An Empirical Model of High Spatial and Temporal Resolution for Radar Rainfall Nowcasting

A short term rainfall prediction algorithm for intense storms is introduced in this work. The algorithm uses high spatial and temporal resolution (0.06 km and 1 min) radar data to predict the evolving distribution of rainfall rate. It is assumed that for a short time period, (10 min) a rain cloud behaves as a rigid object, with all parts moving in the same direction at a constant speed. Thus, the most likely future rainfall areas are estimated by tracking rain cell centroid advection in consecutive radar images. To achieve this estimation, a nonlinear regression model varying in the time and space domain is proposed to predict the most likely rainfall patterns. This model is based on the assumption that the current radar reflectivity is a function of the previous reflectivity observed in surrounding areas centered on the location of a predicted pixel. A second assumption is that the ratio of reflectivity of a given pixel to reflectivity of convective core is a fundamental predictor for rainfall estimation. This rainfall nowcasting algorithm was validated against five rainfall events which occurred over western Puerto Rico between March and October 2012, and in February 2014. Probability of detection, false alarm rate, and the Heike Skill Score were: 0.64, 0.27, and 0.61, respectively. The average root mean squared error was 0.03 mm/hr. The suggested high resolution system can improve the operational forecast especially when convective core is below of 3 km or less over the surface. Results show that the nowcasting algorithm is a potential tool to couple with a hydrological numerical model to forecast inundation areas.