Estimating PM2.5-related premature mortality and morbidity associated with future wildfire emissions in the western US

The annual number and size of wildfires in the western United States (US) have been increasing [1–3]. Recent trends are a significant departure from historic wildfire regimes and are significantly correlated with human-induced climate change [4–8]. Wildfire smoke degrades air quality both near fires and far downwind—a phenomenon readily apparent in the 2020 wildfire season [9, 10]. The intensity of the 2018 wildfires in the western US led to five rural cities in northern California and the Pacific Northwest registering annual average fine particulate matter (PM_2.5) concentrations exceeding 20 μg m⁻³, compared with the National Ambient Air Quality Standard of 12 μg m⁻³ [11].

Climate change affects wildfire activity by modifying patterns of temperature and precipitation, which can increase the number and length of low moisture periods in air, soil, and biomass [1, 3, 5, 7, 12–14]. Climate effects vary geographically depending on whether an area is fuel-limited or flammability-limited [12]. Most studies examining these patterns in the western US have found that the largest increases in wildfire activity are likely to occur in northern California and the intermountain West—areas with high fuel loads that are projected to experience increased aridity [6]. In addition, lightning strikes in the contiguous US (CONUS) are projected to increase at higher temperatures—up to 50% over this century [15]. Other non-climate drivers of wildfire activity in the western US include the role of changing settlement patterns in the wildland–urban interface and human wildfire ignitions [16, 17].

As wildfire activity increases, the effects of wildfire emissions on local and downwind communities will also intensify [9, 18–20]. Exposure to wildfire smoke can lead to significant health effects that impose a substantial cost on society [21–23]. Previous studies have linked exposure to wildfire smoke with a broad range of health effects, including premature mortality, hospital admissions, and emergency department (ED) visits, often focused on respiratory ailments [9, 10, 24–26]. The literature on the potential air pollution-related mortality and morbidity burden from future wildfire smoke exposure in the US remains limited, though appears relatively consistent in estimating worsening health outcomes with some heterogeneity by region, scenario, and climate model used [10, 24].

Here we build on this earlier work to provide a comprehensive analysis of the present and potential future health burden associated with PM_2.5 exposure from wildfire smoke in the western US. Specifically, we examine the health burden of projected future wildfire activity on P_2.5-attributable premature mortality and a range of morbidity outcomes, including hospital admissions, ED visits, asthma exacerbations, and upper and lower respiratory symptoms (see table 1 and, for details on specific endpoints and ICD codes, table S7 in the supplemental material (available online at stacks.iop.org/ERL/16/035019/mmedia)). We use methods compatible with the US Environmental Protection Agency's (US EPA) Climate Change Impacts and Risk Analysis (CIRA) framework, developed as an input to the US Global Change Research Program's National Climate Assessment, to achieve two distinct objectives: (a) to allow comparison of the results with the economic impacts of climate change on other sectors in the US; and (b) to assess wildfire impacts across two future climate forcing scenarios simulated in five climate models [27, 28]. Assessing a range of outcomes across greenhouse gas emissions (GHGs) scenarios and alternative climate models adds new insights relative to previous wildfire literature, which have typically relied on a single general circulation model (GCM—a model with a global domain that projects future multi-dimensional climate change) or single GHG emission scenario. Our results also complement other recent CIRA estimates that use similar methods to show climate change could also alter meteorological conditions in ways that still further increase US population exposure to PM_2.5 beyond that estimated here [29].

We estimate wildfire-attributable PM_2.5 and associated mortality and morbidity in the US under two climate scenarios and five GCMs throughout the 21st century. The two climate scenarios are Representative Concentration Pathway (RCP) 8.5, a higher GHGs scenario, and RCP4.5, a lower scenario. We do not assign probabilities to the relative likelihood that one of these scenarios may describe better the pathway of future emissions. We employ spatially downscaled climate projections from the Localized Constructed Analogs dataset for five Earth system models participating in the Coupled Model Intercomparison Project Phase 5: CanESM2, CCSM4, GISS-E2-R, HadGEM2-ES, and MIROC5, consistent with the CIRA framework (described further in section 1 of supplemental material) [27, 28]. We average across 10 year climatic time periods to smooth interannual variability: a baseline of 1996–2005, and projections for 2050 (2046–2055) and 2090 (2086–2095). For each climate scenario, GCM, and year, we: (a) estimate ecoregion-specific wildfire area burned and black carbon (BC) and organic carbon (OC) emissions (see supplemental material sections 2 and 3); (b) simulate ambient PM_2.5 concentrations resulting from these emissions; and (c) quantify wildfire-attributable PM_2.5 mortality and morbidity (figure 1). We estimate wildfire activity in the western US only, because the high prevalence of prescribed burns in the east obscures statistical relationships between climate and wildfires. However, we estimate air quality and health impacts stemming from western wildfire-related PM_2.5 for the entire CONUS in order to provide a more comprehensive estimate of the domestic effects of wildfire changes, since smoke can travel long distances, affecting air quality on continental scales.

