Spatiotemporal Machine Learning for Wildfire Spread Behavior Prediction
Loading...
Authors
Roth, Ahren Daniel
Issue Date
2024
Type
Thesis
Language
Keywords
Alternative Title
Abstract
Wildfire is one of the most destructive ecological processes of our natural world. It is an integral part of our ecosystem, and the life cycle of many species depend on it for propagation. In order to manage fire, we must understand its behavior and properly model the condition of the surrounding ecosystem. Predicting wildfire behavior is essential to effectively managing the wildland environment. Wildland management includes suppressing fires where necessary, promoting them where advantageous, all while protecting people, property, and resources. Fire behavior is influenced by the interplay of many factors. Topography, fuel type and load, wind, humidity, and temperature all effect the direction and rate of growth of wildland fires. Wildfires can create their own weather, amplifying the effects prior fire behavior have on future fire behavior. Spatiotemporal machine learning techniques may be utilized to model complex real-world fire behavior dynamics and produce accurate predictions. Anticipating wildfire behavior is the first step to effectively managing its effects. Leveraging the ability of deep learning models to infer future fire behavior from past fire behavior is a novel approach to wildfire behavior prediction. Spatiotemporal modeling has proven effective at accurately predicting wildfire spread behavior days and weeks in advance. Current performance is limited by the availability of high-resolution, high frequency data. As remote-sensing data improves, and the availability of computational resources grows, temporal-spatial modeling of wildfire spread behavior will contribute more to effective wildland management practices.