Use of Flood Modeller Pro to Develop Linked, Alternating 1-D and 2-D Models of Overland and in-river Flows for Breach of a Large Off-Channel Ring Dam

Transcription

1 Use of Flood Modeller Pro to Develop Linked, Alternating 1-D and 2-D Models of Overland and in-river Flows for Breach of a Large Off-Channel Ring Dam Tyler L. Jantzen, P.E., Water Resources Project Technologist, CH2M; and Duane M. McClelland, P.E., Water Resources Senior Technologist, CH2M; and Jason A. Eichler, P.E., PMP, Senior Engineer, Lower Colorado River Authority; and Nathan M. Gullo, P.E., Senior Engineer, Safety of Dams Lead, Lower Colorado River Authority Abstract--CH2M designed a new 45-foot-high, 27,000-foot-long ring dam that impounds the 40,000-acre-foot offchannel Lane City Reservoir in Texas. Dam-failure inundation maps reflect a 360-degree breach potential, extensive overland flow split between multiple watersheds, floodplains and channels, integration with an accepted 1-D HEC-RAS model of the Colorado River, and momentum effects that cross the river. It was imperative to model and display flood depths and timing in a way that is complex, yet easily understood. CH2M selected Flood Modeller Pro, a suite of publically available integrated 1- D and 2-D hydraulic software developed, owned and offered by CH2M, to provide 1-D and 2-D model integration, flexibility, stability and reliability. This paper provides a case-study in a hydraulically complex dam-failure context, compares results to those from HEC-RAS, describes selection and evaluation of failure scenarios, describes development and linkage of multiple 1-D and 2-D models, and illustrates flood-modeling and mapping challenges and solutions. I. INTRODUCTION Lane City Reservoir (LCR) is a proposed off-channel reservoir owned and operated by the Lower Colorado River Authority (LCRA) and designed by CH2M. Construction should begin in late 2015 or 2016 and should be completed by Existing and proposed conveyance facilities will move water in and out of the reservoir, including pump stations, canals, and a river outlet. The undeveloped project site is relatively flat, and the reservoir will be ringed on four sides of a rectangle by a 45-feet-high embankment dam that encloses nearly 1,100 acres of farmland to store 40,000 ac-ft of water 1. By optimizing reservoir operations, LCRA projects that the new reservoir will add 90,000 ac-ft of firm water supply. During design, CH2M performed dam-breach modeling to confirm hazard classification of the dam and develop damfailure inundation maps for an Emergency Action Plan (EAP). Because of flat terrain, a variety of split flow paths, and complex hydraulic interactions between multiple watersheds, channels and floodplains, dam-failure floods were modeled for breaches on each side of the rectangular ring-embankment using the combined strengths of linked one-dimensional (1-D) and two-dimensional (2-D) models. Modeling was accomplished using Flood Modeller Pro 2, a publically available integrated suite of 1-D and 2-D hydraulic models developed, owned and offered by CH2M. II. PROJECT DESCRIPTION LCR is located in Wharton County, Texas, beginning approximately 0.6 to 0.8 miles east of the Colorado River, and extending a little over 2 miles northeast where it is bounded by Texas State Highway 60 along its eastern embankment and Lane City immediately to the northeast, as shown on Figures 1 and 2. Figures 2 and 3 show artist renderings of the proposed LCR and its embankment. LCR will be impounded by a rectangular, earthen ring embankment dam constructed of native soils excavated within the reservoir. Figure 4 shows a typical section from the drawing set to provide a feel for the embankment configuration. 1 As of this writing, LCRA intended to solicit bids in early summer 2015 for both the full reservoir capacity and a scaled-back reservoir capacity of about 35,000 ac-ft. The capacity of the reservoir will likely be finalized in fall 2015, at which point the dam-failure hydraulic modeling described in this paper will also be finalized. The information presented in this paper pertains only to the original, larger reservoir size. 2 More information about Flood Modeller Pro, including freely downloadable installation files, can be found at The spelling of Modeller is British. Either Flood Modeller Pro or Flood Modeller Free could have been used for this project. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 1 of 18

2 Figure 1. Project site and vicinity Figure 2. Artist s rendering of LCR ring-shape embankment Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 2 of 18