Zoom In Zoom Out Reset image size

2.1. Estimating annual area burned and wildfire emissions based on climate data

2.2. Wildfire-associated PM_2.5 concentrations

2.3. Estimation of wildfire-associated PM_2.5 health impacts

We estimate wildfire PM_2.5-associated health impacts and economic values using the US EPA's Environmental Benefits Mapping and Analysis Program—Community Edition (BenMAP—CE) version 1.5.1.0 [35, 36]. General forms of the epidemiological concentration-response functions used to calculate wildfire PM_2.5-attributable health impacts are provided below:

$$\begin{align}{y_{ika}} & = {\text{Incidenc}}{{\text{e}}_{ika}} \times \left( {1 - {{\text{e}}^{\, - {\beta _i} \times {\text{WildfirePM}}{{2.5}_k}}}} \right) \nonumber\\[4pt] & \quad \times {\text{Populatio}}{{\text{n}}_{ka}}\end{align} \tag{ 1a }$$

$$\begin{equation}{y_{ika}} = {\text{Incidenc}}{{\text{e}}_{ika}} \times \left( {1 - \frac{1}{{\left( {1 - {\text{Incidenc}}{{\text{e}}_{ika}}} \right) \times {{\text{e}}^{\,{\beta _i} \times {\text{WildfirePM}}{{2.5}_k}}} + {\text{Incidenc}}{{\text{e}}_{ika}}}}} \right) \times {\text{Populatio}}{{\text{n}}_{ka}}\end{equation} \tag{ 1b }$$

$$\begin{equation}{y_i} = \sum\limits_k {\sum\limits_a {{y_{ika}}} } .\end{equation} \tag{ 2 }$$

For each health endpoint, the use of either the log-linear form (1a ) or the logistic form (1b ) is determined by the functional form in the epidemiological study used to derive the health impact function. In these equations, i signifies a specific health endpoint, k signifies a 0.5° × 0.625° GEOS-Chem grid cell in the western US, and a represents a population subgroup spanning 5 years of age. In equations (1a ) and (1b ), y_ika represents the wildfire-attributable PM_2.5 health impacts for health endpoint i in grid cell k for population subgroup a; Incidence_ika is the baseline incidence of health endpoint i in grid cell k for population subgroup a; β_i is the effect estimate relating a change in PM_2.5 concentration to a change in the risk of health impact i; WildfirePM2.5_k is the annual mean wildfire-attributable PM_2.5 concentration in grid cell k, and Population_ka is the population subgroup in grid cell k. To calculate total wildfire-attributable PM_2.5 health impacts for a particular health endpoint during the baseline, 2050, or 2090 eras, health impacts are aggregated across all grid cells and all population subgroups as shown in equation (2). In equation (2), y_i signifies the total additional wildfire-attributable health impacts for a given health endpoint during the baseline, 2050, or 2090 eras when compared to the baseline without wildfires. We follow the US EPA methodology for estimating health risks from PM_2.5 using concentration-response functions for total, not source-specific, PM_2.5 mass. Thus, we apply total PM_2.5 based health impact functions to the fine particles emitted as a result of wildfires; however, we limit our analysis of health impacts to those specific endpoints that have to date been associated with wildfire PM_2.5 (listed in table 1).

For the baseline period, we use year 2000 population from BenMAP-CE. For the 2050 and 2090 eras, we use population projections derived for each 5 year age group at the county level using the EPA's Integrated Climate and Land Use Scenarios v.2 (ICLUSv2) model [37, 38]. This ICLUSv2 scenario was developed as an intermediate growth population projection, with underlying assumptions about fertility, migration rate, and international immigration derived from the storyline for IPCC Shared Socioeconomic Pathway 2. County-level incidence rates were used as provided in BenMAP-CE for 5 year age groups and were originally derived from CDC WONDER [35]. We applied 2000 mortality rates for the baseline period, and matched the appropriate mortality rate projections for the 2050 era. As the CDC mortality rates were projected only through 2060, we apply 2060 mortality rates to the 2090 era analyses. Morbidity incidence rates are available only for 2014, so these rates are assumed to be constant. Results would scale approximately linearly with different projections of mortality rates.

