Accuracy Evaluation of BCC_S2S Summer Precipitation Forecast in the Upper and Middle Reaches of the Yangtze River

In order to evaluate the forecast of the sub-seasonal to seasonal (S2S) forecasting system of the Beijing Climate Center (BCC),called BCC_S2S model for short, for daily precipitation and summer extreme precipitation events in the upper and middle reaches of the Yangtze River (UMRYR), based on the products of the BCC_S2S model, which reforecasts 60-day precipitation twice a week from 2005 to 2020, bilinear interpolation is implemented to downscale grid point data to station data. Then, various comparison is applied to evaluate the model performance. Three indexes are used to evaluate the daily precipitation forecasting, as correlation coefficient (CC), root mean square error (RMSE) and mean error (ME). For extreme precipitation events defined by percentile method, HSS index is used to evaluate individual station precipitation extremes. Hierarchical clustering method is used to classify regional precipitation extremes, and different precipitation extreme pattern is evaluated by comparing to the observations. The results show that, for the daily precipitation, the BCC_S2S model performance decreases with the increase of the leading period in each season, and the forecast skill stays low when the leading period is longer than 5～10 days. Forecast skill is evaluated detailly in six sub-regions. The CC indexes show similar trend as decreasing with the leading period increasing. And the ME indexes indicate that the model may over estimate precipitation in the middle-east parts of UMRYR, while under estimate that in Jinsha River Basin. The RMSE indexes are high in the middle-east parts of UMRYR. The box plots show that in most parts of UMRYR, the BCC_S2S model performs more stable in June due to the small range of errors. For the precipitation extremes, the HSS decreases with the leading period increasing in forecasting station precipitation extremes. In most parts of the UMRYR, the model performs better in high total-precipitation months. In the perspective of spatial distribution, the model can well represent the strong rainfall pattern for the four regional precipitation extremes in short leading period (0～10 d), but not in long leading period over 10 days. Generally, the BCC_S2S model performance for the daily precipitation and precipitation extremes in the UMRYR shows decreasing trend with the leading period increasing. And it performs better in June, which may be related with the better model performance in representing large scale synoptical systems such as the water vapor transportation or the front. © 2022 Editorial Department of Journal of Sichuan University. All rights reserved.