A Deep Model For Forest Fire Risk Estimation In Algeria

Abstract

Algeria is one of the most affected Mediterranean countries by wildfires, often leading to devastating consequences. However, unlike its neighbors (Spain, Portugal, Greece, and Italy) it has not yet developed a sophisticated and reliable wildfire prevention system. In this study, we are going to use MODIS Collection 6.1 fire data, GLOH2O weather data, elevation data from DIVA-GIS, Land Cover Map from ESA, and the study area’s shape file from GADM to estimate wildfire risk. Our approach relies on the application of different algorithms (Logistic Regression, Support Vector Machines (SVM) with an RBF kernel, long short term memory networks (LSTMs), Random Forest, and XGBoost) to predict wildfire occurrences based on the available data. The study aims to find the optimal combination of predictor model and preprocessing techniques that gives the most accurate wildfire predictions.