High-resolution modeling and prediction of urban floods using WRF-Hydro and data assimilation

Abstract

We assess the impact of increasing the resolution of hydrologic modeling, calibration of selected model parameters and assimilation of streamflow observation toward event-based urban flood modeling and prediction using WRF-Hydro in the Dallas-Fort Worth area (DFW). We use quantitative precipitation estimates at 500-m 1-min resolution from the Collaborative Adaptive Sensing of the Atmosphere radar network for observed rainfall, Stepwise Line Search for calibration, and fixed-lag smoothing for data assimilation (DA). The model domain is a 144.6 km2 area comprising 3 urban catchments in Arlington and Grand Prairie in the middle of DFW. It is shown that event-specific calibration of 6 WRF-Hydro parameters is largely successful in simulating hydrographs at the catchment outlets particularly for the most important rising limbs, but less so for attenuated peaks or fast-receding falling limbs. A spatial resolution of at least 250 m was necessary for the land surface model (LSM) to delineate small catchments and hence to capture catchment-wide rainfall with acceptable accuracy. Simulations at selected combinations of resolutions, 250 and 125 m for the LSM and 250, 125, 50 m for the routing models, showed mixed results. The overall results indicate that, in the absence of resolution-specific prescription and calibration of channel routing parameters, a resolution of 250 m for both the LSM and routing models is a good choice in terms of performance and computational requirements, and that, in the absence of high-quality calibration and continuous simulation of streamflow, DA is necessary to initialize WRF-Hydro for event-based high-resolution urban flood prediction.