Wildfire Data Analysis: Predicting Risk and Severity in San Diego County. In Data Science & Engineering Master of Advanced Study (DSE MAS) Capstone Projects

Major wildfires have grown in intensity and frequency not only in southern California, but also across the world, in the last few years. Property loss, damages and costs associated with these wildfires can be especially significant to residents, communities, public space, and the environment. Gas and electric companies have begun to proactively take down power grids as a preventative measure, a significant disruptor to everyday life. In this work, we analyze factors that affect wildfires and how changes in these factors over time affect wildfire behaviors and risks. Specifically, we took daily weather observations and associated fire measurements (indication of fire and the acres burned) in San Diego county over the course of the past 20 years, hypothesizing that changing weather conditions are the primary drivers of wildfires. We validate this theory by providing evidence that wildfire risk and severity are correlated with evapotranspiration (a measure of ground dryness), available vegetation, ambient temperature and wind speed. Using machine learning, we develop a robust model that detects the incidence and acres burned of a wildfire with a testing validation Recall score of 77%. Additionally, we pose a method to estimate the economic value of a region with associated “high” wildfire risks. Residents, firefighters, and public policy officials may leverage the results of this project for effective management and mitigation of wildfire risks.