Fusion of multiple radar-based quantitative precipitation estimates (QPE) for high-resolution flash flood forecasting in large urban areas

Abstract:

With increasingly widespread use of weather radars, multiple radar-based QPEs are now routinely available in many places. In the Dallas-Fort Worth Metroplex (DFW), for example, the Multisensor Precipitation Estimator (MPE), Q2 (Next Generation QPE) and CASA (Collaborative Adaptive Sensing of Atmosphere) QPEs are available. Because these products are based on different radar systems, different sources of additional information, and/or processing algorithms, they have different error characteristics and spatiotemporal resolutions. In this work, we explore improving the accuracy of the highest-resolution radar QPE by fusing it with lower-resolution QPE(s). Two approaches are examined. The first is to pose fusion as a Fisher estimation problem in which the state vector is the true unknown precipitation at the highest resolution and the observation vector is made of all radar QPEs at their native resolutions. The second is to upscale the higher resolution QPE(s) to the lowest resolution, merge them via optimal estimation, and disaggregate the merged estimates based on the spatiotemporal patterns of precipitation in the high resolution QPE. In both approaches, we compare Fisher estimation with conditional bias-penalized Fisher-like estimation which improves estimation of heavy-to-extreme precipitation. For evaluation, we compare the precipitation estimates from the two approaches with rain gauge observations in the DFW area.