Ability of the GRAPES Ensemble Forecast Product to Forecast Extreme Temperatures over the Tibetan Plateau

Abstract

Due to climate change, extreme temperature events are receiving increased attention. Based on the climate state deviation and threat score (TS), the ability of the Global/Regional Assimilation and Prediction System (GRAPES) ensemble model to forecast extreme temperature events was examined. The “optimal” Extreme Forecast Index (EFI) was derived for plateau forecasting, and its predictability was examined based on the receiver operating characteristic (ROC) curve method. Meanwhile, the applicability of the Shift of Tails (SOT) index to extreme temperature forecasting was analyzed using extreme temperature cases. Results showed that the GRAPES model has a warm bias for both summer extreme high temperature and winter extreme low temperature, and the warm bias decreases slightly with an increase in the forecasting lead time. The ensemble mean and median forecasts are less effective, and the maximum value is more predictable. However, for the ensemble forecast model, the extreme information in its forecast is more unstable, and the limitation of the extreme temperature forecast in the plateau region is higher. With different forecast lead times, the TS tends to increase and then decrease with an increase in the EFI threshold, which means that there is an optimal EFI. The optimal EFI thresholds for summer extreme high-temperature forecasts are all less than −0.5, while for winter extreme low-temperature forecasts, they are almost all less than 0. From the ROC curves, the EFI has a certain level of predictability for summer extreme high temperatures but poorer forecasting effects. Furthermore, the EFI has some predictability for extreme summer high temperatures, but the prediction effect is poor. For the extremely low temperatures in winter, which are poorly predicted by the model itself, post-processing of the extreme information predicted by the model with the EFI can improve the forecasting effect of the model. Through analysis of individual cases, it was found that the extreme intensity reflected by the SOT_+ (0.9) index of the model was closer to reality for the prediction of extremely high temperatures, whereas for the prediction of extremely low temperatures, the extreme intensity indicated by the SOT_− (0.1) index of the model was weaker. Therefore, the SOT index can play an important auxiliary role in the prediction of the intensity of extreme events based on the EFI.