Abundant probability information from ensemble forecasting has been used to forecast extreme weather such as extremely high/low temperatures. In this study, leveraging the Convection Permitting Ensemble Forecasting System (CPEFS) of North China rather than coarse-resolution global ensemble forecast systems as in previous studies, based on the Anderson–Darling (A–D) test principle, we built an Extreme Weather Forecast Index (EFI) for temperature to forecast extreme temperature events over North China. Using the CPEFS’s 3-yr historical forecast data, a cumulative distribution function (CDF) for temperature in North China was constructed, establishing a refined model climate capable of identifying extreme temperatures with geographical specific features. The temperature EFI was formulated to reflect differences between the ensemble forecast CDF and the model climate CDF, which can be used to forecast extreme temperatures. We conducted simulation experiments for extremely high (low) temperatures in summer (winter) 2023. The results demonstrate that the EFI warning signals for extreme temperature aligned reasonably with the automatic station observations. It is found that the optimal EFI thresholds are 0.7 or 0.8 (−0.8) for extremely high (low) temperatures. When applying the EFI thresholds to the extreme temperature events in China in 2023, EFI showed a strong skill at the 2-day lead time. However, the forecast accuracy decreased as the forecast lead time extended. Comparison between global ensemble-based EFI and CPEFS-based EFI reveals that high-resolution ensemble-based EFI products could in general achieve a better performance, providing strong warning signals and more refined geographical distributions.

This paper describes a fire forecast system—Weather Research and Forecasting-Fire (WRF-Fire)—that is employed to simulate a real wildfire case in Xichang, Sichuan Province, Southwest China on 30 March 2020 at a 100-m resolution over the fire area, in order to provide a fine representation of the terrain and fuel heterogeneities and explicitly resolve the atmospheric turbulence. Four sensitivity experiments were conducted to analyze the impacts of atmospheric model grid spacing and fire–atmosphere interaction on simulated meteorological fields and fire behavior. The results indicate that finer horizontal grid spacing in the atmospheric model improves the accuracy of wind, temperature, and moisture simulations in the near surface layer. Especially, it can better describe local wind field characteristics, capture microscale wind speed fluctuations, and produce more significant effect from fire–atmosphere interaction. The mass and energy released by the fire model and its feedback to the atmospheric model exhibit enhanced heterogeneous characteristics. The simulated fire area aligns well with the observation, with KAPPA coefficient (KC) of 0.56–0.59 and spatial correlation coefficient (SC) of 0.52–0.59. For this real case, the influence of heterogeneous land surface on the fire behavior is much greater than the atmosphere–fire interaction. The study suggests that WRF-Fire holds high potential as a real wildfire simulation tool, offering a new and feasible approach for fire prediction.