Fig 1.
Linear models used for predictions.
(a) LDPS (last day of permanent snow cover) and (b) AAB (annual area burned). (c) Structural Equation Model (SEM) diagram of direct and indirect controls on annual area burned in western North America.
Fig 2.
Temporal trends (1972–2006) in instrumental seasonal climate and snow cover duration.
(a) Winter (JFM) temperature (°C), (b) spring (AMJ) temperature (°C), (c) summer (JAS) temperature (°C), and (d) LDPS (days/decade), based on the Theil-Sen median slope estimator.
Fig 3.
Results of PCA analysis of z coefficients of a complete multiple regression model in each grid cell.
Panels show the first four principal components and percent in variation in AAB explained. Seasonal climate variables correlated with PC loadings at r ≥ 0.5 are listed including the sign of the correlation with AAB. (a) PC1, summer temperature (+) and spring precipitation (-); (b) PC2, spring temperature (+); (c) PC3, winter temperature (-); (d) PC4, preceding year summer temperature (+). Red (blue) colors indicate increases (decreases) in log-transformed AAB with increases in variables correlating positively/negatively with PC scores. Note that PC3 is inverted in sign for ease in interpretation. See S1 Fig for PC5.
Fig 4.
Direct and indirect influences of seasonal temperature and snow cover duration on annual area burned.
Results of Structural Equation Models (SEM;1974–2004) show the direct path coefficients for (a) spring and (b) winter temperature on LDPS, (c) the direct effect of LDPS on AAB, the indirect effects of (d) spring and (e) winter temperature on AAB as expressed by variation in LDPS, and the direct effects of (f) spring and (g) winter temperature on AAB.
Fig 5.
Direct and indirect effects of climate and snow cover on AAB by LDPS regions.
Spring (a) and winter (b) mean (± SE) path coefficients averaged over areas of boreal and western North America with areas of similar snow cover duration (monthly classes of long-term (1972–2006) mean LDPS. Black bars indicate direct effects of temperature on log-AAB; grey bars indirect effects on log-AAB mediated by variation in LDPS. (c) Geographic distribution of monthly mean LDPS.
Fig 6.
Projected change in seasonal climate variables influencing AAB.
Changes are based are for the period 2010–2039 compared to the baseline period 1961–1990 based on an ensemble of A1B emission scenarios: (a) winter (b) spring, and (c) summer temperature (Δ°C); (d) LDPS (last day of permanent snow cover (in Julian date).
Fig 7.
Projections rates are for the period 2010–2039 compared to the period 1961–2004 based on ensemble of A1B emission scenarios. (a) Percent change in AAB resulting from stepwise selection of individual cell-based models, based on AIC model selection criteria; (b) proportion of variance explained (R2). Only significant models (p < 0.05) are plotted.
Fig 8.
Changes in AAB across United States and Canada by state/province.
(a) Boxplot of percent change in AAB (2010–2039 vs.1961-2004, SRES A1B scenario) binned by US state or Canadian province, based on significant models with p < 0.05. Sharing of any letter (below the graph) indicates lack of significant differences in medians of percent change in AAB based on Bonferroni-corrected a posteriori comparisons of a Kruskal-Wallis median test. Colored boxes indicate groups of states/regions with statistically similar medians ordered from low (green) through high median values (red). (b) States/regions ordered by increasing median change in AAB. Histograms are model projections based on the 1976–2006 baseline period; red dots are extrapolated increases in median AAB and bars are 95% confidence intervals estimated from Theil-Sen trends (1972–2015).