Modeling fuel loads dynamics and fire spread probability in the Brazilian Cerrado

Abstract

Fire events substantially influence biodiversity, carbon cycling, hence climate, human health, and the economy on a global scale. To cope with increasingly widespread fires, we need accurate estimates of the chances of fire occurrence across heterogeneous landscapes. Estimates of risk of fire are fundamental to prevent and fight wildfires. To this end, we developed a monthly fire spread probability model for the Brazilian Cerrado biome based on the historical relation between fuel loads and burned areas. To so, we firstly stratified the biome into 16 climatic regions, given the climate influence on fuel loads. We used historical burned areas and fuel loads from remote sensed data between 2015 and 2018 to build a non-stationary model that estimates fuel loads dynamics across the biome. Climate seasonality is the main factor driving fuel loads dynamics. The correlation between fuel loads and the best predictor (monthly mean precipitation) ranges from 0.27 to 0.88 (mean r = 0.61 and deviation of 0.18) across the Cerrado. The average amplitude of fuel loads is 32% between dry and rainy seasons. Our Bayesian fire risk model uses the burned area as a prior probability and fuel loads to estimate the posterior probability of fire spread. Our results show that the probability of fire spread highly correlates with historical burning events (r = 0.87). The recovery of fuel loads post-fire takes, on average, 2.43 years; however, our results point to a downward pathway of the biome’s vegetation biomass due to frequent recurrent fires. The models we developed provide a useful tool for improving the representation of spatial patterns and seasonality of fires in order to support management practices.