SparkNet: Transparent Prediction of Wildfire Potential in Southern California

Overview

In Southern California, millions of dollars of infrastructure are destroyed annually by uncontrollable wildfires. For instance, the L.A. fires caused upwards of $164 billion in damages. Current methods to combat wildfires are reactive rather than proactive, allowing fires to burn ecosystems before they can be contained. We present SparkNet, a machine learning model that identifies regions in SoCal with high potential for a wildfire to spread based on various environmental factors. This project progressed across three main stages: aggregating data, training the model, and interpreting its predictions. To emphasize predictions of larger fires, we logarithmically scaled fire burn areas prior to training, allowing the model to predict the magnitude of a wildfire with an average error of 0.71 log-acres. Explainability tools (SHAP and LIME) were then implemented to transparently break down model predictions on both global and local levels, paired with the creation of our own sensitivity analyzer. The model was able to capture the nuanced patterns of fire spread, and the findings revealed the influence of environmental conditions on the spread of a wildfire as well as the potential for explainable AI to be used in real-time fire modeling solutions. By training a tree-based regression model on past SoCal wildfire and environmental data, we achieved a model that performed well on key evaluation metrics, surpassed our baselines, and met our interpretability standards. Overall, the results supported our hypothesis that AI models can predict wildfire trends in SoCal, though it is limited by the unpredictability of ignition sources.

Competition history

• CSEF 2026 Environmental Engineering (Track 2) (Senior Division) · Entry S-12-10

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Environmental Engineering projects
• CSEF Environmental Engineering projects
• Every CSEF 2026 project