[M-AG32] Satellite Land Physical Processes Monitoring at Medium/High/Very High Resolution

[MAG32-P10] Modelling the Spatial Variability of Wildfires in Guatemala, Central America.

*Miguel Valdez¹、Chi- Farn Chen¹、Liang-Chien Chen¹ (1.National Central University, Center for Space and Remote Sensing Research.)

キーワード：Wildfire probability , Guatemala , Random Forest, Hotspots , MODIS, GIS

Wildfires are part of the ecology of forests in Guatemala, and forests conform the largest proportion of land cover in the country. Nevertheless, limited understanding of wildfire probability and the factors that influence it hinder the planning of intervention strategies. In this research, we used climatic variables such as temperature, land-surface temperature, precipitation, wind-speed and solar radiation with, anthropogenic variables such as proximity to settlements and different road-types in addition to population density, topographic factors such as elevation and slope, and vegetation factors such as vegetation indexes derived from satellite imagery to identify the wildfire probability and determine the most relevant factors. We performed an exploratory analysis to identify important factors using a hotspot analysis and density plotting these areas, and integrated them with wildfire observations acquired from the satellite MODIS-FIRMS data using Random-Forest (RF). We then used the most relevant factors to predict wildfire probability and validated our results using different measures. The results demonstrated satisfactory agreement with the independent data with an Area under the receiver operating curve (AUC) of 0.85, and a biserial correlation of 0.52. Central and northern regions of Guatemala have a very high probability of wildfire occurrence. Human imprint and extreme climatic conditions influence wildfire fire probability in Guatemala and there is an especially high wildfire probability in the northeastern region of the country, within the largest reserve, namely, El Petén. This area is also neighboring Belize. Using machine learning, we identified the major influencing factors and the areas with a high probability of wildland fire occurrence in Guatemala. Results from this research can support local organizations in applying enhanced strategies to minimize wildfires in high-probability areas. Additional efforts may also include using future climate change scenarios, updating of the wildfire probability maps yearly, and increasing the time frame to evaluate the influence of other factors that may show relatively high importance in the wildfire probability occurrence in Guatemala.