Interpretable machine learning reveals daytime and nighttime forest fire point drivers in Guizhou.

Daytime and nighttime forest fire risk reflects contrasting atmospheric, fuel, and human activity conditions, with implications for time specific fire management. We analyzed moderate resolution imaging spectroradiometer (MODIS) active fire points in Guizhou Province from 2009 to 2024 and integrated meteorological, topographic, vegetation, anthropogenic, and socioeconomic variables to model all, daytime, and nighttime fire points occurrence. Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and ensemble models were compared, and SHAP analysis was used to interpret driver contributions and generate relative fire risk grade zoning maps. Fire points showed strong spring concentration and clustering in southern and southwestern Guizhou. Daytime high-risk areas were broader and more continuous, whereas nighttime high-risk areas were concentrated in fewer core zones. Humidity, temperature, rainfall, wind, and normalized difference vegetation index (NDVI) were shared drivers, while human related factors differed between daytime and nighttime models. This study provides evidence for time-specific forest fire monitoring, risk assessment, and prevention in mountainous regions by revealing daytime and nighttime differences in fire points and their drivers.