A Multi-Model Ensemble Approach for Constructing Subseasonal Extreme Forecast Index of Temperature

Current subseasonal forecasting of extreme temperature in China faces challenges due to relatively low capability of the subseasonal-to-seasonal (S2S) models. Benefiting from ensemble forecasting, the extreme forecast index (EFI) is an effective approach to tackling the challenge; however, it usually requires abundant ensemble members (i.e., dynamic models), while individual S2S models usually have a limited sample of members that constrain their performance in constructing EFI. To address this issue, this study proposes a novel subseasonal EFI approach for temperature utilizing the S2S multi-model ensemble (MME) to provide a much larger ensemble size. This approach, applied to both a bias-corrected MME (MME-C) and a direct MME (MME-D), is evaluated for high temperature scenarios in comparison with the ECMWF single model forecasts. The results show that MME-C, by removing the drifted climatology of individual model and replacing it with observed climatology, can maximally use the unbiased large-sample ensemble information to decently construct EFI. For verification of subseasonal forecast of extremely high temperatures over China, both MME approaches for the EFI construction outperform the ECMWF single model, e.g., threat score (TS) of MME-C can reach 0.52 for 27-day forecasts, and MME-C achieves overall higher skills than MME-D. The decreased skills with lead time exhibit variations across different regions and both of the MME approaches show higher temporal correlation coefficient (TCC) skills in the northern regions such as Northwest and Northeast China and achieve smaller errors across the entire China. Using larger-sample MME information, the MME-C approach proves to be an effective tool for improving subseasonal early warnings of extreme high-temperature risks.