Assessing the Capability of a Regional-Scale Weather Model to Simulate Extreme Precipitation Patterns and Flooding in Central Texas

Abstract

A regional-scale weather model is used to determine the potential for flood forecasting based on model-predicted rainfall. Extreme precipitation and flooding events are a significant concern in central Texas, due to both the high occurrence and severity of flooding in the area. However, many current regional prediction models do not provide sufficient accuracy at the watershed scale necessary for flood mitigation efforts. The Weather Research and Forecasting (WRF) model, created with the purpose of improving upon the current fifth-generation Pennsylvania State University?National Center for Atmospheric Research (PSU?NCAR) Mesoscale Model (MM5), is specifically designed for regional grid spacings of 1?10 km. Previous research by the authors resulted in the development of a regional-scale prediction system over the San Antonio River basin, using a geographic information system (GIS) database, a hydrologic model, and a hydraulic model. Observed precipitation drives the prediction system; the authors hypothesize that the WRF model has the potential to predict flooding, at a lead time of several days, with a level of accuracy near that of observed precipitation. Causes of model error are also investigated, to determine the relative errors caused by model physics, initialization interval, buffer zone and domain size, and small-amplitude random errors. Results show that the Betts?Miller?Janji? cumulus and Lin et al. microphysics schemes, 48-h initialization interval, and two-domain configuration covering minimal ocean and having a parent-to-nest area ratio of greater than 10 best simulates a recent (July 2002) large storm event over the San Antonio River basin. This particular storm was selected because it produced extreme rainfall volumes and intensities, and also because its meteorological characteristics are typical of central Texas storm events. Location errors in rainfall are most significant because of their typically nonlinear patterns (increasing location error does not linearly modify streamflow output). Errors in intensity and timing show a more predictable (linear) watershed response that may be useful in the estimation of streamflow ranges for flood forecasting.