Potential Role of Deep Learning in Operational Radar Rainfall Mapping: Rationale and Demonstration Studies

Although polarimetric weather radar measurements provide rich information about precipitation, traditional approaches of using them can only extract part of the information due to limitations of the applied analytical tools. The performance of conventional radar rainfall estimation algorithms, such as the radar reflectivity ( Z ) and rain rate ( R ) relations, is highly dependent on the local raindrop size distributions that vary in different precipitation regimes. This article revisits the surface rainfall mapping problem using operational radar observations, but from a deep learning perspective. For demonstration, a machine learning model driven by a convolutional neural network and a vision transformer is designed to estimate surface rainfall using multiparameter multidimensional observations from U.S. weather surveillance radar-1988 Doppler (WSR-88D). Results from the deep learning model are compared with rainfall estimates from the operational polarimetric radar algorithms implemented by the WSR-88D and the National Oceanic and Atmospheric Administration’s (NOAA) multiradar multisensor (MRMS) system, which indicates the great potential of deep learning for operational applications.