Impact of anthropogenic climate change on wildfire across western US forests

Abstract

Increased forest fire activity across the western continental United States (US) in recent decades has likely been enabled by a number of factors, including the legacy of fire suppression and human settlement, natural climate variability, and human-caused climate change. We use modeled climate projections to estimate the contribution of anthropogenic climate change to observed increases in eight fuel aridity metrics and forest fire area across the western United States. Anthropogenic increases in temperature and vapor pressure deficit significantly enhanced fuel aridity across western US forests over the past several decades and, during 2000-2015, contributed to 75% more forested area experiencing high (>1 σ) fire-season fuel aridity and an average of nine additional days per year of high fire potential. Anthropogenic climate change accounted for ∼55% of observed increases in fuel aridity from 1979 to 2015 across western US forests, highlighting both anthropogenic climate change and natural climate variability as important contributors to increased wildfire potential in recent decades. We estimate that human-caused climate change contributed to an additional 4.2 million ha of forest fire area during 1984-2015, nearly doubling the forest fire area expected in its absence. Natural climate variability will continue to alternate between modulating and compounding anthropogenic increases in fuel aridity, but anthropogenic climate change has emerged as a driver of increased forest fire activity and should continue to do so while fuels are not limiting.

Keywords: attribution; climate change; forests; wildfire.

Fig. 1.

Annual western continental US forest fire area versus fuel aridity: 1984–2015. Regression of burned area on the mean of eight fuel aridity metrics. Gray bars bound interquartile values among the metrics. Dashed lines bounding the regression line represent 95% confidence bounds, expanded to account for lag-1 temporal autocorrelation and to bound the confidence range for the lowest correlating aridity metric. The two 16-y periods are distinguished to highlight their 3.3-fold difference in total forest fire area. Inset shows the distribution of forested land across the western US in green.

Fig. S1.

Multimodel mean anthropogenic climate change signal of 50-y smoothed values for 2015 minus those for 1901 for (Left to Right) Dec–Feb, Mar–May, Jun–Aug, and Sep–Nov for (Top to Bottom) maximum temperature, minimum temperature, vapor pressure, vapor pressure deficit, mean relative humidity, maximum relative humidity, and minimum relative humidity. Black dots show grid cells where at least 20 (>74%) of the 27 models agree on the direction of the trend.

Fig. 2.

Standardized change in each of the eight fuel aridity metrics due to ACC. The influence of ACC on fuel aridity during 2000–2015 is shown by the difference between standardized fuel aridity metrics calculated from observations and those calculated from observations excluding the ACC signal. The sign of PDSI is reversed for consistency with other aridity measures.

Fig. S2.

As in Fig. 2 but for (A–H) ERA-INTERIM and (I–P) NCEP–NCAR reanalysis. The influence of ACC on fuel aridity during 2000–2015 is shown by the difference between standardized fuel aridity metrics calculated from observations and those calculated from observations excluding the ACC signal. The sign of PDSI is reversed for consistency with other aridity measures.

Fig. S3.

As in Fig 3 but for ERA-INTERIM. (A) Time series of (Top) standardized annual fuel aridity metrics and (Bottom) percent of forest area with standardized fuel aridity exceeding one SD. Red lines show observations and black lines show records after exclusion of the ACC signal. Only the four monthly metrics extend back to 1950. Daily fire danger indices are constrained to 1979–2015. Bold lines indicate averages across fuel aridity metrics. (B) Linear trends in the standardized fuel aridity metrics during 1979–2015 for (red) observations and (black) records excluding the ACC signal (black). Asterisks indicate positive trends at the (*) 95% and (**) 99% significance levels.

Fig. S4.

