Hyper Resolution Modeling of Urban Flood Inundation

This report presents an evaluation by the US National Weather Service of two hydrologic models to produce street-level flood predictions. Model selection involved a literature review from English-language water resources journals that identified 18 candidate models. We performed an e-mail survey of model authors, asking three questions: (1) Is your model parallelized to run in an HPC environment?; (2) Does your model have at a minimum two dimensional overland flow coupled to one dimensional channel flow?; and, (3) Is your model open source? Two models were selected for testing based on the survey responses. Model performance evaluation used data from the 679km2 watershed above US Geological Survey (USGS) gage 02146800, Sugar Creek near Fort Mill, SC, which contains the Charlotte, N.C. municipal area. Advantages include few levees and large reservoirs, and a rich flood hydrograph and inundation data set. Only soils and LULC data with national coverage were used to assign parameters. A total of 300 simulated and observed hydrograph pairs were generated and analyzed from simulations of three extreme storm events using data from a nested set of 25 USGS gages. Nearly 2000 observations of high water marks allowed evaluation of inundation predictions using multiple metrics including a newly developed spatial analysis technique accounting for model element size and discretization. Our study also demonstrated a novel technique to evaluate numerous crowd-sourced observations of flooded locations. Finally, results indicated clear performance differences between two very different model formulations. 2020 Grant no. NA16NWS4620043 Grant no. NA18NWS4620043B NWS (National Weather Service) NWC (National Water Center) Submitted https://doi.org/10.25923/9t55-tn77 Public Domain 1931