Spatially continuous mapping of pre-fire fuel characteristics with imaging spectroscopy and lidar for fire emissions modeling

Fuels are a large source of uncertainty in fire emissions estimates due to variability in the physical and chemical properties of fuels and how they are represented. These uncertainties can be addressed using imaging spectroscopy and lidar data, that provide observations of the chemical and physical traits and spatial distribution of vegetation. Combined with ground fuel measurements, these data provide information on fuel distribution and quantity important for mapping and modeling fire effects. In this study, we present a methodology to develop models and continuous maps of pre-fire fuel characteristics for use in fire emissions modeling. We first addressed any spatial gaps over fire areas for Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) chemical trait data using Random Forests regression and for derived fractional cover. We used the AVIRIS fractional cover and chemical traits or AVIRIS estimates alongside lidar, multispectral, and topographic variables to build fuel characteristic models informed by ground measurements with partial least squares regression. We derived maps of predictive uncertainty alongside a suite of uncertainty statistics for each fuel characteristic that inform the use of fuels data within fire effects models. We used two study sites: the Williams Flats wildfire in eastern Washington state, USA and three prescribed crown fires in Utah, USA. The results show similar error between calibration and validation sets and NRMSE of around 20 % or lower for a majority of the fuel models. We present fuel characteristic and uncertainty maps for all fires. This study shows that the use of imaging spectroscopy and lidar data have the potential to represent fuel heterogeneity and continuously map fuel characteristics for fire effects modeling.