Application of neighborhood spatial verification method on precipitation evaluation

Methods for dichotomous forecasts (yes/no) (Threat Score et al.) are always used in the verification of precipitation forecasts. But these standard methods are affected by the well-known double penalty problem. In a point to point comparison, small displacements in space or time with regard to the observation causes double-penalty to the forecast. Spatial techniques as the popular scientific and diagnostic methods have been recently developed to alleviate this problem. The neighborhood spatial verification method Fraction Skill Score (FSS) directly compares the fractional coverage of events in windows surrouding the observations and forecasts. Based on daily forecast products from GRAPES regional model and QPE products from national meteorological information center during July and August in 2016, we study the effects of FSS in precipitation verification and its difference from traditional skill score. It shows that the TS can get a similar trend to FSS with low precipitation threshold and becomes unstable with high threshold. The FSS show a similar trend with different thresholds, and it still represents a significant difference in the period of weak forecast that should distinguish features of daily heavy rainfall forecast. By partitioning verification for North and South, FSS score in the southern region shows the ability to describe the strong rainfall significantly better than the TS score. In the area with relatively severe precipitation (north), the TS score has a low ability to describe the heavy rainfall, and the neighborhood spatial test method can effectively show the difference of daily precipitation forecast.