We employ the open-source REgion-Specific ecosystem feedback Fire (RESFire) model coupled with the Community Land Model version 4.5 and the Community Atmosphere Model version 5 (CAM5) of the Community Earth System Model (CESM) version 1 (Zou et al., 2019; Neale et al., 2013) to perform two-way coupled simulations. RESFire provides state-of-the-art capabilities to simulate the complex fire–climate–ecosystem interactions globally for fires occurring over wildland, cropland, and peatland. Although wildfires dominate in the North American region, RESFire simulates both wildfires and prescribed fires. Moreover, this integrated setup includes climatic feedback from fire-induced aerosol direct and indirect radiative effects and associated weather changes. It also includes feedback from fire-induced vegetation distribution changes and associated biophysical processes such as evapotranspiration and surface albedo. Sofiev et al. (2012) described the fire plume rise parameterization. Other features in CLM4.5 and CAM5, such as the photosynthesis scheme (Sun et al., 2012), the MAM3 aerosol module (Liu et al., 2012), and the cloud macrophysics scheme (Park et al., 2014), allow for more comprehensive assessments of the climate effects of fires through their interactions with vegetation and clouds. Fire–ecosystem interactions are modeled by simulating fire-induced vegetation mortality and regrowth (and associated land cover change) in RESFire. This approach has been introduced in Zou et al. (2019), and the simulated ecological and climatic effects of wildfires have been evaluated in two sets of sensitivity experiments in Zou et al. (2020). Although fire–climate–ecosystem interactions are considered in this study, our focus is on the fire-induced changes in PM2.5 over Canada and the US, so the two vegetation-focused sensitivity experiments reported in Zou et al. (2020) are not included in this paper. Please refer to Zou et al. (2019) and Zou et al. (2020) for more details about the simulation of fire–ecosystem interactions.

We designed two sets of simulations for the present-day and future scenarios to quantify the impacts of fire–climate–ecosystem interactions (Table 1). The spatial resolution is 0.9∘ (lat) × 1.25∘ (long) with a time step of 30 min. In each set of simulations, we conducted a default all-emissions-included control run (XALL, where x=2000 or 2050 indicates the present day or future, respectively) and a sensitivity run with no wildfire emissions to the atmosphere (XWEF, where X is the same as for the control runs). The ALL runs are designed to simulate fully interactive fire disturbances such as fire emissions with plume rise and fire-induced land cover changes in the present day (representative of the 2000s, 2000ALL) and a moderate future emission scenario (representative of the 2050s, 2050ALL) via the RCP4.5. The only difference between the ALL and WEF scenario is that wildfire emissions are absent in the WEF scenario. Specifically, in the WEF runs, the online-simulated fire emissions are not passed to the CAM5 atmosphere model so that the difference between the ALL and WEF runs can be used to isolate the atmospheric impacts of fire–climate interactions.

For the present-day experiments, we used the spun-up states from Zou et al. (2019) as initial conditions for both meteorological and chemical variables. Sea surface temperature (SST) for the present day was obtained from the Met Office Hadley Centre (HadISST). Present-day non-fire emissions from anthropogenic and other sources were based on ACCMIP (Lamarque et al., 2010) for the year 2000. We replaced the prescribed GFED2 fire emissions (van der Werf et al., 2006) in the default setting of CESM with the online-coupled fire emissions generated by the RESFire model. Zou et al. (2019) provided more details of the physics parameterizations and modeling experiment settings used in these simulations. Land use and land cover data for 2000 and 2050 from the Land-Use History A product (Hurtt et al., 2006) are used to initialize the 2000ALL and 2000WEF as well as the 2050ALL and 2050WEF simulations, respectively. Following the above setup, the future scenario 2050ALL experiment accounts for both fuel load changes associated with the projected land use and land cover change (LULCC) in the 2050s and fire weather changes driven by the SST and sea ice forcing from a coupled CESM simulation following the greenhouse gas (GHG) forcing of the RCP4.5 scenario. The global-mean GHG mixing ratios in the CAM5 atmosphere model were fixed at year 2000 levels in all the present-day experiments, and they were replaced by those of the RCP4.5 scenario with the well-mixed assumption and monthly variations. However, the future population and socioeconomic conditions were identical to those of the present day, so there was no explicit impact of human-induced mitigation/enhancement effects on wildfires in the future projection in all the future experiments. Future human impacts were considered implicitly in LULCC-induced fuel load changes in the RCP4.5 scenario.

