Severity impacts reflect the impact of wildfire based on traditional fire behavior modeling techniques. This requires detailed geospatial information about fuel, topography, and weather. LANDFIRE (http://landfire.cr.usgs.gov/viewer) is a nationwide product that provides fuel and topography inputs for fire behavior modeling.
Weather information is derived from publicly available (https://registry.opendata.aws/noaa-hrrr-pds/) maps for 99th-percentile temperature, humidity, and wind speed.
Fire behavior calculations are made using modified BehavePlus and FlamMap code originally produced by the USDA Forest Service. Modifications are focused specifically on the processing of weather input. For each pixel (a 30-meter by 30-meter square portion of the landscape), the fuel type, topography (aspect, slope, and elevation), and weather parameters such as fuel moisture and wind (both at the 99th percentile) are used to predict how a fire would behave. These predictions are then used to produce the overall risk map.
Although a percentile-based approach to weather is used in fire behavior calculations, it is incorrect to assume that this alone is sufficient to address future probability. This is the exact reason why Frequency group variables are also included in Wildfire Risk Extreme.