2.4. Economic valuation of wildfire-associated PM_2.5 health impacts

Overall, our estimates of fire activity capture a substantial amount of interannual variability in area burned that can be attributed to climatic variation, though they generally underestimate peak years across all ecoregions (figure 2 and table S2). Underestimates of peak burn years are an important qualification of our results, particularly as western US burn area records set in the most recent 2020 fire season suggest the possibility of greater frequency of extreme annual wildfire incidence. We estimate that climatic factors increase wildfire BC and OC emissions under both RCP4.5 and 8.5, for all GCMs and future years (figure 3). Averaged across the five GCMs, BC and OC emissions increase 0.40% per year over the 2006–2100 period under RCP4.5 and 0.71% per year for RCP8.5 (p < 0.001). We find large interannual variability in wildfire BC and OC emissions driven by variability in precipitation and soil moisture inputs. Comparing between GCMs, we find that the estimated wildfire emissions increase is more moderate using simulated climate variables from a cooler climate model (GISS-E2-R), with virtually no change in emissions in both the 2050 and 2090 eras for RCP4.5, but increases of 8% in the 2050 era and 17% in the 2090 era relative to the 1996–2005 reference period for RCP8.5. The four other GCMs produce larger increases in emissions. HADGEM2-ES results have the largest magnitude change in emissions, with wildfire BC and OC emissions increasing from 2000 by over a third in the 2050 period for RCP4.5, and more than doubling for the 2090 period for RCP8.5. The results across GCMs largely correspond to variation in individual GCM projections of temperature and precipitation over the western U for our time period, and clearly illustrate the importance of using a wide range of plausible climate drivers when estimating future fire activity and emissions.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 4 shows western wildfire-attributable PM_2.5 concentrations for RCP8.5 in 2090 for all five GCMs. Spatially weighted PM_2.5 concentrations from the CONUS GEOS-Chem results across all models, RCPs, and time periods are provided in the supplemental material. The HADGEM2-ES results show the largest impacts of wildfires, with a maximum change of nearly 4 μg m⁻³ in 2090 relative to the baseline period. Compared to the baseline period value, western wildfire PM_2.5 increases in most GCM/RCP/era combinations, more than doubling for the HADGEM2-ES/RCP8.5/2090 projection scenario. For all GCMs, the largest changes occur in the Nevada Mountain/Semi-Desert and Coastal California regions, and extend southward to Southern California along the Sierra Mountain range. Results in the supplemental material also show substantial differences between the RCP4.5 scenario (associated with lower temperatures, lower levels of wildfire area burned, lower BC and OC emissions, and lower PM_2.5 concentrations) relative to RCP8.5, across nearly the entire western U.S (figure S6). GCMs other than HADGEM2-ES show less spatially uniform and smaller impacts of wildfires on PM_2.5 concentrations, with the GISS-E2-R model showing the smallest increases from the present day.

Zoom In Zoom Out Reset image size

These PM_2.5 changes lead to impacts on mortality and morbidity across the CONUS. Results in table 1 show annual baseline wildfire PM_2.5-attributable mortality and morbidity, and the annual excess PM_2.5-attributable mortality and morbidity burden associated with future western US wildfires, averaged across the five GCMs. The estimates for mortality indicate that the excess burden in the 2090 RCP8.5 scenario is roughly 3.5 times larger than in the baseline period—however, there are 400 fewer premature mortality cases in RCP4.5 compared with RCP8.5 in 2090. For the RCP8.5 case, non-climate factors such as the projected increase in population, change in population composition by age, and projected reduction in baseline mortality rates over our simulation period together account for a large component of the change in health burden relative to baseline. Nonetheless about a 60% increase in mortality health burden in 2090 relative to baseline is attributable solely to the effect of climate change. Additional details are provided in the supplemental material (see table S9).

Our estimates of the health and economic impacts depend in part on the effect of increased wildfire activity, but also the combined non-climate elements of the method, particularly population growth and increase in the valuation of avoided risk of premature mortality over time (associated with income growth). The overall result is a large increase in total health and health-related economic impact of wildfire-related PM_2.5 mortality and morbidity, from roughly $7 billion per year in the baseline period to roughly $36 billion per year in 2090 for RCP4.5, and $43 billion per year in RCP8.5 (figure 5). The total economic burden of wildfire-attributable PM_2.5 health impacts we are able to estimate steadily increases over time. As noted in the methods above, these estimates reflect use of an income elasticity for mortality valuation of 0.4—if we were to use an elasticity of 1.0 instead, estimated costs associated with the mortality burdens would be 87% higher in 2090, suggesting our valuation of the mortality endpoints may be conservative.