The net projected changes by the 2050s in emissions, meteorology, and air quality during summer months (JJA: June, July, August) are estimated by comparing decadal-mean values simulated by 2000ALL with 2050ALL. Wildfire-induced enhancement in PM2.5 concentration in the present day and mid-21st century is estimated by comparing 2000ALL with 2000WEF and 2050ALL with 2050WEF, respectively. Further, the projected increase in wildfire-induced PM2.5 in the future is calculated by comparing the simulated wildfire effect of the 2050s (2050ALL-2050WEF) with that of the 2000s (2000ALL-2000WEF). With large spatiotemporal variability, the projected changes in transported fire-emissions from the western US and Canada to the eastern US by the 2050s and the corresponding impacts are summarized using probability distribution functions. The latter provide information for not only the mean but also variability and extreme values to quantify the simulated changes for the three subregions.

Zou et al. (2019) performed comprehensive evaluation of the RESFire-simulated wildfire-burned area distribution, associated carbon emissions, and terrestrial carbon balance to demonstrate reasonable model skill. Zou et al. (2020) compare global fire simulations by RESFire–CESM with modeling results reported in the literature to show better agreement with the GFED4.1s benchmark data and predict more prominent changes in the future than those predicted by Kloster et al. (2010, 2012). These differences might come from differences in the climate sensitivities of the fire models and scenarios and other input data used to make future projections.

Here, we evaluate the simulated surface PM2.5 against satellite estimates (Fig. 1) over North America. The PM2.5 concentration is calculated as the sum of sulfate, nitrate, fine sea salt (first two size bins), fine dust (first size bin), black carbon (BC), and organic aerosol (OC) at the surface level of the model. OC is the sum of primary organic matter (POM) and secondary organic aerosol (SOA), and SOA is the sum of secondary species formed from toluene, monoterpenes, isoprene, benzene, and xylene. Figure 1 compares the observed and simulated mean annual PM2.5 averaged over 2001–2010. The 10-year-average satellite AOD-derived annual-mean surface PM2.5 concentrations (Van Donkelaar et al., 2018) are regridded to the model grid (Fig. 1a) and then compared with the RESFire simulations in the 2000ALL present-day run (Fig. 1b). The spatial distribution of annual surface PM2.5 is reasonably well simulated but also has some biases. To quantify the biases, we also estimated the correlation coefficient as well as normalized mean biases (NMBs) of the simulated values compared against the satellite-retrieved values over two subregions. Quantitatively, the NMB values over the western US (WUS) and eastern US (EUS) are 18 % and 7 %, respectively (Fig. 1c–d). In addition, the spatial variability in the 2001–2010 averaged annual AOD distribution (Supplement Fig. S1) is also well represented in our simulation, although the model underestimates high AOD values. Similar spatial variability and biases in AOD and PM2.5 were also found when a comparison was performed for only summer months (JJA). The simulated PM2.5 has also been evaluated against the ground-based Interagency Monitoring of Protected Visual Environments (IMPROVE) data, showing similar spatial pattern and biases (10 %–25 %) (Supplement Fig. S2). The biases are smaller over the eastern US and southwestern US region. The simulated PM2.5 values over California match quite well with the observed annual-mean values. However, the biases over the northwestern US region are ∼ 30 %–40 %, a portion of which could be attributed to possible biases in the model's meteorology in the northwestern US region. Nonetheless, both satellite and in situ evaluation indicates that our simulation biases are largely within the uncertainty range among the various satellite- and ground-based datasets, which have normalized mean biases ranging from -3.3 %–33.3 % when benchmarked against the ground-based IMPROVE data over the contiguous US (Diao et al., 2019; Val Martin et al., 2015).

