Probabilistic Radar Nowcasting Based on Time Nuggets

A new probabilistic radar nowcasting technique based on Lagrangian rainfall extrapolation is presented. The technique combines deterministic advection of radar pixels along the direction of apparent motion together with a scale-dependent stochastic innovation term based on the recently proposed concept of “time nugget”. The spatial structure and magnitude of the innovation terms are determined by linking spatial and temporal variabilities of rainfall rates along the direction of apparent motion. The proposed technique is applied to the Dutch 1 x 1 km and 5 min national C-band radar composites for a 100 x 100 km domain covering the city of Rotterdam. Special emphasis is put on analyzing the small-scale variability and predictability of heavy convective rain events and on testing different strategies for extrapolating radar rainfall estimates taking into account anisotropy and possible spatio-temporal trends. The usefulness of the proposed technique is demonstrated by generating 100 ensemble members for different lead times between 5 min and 2h and by comparing the accuracy and structure of the forecasts to observations. Detailed comparisons with deterministic radar-nowcasts provided by the Dutch meteorological office show that the probabilistic rainfall forecasts provide a more accurate depiction of spatio-temporal variability and structure with lead time than their deterministic counterparts.