Zoom In Zoom Out Reset image size

Our results show the potential for a substantial increase in air pollution-related health impacts associated with exposure to PM_2.5 from western US wildfires under plausible scenarios of future changes in climate. The increase is consistent across alternative projections of future climate and across multiple climate models, though the effect varies in magnitude and location across these futures. The increase in wildfire-related PM_2.5 mortality and morbidity grows over time and is larger for the higher GHG scenario (RCP8.5) compared to the lower emission scenario (RCP4.5).

Our results are directly comparable to other estimates of climate change damages in the US that followed the CIRA framework, as we used the same climate scenarios, GCMs, and years of analysis. We estimated that future wildfire-attributable PM_2.5-related mortality could increase by 3.7 times and 4.2 times in 2090 under RCP4.5 and RCP8.5, respectively, translating to $29 and $36 billion in annual total excess economic damages. An estimated 40% (for RCP4.5) and 60% (for RCP8.5) of the mortality increase is attributable to climate change, with the rest driven by population growth, change in population age structure, and mortality rate changes. Compared with other climate health impacts estimated in the CIRA framework, in 2090 under RCP8.5, these projected economic damages from increased wildfire PM_2.5 mortality are larger than the climate-attributable impacts of West Nile Virus and hazardous algal blooms, smaller than those for extreme heat mortality and labor productivity, and comparable to those for southwestern US fugitive dust impacts (see comparison figure and full citations in the Executive Summary of US EPA 2017 [28]). For example, Achakulwisut et al [40] projected that damages related to dust-attributable cardiovascular mortality and asthma could increase from $13 billion/year historically to $47 billion/year by the end of the century, including future changes in population—comparable to the increase from a $7 billion baseline to a future $36 billion to $43 billion estimated here using the same climate scenarios, socioeconomic inputs, population scenarios, baseline health effects, and valuation estimates).

These findings are consistent with other studies in the literature, though they generally show a smaller impact of wildfires, which can be attributed to differences in analysis scope and methodology. Fann et al (2018) found that the baseline (2012) health burden for fire activity in the US is six times higher than our estimates, but that work includes the effect of prescribed fires, which predominate in the populous southeast region, and the effect of eastern U.S wildfires [9], both of which we exclude in our baseline and climate projections. Comparison of our estimates at a more granular level shows that, once we control for these differences in scope, our estimates show good agreement with those by Fann et al [9]. In addition, similar to our work but limited only to the PM_2.5 concentration estimates, Li et al (2020) applied the GEOS-Chem model to western wildfires and found fine PM increases of 53% for RCP4.5 and a doubling under RCP8.5 by 2100.

Yue et al (2013) found higher estimates of both area burned and emissions (see figure 3)[5]. One reason is that our area burned regressions, based on bias-corrected GCM data, do not employ certain climate variables that Yue et al drew directly from GCM output without bias correction (e.g. wind speed is used by Yue et al but not in this work due to lack of data availability for spatially downscaled results from GCMs). Local scale estimates of wind speed could be important to area burned estimates (for example, Jin et al [4] find that about half of the area burned by wildfires in California coincide with Santa Ana winds[41]). However, it remains challenging to project local scale wind characteristics for mid- to late-21st century conditions from global climate models [42]. Another important reason that our results are lower than previous estimates is the spatial scale of analysis—our four times higher spatial resolution for area burned estimates (25 × 25 km versus 50 × 50 km) yields a lower overall mean annual area burned on net across the region (according to our sensitivity testing using a coarser grid). The effect of spatial scale of analysis is strongest for the California Coastal Shrub and Eastern Rocky Mountain regions.

Ford et al (2018) conducted a global scale analysis of wildfire impacts on PM_2.5 that showed higher estimates of area burned, larger effects on PM_2.5, and a larger health burden for the western US than results in our study [10]. We attribute the difference in results to four key differences in the methodology: (a) Ford et al use a single Earth system model (CESM) as opposed to our multi-model ensemble of five GCMs—CESM for the; (b) Ford et al use a coarser grid resolution (0.9° × 1.25°) similar to Yue et al; (3) we hold meteorology constant over time as an input to the GEOS-Chem air quality modeling, while Ford et al use time-varying meteorology; and (4) Ford et al consider changes in forest cover associated with forecast economic activity, which generally increases forest cover in the western US Although the combined directional effects of these methodological differences on results are unknown, in totality their effect is that our estimates are lower. The forecast changes in forest cover in Ford et al, a feature not included in our study, appears to have a large impact on their results. However, neither Ford et al nor we consider changes in vegetation type. Some wildfire studies using dynamic vegetation models [28] suggest that historically-dominant plant species in many Western areas are replaced with those having less-frequent fire return intervals. These shifts can result in lower levels of future burning than would be anticipated in an analysis that does not include these dynamic changes [7].