Discrepancies between the simulated and observed PM2.5 values may be attributed to several potential reasons. First, the satellite-derived data have a non-zero lower bound of PM2.5 concentrations, so the ambient background concentrations for relatively cleaner regions such as the western US may be overestimated (Fig. 1c), also the sampling frequencies between these datasets are different. Second, year-2000-based constant non-fire emissions were used in the RESFire simulation, which may result in overestimation of the PM2.5 concentrations from non-fire sources during 2001–2010 when anthropogenic emissions and PM2.5 concentrations continue to decrease (US EPA, 2018). This overestimation is prominent in regions dominated by non-fire sources such as the southeastern US. Third, large uncertainties in fuel consumption and emission factors preclude an accurate estimation of the primary fire emissions in the model, especially for the eastern US, where large fractions of low-intensity prescribed fires consume only under-canopy fuels such as litter and duff layers. The fire model may fail to capture the subtle distinctions between low-intensity prescribed fires and forest fires, so more fuels are consumed and result in higher emissions. Lastly, comparison of a coarsely resolved simulation against in situ observations also contributes to uncertainty. Differences in the degree to which fire–climate interactions and other physical processes and feedbacks are represented by the models can explain the slight differences in estimating the present-day mean wildfire-induced change in PM2.5 over local and downwind regions between our simulations and previous studies. Nonetheless, reasonable simulation of the spatial distribution of wildfire-burned area, AOD, and near-surface particulate concentration (mean bias of ∼ 10 %–20 %) instills confidence about the fidelity of our model setup in particulate pollution simulation, which is the focus of this study.

The decadal-mean annual fire-burned area simulated for the present day shows widespread wildfires over all of North America (Fig. 2a). Specifically, Canada and the forested areas of the northwestern (> 36∘ N latitude) and southeastern (< 36∘ N latitude) US are most intensely affected by wildfires in the present day. By the mid-21st century, a striking increase of a factor of 2–5 in fire-burned area is projected over Canada, Alaska, the Pacific Northwest, and portions of the western US by the 2050s (Fig. 2b). A distinct positive shift in the probability density function (PDF) of annual fire-burned area is evident in the future, with the decadal-mean difference statistically significant at the 99 % confidence level (Zou et al., 2020). A small and statistically insignificant change in interannual variability (∼ 0.4 Mha yr-1) of fire-burned areas is also simulated between the present and future. Specifically, our model predicts more than a doubling of burned area in boreal regions of Canada in the future, in line with a previous projection for Canada (Wotton et al., 2017). Future enhancement in fire-burned area is ∼ 20 %–50 % in most fire grids over the western coast of the US, which is higher than that over the eastern US (Fig. 2a and c). The increase over the western US is closer to the lower bound of that derived from statistical model ensemble projections for the western US in the mid-21st century (Yue et al., 2013). The statistics-based projections of future burned area over North America were likely too high because fire-induced land cover change, fuel load reduction, and other factors could induce a negative fire feedback, which was not considered in previous fire projection studies (Zou et al., 2020).

Annual fire-burned area in the southeastern US shows a decline in the future (Fig. 2c), as precipitation is projected to increase in that region (discussed later). Note that all future fire changes between 2050ALL and 2000ALL are primarily associated with climate warming in response to the increase in greenhouse gas (GHG) concentrations in the RCP4.5 scenario. No direct impacts of population and socioeconomic changes on wildfires are included in our simulations, although these factors contribute to changes in GHG emissions (via the Representative Concentration Pathway – RCP – scenario) that influence the climate simulated in 2000ALL and 2050ALL. As about 80 % of the projected fire changes in the future are restricted to the summer season (JJA) (Fig. 2d), we focus on analysis of the summer-mean wildfire-induced PM2.5 and its projected future changes over North America.

