Impact of anthropogenic climate change on wildfire across western US forests

Widespread increases in fire activity, including area burned (1, 2), number of large fires (3), and fire-season length (4, 5), have been documented across the western United States (US) and in other temperate and high-latitude ecosystems over the past half century (6, 7). Increased fire activity across western US forests has coincided with climatic conditions more conducive to wildfire (2–4, 8). The strong interannual correlation between forest fire activity and fire-season fuel aridity, as well as observed increases in vapor pressure deficit (VPD) (9), fire danger indices (10), and climatic water deficit (CWD) (11) over the past several decades, present a compelling argument that climate change has contributed to the recent increases in fire activity. Previous studies have implicated anthropogenic climate change (ACC) as a contributor to observed and projected increases in fire activity globally and in the western United States (12–19), yet no studies have quantified the degree to which ACC has contributed to observed increases in fire activity in western US forests.

Changes in fire activity due to climate, and ACC therein, are modulated by the co-occurrence of changes in land management and human activity that influence fuels, ignition, and suppression. The legacy of twentieth century fire suppression across western continental US forests contributed to increased fuel loads and fire potential in many locations (20, 21), potentially increasing the sensitivity of area burned to climate variability and change in recent decades (22). Climate influences wildfire potential primarily by modulating fuel abundance in fuel-limited environments, and by modulating fuel aridity in flammability-limited environments (1, 23, 24). We constrain our attention to climate processes that promote fuel aridity that encompass fire behavior characteristics of landscape ignitability, flammability, and fire spread via fuel desiccation in primarily flammability-limited western US forests by considering eight fuel aridity metrics that have well-established direct interannual relationships with burned area in this region (1, 8, 24, 25). Four metrics were calculated from monthly data for 1948–2015: (i) reference potential evapotranspiration (ETo), (ii) VPD, (iii) CWD, and (iv) Palmer drought severity index (PDSI). The other four metrics are daily fire danger indices calculated for 1979–2015: (v) fire weather index (FWI) from the Canadian forest fire danger rating system, (vi) energy release component (ERC) from the US national fire danger rating system, (vii) McArthur forest fire danger index (FFDI), and (viii) Keetch–Byram drought index (KBDI). These metrics are further described in the Materials and Methods and Supporting Information. Fuel aridity has been a dominant driver of regional and subregional interannual variability in forest fire area across the western US in recent decades (2, 8, 22, 25). This study capitalizes on these relationships and specifically seeks to determine the portions of the observed increase in fuel aridity and area burned across western US forests attributable to anthropogenic climate change.

The interannual variability of all eight fuel aridity metrics averaged over the forested lands of the western US correlated significantly (R² = 0.57–0.76, P < 0.0001; Table S1) with the logarithm of annual western US forest area burned for 1984–2015, derived from the Monitoring Trends in Burn Severity product for 1984–2014 and the Moderate Resolution Imaging Spectroradiometer (MODIS) for 2015 (Supporting Information). The record of standardized fuel aridity averaged across the eight metrics (hereafter, all-metric mean) accounts for 76% of the variance in the burned-area record, with significant increases in both records for 1984–2015 (Fig. 1). Correlation between fuel aridity and forest fire area remains highly significant (R² = 0.72, all-metric mean) after removing the linear-least squares trends for each time series for 1984–2015, supporting the mechanistic relationship between fuel aridity and forest fire area. It follows that co-occurring increases in fuel aridity and forest fire area over multiple decades would also be mechanistically related.

We quantify the influence of ACC using the Coupled Model Intercomparison Project, Phase 5 (CMIP5) multimodel mean changes in temperature and vapor pressure following Williams et al. (26) (Fig. S1; Methods). This approach defines the ACC signal for any given location as the multimodel mean (27 CMIP5 models) 50-y low-pass-filtered record of monthly temperature and vapor pressure anomalies relative to a 1901 baseline. Other anthropogenic effects on variables such as precipitation, wind, or solar radiation may have also contributed to changes in fuel aridity but anthropogenic contributions to these variables during our study period are less certain (22). We evaluate differences between fuel aridity metrics computed with the observational record and those computed with observations that exclude the ACC signal to determine the contribution of ACC to fuel aridity. To exclude the ACC signal, we subtract the ACC signal from daily and monthly temperature and vapor pressure, leaving all other variables unchanged and preserving the temporal variability of observations. The contribution of ACC to changes in fuel aridity is shown for the entire western United States; however, we constrain the focus of our attribution and analysis to forested environments of the western US (Fig. 1, Inset; Methods).

