Radar-based short-term rainfall prediction

Abstract

A radar-based short-term rainfall prediction model is formulated and evaluated. The prediction lead time of interest is approximately 1 h. The model is composed of a physically based component and a statistical component. The physically based part performs mass balancing of mean vertically integrated liquid water content (VIL) under convective warm rainfall situations, using full-volume scan radar data, surface meteorological observations and upper air data. The statistical part performs prediction of residual VIL. Conversion of predicted VIL to rainfall is made using empirical relationships among VIL, rainwater content at cloud bottom, and echo-top height, which is assumed to remain constant over the prediction lead time. To evaluate the model, a comparison is made against advection-based nowcasting using radar data from the National Weather Service Radar Data Processor, version II (RADAP II) system at Oklahoma City. Results from parameter estimation runs show that inclusion of the simple physical and statistical dynamics has potential in improving advection-based nowcasting under convective situations. An apparent bias in mean rainfall prediction, however, suggests room for improvement. Issues concerning possible improvements are described, and future research directions are discussed.