The simulated 10-year-averaged summer-mean wildfire-induced PM2.5 values in 2000ALL are more than 0.5 µg m-3 over a large part of North America in the present day, with noticeably larger values (> 1 µg m-3) in Canada and the northwestern, central, and southeastern US (Fig. 3a). Interestingly, the spatial distribution of wildfire-induced PM2.5 > 1 µg m-3 resembles an inverted horseshoe shape. The inverted horseshoe-shaped spatial distribution is also consistent with the wildfire smoke climatology derived from the satellite-guided operational smoke product of the Hazard Mapping System (HMS) during 2005–2015 (Brey et al., 2018; Kaulfus et al., 2017). By the mid-21st century, the spatial extent of the horseshoe shape for areas with wildfire-induced PM2.5 enhancement > 1 µg m-3 expands significantly to span most regions of North America, with the most pronounced enhancement occurring over Canada (Fig. 3b). The PDFs of the spatial distribution for the three regions can be seen in Fig. 3c–e. Specifically, wildfire-induced PM2.5 in the 2000s over Canada and the WUS and EUS during summer is ∼ 1–3, 1–3, and 0.6–1.2 µg m-3, respectively. Maximum values within the WUS region are found over the Pacific Northwest, with most areas having wildfire-induced PM2.5 values of ∼ 2–3 µg m-3. Similarly, the southern states have relatively high wildfire-induced PM2.5 concentrations of ∼ 2–4 µg m-3 within the EUS in the present-day simulation.

Compared to the 2000s, the wildfire-induced JJA-averaged PM2.5 values are almost doubled to ∼ 3–6 µg m-3 over Canada in the 2050s (Fig. 3b and c). Consistently, the values of wildfire-induced PM2.5 over the WUS (mainly coastal) also doubled in the 2050s compared to the 2000s, with modal values of ∼ 2–2.5 µg m-3 (Fig. 3d). Most interestingly, the enhancement in wildfire-induced summer-mean PM2.5 over the northern EUS is also significant by the 2050s (Fig. 3b). Largely, the summer-mean wildfire-induced PM2.5 concentration over the EUS increases from ∼ 0.8 to ∼ 2 µg m-3 in the mid-century to values of 1.2–3.0 µg m-3 (Fig. 3e). The summer-mean wildfire-induced PM2.5 is thus projected to double in North America by the 2050s compared to the 2000s, with a substantial coverage over the EUS. An important finding from these PDFs appears to be that there are fewer grids with < 1 µg m-3 wildfire-induced PM2.5 or, alternatively, that more regions are being influenced by PM2.5, and many areas that were already seeing wildfire impacts are seeing enhanced impacts. Such enhancement is found not only at the surface but also in an elevated atmospheric layer over the EUS between 900 and 700 hPa. This is nonintuitive given the fact that the increase in fire-burned area by mid-century over the EUS is not substantial.

As anthropogenic- and wildfire-induced PM2.5 concentrations may change differently with time across North America, next, we investigate the relative contribution of wildfire-induced PM2.5 to the total PM2.5 in the future. Prominent enhancement of the wildfire contribution is apparent in the entire domain by the 2050s (Fig. 4a–b). Largely, during the 2000s, the simulated fractional contribution of wildfires to PM2.5 is ∼ 15 %–50 % in Canada (Fig. 4a). Specifically, a bi-modal distribution is simulated over Canada with modal values around 18 % and 30 % (Fig. 4c). Over the WUS, the present-day simulated percentage contributions of wildfire-induced values are 5 %–25 % (Fig. 4a), with modal values between 10 %–20 % (Fig. 4d). Note that many areas located in the Pacific Northwest have higher values of ∼ 30 %–40 % (Fig. 4a). At the same time, the fractional contribution by wildfire-induced PM2.5 is ∼ 5 %–10 % in most areas of the EUS in the present day (Fig. 4f). Nevertheless, some areas in the central US also have higher values of ∼ 10 %–25 % (Fig. 4a).