Anthropogenic increases in temperature and VPD contributed to a standardized (σ) increase in all-metric mean fuel aridity averaged for forested regions of +0.6 σ (range of +0.3 σ to +1.1 σ across all eight metrics) for 2000–2015 (Fig. 2). We found similar results with reanalysis products (all-metric mean fuel aridity increase of +0.6 σ for two reanalysis datasets considered; Methods), suggesting robustness of the results to structural uncertainty in observational products (Figs. S2–S4 and Table S2). The largest anthropogenic increases in standardized fuel aridity were present across the intermountain western United States, due in part to larger modeled warming rates relative to more maritime areas (27). Among aridity metrics, the largest increases tied to the ACC signal were for VPD and ETo because the interannual variability of these variables is primarily driven by temperature for much of the study area (28). By contrast, PDSI and ERC showed more subdued ACC driven increases in fuel aridity because these metrics are more heavily influenced by precipitation variability.

Fuel aridity averaged across western US forested areas showed a significant increase over the past three decades, with a linear trend of +1.2 σ (95% confidence: 0.42–2.0 σ) in the all-metric mean for 1979–2015 (Fig. 3A, Top and Table S1). The all-metric mean ACC contribution since 1901 was +0.10 σ by 1979 and +0.71 σ by 2015. The annual area of forested lands with high fuel aridity (>1 σ) increased significantly during 1948–2015, most notably since 1979 (Fig. 3A, Bottom). The observed mean annual areal extent of forested land with high aridity during 2000–2015 was 75% larger for the all-metric mean (+27% to +143% range across metrics) than was the case where the ACC signal was excluded.

Significant positive trends in fuel aridity for 1979–2015 across forested lands were observed for all metrics (Fig. 3B and Table S1). Positive trends in fuel aridity remain after excluding the ACC signal, but the remaining trend was only significant for ERC. Anthropogenic forcing accounted for 55% of the observed positive trend in the all-metric mean fuel aridity during 1979–2015, including at least two-thirds of the observed increase in ETo, VPD, and FWI, and less than a third of the observed increase in ERC and PDSI. No significant trends were observed for monthly fuel aridity metrics from 1948–1978.

The duration of the fire-weather season increased significantly across western US forests (+41%, 26 d for the all-metric mean) during 1979–2015, similar to prior results (10) (Fig. 4A and Table S2). Our analysis shows that ACC accounts for ∼54% of the increase in fire-weather season length in the all-metric mean (15–79% for individual metrics). An increase of 17.0 d per year of high fire potential was observed for 1979–2015 in the all-metric mean (11.7–28.4 d increase for individual metrics), over twice the rate of increase calculated from metrics that excluded the ACC signal (Fig. 4B and Table S2). This translates to an average of an additional 9 d (7.8–12.0 d) per year of high fire potential during 2000–2015 due to ACC.

Given the strong relationship between fuel aridity and annual western US forest fire area, and the detectable impact of ACC on fuel aridity, we use the regression relationship in Fig. 1 to model the contribution of ACC on western US forest fire area for the past three decades (Fig. 5 and Fig. S5). ACC-driven increases in fuel aridity are estimated to have added ∼4.2 million ha (95% confidence: 2.7–6.5 million ha) of western US forest fire area during 1984–2015, similar to the combined areas of Massachusetts and Connecticut, accounting for nearly half of the total modeled burned area derived from the all-metric mean fuel aridity. Repeating this calculation for individual fuel aridity metrics yields ACC contributions of 1.9–4.9 million ha, but most individual fuel aridity metrics had weaker correlations with burned area and thus may be less appropriate proxies for attributing burned area. The effect of the ACC forcing on fuel aridity increased during this period, contributing ∼5.0 (95% confidence: 4.2–5.9) times more burned area in 2000–2015 than in 1984–1999 (Fig. 5B). During 2000–2015, the ACC-forced burned area likely exceeded the burned area expected in the absence of ACC (Fig. 5B). A more conservative method that uses the relationship between detrended records of burned area and fuel aridity (2) still indicates a substantial impact of ACC on total burned area, with a 19% (95% confidence: 12–24%) reduction in the proportion of total burned area attributable to ACC (Fig. S5).

