Under the global warming, substantial changes have occurred in the intensity and frequency of extreme heat events in China, and accurate prediction of such events is critical to disaster prevention and mitigation. Extreme forecast index (EFI), as an effective method, has been widely used in extreme weather short-range prediction research and operation. This study applies the EFI method to the subseasonal forecasts of summer extreme high temperature in China based on the subseasonal-to-seasonal (S2S) model data from the ECMWF during 2015–2023. The results show that the EFI method can provide skillful predictions of extreme high temperature of surface air at an 8-day lead time in China, with good performance on the subseasonal timescale. Through a new verification formula using normalization, we show that the EFI of high temperature in China has decent prediction skills at 1–19-day lead times, and as lead time extends, the Threat Score (TS) and Factions Skill Score (FSS) decline. In addition, the threshold of extreme temperature has been identified by using a similar normalization method; that is, an observed dimensionless temperature anomaly value of around 1.28, which corresponds to the 90th percentile of the temperature data, can be recognized as an extreme high temperature. The effectiveness of the EFI verification and threshold identification approaches shed lights on the improvement for subseasonal prediction of extreme high temperature in China.

Ensemble prediction play a vital role in numerical weather prediction. Hence, how to effectively extract information from ensemble members to improve deterministic precipitation forecasting skills has always been a challenging issue. Based on precipitation data from the CMA-GEPS (China Meteorological Administration Global Ensemble Prediction System), a stepwise correction method based on Segmented Hierarchical Clustering (SHC) for daily precipitation is proposed. To evaluate the effect of the SHC method, two comparative experiments are further verified in this study based on TS score and ETS score. Effects of SHC are compared with that of an Ensemble Mean (EM) forecast method and a directed Hierarchical Clustering (HC) method. Results indicate that the deterministic forecast by the proposed SHC method can improve the predictability of heavy precipitation. Taking more probabilistic forecasting via the segmented correction scheme in correction method, SHC performs better than EM and HC. Meanwhile, SHC shows a better feasibility based on long-term forecast verification during the summertime in 2021; it also has much better effects on extreme precipitation event such as the case of heavy rainfall in Zhengzhou on 20 July 2021. For operational system, the developed SHC method provides a new tool that can further improve the result of ensemble forecasting.