A Rigorous Comparison of Various Machine Learning Models for Next-Day Wildfire Detection

Wildfires in the western United States pose significant threats to large, barren areas, as well as more populated, urban lands, especially in wildfire-prone areas in the West. This study aims to develop an Artificial Intelligence model that predicts wildfires on a next-day basis, enabling timely mitigation and safety efforts, which is beneficial, especially in high-risk areas like the Western United States. Understand how safety and health-related measures can be implemented by predicting how the wildfire will spread over a day. Environmental factors such as elevation, maximum and minimum temperatures, wind speed, humidity, precipitation, drought indices, vegetation health metrics (Normalized Difference Vegetation Index), population density, energy release, and initial fire masks (spreads) are features used by the model to make predictions about the spread of wildfires over 24-hours. The model's performance was evaluated based on its precision and recall in predicting fire spread within a 64 km x 64 km area. The results demonstrate the neural network’s strong capability to identify high-risk areas for wildfires with an 84% accuracy, improved by boosting with a Random Forest Classifier to 99% accuracy. By providing reliable next-day predictions, this model serves as a valuable tool for wildfire management, enabling authorities to implement preemptive measures that could significantly reduce the impact of wildfires. This research and model development contributes to the broader field of disaster prevention and management, offering a data-driven approach to mitigating wildfires on a next-day basis.