Abstract: This study is based on multi-source satellite hotspot data and ERA5-Land meteorological data from 2001 to 2024. The spatiotemporal distribution characteristics of hotspots were extracted using GIS spatial analysis. Multiple linear regression（MLR) and autoregressive moving average(ARMA) models were employed to analyze the driving mechanisms of meteorological factors, with model parameters optimized using the Akaike Information Criterion(AIC). On this basis, a provincial local dual-layer prediction framework was constructed to overcome the limitations of traditional monitoring methods. Multiple linear regression analysis revealed that climatic factors exert a significant driving effect on regional environmental changes, the annual mean temperature (β = 0.38, p < 0.001) shows a strong positive correlation with the dependent variable; The moisture coefficient (β =-0.29, p = 0.005) exhibits a significant inhibitory effect; and the annual mean wind speed (β = 0.18, p = 0.006), despite a relatively modest increment, remains statistically significant, indicating its sustained contribution to environmental change. The results further show that forest fire hotspots during from 2015 to 2024 exhibit a fluctuating downward trend, mainly concentrated in southwestern and southeastern Yunnan; both the provincial multiple linear regression model and the local model for Kunming demonstrate satisfactory predictive performance. Nevertheless, future work should integrate machine learning methods and terrain data to enhance model applicability in topographically complex regions. This study provides theoretical support and technical reference for forest fire prevention and control in southwestern China, and is of significant value for safeguarding ecological security and reducing economic losses.

Key words: forest fire, satellite monitoring, forest fire hotspots, regression model