Predicting forest fire kernel density at multiple scales with geographically weighted regression in Mexico

Area of Science:

• Environmental Science
• Spatial Analysis
• Fire Ecology

Background:

• Understanding human and environmental drivers of fire occurrence is crucial for effective fire management.
• Limited research exists on multi-scale spatial fire patterns and driver relationships, especially at regional to national levels.
• Spatial non-stationarity in fire occurrence-driver relationships requires further investigation across various scales.

Purpose of the Study:

• To predict spatial fire occurrence patterns at regional and national scales in Mexico.
• To utilize geographically weighted regression (GWR) to model fire density using both regular grid and kernel density approaches.
• To analyze fire density prediction at multiple spatial resolutions (5-50 km) for both dependent and independent variables.

Main Methods:

• Employed geographically weighted regression (GWR) for spatial prediction of fire density.
• Calculated fire density using two methods: regular grid density and kernel density.
• Tested multiple spatial resolutions (5-50 km) for both fire density and predictor variables.

Main Results:

• GWR demonstrated superior performance in predicting kernel density compared to regular grid density, evidenced by better goodness of fit and reduced residual correlation.
• Model performance improved with increasing kernel density search radius (bandwidth), with predictive capacity saturating around 15-20 km.
• This study is the first to apply GWR for predicting fire kernel density and to incorporate multi-scale analysis for both dependent and independent variables.

Conclusions:

• The 15-20 km spatial scale appears optimal for operational fire prevention and suppression decision-making, balancing predictive accuracy and spatial detail.
• Kernel density estimation combined with GWR offers a robust approach for spatial fire occurrence prediction.
• Further research is recommended to validate these findings in different geographical contexts.