3 Figure 3. Artist s rendering of LCR, looking west from the northern corner. Figure 4. Typical LCR embankment section The LCR embankment has a crest width of 12 feet and interior and exterior slopes that vary between 2H:1V and 3H:1V. The embankment height around the embankment perimeter ranges from about 47 feet above existing grade at the south corner where there is a local swale, to about 41 feet near the eastern corner. The embankment consists of low-permeable native clay soil upstream of a 3-foot-wide vertical chimney drain/filter and random fill downstream of the chimney. The upstream slope is protected from erosion by stepped lifts of soil cement. The downstream slope is protected by vegetation seeded in cohesive topsoil. The embankment is keyed into its foundation with a central cutoff trench. The maximum 6-foot grade difference around the reservoir perimeter illustrates the flat nature of the broad floodplain across which the Colorado River historically meandered near the Gulf Coast. A major exception is the deeply incised Colorado River channel itself, which drops nearly 40 feet down steep banks from the eastern bank rim adjacent to LCR. There are also a handful of small local creeks that cross the historical floodplain. One of these, Jarvis Creek, parallels the southwestern edge of LCR and drains to the Colorado River. Others to the east parallel the Colorado River and drain directly to the Gulf of Mexico, located approximately 40 linear miles to the southeast. The area surrounding LCR is rural, consisting primarily of agricultural land dominated by hay fields, pasture, and cultivated crops, as well as less developed deciduous forest. The main populated areas near LCR are Lane City (population of 111 in 2000) and Bay City (population 18,000 in 2010; US Census, 2010), located approximately 14 miles south of LCR. Apart from the small town of Lane City and distant Bay City, structures within the embankment-failure floodplain primarily Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 3 of 18

4 consist of occasional farm houses, barns, oil wells, other widely scattered rural structures, and the local roads that provide access. III. MODELING AND MAPPING CHALLENGES The LCR dam hazard classification could readily be assumed. The LCR embankment is a high hazard dam by inspection due to the proximity of Lane City 1,700 feet from the northeastern corner of the dam and well-travelled Highway 60 offset 400 feet from the northeastern embankment toe. Modeling confirmed incremental flow depths at these locations in excess of 8 feet with local velocities exceeding 6 feet per second. What was at question, however, was the variety of dam-failure flow paths and travel times that were possible, how to realistically model them, and how to communicate these most readily for use by LCRA and emergency responders. It s fair to say that many of the flow paths, flood-wave travel distances and travel times were not fully identified or understood prior to integrated 1-D/2-D modeling. During model development, there were many informative surprises that could not have been determined using 1-D or 2-D modeling alone. Through integrated 1-D and 2-D modeling, it was discovered that breach flows can follow four broadly defined divergent or convergent paths, each of which covers an extensive area with isolated populations at risk. The extent to which any path is taken depends primarily on the breach location, peak breach discharge and antecedent flood conditions. 1. Flow can move west, downslope across the Colorado River floodplain to enter the Colorado River. 2. During a 100-year flood, the deeply incised Colorado River is near bank full on the eastern bank, and spills widely into divergent flow paths left by remnant historical channels on the western floodplain. Under these conditions of concurrent flooding, a dam-failure flood will not only enter the Colorado River, but carry flow across the river onto the western floodplain, eventually to rejoin the Colorado River from the west at distant points downstream. This interconnection can occur at many locations, with excess dam-failure flows leaving and re-entering the primary Colorado River channel. 3. Flow that begins traveling north, east, west or south eventually turns south and southeast to follow local swales, remnant historical meanders, and parallel creeks many miles before entering the Colorado River, or continuing southeast without enter the River at all. The flow can split at many locations, providing multiple tributaries to the Colorado River and local farms. 4. Flow from an eastern breach carries enough depth and momentum for a portion of the flow to cross Highway 60 and enter adjacent watersheds that drain directly to the Gulf of Mexico parallel to the Colorado River. Dam breach flows in these drainages eventually abate prior to reaching the Gulf, but also have the greatest impact on Lane City and Highway 60, where populations are most vulnerable. The LCR configuration and complex surrounding topography creates challenges not easily overcome by typical inundation modeling techniques. These challenges, discussed in more detail later in the paper, can be boiled down to five: 1. How to effectively model dam-failure flooding across three broad floodplains filled with creeks and swales, separated by a large river and a watershed divide. 2. How to provide adequate spatial resolution of 2-D data points for map display over a very large spatial domain without exceeding reasonable computational requirements. 3. How to reduce a large number of possible maps and map details to those most useful to LCRA and emergency responders. 4. How to identify and overcome any limitations in the software itself. 5. How to validate a software package that is new to U.S. users. IV. FLOOD MODELLER PRO After considering a number of popular 2-D hydraulic software packages, the CH2M modeling team selected Flood Modeller Pro to address these modeling and mapping challenges. This integrated software suite was selected because of its integrated 1-D and 2-D modeling capabilities, robust and proven algorithms, ability to accurately and quickly model rapidly changing flow conditions, user-friendly interface, and strong technical support. At times, the modeling team worked directly with the software developers to understand or refine highly technical aspects of the software and code. By using Flood Modeller Pro, the modeling team avoided many model instabilities that can plague or derail complex hydrodynamic damfailure models, and directly integrated multiple 1-D and 2-D models within a single application from a single user interface. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 4 of 18

