Despite growing knowledge of fireenvironment linkages in the western USA, obtaining reliable estimates of relative wildfire likelihood remains a work in progress. The purpose of this study is to use updated fire observations during a 25-year period and a wide array of environmental variables in a statistical framework to produce high-resolution estimates of wildfire probability. Using the MaxEnt modelling technique, point-source fire observations that were sampled from area burned during the 19842008 time period were related to explanatory variables representing ignitions, flammable vegetation (i.e. fuels), climate and topography. Model results were used to produce spatially explicit predictions of wildfire probability. To assess the effect of humans on the spatial patterns of wildfire likelihood, we built an alternative model that excluded all variables having a strong anthropogenic imprint. Results showed that wildfire probability in the western USA is far from uniform, with different areas responding to different environmental drivers. The effect of anthropogenic factors on wildfire probability varied by region but, on the whole, humans appear to inhibit fire activity in the western USA. Our results not only provide what appear to be robust predictions of wildfire likelihood, but also enhance understanding of long-term controls on wildfire activity. In addition, our wildfire probability maps provide better information for strategic planning of land-management activities, especially where fire regime knowledge is sparse.
- MaxEnt algorithm
- spatial modelling