Evaluating sensitivity to input fuel resolution in popular fire behavior models

Abstract

Fire behavior models ingest a variety of inputs such as weather, topography, and fuel maps to generate predictions of how a fire will behave. Model prediction accuracy is thus to some degree dependent on the fidelity of the input data sources. For many widely used fire models, however, the exact relationship between fuel input quality and model performance is not well understood. This paper seeks to quantify the relationship between input fuel data and output prediction accuracy in popular fire models based on the Rothermel fire spread equation. In particular, it examines how granularity of fuel classes, spatial resolution, and temporal resolution affect the accuracy of fire behavior predictions. Fuel maps used in the study are generated from remote sensing images using machine learning to map between satellite and ground conditions. Prediction accuracy is evaluated with multiple metrics including rate of spread (ROS) and fire front shape. The outcomes of this study will provide important guidance as to the benefit of producing high fidelity fuel maps when utilizing the Rothermel spread equation to predict fire behavior.