5 Flood Modeller Pro and its predecessor software 3 have become the industry standard for hydraulic modeling in portions of Europe and selected non-european countries, although its introduction to the United States has been relatively recent. The LCR project provided an opportunity to leverage the software s integrated 1-D and 2-D modules, benefit from a stabilizing engine intended to model shock waves produced by a rapid dam breach, and validate the software by comparing its 1-D output to parallel results using HEC-RAS, the U.S. industry standard for dam-failure flood routing in channels. A useful feature of Flood Modeller Pro is the inclusion of multiple 2-D solvers, including the Total Variation Diminishing (TVD) solver, which has been developed specifically to provide accurate representation of two-dimensional shocks. It allows the complex hydraulics of steep changes in velocity and water level to be calculated more accurately, and provides increased stability when compared to other 2-D solvers. This feature is especially important for modeling a dam breach flood wave across a flat terrain surface, like that surrounding LCR. V. MODEL DEVELOPMENT Within Flood Modeller Pro, two 1-D and two 2-D hydrodynamic models were coupled, as shown on Figure 5. The 1-D models represented discharge through the dam breach and flows within the Colorado River, exclusive of its floodplains. The 2-D models represented flow across overland areas east and west of the Colorado River, including its floodplains and the adjacent eastern watershed. The 1-D breach model was influenced by 2-D tailwater effects; 2-D overland flow on both sides of the Colorado River were influenced by 1-D elevations within the Colorado River channel; 1-D flow within the Colorado River channel exchanged 2-D flows with its eastern and western floodplains; and regions east and west of the eastern watershed divide communicated internally within the eastern overland 2-D model. Each model seamlessly passed flows to the connecting models, responding dynamically to adjacent water depths and velocities. Figure 5. Areas modeled using 1-D and 2-D solvers. 3 Flood Modeller Pro has been developed and supported for 40 years, although the name Flood Modeller Pro is new with the April 2015 updated version release. Future upgrades are also anticipated. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 5 of 18

