Biomass burning is an important source of global aerosols that has a great impact on air quality. Figure 2 shows how wildfire smoke affects local and downwind air quality (Koppmann et al., 2005; Seinfeld and Pandis, 2016; Schlosser et al., 2017). Wildfire emissions include primary aerosols (direct emission) and large amounts of gases that can be oxidized to form secondary aerosols (generated after emission). In the biomass burning input of our model, the major components of the primary aerosols are organic carbon, non-carbon organic matter, elemental carbon, chloride, and potassium. The other wildfire emissions like SO2, NOx (NO + NO2), NH3, and volatile organic compounds (VOCs) may form secondary aerosols such as sulfate, nitrate, ammonium, and secondary organic aerosols (SOAs) after being emitted. The temporal and spatial impacts of plume rise on different primary or secondary aerosol species may be different, as the generation of the secondary aerosols usually takes time. The difference in the dispersion of primary and secondary aerosols will contribute to further differences in photochemistry and health impacts. Therefore, it is important to discuss the impact of plume rise on each primary and secondary aerosol species.

To evaluate the impact of different plume rise schemes on aerosol dispersion and photochemistry modeling during the 2020 record-breaking wildfire season, four CMAQ simulations were conducted. In the first run (B69), we used the CMAQ default plume rise scheme based on Briggs (1969). In the second run (F07), we implemented the Freitas et al. (2007) scheme into the CMAQ model and used it to calculate the plume injection height. In the third run (S12), we used the Sofiev et al. (2012) plume rise scheme, as implemented in Li et al. (2021). In the fourth run (NoFire), we turned off all types of biomass burning emissions. The wildfire impact is represented by the difference between the simulation with fire and the NoFire run. More information on the three plume rise schemes is provided in Sect. 2.3. Besides the difference in the plume rise scheme, the setups for these three runs are the same. More details about the CMAQ setup are given in Sect. 2.2. Comparing results from these three simulations elucidates the impacts of plume injection height predictions on the distribution of each aerosol species (Sect. 3.2), AOD, and photochemistry (Sect. 3.3), as well as surface air quality and PM2.5 exceedances (Sect. 3.4).

The George Mason University (GMU) air quality modeling system was employed to simulate the 2020 summer wildfire season from 1 August to 30 September 2020 over the contiguous United States (CONUS) domain. This system uses CMAQ version 5.3.1 (United States Environmental Protection Agency, 2020a) as the chemical transport model and the Weather Research and Forecasting (WRF; Skamarock et al., 2019) model (Version 4.2 output) as the meteorology inputs for the CMAQ model. The model domain is configured with 12 km by 12 km horizontal resolution and 35 vertical layers (the same horizontal and vertical resolution as NOAA's operational National Air Quality Forecasting Capability). The initial and boundary conditions for WRF are from the Global Data Assimilation System (GDAS) 0.25∘ analysis and forecast. The main physics choices were the Grell–Freitas scheme (Grell and Freitas, 2014) for parameterized cumulus processes, the Mellor–Yamada–Janjic scheme (Janjic, 1994) for planetary boundary layer (PBL) processes, the two-moment Morrison microphysics (Morrison et al., 2009) for cloud physics processes, the rapid radiative transfer model for general circulation models (RRTMG) scheme (Iacono et al., 2008) for longwave and shortwave radiation, and the Noah scheme (Koren et al., 1999) for land surface processes. The biomass burning emissions product used in this study is the 0.1∘ daily blended Global Biomass Burning Emissions Product (GBBEPx) from the Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS; GBBEPx V3; Zhang et al., 2012, 2019). The GBBEPx fire radiative power (FRP) in the GBBEPx is averaged from observations from MODIS on Terra and Aqua MODIS and VIIRS M band on the Suomi National Polar-orbiting Partnership (SNPP) and the Joint Polar Satellite System-1 (JPSS) VIIRS. A climatological diurnal cycle from the Western Regional Air Partnership (WRAP) work was applied to the daily GBBEPx emission to derive hourly model-ready emission input. Anthropogenic emissions were prepared with the 2016v1 emissions modeling platform, using the baseline emissions taken from the National Emissions Inventory (NEI) 2016 collaboration (Eyth et al., 2021). We then shifted the base year emission to the prediction year 2020 using representative days of each month. The model-ready emission files are processed and generated by the Sparse Matrix Operator Kernel Emissions (SMOKE) model (Baek and Seppanen, 2019) V4.7. The carbon bond 6 (CB6) gas-phase chemical mechanism (Luecken et al., 2019), CMAQ aerosol module 7 (AE07) aerosol scheme (Pye et al., 2015; Xu et al., 2018), and aqueous chemistry (Fahey et al., 2017) are used in the CMAQ system. Details about the system setup are shown in Table S1 of Li et al. (2021).

The evaluation of the model performance with the Sofiev et al. (2012) plume rise scheme has been discussed by Li et al. (2021). The average correlation between observed (from AirNow) and simulated daily PM2.5 concentrations is 0.55. The averaged normalized mean error in the simulated surface PM2.5 is 3.9 % for the year 2020. The average area hit ratio for exceedances is 0.68 (Fig. 2a in Li et al., 2021). A high area hit ratio represents a good capture of the region impacted by smoke. During the peak pollution days (from 12 to 16 September), the area hit ratios were higher than 0.96, with a maximum of 1.0 on 13 September 2020. This suggests that the model could predict more than 96 % of the observed exceedances when the smoke pollution was at its peak. Also, the simulated PM2.5 vertical profiles along the West Coast and in central USA matched the vertical profiles of backscatter from Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) near the wildfire source region and downwind area (Fig. S2 in Li et al., 2021). Overall, the model can reproduce wildfire smoke dispersion, especially when the smoke is thick.