The wildfire contributions in the 2050s show a clear shift towards higher values in all subregions compared to the 2000s (Fig. 4b). Over Canada, the values shifted from 15 %–30 % in the 2000s to ∼ 30 %–60 % in the 2050s; a nearly 2-fold increase in the fractional contribution of wildfire emissions to the total PM2.5 concentration is simulated (Fig. 4b and corresponding PDF in Fig. 4c). Similarly, the contribution values increased to ∼ 10 %–35 % in the 2050s, compared to 10 %–20 % in the 2000s over the WUS (Fig. 4b), thereby featuring a broadening of the bi-modal distribution of wildfire contribution (Fig. 4d). The shift in the percentage contribution is most prominent for the higher values, corresponding to some areas located in the Pacific Northwest and on the west coast of the US (Fig. 4b). Consistent with Fig. 3b, the shift in the contribution values over the EUS is also very distinct, revealing an increase in the mode values from 6 %–10 % in the 2000s to ∼ 16 %–20 % by the 2050s (Fig. 4b and e). Thus, our results underscore a large increase in not only absolute values but also the contribution of wildfire emissions over the EUS in the future.

The larger enhancement in the relative contribution of wildfire emissions to the total surface PM2.5 in the EUS in the 2050s can be explained by three mechanisms. First, due to the increase in Canadian and western US wildfires, downwind transport of wildfire smoke plumes to the EUS will be enhanced by the 2050s. This long-range transport to the atmospheric column of the EUS can happen within a few days of the fire occurrence (Supplement Fig. S3a and b). Using HMS-detected smoke plumes, recent studies identified a strong positive association between the transported smoke plumes in the atmospheric column and collocated surface PM2.5 enhancement in the EUS (Brey et al., 2018; Wu et al., 2018; Gunsch et al., 2018; Kaulfus et al., 2017; Larsen et al., 2017; Dempsey, 2013). HMS is an operational smoke detection product over North.

America developed by the National Oceanic and Atmospheric Administration (NOAA) and operated by the National Environmental Satellite, Data, and Information Service (NESDIS), available at http://satepsanone.nesdis.noaa.gov/FIRE/fire.html (last access: 20 November 2022). Specifically, these studies found that the smoke plumes transported from Canada are located at an altitude of ∼ 1–3 km over the EUS (Colarco et al., 2004; Wu et al., 2018). Due to mixing by the daytime boundary layer and deposition, the smoke plumes enhance the surface PM2.5 concentration over the EUS (Wu et al., 2018; Colarco et al., 2004; Rogers et al., 2020; Dreessen et al., 2015). Hence HMS smoky days may be a useful proxy for wildfire-induced surface PM2.5 over North America. In agreement, Brey et al. (2018) showed that the HMS-based smoke plumes observed over the EUS are significantly aged, suggestive of their long-range-transport origin. Consistent with the observed temporal change in HMS pattern, Xue et al. (2021) estimated using the mid-visible Multi-Angle Implementation of Atmospheric Correction (MAIAC) satellite-derived aerosol optical depth (AOD) that Canadian and western US fires have caused an increase in the daily PM2.5 over Montana, North Dakota, South Dakota, and Minnesota by 18.3, 12.8, 10.4, and 10.1 µg m-3, respectively, between August 2011 (a low-fire month) and August 2018 (a high-fire month). In summary, the visually apparent satellite-based signatures of wildfire smoke across Canada and the EUS provide necessary, though not sufficient, support for the influence of Canadian smoke plumes on EUS air quality. Although, the change in burned area over the northeastern EUS is negligible compared to the western US and Canadian regions, there are some enhancements seen over the east coast of the US, which can also contribute to enhanced fire emissions.

In the future, the wildfire-induced change in speed of the westerly jet flows over Canada wildfire regions is increased (Fig. 5a–b). It indicates that the westerly-induced transported wildfire emissions from Canada boreal forests to the eastern half of North America and the EUS will be enhanced in the future compared to that in the present era. On the one hand, the wildfire-induced changes in wind speed over the EUS are reduced in the future, which implies that the local emissions over the EUS are less dispersed. Simultaneously, this will also cause the transporting smoke plumes to slow down and be subjected to relatively more boundary layer mixing over the EUS and dry deposition/settling enhancements, thereby contributing to the enhanced PM2.5 values at the surface. The westerly winds over the western US below 40∘ N are also strengthened in the future (Fig. 5a–b) compared to the present day, which indicates more advection flux wildfire emissions to the EUS. Thus, the net effect is more removal of wildfire-emitted PM2.5 from the WUS and more influx of wildfire-emitted PM2.5 in the EUS.