6 The 1-D Flood Modeller Pro model of the Colorado River channel was developed by importing cross sections and other data from two existing, sequential 1-D HEC-RAS models of the Colorado River. The first model began well upstream of LCR and extended downstream to Bay City, and the second model extended from there to the Gulf Coast. The first HEC-RAS model used results from the second model as a downstream boundary condition, so the two models were readily concatenated without affecting their hydraulic performance. The 2-D models were developed using digital elevation models (DEMs) with a combined spatial domain that covered over 300 square miles. The finest DEMs were available with a 3-m grid size, but a coarser resolution was selected for modeling most areas, as described below. VI. BREACH PARAMETERS Ultimately, eight breach scenarios were evaluated: breaches on each side of the four-sided reservoir for both sunny day and PMF hydrologic conditions. Following a sensitivity analysis of failure modes and breach parameters, an overtopping failure using parameters developed from Von Thun s (Von Thun, J. L. and Gillette, D. R., 1990) empirical equations were selected. These parameter sets were selected to model both the PMF and sunny day breach after comparing breach parameters and breach discharge from a number of different methods, including equations developed by Froehlich (1987, d1995a, 1995b, 2008). A northeastern embankment breach is assumed to be the most hazardous because of the proximity of Lane City and State Highway 60. The southwestern breach resulted in the largest and most rapid discharge to the Colorado River, which may be pertinent to evacuation along the riverbanks downstream. The two remaining breaches produced results that were hybrids of the first two breaches. High-level maps were developed for all four breaches. Informal and formal sensitivity analyses were used to help finalize the choice of failure mode, breach parameters and breach locations. The first choice was failure mode, for reasons explained later. Both piping and overtopping failure modes were considered and were simulated using a simple 1D breach model in HEC-RAS using identical breach parameters. Reservoir and breach parameters for this sensitivity analysis were similar but not identical to those used for final inundation mapping. The peak discharge from a piping and overtopping failure were nearly identical (119,420 cfs for overtopping versus 119,299 cfs for piping). In light of negligible sensitivity and to simplify analyses, an overtopping failure mode was selected for all breach scenarios for further modeling. To identify the range of breach parameters that may be credible, a breach parameter sensitivity analysis was performed on three separate sets of breach parameters that were derived from distinct methods and were considered reasonable, internally congruent and representative of calculated best-estimates. A fourth set of breach parameters was also selected to represent a conservative synthesis of remaining methods. The Von Thun and Gillette (1990) method was chosen for use in inundation modeling and mapping for the following reasons: 1. The sensitivity of peak discharge to choice of method was found to be relatively low. 2. The Von Thun and Gillette equations produced the highest peak discharge from among congruent parameter sets, and so was considered conservative, yet reasonable. 3. The Von Thun and Gillette equations produced peak discharge that was lower than produced using an intentionally conservative synthesis of remaining methods, as expected, but only slightly lower, thereby confirming conservatism. To further validate the selected method, inundation models prepared for all four methods show that the Von Thun and Gillette method produces the highest depths at the nearest populated area, Lane City. In addition, the Von Thun and Gillette method uses the shortest breach development time, resulting in the shortest flood wave travel time and shortest warning time from breach initiation until flood arrival at Lane City. Given Lane City s proximity to the embankment, a short breach development time has a large impact on evacuation time, and is thus considered appropriately conservative for the intended purpose of EAP and emergency response planning. VII. MODELING AND MAPPING SOLUTIONS The five modeling and mapping challenges listed previously were each addressed using the integrated Flood Modeller Pro model developed for the project. These challenges and how they were addressed are described sequentially. A. Linked 1-D/2-D Modeling The first challenge was how to effectively model dam-failure flooding across three broad floodplains filled with creeks and swales, separated by a large river and a watershed divide. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 6 of 18

7 Depending on the location of the breach and antecedent flood conditions in the Colorado River, an LCR dam-failure flood wave can rush down the Colorado River, move slowly parallel to the River on the east or west floodplains, leave the Colorado River drainage completely and pass down an adjacent watershed, or do all of the above concurrently. Along the Colorado River, the main channel and adjacent floodplains pass bi-directional flow for many miles. While the existing HEC-RAS models of the Colorado River were important, 1-D modeling could not credibly represent the complex hydrodynamics outside of the main channel, where the population at risk resides, and therefore could not be relied on for reliable inundation limits, flood paths, flows, depths or travel times. 2-D modeling was essential to capture overland flow dynamics. Even close to the reservoir, 2-D modeling revealed tremendous multidirectional flow that profoundly affected time to flood peak, local velocities, and flood depths, especially where water was released to flow uphill. Figure 6 provides an example of maximum flood depths and velocity vectors for a breach near Lane City. Based on the velocity vectors, flow followed virtually every compass point, with almost no channelization typical of 1-D hydraulics. Figure 6. Example of maximum flood depth and velocity vectors for the northeastern breach location. Without 2-D modeling, the modeling team would not have discovered that certain breach locations produced enough local flow depth and momentum to cross the watershed divide to the east of the reservoir. This early discovery required a large increase in the model domain, and refined the team s understanding of alternative, parallel flow paths. The ability to display 2-D velocity vectors greatly improved understanding of the flat underlying terrain and its effects on flood propagation. As important as 2-D modeling was for this project, 1-D modeling was also essential. A large map polygon, known as the model domain, was required to encompass the diverging flow paths that extended miles beyond the reservoir. A coarse DEM grid size was required to represent such a large model domain, and this grid size was too large to adequately represent the channel slopes, widths and side boundaries of a confined, steeply-incised, meandering river channel. Early trials using only a 2-D model did not properly characterize the complex bi-directional interaction between river and floodplain. Modeling the breach itself is also a 1-D modeling problem, affected by 2-D tailwater conditions. The integrated 1-D and 2-D modeling provided by Flood Modeller Pro allowed the project to take advantage of the inherent strengths of each model type, producing a better and more credible result than either method could produce on its own. B. Balancing High 2-D Computational Requirements with Adequate Spatial Resolution The second challenge was how to provide adequate spatial resolution of 2-D data points for map display over a very large spatial domain without exceeding reasonable computational requirements. 2-D modeling is considerably more computationally demanding than 1-D modeling. Considering a full range of breach locations, a 2-D model domain exceeding 300 square miles was needed for LCR inundation mapping. To keep computational times reasonable (less than one day per scenario), each grid cell within the model domain was 100m square, uniformly representing the elevation at the grid center. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 7 of 18