Regarding time-varying meteorology, in our study running GEOS-Chem with future meteorology was not possible since the model was not configured to use outputs from global climate models. Recent evidence suggests, however, that our use of present-day meteorological fields when simulating wildfire emissions effects on PM_2.5 concentrations likely leads to underestimated impacts on wildfire-related PM_2.5 due to a shorter lifetime of PM_2.5 than might be expected in the future in areas projected to experience reductions in precipitation [29, 43, 44]. The net effects of climate change on meteorology remain important for estimating PM_2.5 concentrations. Results of a parallel effort using a different air quality model, but otherwise similar methods and inputs with the exception of climate-driven wildfire emissions, show that changes in meteorology alone could increase population-weighted PM_2.5 exposure in CONUS by 0.08–0.18 μg m⁻³ by mid-century, and 0.13–0.65 μg m⁻³ by end century, leading to an additional 15 000–21 000 annual PM_2.5-attributable premature deaths by the end of the century [29]. Other literature suggests the effect of changes in future meteorology on PM_2.5 concentrations may remain uncertain or at least highly context-specific, so future research that integrates forecast meteorology with changes in wildfire emissions attributable to climate change remains a high priority [45].

There are also other reasons to suggest that our results underestimate the full air pollution-related health burden of wildfire. We focus on the annual average concentration of PM_2.5, but increasingly the mixture of air pollutants that constitute wildfire smoke has been implicated in health burden, particularly short-term effects associated with pollutants other than PM_2.5 (e.g. air toxics) [21, 23]. In addition, the overall economic burden of wildfire clearly extends beyond the health impact. The most well documented costs are those reported right after a major fire and include the direct costs of fire suppression and firefighting, as well as evacuations, equipment damage, damaged property, school and business closures, and short-term public health alerts (Association for Fire Ecology, 2015). The immediate annual and cumulative wildfire response costs alone (excluding property damage and business closures) are roughly proportional to acres burned, and for our study would represent an additional excess annual economic burden of $640 million to $730 million in 2050, and $710 million to $890 million in 2090 (see calculations in supplemental material). These costs fail to include more nuanced and long-lasting impacts that wildfires can have on things as far reaching as public water supplies and the performance of photovoltaic installations [46–48]. In total, Thomas et al (2017) find that the annualized economic burden from wildfire is between $71.1 billion to $347.8 billion (2016 US$) [49]. Table 2 provides a summary assessment of some of the key assumptions involved in this complex multi-disciplinary, multi-model study, and the potential impact of the assumption on the overall estimates. As noted in the third column, if information exists to assess the assumptions, in most cases the impact is to underestimate impacts. Improving the basis and ability to conduct quantitative uncertainty and sensitivity analysis for these assumptions in particular represent important areas for future research.

We demonstrate that climatic change over the 21st century has the potential to greatly increase the overall air pollution-related health burden of western wildfires. Our work extends and complements prior work in this area. In particular, a key contribution of our work to this literature is providing detailed estimates of the impact of wildfire PM_2.5 on premature mortality, and various forms of morbidity, in communities near the source of wildfire smoke but also extending to areas far downwind (recently evidenced in the 2020 wildfire season [50]), using more highly resolved emissions and air quality estimation tools than prior work, and for a strategically chosen wide range of climate projections (five GCMs and two RCPs).

Opinions expressed in this article are those of the authors and do not necessarily represent those of their employers or affiliated institutions, including the US EPA. The authors are grateful to Alisa White for wildfire activity and emissions modeling; Jeff Pierce and Neal Fann for review of some early results; Xu Yue for guidance regarding application of his model; Loretta Mickley and Allison Crimmins for general analytical guidance; Mike Long and Daven Henze for GEOS-Chem modeling related analytical support; and Arash Mohegh for computing and IT support. This research was funded by the US Environmental Protection Agency under contract EP-D-14-031. The authors declare that they have no competing financial interests. Some data used are listed in the references and supporting information. All other data are posted on publication at the EPA Environmental Dataset Gateway (www.edg.epa.gov).

The data that support the findings of this study are available upon reasonable request from the authors.