Estimations of fuel moisture content for improved wildland fire spread prediction
Presenting Author: Branko Kosovic
Organization: National Center for Atmospheric Research
Co-Author(s): Tyler McCandless, Bill Petzke, Pedro Jimenez, Steven Massie, Amanda Anderson, Amy DeCastro, Domingo Muñoz-Esparza, and Sue Haupt
The rate of spread of a wildland fire significantly depends on the fuel moisture content (FMC) of the fire's fuel. Currently, FMC data sets included in the National Fuel Moisture Database are based on surface observations and sparse sampling. Surface observations provide relatively sparse coverage over the Continental United States (CONUS). Furthermore, live FMC is sampled relatively infrequently. An operational wildland fire prediction system requires a high resolution, gridded, real-time FMC data set. We are therefore developing such a data set using surface observations of FMC to train machine learning models based on satellite observations from Moderate Resolution Imaging Spectrometer (MODIS) Terra and Aqua platforms. MODIS reflectances and derived vegetation index products are used to predict FMC and compared to measurements by the Wildland Fire Assessment System (WFAS) and Remote Automated Weather Stations (RAWS). Both live and dead FMC are considered. We have explored different machine learning techniques including: Multiple Linear Regression, Random Forest, Gradient Boosted Regression and Artificial Neural Network. We used these algorithms to develop machine learning models relating satellite reflectances and satellite derived vegetation index predictors and FMC. In addition to satellite observations we have also used static data: elevation and terrain slope, as well as gridded surface products from the National Water Model including: accumulated evapotranspiration and soil moisture. Machine learning algorithms are applied across the entire spatial grid, populated by all the 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 live and dead FMC data sets are used in simulations of eleven wildland fires observed in Colorado during 2016. Results obtained using the new FMC data set are compared to results obtained using uniform FMC over the wildland fire domain.