Estimation of Live Fuel Moisture Content From Multiple Sources of Remotely Sensed Data

Live fuel moisture content (LFMC) is one of the key variables affecting wildfire ignition and behavior. Remote sensing data-derived vegetation indices, meteorological indicators, and soil moisture have been reported to estimate LFMC in previous studies, but LFMC estimation from all these sources has not yet been attempted. This study pooled all these remotely sensed data in the construction of LFMC estimation models to this end. Based on the XGBoost (Extreme Gradient Boosting) algorithm, we built an empirical model and reached reasonable LFMC estimates (R ² =0.56, RMSE=27.16%) across the western U.S. states. The best LFMC estimate was found for the closed shrublands (R ² =0.66, RMSE=24.38%), followed by open shrublands (R ² =0.60, RMSE=30.04%), grasslands (R ² =0.58, RMSE=27.02%), savannas (R ² =0.50, RmSe=24.23%) and woody savannas (R ² =0.44, RMSE=25.94%). This study advances from previous research as it involved the combination and analysis of multi-source indicators and required the improvement of LFMC estimation accuracy using meteorological long-term temporal characteristics data for multiple vegetation cover types in a large-scale regional context.