Cyberinfrastructure to Support Flood Modeling from Catchment to Regional Scales
Morsy, Mohamed, Civil Engineering - School of Engineering and Applied Science, University of Virginia
Goodall, Jonathan, Civil & Env Engr, University of Virginia
Flooding events are projected to increase in frequency and intensity in coming years due to climate change. New tools and approaches are needed to assist decision makers in better understanding and addressing societal impacts due to flooding and how to mitigate these impacts. This research addressed three challenges related to flooding impacts: (i) better understanding how distributed stormwater infrastructure can mitigate flooding in urban catchments, (ii) designing and building spatially-detailed, real-time flood warning systems for emergency management purposes, and (iii) designing and building cyberinfrastructure to support reuse and transparency in both flood modeling and hydrologic modeling more broadly. The goal of this research was to address these challenges by conducting three studies.
The first study explored building a catchment-scale flood model used to improve understanding of how distributing low impact development (LID) practices at the parcel level in an already urbanized watershed reduces runoff and, therefore, flood risk. At this scale and in an urban environment, spatially detailed descriptions of the physical environment are required. A physically-based modeling approach was used in order to answer "what if" hypothetical scenarios of rain garden adoption rates and their impact on watershed-scale runoff generation.
The second study explored building an automated cloud-based system for forecasting flooded roadway and bridge locations at a regional-scale. Because the study area has very low topographic relief, a two-dimensional (2D), computationally-expensive hydrodynamic model is required. This study demonstrated the ability of using instances in a public cloud with powerful graphical processing units (GPUs) to run a large (average of 4 million nodes) 2D hydrodynamic model in a time frame relevant to real-time emergency management applications. The steps required to build this system were (i) creating an automated workflow for obtaining and processing forecast rainfall data, (ii) running the 2D model in the cloud, (iii) using geospatial analysis tools to identify flooded bridges, and (iv) presenting the results online for decision makers. The system automates forecast data access and pre-processing, execution of a high-resolution 2D hydrodynamic model, and map-based visualization of model outputs using Amazon Web Services (AWS).
The third study advanced approaches for sharing hydrologic models, such as the models created in this dissertation, through community supported cyberinfrastructure. Sharing models is important for scientific reproducibility, reuse, and fidelity. In this study, the first task was to design a metadata framework for hydrologic models that is flexible and applicable across the wide variety of models used by hydrologists. Then the study demonstrated the utility of this framework for sharing, publishing, and reusing models through an implementation within the HydroShare cyberinfrastructure system.
In the first study, the results suggest that rain gardens with 30 cm berm heights and a total area equal to 20% of the impervious surfaces within the watershed should provide sufficient storage to mitigate flooding for rain events up to and including a 10 year return period storm event. The results also suggest approximately 15%, 27%, and 38% of the runoff generated from impervious surfaces should be diverted to the rain gardens to mitigate flooding from 2, 5, and 10 year return period storm events, respectively. Given prior work on the adoption of LID approaches for other watersheds, rain gardens could effectively mitigate up to a 5 year return period storm event within the watershed, although further research on possible adoption rates in the study watershed is needed to more fully support this conclusion. In the second study, an 80x speed-up in execution time of the 2D models was achieved by using GPUs rather than a central processing unit (CPU). A prototype deployment system was built within the Amazon Web Services (AWS) cloud that includes a web front-end, execution for the model engine, and storage of the model output data. The system is designed to run automatically during extreme weather events, produce near real-time results, and consume few computational resources until triggered by an extreme weather event. Although the model is built for a specific region of Virginia, the architecture serves as an example that could be replicated to other regions where 2D hydrodynamic models are required for real-time flood warning applications. In the third study, a general approach for representing environmental model metadata that extends the Dublin Core metadata framework was proposed. The framework was implemented within the HydroShare system and applied for a hydrologic model sharing use case. This example application demonstrates how the metadata framework implemented within HydroShare can assist in model sharing, publication, reuse, and reproducibility.
PHD (Doctor of Philosophy)
Cyberinfrastructure, Stormwater management, Urban hydrology, Urban flooding, Storm water management model (SWMM), Low impact development (LID), Environmental modeling, Hydrologic modeling, Model metadata, Linked data, Dublin core metadata initiative, Reproducibility, Flood warning applications, cloud computing, graphical processing units (GPUs), Amazon Web Services (AWS)