Fig 1.
Conceptual model of major factors affecting fire occurrence, size, and severity.
Red lines: human influences; green lines: direct climate influences; black lines: indirect climate (vegetation) influences; blue lines: topographic influences. Bold text represents groups of variables included in the analysis. Non-bold text represents implicit relationships that were not directly analyzed.
Fig 2.
Spatial distributions of (a) smoothed density of large fires (fires per million ha per year), (b) smoothed mean fire size (ha), (c) smoothed mean high severity fire size (ha), and (d) smoothed percent of high severity burning (%) in the western US.
Non-burnable areas are displayed in white.
Fig 3.
Spline correlograms illustrating the non-parametric spatial covariance function and 95% confidence intervals (gray area) for (a) large fire occurrence; (b) fire size; and (c) percent of high severity burning.
Fig 4.
Relative influences of variables that explained greater than 5% of the variation and marginal effects (red trend lines within each bar) from boosted regression tree models for (a) large fire occurrence, (b) fire size, and (c) percent of high severity burning.
Values are specified for truncated bars. Abbreviations of predictor variables and their corresponding full names are described in Table 1.
Table 1.
Environmental variables summarized for each fire and used in the analysis.
Fig 5.
Simplified versions of the first tree of (a) fire occurrence model (b) fire size model, and (c) percent of high severity burning model computed with the boosted regression tree algorithm.
The first three splits of each tree are shown to illustrate the interactions between key variables. The splitting variable and its corresponding splitting value are shown in oval above the node (variable abbreviation and units are provided in Table 1). The values in the rectangles at the terminal nodes represent the mean prediction and number of the records in the terminal nodes (n). The total number of records in the terminal nodes equals 0.75 (the bag fraction of the BRT model) of the total number of fires. Abbreviations: P: relative probability of fire occurrence; MFS: mean fire size; PHS: percent of high severity burning.