Fig 1.
Regions of California, corresponding to CalVeg ecosystem provinces [45].
Table 1.
Description of variables used to estimate annual fire probability.
Fig 2.
(a) Observed mean annual fire return probabilities, (b) predicted mean annual fire probabilities from 1970–2016 produced by the statewide model, and (c) by a composite of all regional models. Boundaries between regions are delineated by black lines in predictions developed by regional models.
Table 2.
Parameter significance of the statewide Generalized Additive Model.
Table 3.
Mean ROC/AUC across iterations of both statewide and regional models (averaged across all regions within California) in predicting fire probabilities in novel years, novel locations, and in novel years at novel locations not used in model training.
Table 4.
Pearson correlation between observed 1970–2016 (and 1930–2016) fire probability and predicted fire probability from 1970–2016, as well as Pearson correlation among predictions of fire probability generated using a full model, using only local climate conditions, only local climate normals, human development (consisting of local housing density, distance to electrical infrastructure, and distance to roads), cultivation (consisting of % cultivated area within each pixel), and time since the most recent fire within each pixel.
Fig 3.
Predicted changes in mean annual fire probability after (a) eliminating the effects of human activity and after (b) eliminating the effects of variation in climate conditions throughout California.
Fig 4.
Predicted changes in 1970–2016 mean annual fire probability after eliminating the effects of short-term climate deviations.
Fig 5.
Smoothed coefficients for the statewide generalized additive model.
Coefficients include (a) 1951–1980 normal actual evapotranspiration, (b) 1951–1980 normal climatic water deficit, (c) annual deviations from 1951–1980 normal actual evapotranspiration, (d) annual deviations from 1951–1980 normal climatic water deficit, (e) proportion of cultivated area, (f) annual mean housing density, (g) years since fire, (h) distance from roads, and (i) distance from electrical infrastructure.