WorksComputer Science
Machine Learning for Near Real-Time Identification of Wildfire Burned Areas
Abstract
In this article, we present a near real-time method that estimates burned probability anywhere in the world using satellite images. Our approach takes advantage of the state-of-the-art pix2pix model to translate a preprocessed index calculated from Landsat 8/9 images into an approximate of USGS Burn Probability indicating per pixel burn probability. Since this is a machine learning-based approach, it can be automated with a minimum amount of human intervention with significantly less time delays and geographic restrictions compared to other contemporary methods. We include results from our experiments and compare our approximation with the USGS Burn Probability using RMSE score.