As in Fig 3 but for NCEP–NCAR reanalysis. (A) Time series of (Top) standardized annual fuel aridity metrics and (Bottom) percent of forest area with standardized fuel aridity exceeding one SD. Red lines show observations and black lines show records after exclusion of the ACC signal. Only the four monthly metrics extend back to 1950. Daily fire danger indices are constrained to 1979–2015. Bold lines indicate averages across fuel aridity metrics. (B) Linear trends in the standardized fuel aridity metrics during 1979–2015 for (red) observations and (black) records excluding the ACC signal (black). Asterisks indicate positive trends at the (*) 95% and (**) 99% significance levels.

Fig. 3.

Evolution and trends in western US forest fuel aridity metrics over the past several decades. (A) Time series of (Upper) standardized annual fuel aridity metrics and (Lower) percent of forest area with standardized fuel aridity exceeding one SD. Red lines show observations and black lines show records after exclusion of the ACC signal. Only the four monthly metrics extend back to 1948. Daily fire danger indices begin in 1979. Bold lines indicate averages across fuel aridity metrics. Bars in the background of A show annual forested area burned during 1984–2015 for visual comparison with fuel aridity. (B) Linear trends in the standardized fuel aridity metrics during 1979–2015 for (red) observations and (black) records excluding the ACC signal (differences attributed to ACC). Asterisks indicate positive trends at the (*) 95% and (**) 99% significance levels.

Fig. 4.

Changes in fire-weather season length and number of high fire danger days. Time series of mean western US forest (A) fire-weather season length and (B) number of days per year when daily fire danger indices exceeded the 95th percentile. Baseline period: 1981–2010 using observational records that exclude the ACC signal. Red lines show the observed record, and black lines show the record that excludes the ACC signal. Bold lines show the average signal expressed across fuel aridity metrics.

Fig. 5.

Attribution of western US forest fire area to ACC. Cumulative forest fire area estimated from the (red) observed all-metric mean record of fuel aridity and (black) the fuel aridity record after exclusion of ACC (No ACC). The (orange) difference is the forest fire area forced by anthropogenic increases in fuel aridity. Bold lines in A and horizontal lines within box plots in B indicate mean estimated values (regression values in Fig. 1). Boxes in B bound 50% confidence intervals. Shaded areas in A and whiskers in B bound 95% confidence intervals. Dark red horizontal lines in B indicate observed forest fire area during each period.

Fig. S5.

Relationships between all-metric mean fuel aridity anomalies and burned area in western US forests (A and B) are used to model the annual response of forest fire area to fuel aridity (C and D) under observed fuel aridity conditions and those recalculated after the removal of ACC. Two methods are used to derive the response of forest fire area: (A) derived from raw data (as presented in the article) and (B) derived from detrended data for 1984–2015. This alternate approach is more conservative because it reduces risk of assuming an artificially strong relationship caused by common but unrelated trends. (E) The estimated relative forcing of ACC on cumulative burned area, calculated as the relative difference between burned area modeled from observed fuel aridity and burned area modeled in the absence of ACC. In A–D, areas bounding the central lines correspond to 95% confidence intervals around the regression lines. In E, boxes and whiskers indicate 50% and 95% confidence intervals, respectively.

Fig. S6.

Observations (blue) versus CMIP5 projections (black and gray) of March–September vapor pressure anomalies (relative to 1948–1990 mean) in western US forest areas. Thick black line is the multimodel (n = 27) mean and gray area bounds the interquartile values. CMIP5 projections have had a 50-y low-pass filter applied to exclude high-frequency variations caused by natural climate variability.

Fig. S7.

Linear trend in (A) March–May, and (B) June–September (contours) 250-hPa geopotential height (in meters, data source: ERA-INTERIM) and (background) precipitation (percent of 1979–2015 average, data source: PRISM an81m) during 1979–2015. Only precipitation trends significant at the P < 0.1 level are shown. Lower shows CMIP5 ensemble-mean trends for the same variables during 1979–2015 for (C) March–May and (B) June–September (n = 39 models). For precipitation, trends are only shown if at least 75% of models agree on the sign of the trend. Trends are reported in units per 37 y. The location of western US forests is shown in gray in A and B.