A new paper was published in Environmental Research Letters using pre-fire season climate conditions.
LETTER • OPEN ACCESS
Winter and spring climate explains a large portion of interannual variability and trend in western U.S. summer fire burned area
Citation Ronnie Abolafia-Rosenzweig et al 2022 Environ. Res. Lett. 17 054030
This study predicts summer (June–September) fire burned area across the western United States (U.S.) from 1984 to 2020 using ensembles of statistical models trained with pre-fire season climate conditions. Winter and spring climate conditions alone explain up to 53% of the interannual variability and 58% of the increasing trend of observed summer burned area, which suggests that climate conditions in antecedent seasons have been an important driver to broad-scale changes in summer fire activity in the western U.S. over the recent four decades. Relationships between antecedent climate conditions with summer burned area are found to be strongest over non-forested and middle-to-high elevation areas (1100–3300 m). Statistical models that predict summer burned area using both antecedent and fireseason climate conditions have improved performance, explaining 69% of the interannual variability and 83% of the increasing trend of observed burned area. Among the antecedent climate predictors, vapor pressure deficit averaged over winter and spring plays the most critical role in predicting summer fire burned area. Spring snow drought area is found to be an important antecedent predictor for summer burned area over snow-reliant regions in the nonlinear statistical modeling framework used in this analysis. Namely, spring snow drought memory is realized through dry anomalies in land (soil and fuel) and atmospheric moisture during summer, which favours fire activity. This study highlights the important role of snow drought in subseasonal-to-seasonal forecasts of summer burned area over snow-reliant areas.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Since the beginning of the 21st century, the Intergovernmental Panel on Climate Change has projected increases in summer wildfire risk in North America associated with decreased snowpack and increased summer drying caused by human-induced global warming [1, 2]. This projection holds true to a large degree in the western United States (U.S.) as observations have revealed persistent increasing trends in temperature, aridity and wildfire burned area since the mid-1980s [3–8]. It is well established that the observed increase in fire activity in the western U.S. is predominately explained by warmer and drier springs and summers caused by natural and human-induced climate changes [3, 8–11]. This is consistent with analyses of global paleo fire and climate records that indicate rapid warming has historically played an important role in modulating broad-scale fire activity over the past two millennia [12, 13]. Climate simulations considering a range of greenhouse gas concentration trajectories have projected that abrupt human-caused climate changes will likely contribute to further increases in fire hazard over the next century in the U.S. [14, 15]. This projection has striking societal and environmental consequences because fires cause thousands of smoke-related deaths per annum which is projected to increase throughout this century [15, 16], increased COVID-19 mortality , permanent changes to ecosystems , persistent changes to streamflow , and multi-billion-dollar suppression expenditures (www.nifc.gov/fire-information/statistics/suppression-costs). A robust understanding of the relationships between fire activity and climate conditions susceptible to change can enable better preparation for these consequences via implementation of policy and management that addresses these relationships as well as models that more accurately predict fire activity to inform resource allocation.
Since the early 2000s, a suite of statistical analyses has quantified relationships between summer fire activity and various climate conditions in the western U.S. [3, 5, 9–11, 20–22]. Many of these relationships have been reviewed by Littell et al [23, 24]. These statistical models primarily differ based on spatial and temporal resolutions and domains, model complexity, and climate predictors. However, they each support that broad-scale fire activity across the western U.S. has been mostly explained by fluctuations in antecedent and fire season climate conditions since the mid-1980s. For instance, Westerling et al  concluded that wildfire activity in the western U.S. made an abrupt transition in the mid-1980s from infrequent and short-burning wildfires to more frequent longer-burning fires due to unusually warm springs and longer and drier summers. A series of follow-up studies have shown that after this transition, the majority of interannual variability in burned area can be explained by climate conditions [5, 9, 11].
The above body of work provides valuable insights on relationships between climate variability and fire. However, they do not provide in-depth analyses that relate only pre-fire climate conditions to summer fire activity, which are essential for lead-time (e.g. subseasonal-to-seasonal) fire forecasts . Previous studies that have compared antecedent climate conditions with fire season burned area have reported significant relationships between pre-fire climate and fire activity, but have not comprehensively explored how much of the interannual variability or trend in fire season burned area can be explained by pre-fireseason climate alone [9, 10, 26–30]. Multiple of these studies have reported the effects of antecedent soil moisture conditions. Namely, wetter conditions 1–3 years before fire seasons in fuel-limited regions correspond with greater fire activity, and drier antecedent conditions in the months preceding the fire season tend to favour more fire activities [26–28]. A recent study that explored relationships between antecedent climate and fire season activity in a multivariate analysis showed machine learning models that predict burned area using both pre-fire and fire season conditions outperform models that use only fire season conditions . Although previous research underlines that antecedent conditions contain unique information for enhancing fireactivity predictability, it has not yet established a comprehensive quantitative relationship between fire season severity and pre-fire climate alone. Moreover, interannual changes in winter and spring snowpack have important relationships with fires [3, 10], motivating our hypothesis that a large portion of summer burned area variability and trend can be explained by pre-fire climate conditions alone, with an important contribution from pre-fire snowpack conditions in snow-reliant areas over the western U.S.