Along with these dynamical changes, other climatic feedbacks simulated can also contribute to enhancement of EUS pollution. Specifically, the enhancement of wildfire-induced smoke aerosols increases solar absorption and scattering in the future (Fig. 6a). This reduces the incoming solar radiation reaching the surface (Fig. 6b) and induces surface cooling. With atmospheric warming and surface cooling, lower-tropospheric stability is enhanced by wildfire aerosols in the future (Fig. 6c). The smoke plumes which reach the eastern US are at an elevated altitude due to the self-lofting property of absorbing aerosols as they travel downwind, but the smoke over the western US is at near-surface elevation as it is at its source region. This can explain the more significant atmospheric stability simulated over the eastern US compared to the source regions in western US and boreal forests of Canada. Relatively stronger atmospheric stability over the eastern US imposes a stronger thermal capping that traps more anthropogenic aerosols and particulate matter near the surface over the EUS (already an emission hotspot). At the same time, future increase in wildfire emissions also leads to greater reduction in monthly rainfall (Fig. 6d) over the EUS, which may additionally strengthen the positive feedback to surface PM2.5 over the EUS by reducing wet scavenging of transported wildfire smoke to the EUS. Thus, wildfire-emitted aerosols induce positive feedback on the surface PM2.5 concentration over the EUS through fire–climate interactions that vary on a regional scale. Moreover, the above-discussed dynamical changes in the future can also feedback these simulated thermodynamical and precipitation changes, exaggerating the enhancement in PM2.5 values over the EUS in the future. However, due to computational constraints, no direct quantification of the magnitude of these feedbacks (with aerosol–radiation and aerosol–cloud interactions turned off) on PM2.5 is performed and would be taken up in future studies.

Lastly, the reason for why the contribution of wildfire emissions to the total surface PM2.5 in the EUS is so substantial in the 2050s is the drastic reduction in anthropogenic contribution to the surface PM2.5 over the EUS in the future primarily due to policy-driven reduction in anthropogenic emissions under the RCP4.5 scenario. Specifically, the simulated ambient summer mean PM2.5 concentration exhibits widespread declines in the future (Supplement Fig. S4), with reduction in PM2.5 concentration over the eastern US in the range of 4–15 µg m-3, which is greatest within North America. Thus, a large reduction in anthropogenic contribution combined with increased downwind advection of Canadian smoke and the associated positive feedbacks can explain the projected dominance of wildfire emissions over the EUS in the future.

However, is the simulated future enhancement in wildfire contribution over the EUS substantial enough to affect the surface PM2.5 values over the EUS in the future? The World Health Organization (WHO) air quality guidelines for annual and daily PM2.5 concentration are 10 and 25 µg m-3, respectively. As no specific guideline for seasonal-mean PM2.5 in the summer is available, we use the annual guideline value as a reference to understand the implication of wildfire emissions for ambient PM2.5 concentration in the future. Interestingly, the mean summertime PM2.5 concentration in the wildfire-emission-free (WEF) scenario is projected to remain within 10 µg m-3 over most of North America, except for the southeastern US (∼ 15 % of the domain) (Fig. 7a). However, the ALL scenario projects an increase in the exposure concentration level such that values > 10 µg m-3 are common in Canada and the EUS in the future (Fig. 7b). Quantitatively, over Canada, the entire PDF of PM2.5 concentration shifts towards higher values by ∼ 5–6 µg m-3. Specifically, the modal value shifts from ∼ 6 µg m-3 in 2050WEF to 11–12 µg m-3 in 2050ALL (Fig. 7c), so PM2.5 concentration is projected to surpass the WHO guidelines over a large fraction of Canada in the future. Similarly, the entire PDF of PM2.5 concentration shifts towards higher values by ∼ 2–3 µg m-3 over the EUS, with the mode of the PDF increasing from ∼ 7–8 µg m-3 in 2050WEF to ∼ 10–11 µg m-3 in 2050ALL (Fig. 7e). The modal value of summer mean PM2.5 over the WUS increases from ∼ 6 µg m-3 in 2050WEF to ∼ 7–8 µg m-3 in 2050ALL (Fig. 7d), although a few grid cells show PM2.5 values greater than 10 µg m-3 (Fig. 7b).

