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

Area of Science:

• Environmental Science
• Remote Sensing
• Forestry

Background:

• Forest fire risk assessment traditionally overlooks diurnal variations.
• Daytime and nighttime conditions present distinct challenges for fire management.
• Understanding spatio-temporal fire dynamics is crucial for Guizhou Province's mountainous terrain.

Purpose of the Study:

• To model and compare daytime and nighttime forest fire occurrence.
• To identify and analyze the drivers of forest fires at different times of day.
• To create time-specific forest fire risk zoning maps for Guizhou Province.

Main Methods:

• Analysis of Moderate Resolution Imaging Spectroradiometer (MODIS) active fire points (2009-2024).
• Integration of meteorological, topographic, vegetation, and socioeconomic data.
• Application of machine learning models (Random Forest, XGBoost, LightGBM, ensemble) and SHAP analysis.

Main Results:

• Forest fires concentrated in spring, primarily in southern and southwestern Guizhou.
• Daytime high-risk areas were extensive, while nighttime risk was localized.
• Shared drivers included humidity, temperature, rainfall, wind, and NDVI; human factors varied diurnally.

Conclusions:

• Distinct drivers influence daytime and nighttime forest fires, necessitating time-specific management.
• The study provides critical insights for targeted forest fire monitoring and prevention in mountainous regions.
• Differentiated risk assessment can enhance the efficacy of forest fire management strategies.