Toward high-resolution flash flood prediction in large urban areas - Analysis of sensitivity to spatiotemporal resolution of rainfall input and hydrologic modeling

Summary Urban flash flooding is a serious problem in large, highly populated areas such as the Dallas–Fort Worth Metroplex (DFW). Being able to monitor and predict flash flooding at a high spatiotemporal resolution is critical to providing location-specific early warnings and cost-effective emergency management in such areas. Under the idealized conditions of perfect models and precipitation input, one may expect that spatiotemporal specificity and accuracy of the model output improve as the resolution of the models and precipitation input increases. In reality, however, due to the errors in the precipitation input, and in the structures, parameters and states of the models, there are practical limits to the model resolution. In this work, we assess the sensitivity of streamflow simulation in urban catchments to the spatiotemporal resolution of precipitation input and hydrologic modeling to identify the resolution at which the simulation errors may be at minimum given the quality of the precipitation input and hydrologic models used, and the response time of the catchment. The hydrologic modeling system used in this work is the National Weather Service (NWS) Hydrology Laboratory’s Research Distributed Hydrologic Model (HLRDHM) applied at spatiotemporal resolutions ranging from 250 m to 2 km and from 1 min to 1 h applied over the Cities of Fort Worth, Arlington and Grand Prairie in DFW. The high-resolution precipitation input is from the DFW Demonstration Network of the Collaborative Adaptive Sensing of the Atmosphere (CASA) radars. For comparison, the NWS Multisensor Precipitation Estimator (MPE) product, which is available at a 4-km 1-h resolution, was also used. The streamflow simulation results are evaluated for 5 urban catchments ranging in size from 3.4 to 54.6 km 2 and from about 45 min to 3 h in time-to-peak in the Cities of Fort Worth, Arlington and Grand Prairie. The streamflow observations used in evaluation were obtained from water level measurements via rating curves derived from 1-D steady-state non-uniform hydraulic modeling. The results indicate that a spatiotemporal resolution of 500 m and 15 min or higher is a good choice for streamflow prediction using HLRDHM and CASA QPE in the study area, but that, due to the nonlinear accretion of random errors in QPE and imperfect model dynamics, there are trade-offs to consider among resolution, timeliness of prediction and prediction accuracy.