These data were generated to map spatial burn severity and emissions of each historically observed large wildfires (>404 hectares (ha)) that burned during 1984–2020 in the state of California in the US. Event-based assessments were conducted at 30-m resolution for all fires and daily emissions were calculated at 500-m resolution for fires burned since 2002. A total of 1697 wildfires were assessed using the Wildfire Burn Severity and Emissions Inventory(WBSE) framework developed by Xu et al 2022. The comprehensive, long-term event and daily emissions records described here could be used to study health effects of wildfire smoke, either by combining them with transport modeling to model air quality and estimate exposures, or by incorporating them into statistical models predicting health impacts as a direct function of estimated emissions. These data will also facilitate analyses of changing emissions impacts on the carbon cycle over the last three decades. High resolution severity and emissions raster maps are generated for each fire event to support further spatial analysis. While the emissions calculated for California with WBSE are not a substitute for real-time daily emissions estimates, it is designed to extend the estimated emissions record back to 1984 with a finer spatial resolution and provide more up-to-date estimates on emissions factors reflecting information from California's recent extreme fires.

This dataset provides estimates of 30 m resolution burn severity, and emissions of CO2, CO, CH4, non-methane organic compounds (NMOC), SO2, NH3, NO, NO2, nitrogen oxides (NOx = NO + NO2), PM2.5, OC, and BC. WBSE was implemented  for California large wildfires on a per-fire event scale since 1984 and also a daily scale since 2002. The inventory implementation steps, input datasets, and output data are summarized in figure 1 in Xu et al, 2022.

Burn severity calculation

Fire records for California from 1984 to 2019 were retrieved from MTBS (https://mtbs.gov/viewer/index.html) via interactive viewer on 8 May 2021, resulting in a dataset with a total of 1623 wildfires. We also acquired fire perimeters for 74 large wildfires in 2020 from CAL FIRE (https://frap.fire.ca.gov/frap-projects/fire-perimeters/) and calculated dNBR for each 2020 fire using the dNBR calculation tool with Google Earth Engine (GEE). This process first selects either initial assessment or extended assessment for each fire. The initial assessment utilizes Landsat images acquired immediately after a fire to capture first-order fire effects. The extended assessment uses images obtained during the growing season following the fire to identify delayed first-order effects and dominant second-order effects (Eidenshink et al 2007). We utilized LANDFIRE Biophysical Settings (BPS) to determine which assessment type to apply for each fire burned in 2020. After Picotte et al (2021), we used extended assessment if the majority of general vegetation groups within the fire perimeter are forests, while initial assessment is used when the majority of general vegetation groups are grassland/shrubland. By contrast, MTBS uses extended assessment for forest and shrubland types. We did not delineate grasslands into burn severity categories. Instead, we classified them as burned ('grass burn') because of difficulties in assessing vegetation change. Post-fire images for extended assessment were selected during the next peak of the green season (June–September) using the mean compositing approach suggested by Parks et al (2018). Composite post-fire images acquired immediately within two months after the fire containment dates were used for the initial assessment. Composite pre-fire images for extended and initial assessments were acquired with the matching periods from the preceding year. The dNBR images were produced by quantifying the spectral difference between composite pre-fire and post-fire Landsat scenes. We calculated the unitless, continuous CBI variable from dNBR/NBR values using the linear and Sigmoid B regression models developed for the CONUS by Picotte et al (2021). CBI values were then classified following thresholds modified based on Crotteau et al(2014) into six severity classes: unburned, low severity, moderate severity, high severity, grass burn, and non-processing area.

Emissions calculation

Emissions of all species are calculated as a function of area burned, fuel loading, the fraction of vegetation burned based on burn severity, and an emissions factor specific to each vegetation type using the following equation modified from the FINN model (Wiedinmyer et al 2011). Fuel categories were assigned from LANDFIRE EVT products. For emissions calculations, EVT data were then categorized into five general vegetation categories: grass, shrub, forest under 5500 feet (1676 m), forest between 5500–7500 feet (1676–2286 m), and forest above 7500 feet (2286 m), updated for California ecosystems. Fuel consumption was determined following Hurteau et el 2014 assigning fuel loading and consumption values for each severity class for the five general vegetation categories based on the First Order Fire Effects Model v5 (Reinhardt et al 1997). Emission factors for greenhouse gases, particulate matter, and reactive trace gases were updated with recent data for each general vegetation class using results from recent field campaigns and studies specific for California ecosystems and Western U.S. ecosystems.

Day of burning and daily emissions

To assign the day of burning for individual pixels, NASA fire information for resource management system (FIRMS) active fire products from MODIS (Collection 6) within 750 m of the fire perimeter shapefiles supplied by MTBS or CAL FIRE were selected for interpolation to account for detections that might be outside the boundary due to detection radius. VIIRS 375 m data, when available since 2012, was added to complement MODIS data with improved performance to assign burn dates using the fire progression raster tool (figure 4). We filtered the MODIS/VIIRS detection points to the date range of interest and created a 500 m buffer around each point. Points were then converted to circle polygons to represent each point's detection extent properly. The average date was selected as the proper date in regions of overlapping buffers. We then calculated daily emissions and assigned them to the centroids of the aggregated daily progression polygons.

Please refer to README file.