Wildfire Risk Evaluation using Machine Learning Techniques on Satellite Multispectral Data

Since climate change has continued to raise global temperatures, wildfires are becoming a greater hazard with increased severity and frequency. To help fire management agencies better plan their preventive measures and increase the effectiveness of suppression, it is crucial to monitor and map fire-susceptible areas through a deep analysis of the regional fire trends. The aim of this thesis is to test the ability of supervised machine learning methods to map and measure the risk of fires in the vegetative areas of Sicily in 2021 using data from 2016 to 2020. This task was performed on areas of 200x200 square meters by creating a novel dataset with variables representing potential fire drivers: weather, topography, and fuel. Fuel type and moisture content are modeled using a collection of spectral indices taking advantage of open source multispectral data from Sentinel-2 mission (ESA). Despite the low spatial resolution, the results are encouraging and this work could represent a starting point to build a solid fire management Italian system.