Quantifying contributions of natural variability and anthropogenic forcings on increased fire weather risk over the western United States

Abstract

Previous studies have identified a recent increase in wildfire activity in the western United States (WUS). However, the extent to which this trend is due to weather pattern changes dominated by natural variability versus anthropogenic warming has been unclear. Using an ensemble constructed flow analogue approach, we have employed observations to estimate vapor pressure deficit (VPD), the leading meteorological variable that controls wildfires, associated with different atmospheric circulation patterns. Our results show that for the period 1979 to 2020, variation in the atmospheric circulation explains, on average, only 32% of the observed VPD trend of 0.48 ± 0.25 hPa/decade (95% CI) over the WUS during the warm season (May to September). The remaining 68% of the upward VPD trend is likely due to anthropogenic warming. The ensemble simulations of climate models participating in the sixth phase of the Coupled Model Intercomparison Project suggest that anthropogenic forcing explains an even larger fraction of the observed VPD trend (88%) for the same period and region. These models and observational estimates likely provide a lower and an upper bound on the true impact of anthropogenic warming on the VPD trend over the WUS. During August 2020, when the August Complex "Gigafire" occurred in the WUS, anthropogenic warming likely explains 50% of the unprecedented high VPD anomalies.

Keywords: anthropogenic warming; atmospheric circulation; attribution; fire weather; western United States.

Fig. 1.

(A) Annual mean burned areas (10⁵ acres/yr) in the warm season during the period 1984 to 2000. Results for the average of each state are given by shading and with a numerical value. The averaged burned areas over the whole WUS are shown in the Lower Left corners. (B) Same as A but for the period of 2001 to 2018. The percentage changes of burned areas relative to those of the 1984 to 2000 period are shown below the annual mean burned areas. (C) Average days with high VPD (percentile VPD′ over 90% in a year) for the 1984 to 2000 period. (D) Same as B but for the averaged days with high VPD.

Fig. 2.

(A) Average time series of VPD′ from the gridMET dataset (solid line) and burned areas from the MTBS dataset (bars) for all warm season days. The VPD′ trend is the slope of the regressed line (dashed line) of the time series for all available years (1979 to 2020). The VPD′ trend for the shorter period 1984 to 2018 shows a similar result (SI Appendix, Fig. S1). (B and C) Same as A but for time series of e_s and e_a. (D–F) Trend map of these anomalies for the WUS (all warm season days). The absence of hatching denotes regions where the trends are significant at the P < 0.05 level.

Fig. 3.

(A) VPD′ time series in 2020 warm season over the WUS from both observations (black line) and analogues (blue line for mean analogue; shading for IQR). Starting days of the August Complex fire and California Creek fire are labeled. Dashed horizontal lines are the warm season mean values. (B) PDF of August VPD′ for the observations from the climatological period of 1979 to 2010 (black curve), 2020 observations (red bars, shaded dark gray where they overlap with blue bars), and 2020 analogues (blue bars). The three vertical lines in each box plot represent the 25th, 50th, and 75th percentiles, the dot represents the mean value, and the whiskers extend to two SDs from the mean. (C) Map of Z500 (contours) and its standardized anomalies relative to 1979 to 2010 climatology (shading) averaged over four reanalysis datasets (the fifth generation of the European Centre for Medium-Range Weather Forecasts [ECMWF] atmospheric reanalysis [ERA5], the Modern Era Retrospective analysis for Research and Applications version 2 [MERRA-2], the National Centers for Environmental Prediction [NCEP] Climate Forecast System Reanalysis [CFSR], and the Japanese 55-y Reanalysis [JRA55]) on August 16, 2020, the start date of the August Complex fire. (D) Percentile VPD map on the same date as C, overlaid with the 95, 99, and 100% contours. (E) Same as D but for constructed analogue VPD map.

Fig. 4.

(A) Warm season mean VPD′ time series over the WUS from observations (black line), analogues (blue line; shading represents IQR for VPD′ from the 180 analogue schemes described in Methods), and residuals (observations minus analogue, red line, shading represents IQR). (B) PDF of the residual VPD anomalies for the periods 1979 to 2000 and 2001 to 2020, respectively, and box plots (see Fig. 3B for explanation). (C) Analogue VPD′ trend (1979 to 2020) in each state. The value shown by bold black font within each state shows the VPD trend of that state. The value shown by bold black font in the Lower Left corner is the VPD trend averaged over the entire WUS. One, two, or three asterisk(s) next to these trend numbers denotes trend significance at P < 0.1, 0.05, and 0.01, respectively. Numbers inside brackets are IQR of the trends calculated from 180 individual analogue schemes. (D) Same as C but for residual VPD′ trend (observations minus analogue). (E) Percentage of the analogue VPD trend relative to the observed VPD trend (IQR in brackets). Montana, Wyoming, and Washington have nonsignificant observed VPD trends at the P < 0.05 level (SI Appendix, Table S4), and the corresponding regions are therefore hatched in C–E.

Fig. 5.

(A) Warm season mean VPD′ time series averaged over the WUS region and the trends during 1979 to 2020 calculated with climate models and observations (daily gridMET and monthly PRISM). The orange and blue line represents observed and residual VPD′ from gridMET, respectively; the yellow line represents observation from PRISM (a longer term monthly observational dataset covering 1895 to present that gridMET is based on); the black and cyan solid lines represent CMIP6-ALL and CMIP6-NAT simulations, and the thin gray and cyan lines are for all ensemble members from CMIP6-ALL and CMIP6-NAT, respectively. For the purposes of visual display, the VPD′ lines for ALL, NAT, and PRISM are forced to have the same mean value during 1979 to 2010 as gridMET. The VPD trends, 95% CI, and IQR (only for residual VPD′) labeled in the Upper Left corner are calculated for the 1979 to 2020 period. (B) PDF of VPD trend for PRISM observations and CMIP6-NAT; the vertical lines, dots, and whiskers for the box plots are defined as in Fig. 3B; VPD trend is calculated for every consecutive 42-y period within the periods listed above. (C) Same as B but for CMIP6-ALL. When calculating the ensemble-mean VPD trends and their PDFs in the CMIP6 simulations, VPD trend is first calculated for each ensemble member of each model, and weights are given to all the members in a way that all members from the same model are equally weighted and all models are also equally weighted.