Abstract: Based on meteorological data from 353 ground stations in Yibin City from 1990 to 2020, three correction methods were applied to localize the intelligent grid forecast products, and a localized forest fire weather prediction model was established. The results show that: (1) After interval classification processing, the intelligent grid temperature forecasts with values exceeding historical extremes were effectively corrected, with improvements within 1 ℃ and a bias reduction of 0.1~0.4 ℃. After extrapolation correction using observed data, the error range decreased by approximately 1~2 ℃, the correlation coefficient increased by 5%, and the root-mean-square error decreased by approximately 0.2~1.5 ℃. (2) For the intelligent grid precipitation forecasts, false alarms of weak rainfall were eliminated after extrapolation correction combined with false alarm suppression. (3) The model can provide 1 km × 1 km grid-based forecasts of forest fire risk probability and fire spread trends. It distinguishes variations in fire danger levels at different times and identifies high-risk periods within a day. Additionally, the 24-hour precipitation forecasts offer scientific guidance for firefighting operations. (4) Historical case validation demonstrates that the model exhibits advanced fire risk early warning capabilities, assisting relevant authorities in formulating more scientific and rational fire prevention measures.