Developing an Hourly Fire Progression Database for Large California Wildfires: Application for Modeling Fire Spread Rate

In the western United States, prolonged droughts, the warming climate, and historical fuel build-up have contributed to larger and more intense wildfires, as well as longer fire seasons. As these costly wildfires become more common, new tools and methods are essential for improving our understanding of the evolution of fires and how extreme weather conditions, including heatwaves, windstorms, and droughts, and varying levels of active fire suppression drive or mitigate fire spread. Here we develop an hourly fire progression database for 28 large wildfires (> 50,000 acres) in California from 2019-2021. We combine GOES-16 and GOES-17 geostationary satellite detections of active fires to derive hourly fire perimeters, active fire lines, and fire spread rates with parameter optimizations for defining the burned to unburned boundary and correcting for the parallax effect from elevated terrain. We evaluate GOES perimeters with the VIIRS-derived Fire Event Data Suite (FEDS), version 2, and Monitoring Trends in Burn Severity (MTBS). Relative to MTBS perimeters, GOES (mean IOU = 0.76) performs slightly worse than FEDS (mean IOU = 0.84) in terms of spatial accuracy for fires in 2019-2020. At a coarser spatial resolution of 2 km at the equator, GOES struggles to accurately delineate early fire perimeters if the fire is relatively cool and smoldering and classify incremental growths in the fire perimeter later in the fire’s lifetime. However, GOES fills in temporal gaps present in FEDS, which is only available at 12-h intervals; this is particularly relevant when a fire spreads rapidly and continually. As an application of the resolved GOES fire diurnal cycle, we use random forest to model how meteorology and active suppression drive fire spread rate at an hourly scale. Preliminary results show that random forest models with only mean temperature, relative humidity, and wind speed as independent variables can explain on average 25% (3-59%) of variance in fire spread rate for 10 large wildfires > 100,000 acres in 2020. Our method of using GOES to derive hourly fire progression database is applicable to large wildfires across the contiguous United States and has a broad set of applications including predictive modeling of fire spread and atmospheric transport of smoke.