Clearly, the climate-induced enhancement in fires and its influence via the advected wildfire smoke to the EUS can have significant implications for air quality management in the future. The PM2.5 enhancement in the future over the southern states within the EUS is large (Fig. 7a–b), which is consistent with the results of Figs. 3 and 4. However, the future change in burned area over the same region is negligible or mostly reducing (Fig. 1c–d). Thus, it can be argued that the simulated enhancement is mostly related to the dynamic perturbations and thermodynamical feedbacks due to wildfire emissions (Fig. 6). As the rate of anthropogenic emissions is also the highest regionally over the southeastern states, the impact of these wildfire-induced climatic feedbacks on local air quality is distinctly seen over the EUS.

Note that our simulated present-day estimates of wildfire-induced PM2.5 values as well as the percentage contribution of wildfire emissions are within the range of reported values in previous studies over the domain, which augment the fidelity of our future projections. Specifically, our simulated present-day estimates of wildfire-induced PM2.5 values are also within the range of reported values in previous studies over the domain. Reported values of wildfire-induced PM2.5 over the WUS during summertime vary from ∼ 1 µg m-3 (Jaffe et al., 2008) to ∼ 2 µg m-3 (Park et al., 2007) and ∼ 3 µg m-3 (Ford et al., 2018), with the highest values documented over the Pacific Northwest and west coast regions (∼ 1–4 µg m-3) (O'Dell et al., 2019). The wildfire-induced PM2.5 over the EUS during summertime varies from ∼ 1 µg m-3 (Park et al., 2007) to ∼ 2.5 µg m-3 (∼ 3 µg m-3 in the southeastern US) (Ford et al., 2018). Consistently, our simulated present-day estimates of wildfire contribution values are also within the range of reported values in previous studies. For example, Meng et al. (2019) found that wildfires can be the largest sectoral contributor (∼ 18 %–59 %) to the population-weighted PM2.5 in various subregions of Canada. Over the WUS, the present-day percentage contribution of wildfire-induced PM2.5 to the total PM2.5 is reported to be ∼ 12 % (Liu et al., 2017), ∼ 15 % (Park et al., 2007), and ∼ 30 % (Ford et al., 2018), with higher values of ∼ 40 % in the Pacific Northwest (O'Dell et al., 2019). Over the EUS our simulated values are also within the range of previously reported values of ∼ 5 % (Park et al., 2007) and ∼ 15 %–18 % (Ford et al., 2018). However, our two-way coupled simulations illustrate that future enhancement in the wildfire-associated PM2.5 over the EUS could be greater compared to the western US, which is not emphasized explicitly in any of the previous studies (although Ford et al., 2018, illustrated an increase in PM2.5 over the mid-US and central US from Canadian fires). These could be since inclusion of the wildfire-induced climatic feedbacks in our simulation is an unprecedented exercise. Please also note that our study is focused on the JJA period, and the wildfires in the western US mainly occur during August–September, so the results should be compared consciously.

Nonetheless, inherent limitations in our simulations may introduce uncertainties in the projected future changes. For example, our reported changes in PM2.5 concentrations based on relatively coarse-resolution simulations and decadal averages likely represent a low-end estimate compared to changes at regional and daily/weekly scales. Moreover, our experiments do not consider the direct human influences such as population change and socioeconomic development on wildfires, which may aggravate the increase in PM2.5 concentrations over the densely populated EUS in the future. Common sources of uncertainty in modeling burned area and fire emission and fire aerosol and smoke are also present in our model. Fire smoke, in particular, is extremely hard to measure and evaluate. Lastly, inherent uncertainties in the physics parameterizations used in the model, sensitivity of climate to GHG emissions, and the RCP scenarios should also be noted. Thus, ensemble modeling considering different emissions scenarios, population, and future time periods and the use of a finer spatial resolution may provide a more robust and better quantification of the wildfire-induced impact on policy-regulated improvements in PM2.5 over the EUS.