Assessing the Accuracy of a Machine Learning Model for Wildfire Spread Prediction in Regional Air Quality Forecasting

Abstract

Due to its significant fire emissions, wildfire spread, characterized by its spatiotemporal variations, is crucial for regional-scale air quality forecasting. The process involves various physical and dynamic phenomena, making it complex and computationally intensive to simulate these interactions accurately at regional scales. Recent advancements in machine learning have shown promise in predicting fire spread, effectively capturing the spatial distribution of fires while being computationally efficient once trained.

In this study, we developed a machine-learning wildfire spread model at NOAA-ARL to enhance next-generation regional air quality forecasting (AQF) applications. The model seeks to increase the accuracy and efficiency of wildfire-related air quality predictions. Preliminary results indicate the validation of the model's predictions compared to data obtained in the March 2024 Page County, Virginia wildfires. These outcomes suggest that the machine learning model is a powerful tool for forecasting wildfire spread and its impact on regional air quality. Ultimately, this tool can be used as a short-term prediction for the growth of future wildfires.