Wed, Aug 04, 2021:On Demand
Background/Question/Methods
Wildfires are becoming more destructive around the world due to climate change and growth of the wildland-urban interface (WUI). Efforts to model wildfire trends have largely focused on seasonally dry regions that are prone to extremely large fires, such as the western US. Similar efforts are needed for regions where wildfires are not typically as large, but where dense development in the WUI contributes to frequent ignitions and where many homes are exposed to fire. We investigated trends in wildfire occurrences and areas burned in the eastern U.S. since 1984, and related these trends to spatial and temporal patterns in climate, vegetation, and human land use. Our analysis consisted of two components: 1) using machine learning methods to model burn probabilities and numbers of structures affected by fires within ecoregion boundaries at a monthly time step, and 2) using regression models to determine relationships between seasonal fire activity and climate variables at annual time steps. Machine learning models predicted the areas burned by large fires, using burn perimeters from the Monitoring Trends in Burn Severity dataset, while regression models predicted numbers of smaller fire occurrences using incident reports from a US Forest Service database. We used our results to project future changes in wildfire activity in the eastern U.S. under different scenarios for climate and WUI growth.
Results/Conclusions Climate variables, particularly relative humidity and vapor pressure deficit, were important predictors of both the area burned by large fires and total numbers of fire occurrences. Variables representing human land use, such as distance from roads/railroads, WUI classes, and distance from protected areas were also important in many ecoregions. Model predictors and accuracy scores varied by ecoregion, but consistently selected humidity variables as top predictors. Regression models had significant positive correlations between number of fires and vapor pressure deficit, particularly in the summer and fall months. Projections of future climate change in the eastern U.S. suggest that precipitation and relative humidity will increase, but seasonal variations in these trends could result in periods of increased fire activity. Projections of future fire activity varied across the region and among climate and WUI growth scenarios, depending on the magnitude of projected increases in precipitation. Growth of the WUI is projected to increase the number of structures affected by fire. Communities interested in adaptive wildfire management in a changing climate may therefore want to consider land use planning as one route to reduce future hazard.
Results/Conclusions Climate variables, particularly relative humidity and vapor pressure deficit, were important predictors of both the area burned by large fires and total numbers of fire occurrences. Variables representing human land use, such as distance from roads/railroads, WUI classes, and distance from protected areas were also important in many ecoregions. Model predictors and accuracy scores varied by ecoregion, but consistently selected humidity variables as top predictors. Regression models had significant positive correlations between number of fires and vapor pressure deficit, particularly in the summer and fall months. Projections of future climate change in the eastern U.S. suggest that precipitation and relative humidity will increase, but seasonal variations in these trends could result in periods of increased fire activity. Projections of future fire activity varied across the region and among climate and WUI growth scenarios, depending on the magnitude of projected increases in precipitation. Growth of the WUI is projected to increase the number of structures affected by fire. Communities interested in adaptive wildfire management in a changing climate may therefore want to consider land use planning as one route to reduce future hazard.