8 While a smaller grid size is preferable from the standpoint of both precision and accuracy, it is the enemy of computational efficiency. A 300-square-mile model domain contains approximately 80, m grid cells which must be processed at a fine modeling time step. Using a 10-second modeling time step, it requires approximately 3 billion calculations to assign a value to each grid cell at every time step for a 4-day flood. When iterative solutions are considered across the model domain, the number of calculations is difficult to imagine. Thanks to modern computer processors, a model run may be completed overnight. However, if one then reduces the grid size by half from 100m squares to 50m squares, the number of grid cells quadruples and the modeling time step may need to be reduced to maintain stability. This might increase processing time 10 fold. If the grid size is instead reduced to 30m squares, the number of grid cells increases about 10 times, the modeling time step must be further reduced, and the number of calculations per run may increase by 30 or 100 fold, increasing run times from a day to a month or longer. While this might be acceptable for a single, final run, it is impractical when exploring multiple breach scenarios and locations around a ring dam. Using the finest available grid size of 3m would work well locally, but would not be considered for a model domain this large. The relatively coarse grid size selected was considered appropriate given the scale of peak LCR breach flows (200,000 cfs) and the sparsely-developed flat, rural floodplain. However, the 100m grid was too coarse for meaningful results near populated areas, such as Bay City (population 18,000). In order to meet both the need for reasonable computational time and refined results in populated areas, a fine grid sub-model of the relatively small populated area was created using boundary inputs from the coarse model. A user-friendly interface and excellent workflows in Flood Modeller Pro allowed for a relatively seamless integration of the larger coarse grid model with smaller fine grid models. Figure 7 shows a comparison between the coarse and fine grid results for Bay City. Figure 7. Comparison of inundation results using (a) a coarse grid (100m) and (b) a fine grid (5m) C. Producing Useful Maps: numerous permutations of breach location, spatial resolution, and temporal resolution. The third challenge was how to reduce a large number of possible maps and map details to those most useful to LCRA and emergency responders. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 8 of 18

