Applying Scale Decomposition Method to Verification of Quantitative Precipitation Nowcasts

Verification of quantitative precipitation forecast has always been a challenge due to the high discontinuity of rainfall in spatial and temporal scales. Conventional methods (TS and MSE score, etc.) cannot meet the need for verification of high resolution (1—2 km) forecasts. In recent years, several new spatial verification approaches which can give more information about the complex spatial structure of forecast field have been developed, such as the intensity-scale verification method, the feather based (or object-oriented) verification method, neighborhood (or fuzzy) verification method and field deformation approaches etc. The attempt to use the intensity scale technique is introduced by Casati et al. to assess radar based 1 hour quantitative precipitation forecasts for 4 different nowcasting systems (BJANC, GRAPES SWIFT, STEPS, and CARDS) which attended the World Weather Research Program for the Beijing 2008 Olympic Games (WWRP B08FDP). The intensity-scale verification approach accounts for the spatial structure of the forecast field and allows the skill score to be diagnosed as a function of the spatial scale of the forecast error and intensity of the precipitation events. Different precipitation types (convective, stratiform and mixed type) during B08FDP demo period are selected to get representative results, it shows that these nowcasting systems exhibit forecasting skills only when the precipitation system is above 32 km spatial scale and last longer than 1 hour despite the using of most advanced systems in the world with the high resolution of 1 to 2 km. For spatial scale lower than that, the forecasting ability is very limited, which indicates that product performance characterized at different spatial scale should be considered in the applying of nowcasting products. When analyzing the forecasting errors with different spatial and temporal scales, it's found that more than 60% and more than 85% of the forecasting error come from spatial scales smaller than 8 km and time scales smaller than 1 hour respectively. Improvements in smaller scales precipitation forecasts are important. Most nowcasting systems explore the linear extrapolation technique to make 0 to 2 hours nowcasts, yet the valid extrapolation time is very limited, especially for smaller scale (less than 1 hour) convective systems, which is mainly caused by the nonlinear development of the convections. Therefore, more information about the circulation which has close relation to the movement and development of the storm should be considered. So far, the most popular way is to blend the radar based extrapolation with the dense observations and numerical model based potential forecast.