Our attribution explicitly assumes that anthropogenic increases in fuel aridity are additive to the wildfire extent that would have arisen from natural climate variability during 1984–2015. Because the influence of fuel aridity on burned area is exponential, the influence of a given ACC forcing is larger in an already arid fire season such as 2012 (Fig. 5A and Fig. S5C). Anthropogenic increases in fuel aridity are expected to continue to have their most prominent impacts when superimposed on naturally occurring extreme climate anomalies. Although numerous studies have projected changes in burned area over the twenty-first century due to ACC, we are unaware of other studies that have attempted to quantify the contribution of ACC to recent forested burned area over the western United States. The near doubling of forested burned area we attribute to ACC exceeds changes in burned area projected by some modeling efforts to occur by the mid-twenty-first century (29, 30), but is proportionally consistent with mid-twenty-first century increases in burned area projected by other modeling efforts (17, 31–33).

Beyond anthropogenic climatic changes, several additional factors have caused increases in fuel aridity and forest fire area since the 1970s. The lack of fuel aridity trends during 1948–1978 and persistence of positive trends during 1979–2015 even after removing the ACC signal implicates natural multidecadal climate variability as an important factor that buffered anthropogenic effects during 1948–1978 and compounded anthropogenic effects during 1979–2015. During 1979–2015, for example, observed Mar–Sep vapor pressure decreased significantly across many US forest areas, in marked contrast to modeled anthropogenic increases (Fig. S6) (34). Significant declines in spring (Mar–May) precipitation in the southwestern United States and summer (Jun–Sep) precipitation throughout parts of the northwestern United States during 1979–2015 (Fig. S7 A and B) hastened increases in fire-season fuel aridity, consistent with observed increases in the number of consecutive dry days across the region (10). Natural climate variability, including a shift toward the cold phase of the interdecadal Pacific Oscillation (35), was likely the dominant driver of observed regional precipitation trends (36) (Fig. S7 B and D).

Our quantification of the ACC contribution to observed increases in forest fire activity in the western United States adds to the limited number of climate change attribution studies on wildfire to date (37). Previous attribution efforts have been restricted to a single GCM and biophysical variable (14, 16). We complement these studies by demonstrating the influence of ACC derived from an ensemble of GCMs on several biophysical metrics that exhibit strong links to forest fire area. However, our attribution effort only considers ACC to manifest as trends in mean climate conditions, which may be conservative because climate models also project anthropogenic increases in the temporal variability of climate and drought in the western United States (34, 38, 39). In focusing exclusively on the direct impacts of ACC on fuel aridity, we do not address several other pathways by which ACC may have affected wildfire activity. For example, the fuel aridity metrics that we used may not adequately capture the role of mountain snow hydrology on soil moisture. Nor do we account for the influence of climate change on lightning activity, which may increase with warming (40). We also do not account for how fire risk may be affected by changes in biomass/fuel due to increases in atmospheric CO₂ (41), drought-induced vegetation mortality (42), or insect outbreaks (43).

Additionally, we treat the impact of ACC on fire as independent from the effects of fire management (e.g., suppression and wildland fire use policies), ignitions, land cover (e.g., exurban development), and vegetation changes beyond the degree to which they modulate the relationship between fuel aridity and forest fire area. These factors have likely added to the area burned across the western US forests and potentially amplified the sensitivity of wildfire activity to climate variability and change in recent decades (2, 22, 24, 44). Such confounding influences, along with nonlinear relationships between burned area and its drivers (e.g., Fig. 1), contribute uncertainty to our empirical attribution of regional burned area to ACC. Our approach depends on the strong observed regional relationship between burned area and fuel aridity at the large regional scale of the western United States, so the quantitative results of this attribution effort are not necessarily applicable at finer spatial scales, for individual fires, or to changes in nonforested areas. Dynamical vegetation models with embedded fire models show emerging promise as tools to diagnose the impacts of a richer set of processes than those considered here (41, 45) and could be used in tandem with empirical approaches (46, 47) to better understand contributions of observed and projected ACC to changes in regional fire activity. However, dynamic models of vegetation, human activities, and fire are not without their own lengthy list of caveats (2). Given the strong empirical relationship between fuel aridity and wildfire activity identified here and in other studies (1, 2, 4, 8), and substantial increases in western US fuel aridity and fire-weather season length in recent decades, it appears clear from empirical data alone that increased fuel aridity, which is a robustly modeled result of ACC, is the proximal driver of the observed increases in western US forest fire area over the past few decades.