9 1) Global Maps As opposed to a linear dam in a river valley, the LCR ring dam allows for a potential breach with equal probability on any side of the reservoir. This complicates both inundation modeling and mapping. A breach of the eastern embankment sends a large portion of the breach volume into an adjacent watershed, and very little water into the Colorado River. Conversely, a breach on the western embankment sends a large portion of water into the Colorado River, and may flood its western banks if the breach is coincident with a high river stage. Breaches on the north and south embankments also result in unique inundation characteristics, allowing for numerous permutations in breach location, and spatial and temporal resolution. The rectangular shape of LCR and a sparsely populated floodplain allowed the team to limit the number of breach locations to one on each side, selecting specific breach locations so as to differentiate flooding or maximize impacts on developed areas. Figure 8 shows maximum inundation depths for all 4 breach locations (a through d). Even with only four breach locations, it was challenging to display inundation maps with both the temporal and spatial resolution needed by emergency responders to rapidly select and utilize maps in an intuitive manner. Compounding map complexity, each breach location was modeled as a sunny day breach and PMF breach, and the PMF breach was postprocessed to show incremental flooding beyond antecedent flood conditions. Antecedent flood conditions for a PMF breach included a 100-year flood in the Colorado River and 6 inches of rain applied overland throughout the model domain. Combined, this resulted in ten separate model simulations, each of which took up to a day to run, and each of which produced diverse and unique spatial and temporal inundation. Each of these map sets would have had dozens of pages to cover the inundation limits, making rapid, proper map selection a challenge. The functionality of Flood Modeller Pro helped resolve these modeling challenges. Flood Modeller Pro s interface allows the user to easily produce a new model from an existing model where limited model elements (such as breach location or antecedent flooding) change between scenarios. In addition, because Flood Modeller Pro is built on an open file format, the user can build project-specific post-processing scripts and routines to better keep track of result files. Also, Flood Modeller Pro includes built-in batch processing capabilities, allowing for efficient use of computational resources. To reduce the confusion of map permutations and simplify their use for LCRA and emergency responders, results were combined in some maps to show the maximum flooding depth associated with a breach from ANY of the four selected breach locations, as in Figure 8.e). This drastically reduced the number of maps needed to show the meaningful spatial resolution of results. High temporal resolution maps were provided for the northeastern breach location only, because response time for this scenario would be most critical given the proximity of Lane City. Figure 9 shows an example of the temporal progression of the north breach inundation. 2) Index Map Global map sets like Figures 8 and 9 were useful to provide a quick overview of flood diversity and progression, but the spatial domain was too large to depict individual structures at risk. Once emergency planners were familiar with the flood progression, a set of detailed maps was needed at a closer scale to inform emergency response for specific houses and neighborhoods. The first step was to provide a clear index map across the broad 2-D flood footprint to facilitate rapid map orientation and location. An example is shown on Figure 10. A few principles helped ensure a useful index. 1. The index map should include global information, such as maximum depth throughout the inundation footprint, to help quickly identify priority detail maps. 2. Only cover locations where current or future populations may be at risk. Using Figure 10 as an example, there are gaps between some maps, and there are wide spaces between maps in the southern reaches of the Colorado River. 3. Follow an intuitive progression for map numbering. 4. Subdivide a detailed map if still more detail is justified. Examples on Figure 10 are maps sets 3, 8, 12, and 13, all of which are divided into four quadrants sub-labeled a to d. In these cases, the larger detail map was provided, immediately followed in the map set by the four maps at a finer scale. 5. Repeat the index map as a key on the detailed maps to ensure proper orientation. See Figures 11 and Reference the maps on each side of each detail map, keyed to the index map. As examples, adjacent map numbers accompany arrows on the eight compass points in the margins of Figures 11 and 12. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 9 of 18

10 Figure 8. Comparison of maximum inundation depth for the Sunny Day failure from four different embankment breach locations: a) Southwest, b) Northwest, c) North, and d) Southeast. Panel e) shows a composite of maximum depth from all four breach locations. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 10 of 18

11 Figure 9. Example of the temporal progression of the north embankment breach. NOTE: Times shown on figures are from the initiation of the breach, in hours. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 11 of 18

