Using multi-decadal satellite records to identify environmental drivers of fire severity across vegetation types

To date, most studies of fire severity, which is the ecological damage produced by a fire across all vegetation layers in an ecosystem, using remote sensing have focused on wildfires and forests, with less attention given to prescribed burns and treeless vegetation. Our research analyses a multi-decadal satellite record of fire severity in wildfires and prescribed burns, across forested and treeless vegetation, in western Tasmania, a wet region of frequent clouds. We used Landsat satellite images, fire history mapping and environmental predictor variables to understand what drives fire severity. Remotely-sensed fire severity was estimated by the Delta Normalised Burn Ratio (ΔNBR) for 57 wildfires and 70 prescribed burns spanning 25 years. Then, we used Random Forests to identify important predictors of fire severity, followed by generalised additive mixed models to test the statistical association between the predictors and fire severity. In the Random Forests analyses, mean summer precipitation, mean minimum monthly soil moisture and time since previous fire were important predictors in both forested and treeless vegetation, whereas mean annual precipitation was important in forests and temperature seasonality was important in treeless vegetation. Modelled ΔNBR (predicted ΔNBRs from the best-performing generalised additive mixed model) of wildfire forests was higher than modelled ΔNBR of prescribed burns. This study confirms that western Tasmania is a valuable pyrogeographical model for studying fire severity of wet ecosystems under climate change, and provides a framework to better understand the interactions between climate, fire severity and prescribed burning.