Evaluation of the 2.5-km Cloud-Resolving Storm Simulator in Predicting Local Afternoon Convection during the Summer in Taiwan

The forecast of summertime afternoon convection by numerical weather prediction models is highly challenging because of its weak dynamical forcing, small scale, and low predictability. To assess such an ability for future improvement, we evaluated the performance of the 2.5-km Cloud-Resolving Storm Simulator (CReSS) in predicting afternoon convection in Taiwan under weak synoptic conditions during the summers (May–October) of 2011 and 2012. For a total of 89 target days, daily CReSS forecasts in three ranges, starting at 0000 UTC (0800 LST) on the same day (D0), the day before (D-1), and 2 days before (D-2), were examined. With regard to the occurrence of afternoon convection anywhere in Taiwan, the predictive skill of the model was considerably high, as evidenced by the threat score (TS) and post-agreement (PA) of D0 (D-1) forecasts reaching 0.75 and 0.90 (0.50 and 0.79), respectively. While the score values decrease when Taiwan is divided into four regions (i.e., the forecast must be in the correct region to be considered a hit), the TS and PA for D0 (D-2) forecasts remain respectable at 0.44 and 0.73 (0.29 and 0.59). Among the four regions, the TS (0.48–0.65) and PA (0.77–0.83) for Central Taiwan are the highest with the best predictive skill. Overall, while the prediction of afternoon convection in the correct region is challenging, the 2.5-km CReSS model has considerable skill (TS ~ 0.30) even 2 days in advance, and can provide useful guidance for afternoon convection in Taiwan.