12 Figure 10. Sample Index Map. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 12 of 18

13 Figure 11. Detail map of Lane City. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 13 of 18

14 Figure 12. Detail map where sparsely developed. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 14 of 18

15 3) Detailed Maps A number of guiding principles helped improve the usefulness of detailed maps, as well. Examples of detailed maps are shown on Figures 11 and Provide as much detail as needed at critical locations. Avoid detail where it s not needed. 2. Key information is maximum depth, maximum velocity, flood wave arrival time (First Flood Time), and time to peak (Peak Time). In some cases, time to impasse, such as when a bridge overtops, as also useful for evacuation planning. 3. Key information should be specific to structures. Although common in the past with 1-D inundation maps, maximum flood depth and velocity in the center of a channel has little relevance to the flood depth and velocity experience by a house on the flood fringe. Two-D inundation maps are particularly well suited to differentiating flood depth and velocity by location across the floodplain. 4. Color gradients provide a useful means to differentiate flood depths, both by location (detailed maps) and over time (global maps). However, it is difficult to simultaneously display depth and velocity with color, unless the product of depth and velocity are provided a color gradient, which loses some important information. 5. We found that it was clearest to communicate detailed flood depth, velocity and timing using callouts, especially where development was clustered or sparse. In addition to providing information, the callouts help lead the eye to key structures. Flood wave arrival was provided for selected evacuation routes. D. Advancing the Software The fourth challenge was how to identify and overcome any limitations in the software itself. Two modeling challenges were encountered that helped the Flood Modeller Pro software development team advance the software. In the interim, a strong Flood Modeller Pro technical support crew responded quickly to find temporary workaround solutions and permanent solutions in the form of a software patch and objectives for future software enhancements. One of the software challenges was the inability of Flood Modeller Pro to simulate piping breaches; only overtopping breaches are currently supported. The temporary work-around was to use HEC-RAS to compare breach hydrographs for an overtopping and piping failure, and determine how to modify the overtopping failure so as to most closely mimic the piping failure hydrograph. The plan was to then use the corresponding overtopping breach parameters in Flood Modeller Pro to mimic a piping failure within the linked 1-D 2-D composite modeling framework. Comparing the two breach hydrographs in HEC-RAS, it was discovered that for this specific project, which has a large reservoir of modest height, the overtopping breach and piping breach deliver similar flood results, and the overtopping breach is slightly more conservative with respect to flood peak and timing, and thus appropriate to use in lieu of a piping breach. Also, a direct comparison of the overtopping breach in HEC-RAS to the overtopping breach in Flood Modeller Pro confirmed nearly identical results, validating confidence in the Flood Modeller Pro software. As a long-term solution, Flood Modeller Pro developers intend to incorporate piping-failure functionality in future versions. A second software challenge was loss of breach inundation volume when flowing between the 1-D Colorado River and the 2-D floodplain. Flood Modeller Pro has two alternative solvers and three alternative types of 1-D/2-D links for use in different contexts. The observed loss of flood volume was unique to the 2-D TVD solver used in shock-wave applications, and was unique to only one of the three types of 1-D/2-D links. Flood Modeller Pro technical support quickly determined the root cause, and suggested use of a different, more applicable 1-D/2-D link type as a short-term work-around solution. Within days, technical support released a patch correcting the issue, which has since been incorporated directly in the current version of the publicly available software. E. Model Validation with HEC-RAS The fifth challenge was how to validate a software package that is new to U.S. users. Because HEC-RAS has long been a 1-D modeling standard in the United States, the Flood Modeller Pro 1-D dam breach hydrograph results were compared with HEC-RAS dam breach results for confirmation. For breach discharge, the Flood Modeller Pro 1-D model results were 3 percent higher than the HEC-RAS 1-D results for a 120,000 cfs discharge, indicating comparable performance. The HEC-RAS Colorado River model geometries were imported directly into Flood Modeller Pro, allowing for direct comparison of the two 1-D river model hydraulic profiles and dynamic flood routing hydrograph transformation. Comparison for equivalent performance was only possible at lower river flows that were confined within the incised channel, since the overbank floodplain portion of the broad HEC-RAS cross sections was replaced within Flood Modeller Pro by linked 2-D models. Therefore, a lower-flow 2-year unsteady hydrograph was used to compare model performance within identical cross sections, and a 100-year unsteady hydrograph was used to contrast results for the linked 1-D 2-D Flood Modeller Pro models with the 1-D HEC-RAS model. The 2-year comparison shows nearly identical model performance, confirming comparable hydraulic functionality with identical geometries. Results were also similar, but with slightly larger differences in Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 15 of 18

16 Water Surface Elevation (feet) Breach Flow (cfs) Dam Safety 2015, Proceedings of the 32nd annual conference of the Association of State Dam Safety Officials, Sept , 2015, New Orleans, Louisiana. the 100-year-flood comparison due to the difference in handling of overbank flows between the 1-D HEC-RAS model and the 1-D/2-D Flood Modeller Pro model. Presumably, the 1-D/2-D results would be more realistic. A comparison of the two test cases is shown in Figures 13 and 14, respectively. 140, , ,000 80,000 60,000 40,000 20, :00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 Time (hours, breach starts at t=1:00) Flood Modeller - Overtopping HEC-RAS - Overtopping HEC-RAS - Piping Figure 13. Comparison of modeled breach flows, from Flood Modeller Pro and HEC-RAS Channel Distance (miles) 100-yr Flood Modeller 100-yr HEC-RAS 2-yr Flood Modeller 2-yr HEC-RAS Figure 14. Comparison of modeled 2-year and 100-year flows, from Flood Modeller Pro and HEC-RAS VIII. CONCLUSIONS Because of its configuration and complex surrounding topography, modeling and inundation mapping of an LCR embankment breach creates unique challenges. These challenges include complications due to multiple interconnected flow paths, balancing spatial resolution and computational requirements, creating intuitive map products, advancing the software, and introducing a new tool to U.S. users. The CH2M modeling team met these challenges using innovative solutions to effectively and accurately model complex hydraulics and simplify the display of multifaceted and lengthy results. A key part of the solution was the selection and use of a tool that allows integrated 1-D and 2-D hydraulic modeling, Flood Modeller Pro. This tool capitalizes on the strengths of both 1-D and 2-D modeling environments. The ultimate product is a series of maps that shows an appropriate range of spatial and temporal resolution for multiple distinct scenarios. The use of Flood Modeller Pro was further validated through comparison with HEC-RAS, a common standard for breach modeling in the United States. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 16 of 18

