Linking burn severity to pre-fire forest structure and weather on the Dixie fire offers potential to map prospective burn severity

Seeking to understand controllable stand structure drivers of high burn severity, contingent on weather, 35 candidate predictors were derived from Forest Inventory and Analysis observations and ancillary data on managerially uncontrollable factors—fire history, climate, weather, and topography—within California’s massive 2021 Dixie fire. Logistic regression models were fitted to evaluate the effects of these predictors, with remotely sensed burn severity as the response. Using multiple model selection strategies to choose from among parsimonious predictor subsets that we structured to control for collinearity, high burn severity was consistently predicted (area under the receiver operating curve = 0.72–0.73) from just six variables: three stand characteristics—basal area of standing dead trees, basal area of mid-size or all live trees, and ladder fuel abundance—and three uncontrollable predictors—30-year mean maximum annual temperature and maximum wind gust speed and precipitation for the hour at which fire arrived at the stand. To demonstrate how these models might inform landscape treatment priority, we applied the models to publicly available, imputed rasters of the stand characteristics and to the uncontrollable factors listed above, evaluating map accuracy against Relative differenced Normalized Burn Ratio derived severity. Disappointing accuracy of the resultant maps might be improved as forest attribute imputation products begin to incorporate fine scale (e.g., Light Detection and Ranging) information capable of representing ladder fuels and as pixel scale models of weather at time of fire arrival become available.