Fuels are highly variable and dynamic in space and time, and fuel loading can vary considerably even within fine spatial scales and within specific fuel types, such as downed wood or organic soils. Given this inherent variability in fuel loadings, it is not good practice to represent all instances of a fuel type by the same set of fuel loadings, as these vary at multiple spatial scales and are generally independent of each other. The best practice for producing emissions estimates from data with inherent variability is to represent the underlying uncertainty in the base fuels data. This measure of uncertainty can then be used in understanding the reliability of the fuel-loading estimates and also to evaluate how that uncertainty propagates to variability in emissions estimates. Models for emissions inventories are becoming increasingly sophisticated and require corresponding complexity in input datasets – the appropriate probability distributions of their base layers rather than just their means. No such datasets exist for fuels despite their acknowledged variability at multiple spatial scales. In response to the JFSP 2015 FON Task Statement 1 “Fuels mapping for emissions inventories”, our multi-institutional team completed a project to improve information on fuel loadings valuable for fire and smoke management by compiling data on fuel loadings for fuels across the Conterminous US and Alaska and then calculating the variability in fuels found in this existing field data.

In this project, we: (1) developed probability distributions of fuel loadings for US fuelbeds using existing field data; (2) created geospatial fuel layers with enhanced fuel loading information that can be used by the emissions modeling and inventory communities in the United States; and (3) conducted a sensitivity analysis based on the compiled data to evaluate sources of uncertainty and data gaps for emissions estimates. This study was informed by results of the Smoke and Emissions Model Intercomparison Project (SEMIP) funded by the Joint Fire Science Program, which showed that fuel loadings introduce most of the uncertainty in emissions modeling. The final dataset was created by consolidating existing data records on fuel loadings from many sources of field data describing wildland fire fuels in North America. The resulting database, called the North American Wildland Fuels Database (NAWFD; French et al. 2020) can be accessed through a data access application web site available at https://fuels.mtri.org. NAWFD aggregates fuel loading information from 26,620 field sites compiled from 271 data sources. Each data point is assigned to a LANDFIRE Existing Vegetation Group ID (EVT; https://www.landfire.gov/evt.php). Probability distributions have been generated for each fuel stratum within each fuelbed. The ETV Groups served as a means to combine vegetation types that did not have enough data for statistical description. NAWFD was developed to enable best practices for modeling national- and regional-scale fire emissions by incorporating uncertainty into fuels estimates used in smoke emissions modeling. The database and probability density functions created for each EVG were used to analyze the sensitivity of emissions estimates to fuel loading variability. Additional work included a comparison of the fuels distributions to data held out of the database to validate the distributions against field-collected data for a selection of fuelbeds, and an effort to identify gaps in existing data on fuel loading to prioritize field data collection for minimizing uncertainty in emissions modeling.