17 IX. REFERENCES CH2M Flood Modeller Suite. Accessed May 1, Froehlich, D.C "Embankment Dam Breach Parameters." Hydraulic Engineering. p Froehlich, D.C. 1995a. "Embankment Dam Breach Parameters Revisited." Proc., 1st Int. Conf. Water Resources Engineering, ASCE. New York. p Froehlich, D.C. 1995b. "Peak Outflow from Breached Embankment Dam." Journal of Water Resources Planning and Management. Vol. 121, No. 1, Jan./Feb. p Froehlich, D.C "Embankment Dam Breach Parameters and Their Uncertainties." Journal of Hydraulic Engineering, Vol. 134, No. 12. Dec. 1. p United States Census Bureau (US Census) American Fact Finder. Accessed 10/10/ Von Thun, J. L. and Gillette, D. R "Guidance on Breach Parameters," Internal Memorandum, U.S. Dept. of the Interior, Bureau of Reclamation, Denver, March 13. X. AUTHOR BIOGRAPHIES Tyler L. Jantzen, P.E. Water Resources Project Technologist CH2M th Avenue NE, Suite 500 Bellevue, WA Phone/ (425) / Tyler.Jantzen@ch2m.com Tyler Jantzen is a registered water resources engineer with CH2M in Seattle, Washington. Over the past 8 years, he has gained considerable experience with complex hydrologic and hydraulic models, dam-break analysis, stream restoration, climate change adaptation, stormwater planning, roadway drainage design, and combined sewer system analysis. He is especially adept at using state-of-the art tools, including GIS, to communicate complex ideas to a wide range of audiences. He enjoys managing and displaying data to tell a story, to inform decisions, and support solutions to complex problems. Duane M. McClelland, P.E. Water Resources Senior Project Technologist CH2M th Avenue NE, Suite 500 Bellevue, WA Phone/ (425) / Duane.McClelland@ch2m.com Duane McClelland is a registered water resources engineer with CH2M in Seattle, Washington, who has provided leadership in design, planning and management for more than 25 years. For the past 18 years, he has specialized in water resources engineering, including hydraulic and hydrologic modeling and design, dam design and safety, feasibility and alternatives assessments, emergency preparedness, and multifaceted planning and decision support. Mr. McClelland has engineered small to large storage projects and hydraulic structures, coordinated teams of consultants in multidisciplinary water-resources planning, characterized project benefits and risks, and ensured the quality of complex hydraulic and hydrologic analyses. He offers strengths in group and process facilitation to help achieve solutions to multidimensional challenges. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 17 of 18

18 Jason A. Eichler, P.E., PMP Senior Engineer Lower Colorado River Authority 3700 Lake Austin Blvd. Austin, TX Phone/ (512) / Jason.Eichler@lcra.org Jason Eichler is a registered civil engineer at the Lower Colorado River Authority (LCRA) in Austin, Texas with 19 years of experience performing preliminary engineering, design, and operations of water resources, water, and wastewater facilities. He has served as the lead engineer for LCRA s Lane City Reservoir project for the past three years. As lead engineer, Mr. Eichler has been responsible for coordinating various engineering and operations teams during the design phase of the project. Nathan M. Gullo, P.E. Senior Engineer Safety of Dams Lead Lower Colorado River Authority Box 220, L300 Austin, TX Phone/ (512) / Nathan.Gullo@lcra.org Nathan Gullo is a registered civil engineer at the Lower Colorado River Authority (LCRA) in Austin, Texas with 16 years of planning, design, and operations experience on dams, raw water, treated water, and wastewater facilities. He has served as the lead engineer managing the LCRA s safety of dams program for the past four years. Mr. Gullo has extensive experience as the owner s engineer for planning, design, operation, and management of water-resources systems and facilities for public utilities. This work has included independent hydrologic and hydraulic modeling as well as collaborative and review efforts with engineering consulting firms. Copyright 2015 Association of State Dam Safety Officials, Inc. All Rights Reserved Page 18 of 18