Environmental controls on the distribution of wildfire at multiple spatial scales

Despite its widespread occurrence globally, wild. re preferentially occupies an environmental middle ground and is significantly less prevalent in biomes characterized by environmental extremes (e. g., tundra, rain forests, and deserts). We evaluated the biophysical "environmental space'' of wild. re from regional to subcontinental extents, with methods widely used for modeling habitat distributions. This approach is particularly suitable for the biogeographic study of wild. re, because it simultaneously considers patterns in multiple factors controlling wild. re suitability over large areas. We used the Maxent and boosted regression tree algorithms to assess wild. re-environment relationships for three levels of complexity (in terms of inclusion of variables) at three spatial scales: the conterminous United States, the state of California, and five wild. re-prone ecoregions of California. The resulting models were projected geographically to obtain spatial predictions of wild. re suitability and were also applied to other regions to assess their generality and spatial transferability. Predictions of the potential range of wild. re had high classification accuracy; they also highlighted areas where wild. res had not recently been observed, indicating the potential (or past) suitability of these areas. The models identified several key variables that were not suspected to be important in the large-scale control of wild. res, but which might indirectly affect control by influencing the presence of inflammable vegetation. Models transferred to different areas were useful only when they overlapped appreciably with the target area's environmental space. This approach should allow exploration of the potential shifts in wild. re range in a changing climate, the potential for restoration of wild. re where it has been "extirpated,'' and, conversely, the "invasiveness'' of wild. re after changes in plant species composition. Our study demonstrates that habitat distribution models and related concepts can be used to characterize environmental controls on a natural disturbance process, but also that future work is needed to re. ne our understanding of the direct causal factors controlling wild fire at multiple spatial scales.