 949 824 4327; fax: +1 949 824 3672.