Wildfire Ignition Prediction using Machine Learning

In recent years, wildfires have gained significant attention amidst growing concerns over climate change. The state of California, within the frame of the United States, has become a focal point of this discussion due to the increasing frequency and intensity of wildfires. The availability of extensive datasets pertaining to these events provides a valuable resource for understanding their dynamics and implementing proactive measures. Developing models that effectively synthesize such data to predict wildfire ignition points will aid in proactive firefighting strategies. The primary objective of this study is to develop machine learning models that utilize synthesized data at various resolutions to predict wildfire ignition points. A byproduct of this is a comprehensive study that discern trends and causal factors contributing to reported fires. This study analyzes 23 years of historical fire data from California to construct a machine learning-based model tailored to identify potential wildfire ignition locations, with a focus on their proximity to power lines. The results of the model indicate significant efforts in identifying areas that are more susceptible to ignition events at the county level. Additional model results show a nested prediction framework to improve model results at the grid cell level. The findings of this study will contribute substantively to wildfire management strategies by offering a proactive tool for identifying potential ignition hotspots in advance.