A long history of fire suppression in the western United States has interrupted the fire regimes of many forest types. This interruption has significantly changed forest structure and ecological function and led to increasingly uncharacteristic fires in terms of size and severity. Identifying areas at risk for fires whose severity is outside the natural fire regime will allow for targeted fuel reduction and mitigation to preserve ecosystem integrity. In order to identify areas at risk, we must be able to quantify conditions that influence fire severity.
Our objective was to examine fire regime change, vegetation, and climate for their utility in predicting fire severity. How do these factors act separately to influence fire severity and what is their relative importance as co-determinants? Is fire severity controlled more by bottom-up (vegetation and fuels) or top-down (climate) factors?
We have a mapped fire severity dataset of 4591 large fires from 1984-2007 for the western US as well as a suite of topographic and vegetation data layers from the Landfire project. Our hydro-climate dataset was developed using the VIC hydrologic model. We used conditional logarithmic regression analyses to predict occurrence of high severity fire fraction. We then used a generalized pareto distribution model to predict annual area burned with high severity. All models were cross-validated.
Results/Conclusions
Prediction of the probability of high fire severity fraction is robust and largely driven by large scale biophysical factors and only three variables specific to climate in the year of fire occurrence—average Spring temperature, average temperature in the month of fire, previous November moisture deficit. Additionally, our models predict the number of large fires that will reach the 0.168 high severity fraction threshold each year. Initial results from the general pareto distribution are robust in predicting the area burned with high severity fire each year in our dataset.
Removing within-year climate variability decreases our skill in predicting the probability of fires with high fractional fire severity, indicating that these variables are important for very severe fires. Removing vegetation predictors constricts the range of potential probability and decreases skill at its maximum probability. Removing both vegetation and concurrent climate severely diminishes the skill and predictive power of our model.
Our results indicate that though large scale biophysical parameters dominate the predictive models, the importance of vegetation and concurrent climate variables is critical for predicting occurrence of high fractional fire severity.