The influence of observational probability of precipitation on the verification of probability of forecasted precipitation

Based on the precipitation forecast data of ECMWF ensemble prediction system (EPS) and the hourly precipitation observational data of the national meteorological stations between June and August from 2015 to 2017, taking the verification of ensemble prediction in the central and eastern China from the EPS as an example, we first analyzed the reliability and deviation sample characteristics of the ensemble prediction system on the precipitation forecast at different levels. Then, based on whether the observed events occur, the effects of different observational conditional probabilities on the ensemble forecast BS score (BS) and ensemble precipitation forecast discrimination, respectively, were discussed. The results show that (1) EPS has the best performance in forecasting moderate rainfall, but it overestimates the precipitation of light rain level, and most overestimation appears in the small precipitation range of 0-2 mm. For precipitation above heavy rain, EPS forecasted precipitation is dominated by smaller precipitation, and misses almost all of the precipitation above 85 mm. (2) When precipitation is observed, the negative bias of forecasted probability against the observation gradually increases with the increase of precipitation level. When precipitation is not observed, the positive bias of forecasted probability against the observation gradually decreases with the increase of precipitation level. Through calculating the BS of forecasted probability separately for the observational probability of 0 or 1, the defects, that the BS cannot be used to evaluate objectively the forecast ability of probabilistic precipitation due to uneven sample distribution, can be eliminated to some extent. In addition, it can be directly obtained from the validation that the main reason of prediction bias is due to insufficient or excessive forecast.