Improved precipitation nowcasting: Model errors and their correction in operational NWPs at different scales

Recent developments in various data assimilation techniques significantly improved the performance of precipitation nowcasting from numerical weather prediction models. However, these precipitation forecasts still suffer with inherent errors in terms of location and intensity. Nevertheless, their errors are persistent in time and space.

The Adjustment of Rain from Models with Radar (ARMOR) algorithm was developed to correct these persistent errors and consequently to improve the model forecast. The ARMOR assumes the linear variation of model phase errors in time and persistency of intensity errors along the storm motion. This algorithm uses the variational echo tracking used in MAPLE to determine the two-dimensional phase error vectors at different time steps between radar composites and model precipitation forecasts. Furthermore, their linear tendency is derived along with intensity errors at each model pixels. Finally, the time-dependent phase error and intensity error are extrapolated to the future model forecasts and are applied to the original model forecasts.

The ARMOR is applied at three operational models (Canadian GEM, WRF, GSD HRRR, and MM5 at the scales of ~ 700 km to ~ 7000 km model domains) and different forecast products (hourly precipitation accumulation, instantaneous rainfall intensity, and vertically integrated liquid water content). Results are extremely promising. A long-term verification shows that regardless of products the skill scores are significantly improved against the original forecast. Comparison with MAPLE shows the base skill of the original model is a key factor to overcome MAPLE performance. Thus, a modification of ARMOR is essential in cases of poor performance of original NWP models. The “tapered” version of ARMOR is implemented in real-time operation. This presentation will summarize findings and pros and cons of the ARMOR technique.