Development of Machine Learning Models for Estimation of Fuel Moisture Content Based on MODIS Satellite Observations

Wildland fire decision support systems require accurate predictions of wildland fire spread. Fuel moisture content (FMC) is one of the important parameters controlling the rate of spread of wildland fire. However, FMC is a sparsely and relatively infrequently measured surface variable compared to most atmospheric variables. A high resolution, gridded, real-time FMC data set does not currently exist for assimilation into operational wildland fire prediction systems. We use surface observations of live and dead FMC to train machine learning models to estimate FMC based on satellite observations. Moderate Resolution Imaging Spectrometer (MODIS) Terra and Aqua reflectances and derived vegetation index products are used to predict the live and dead FMC measured by the Wildland Fire Assessment System (WFAS) and Remote Automated Weather Stations (RAWS). Machine learning techniques, such as regression trees, random forests, and artificial neural networks, can learn the non-linear relationships between the satellite derived vegetation index predictors and FMC. This allows the methods to be trained on MODIS satellite data corresponding to the temporally and spatially nearest grid points to the irregularly spaced surface FMC observations. Machine learning algorithms are applied across the entire spatial grid, populated by MODIS predictors, to achieve a gridded, real-time FMC dataset. Algorithms are first calibrated on the training data and then applied to a test dataset for Colorado. The results of the test dataset for Colorado show improvements in accuracy for both live and dead FMC estimation compared to persistence and linear regressions.