Mapping Spatial Drivers of Predicted Active Fires Kernel Density with Geographically Weighted Regression in Mexico

Despite the need to understand the spatial variation in human and biophysical drivers of fire spread, studies aimed at predicting remotely sensed fire kernel density at multiple scales are still relatively scarce. The current study aimed at the prediction of MODIS and VIIRS active fire kernel density (AFKD) using region-specific geographically weighted regression (GWR) in Mexico. GWR models overcame stationary models to predict AFKD at a selected 20 km optimum active fire kernel density bandwidth. Our observations confirmed, for the first time for remotely-sensed fire records, the regional variation and the non-stationarity of the human and fuel-related drivers of AFKD in Mexico. Aboveground biomass mainly showed positive relationships in low productivity areas and humped relationships in more productive areas. Contrary to previous observations for fire suppression records, road density mainly showed a negative relationship, and slope mainly showed a positive relationship with AFKD. This highlights the importance of monitoring fire spatial activity, not only from human-based suppression records, but also considering remotely-sensed AFKD, potentially improving fire prevention planning. The methodology shown in the current study can be replicated elsewhere for improving our understanding of the spatial drivers of remotely sensed fire activity.