WCD-08. Deriving Fire Radiative Power (FRP) with Weather Model Variables and Satellite FRP using Random Forest (RF) Models

Abstract
The Fire Radiative Power (FRP) from wildfires is important in calculating a variety of fire-weather variables, such as aerosol calculations and/or smoke forecasts. FRP is derived from satellite data and raw data can include measurements from wildfires, controlled burns, and false-detections. Sometimes, due to clouds or their own smoke plumes, satellite derivations can miss FRP measurements from active fires. In this work, daily mean FRP values collected from satellites in 2018 are used in combination with weather model variables to derive a forecasted hourly FRP value. Random Forest (RF) models were trained with inputs including yesterday averaged FRP satellite values, UTC time of day, latitude, longitude, temperature, wind speed, and relative humidity to produce an hourly FRP value on the Rapid Refresh (RAP) model resolution. The CONUS region was used as a training domain, but to further isolate larger wildfires, a subdomain of longitudes west of 105W were selected to train separate models, labeled “west” in this study, to better model wildfire hourly FRP. In addition to training models of two different domains, models were also trained on separate satellite FRP inputs given the difference in satellite sensors. One variation of the RF models was trained using Geostationary Operational Environmental Satellite (GOES) FRP inputs, and another using a collection of polar orbiting satellite FRP. Results are shown on this poster.