Indeed, dramatic temperature-driven declines in snowpack across the western U.S.  are considered an important link between climate trends and increasing fire hazard [3, 10]. Namely, reductions in winter and spring snowpack (particularly during snow drought) drive earlier spring melt, increase surface temperature and evaporation due to snow-albedo feedback, and more quickly deplete moisture of vegetation and soil, leading to longer and drier summers . The historically observed trend of declining snowpack has been projected to continue through the remainder of this century [33, 34]. Thus, understanding relationships between snow drought and fire over contemporary landscapes is imperative to project how future snowpack changes will impact fire activity, particularly in the western U.S. However, the snow-drought-fire interactions and the predictive ability of snow drought combined with other winter and spring climate conditions in predicting summer fire activity has not been fully understood and evaluated.
In this study, we use ensembles of nonlinear statistical models to predict summer fire burned area across the western U.S. based on pre-fire (winter and spring) climate including snow drought conditions. A unique novelty of this study is that we explore the predictability of burned area with combinations of pre-fire climate conditions and the role of snow-drought-fire interactions in modulating summer fire hazard. The goal of this study is to answer three following science questions. (i) What percent of interannual variability and trend in summer burned area can be explained by pre-fire climate conditions alone? (ii) What predictive information is contained in summer climate conditions that is not provided by winter and spring climate conditions? (iii) To what degree can the inclusion of snow drought enhance the predictability of summer fire activity relative to other traditionally used climate predictors? The overarching objectives of this study are to advance the understanding of seasonal climate-fire relationships, explore a methodology capable of predicting broad-scale fire burned area across the western U.S., and thus better inform policy and resource allocation.
2.1. Study domain
The spatial domain considered in this study includes all areas classified by MODIS satellite observations  as forest, grassland, or savanna in the western U.S. (north of 32° latitude and west of −104° longitude) below typical tree lines (<3300 m)  that have received adequate snowfall in winter and spring (peak snow water equivalent (SWE) > 100 mm) (figure 1(a)). The peak SWE threshold is imposed to limit the study domain to areas potentially affected by snow drought following Livneh and Badger  which supports assessing the importance of spring snow drought as a predictor of summer fire activity. The main results and conclusions presented herein are qualitatively insensitive to the choice of this peak SWE threshold (see sensitivity analyses in figures S1–S3 available online at stacks.iop.org/ERL/17/054030/mmedia). The preceding elevation and vegetation screening are performed to confine the domain to areas susceptible to fires, which reduces the domain area by 8% while retaining 99% of the burned area relative to all areas that meet the peak SWE threshold. After applying the aforementioned spatial screening, the study domain contains 43% land area and 61% burned area during 1984–2020 relative to the entire western U.S. (figure 1(a)). The interannual variability of burned area for the study domain is representative of the burned area over the entire western U.S., indicated by a very high correlation (r = 0.97) between fireseason burned area across the study domain and the entire western US (figure 1(c)). We select summer months (June–September) as the fire season in this analysis, which contains 94% of the total burned area within the study domain (figure 1(b)). Figure S10 shows that including May in the summer months (i.e. May–September) does not affect the qualitative relationships between pre-summer climate and summer burned area presented herein. Note that the domain has experienced an increasing trend in annual burned area from 1984 to 2020 for each calendar month (figure S4 and table S1).
Figure 1. Study domain details. (a) White shaded regions represent the snow-effected area used for the analysis. 500 m resolution observed burned fractions from 1984 to 2020 (in yellow to red shades) are overlain. (b) Box and whisker plots of burned area by month highlight that 94% of the burned area from 1984 to 2020 has occurred during summer months. Variability in boxplots is from the spread in burned area across different years, and whisker lengths are equivalent to the interquartile range. Outliers are plotted as red ‘+’. (c) Scatter plot of summer burned area for the study domain shown in (a) (horizontal axis) and the entire western U.S. (vertical axis) during 1984–2020 reveals that interannual variability of summer burned area in the study domain is representative of the entire western U.S.