BlazeAlert: A Novel Spaciotemporal Deep Convolutional Neural Network Model for Real Time Wildfire Risk Assessment

Overview

In January 2025, severe drought conditions and strong Santa Ana winds powered a series of 14 wildfires that proved to be the most destructive in California history. These fires caused 16,000 structural burnings, $53.8 billion in damage, and 28+ lost lives, more of which were a result of the lack of sufficient evacuation notice. This served as an inspiration to harness the power of AI/ML to predict wildfire behavior for inevitable future situations. This research revolves around the creation of a spaciotemporal deep convolutional neural network model that utilizes 15 scientifically supported factors, including weather, air quality, PM2.5 concentrations, NASA satellite imaging (heat anomalies), vegetation, fuel types, land-movement susceptibility, and topography to predict the chance that an active wildfire would reach one’s location. Along with dynamic and static factors, the model incorporates over 25 years of California-specific historical data, allowing predictions to be made with historical context, completely outperforming current manual efforts. Over five iterations, the model was tested on 14,706 unseen data values (2000-2014) and yielded strong results with an accuracy of 91.85%, 77.46% precision, 99.98% recall, 87.29 F1-score, and a 99.98% AUC-ROC score. This research was presented to the City of Rolling-Hills-Estates to explore the feasibility of incorporating the model in their emergency preparedness efforts. Calculations from the City of LA reveal the financial toll of the January wildfires is $53.8 billion. With the usage of this model, even if only 1% of that financial loss is minimized, that is $538 million saved, along with numerous invaluable lives.

BlazeAlert: A Novel Spaciotemporal Deep Convolutional Neural Network Model for Real Time Wildfire Risk Assessment

Overview

Competition history

Related projects

A Multi-Modal Wildfire Prediction and Personalized Early-Warning System

SparkNet: Transparent Prediction of Wildfire Potential in Southern California

It's Flaming Out: Using Artificial Intelligence To Emulate Critical Aspects of Wildfire Growth

A Novel AI Pipeline for Predicting Next Day Wildfire Spread

Predicting Wildfires with Neural Networks: An Approach to Preventing California Wildfires

Developing a Deep Learning Model to Approximate the Convolutional Geometry of Wildfires

Risk-Aware Early Wildfire Detection from Satellite Imagery Guided by Fire Weather Indices

An Accessible, Low Cost Tool for Citizen Scientists: Using Remote Sensing Techniques to Predict Fire Damage Propensity

Browse more like this