Assessment of fire hazard in Southwestern Amazon

Abstract

Fires are among the main drivers of forest degradation in Amazonia, causing multiple socioeconomic and environmental damages. Although human-ignited sources account for most of the fire events in Amazonia, extended droughts may magnify their occurrence and propagation. The southwestern Amazonia, a transnational region shared by Brazil, Peru, and Bolivia and known as the MAP region, has been articulating coordinated actions to prevent disasters, including fire, to reduce their negative impacts. Therefore, to understand the fire patterns in the MAP region, we investigated their main drivers and the changes in the suitability of fire occurrence for the years 2005, 2010, 2016, and 2020. We used a maximum entropy (MaxEnt) model approach based on active fire data from satellites, climatic data, and land use and land cover mapping to spatially quantify the suitability of fire occurrence and its drivers. We used the year 2015 to calibrate the models. For climatic data and active fire count, we only considered grid cells with active fire count over the third quartile. All our models had a satisfactory performance, with values of the area under the curve (AUC) above 0.75 and p < 0.05. Additionally, all models showed sensitivity rates higher than 0.8 and false positive rates below 0.25. We estimated that, on average, 38.5% of the study region had suitable conditions for fire occurrence during the study period. Most of the fire-prone areas belong to Acre, representing approximately 74% of the entire MAP region. The percentage of deforested areas, productive lands, forest edges, and high temperatures were the main drivers of fire occurrence in southwestern Amazonia, indicating the high vulnerability of fragmented landscapes extreme climatic conditions to fire occurrence. We observed that the modeling approach based on Maxint is useful for useful for evaluating the implications of climatic and anthropogenic variables on fire distribution. Furthermore, because the model can be easily employed to predict suitable and non-suitable locations for fire occurrence, it can to prevent potential impacts associated with large-scale wildfire in the future at regional levels.