Three plume rise schemes are used in this study, namely Briggs (1969), Freitas et al. (2007), and Sofiev et al. (2012).

The simulated plume heights from three simulations, i.e., B69, F07, and S12, are compared with MISR observations for the Milepost 21 fire on 15 August 2020 and the August Complex fire on 31 August 2020 (Fig. 3) at the MISR local overpass time of around 19:00 UTC. The smoke heights from the model 3-D fields were interpolated to the MISR observation pixels using the nearest-neighbor approach. The performance of different schemes varies by fire cases and weather conditions. For the Milepost 21 fire, the plume heights simulated by B69 and S12 are similar but 25 % and 3 % lower than that by F07 for the easterly and westerly plume. In the case of the August Complex fire northerly plume, the plume heights simulated by S12 are 4 % and 8 % higher than that by B69 and F07, respectively. For the southerly plume, the plume heights simulated by B69 are 16 % and 5 % higher than that by F07 and S12. The simulated PBL heights are displayed in Fig. 3 as a reference. When the fire injection height is lower than the PBL height, then the pollution could become confined to the PBL (Sofiev et al., 2012; Thapa et al., 2022). However, when the plume height is higher than PBL, then the fire smoke can be dispersed into the free troposphere where wind speeds are stronger, leading to a wider range of pollution dispersion. In all four cases analyzed in Fig. 3, the simulated plume heights from the three schemes surpassed the model PBL. Previous studies found that, for large fires that are injected above the PBL, the plume height calculated by S12 is less sensitive to FRP than that calculated by B69 (Li et al., 2020). Some of the fire points during the August Complex fire had higher FRP than that during the Milepost 21 fire, so the estimated plume height by B69 is higher than that by S12. For the F07 scheme, the plume injection height is higher when it is wetter (Freitas et al., 2007). The water vapor mixing ratio on 15 August is higher than on 31 August, which contributes to the higher plume height during the Milepost 21 fire than during the August Complex fire. According to the box-and-whisker chart shown in the right panel of Fig. 3, the simulated plume heights are all within the range of MISR observation. Overall, the simulated plume heights with all three schemes are reasonably comparable to the MISR observations.

The vertical profiles of CMAQ-simulated PM2.5 are also compared to the CALIPSO daytime aerosol vertical profile. The daytime CALIPSO overpass occurs around 13:30 LT, closer to the peak fire behavior in the afternoon than the MISR observations. Figure 4 shows the comparison between the CALIPSO plume height results and the estimated plume heights from the three plume rise schemes for West Coast fires. The mean bias for the three schemes is -0.60 for B69, -0.67 for F07, and 0.13 for S12. In most cases, the plume heights from the three schemes are close to each other, especially for the cases with plume tops under 4 km. For strong fires with plume tops higher than 4 km, S12 seems to be more skillful than B69 and F07. It successfully simulates the high plume top observed by CALIPSO, whereas B69 and F07 tend to underestimate the plume heights.

In this section, we investigate the impact of plume injection height on different PM2.5 chemical components. Figure 5 shows the vertical profile of the 2-month average PM2.5 concentration for the three experiments. Over 2 months, B69 simulated a higher average plume height and injected more PM2.5 in the free troposphere than F07 and S12. Meanwhile, F07 simulated a lower average plume height and therefore kept more PM2.5 in the boundary layer than B69 and S12.

Figure 6 shows the distribution of simulated PM2.5 components from both direct emissions and secondary formation (the impact of other PM2.5 sources was removed by subtracting the results of NoFire run) from B69, F07, and S12 in three different regions, namely western USA (west of 120∘ W), central USA (between 120 and 100∘ W), and eastern USA (between 100 and 80∘ W). For all three schemes, organic matter (OM) dominates the chemical composition of PM2.5, with 63 %–64 % near the source region in western USA, and remains dominant in the downwind regions at ∼61 %, between 120 and 100∘ W, and 57 %, between 100 and 80∘ W. A high-OM portion highlights the predominant effect of wildfire emissions on air quality during the gigafire period (Li et al., 2021). The second most abundant component is nitrate (NO3), with 11 %–12 % near the source region and 13 %–16 % in the downwind region. A higher portion of NO3 in the downwind region than in the source region reflects the decreased contribution of primary aerosols and increases in secondary aerosols. The other component with a similar spatial gradient is ammonium (NH4), which contributes 3 % to PM2.5 near the source region and 5 %–6 % in the downwind region. Elemental carbon contributes 10 % to PM2.5 concentration near the source region and 8 %–9 % in the downwind region. Potassium (K), a fingerprint element to indicate fire contribution, accounts for 3 % of PM2.5 near the source region and 2 %–3 % in the downwind region. Sulfate (SO4) contributed 3 % near the source region and 6 %–8 % in the downwind region. In summary, PM2.5 species that are not significantly affected by secondary aerosol formation, such as elemental carbon and potassium, see a decrease in their contributions when transported downwind. For the PM2.5 species that are affected by secondary aerosol formation (e.g., nitrate, sulfate, and ammonium), the contribution increases when transported downwind. These results show that the PM2.5 composition, integrated over all vertical layers, is not sensitive to the choice of plume rise scheme.

Figure 7 shows the difference in the zonal mean (average for each latitude) concentrations of six major PM2.5 species (i.e., organic matter, nitrate, elemental carbon, ammonium, potassium, and sulfate) and total PM2.5 when using different plume rise schemes over the whole domain. Overall, most of the differences are found over the West Coast region (to the west of 115∘ W). The simulation with B69 produces a higher plume height on average, resulting in greater transport of smoke aloft, and hence higher downwind PM2.5 than that with the F07 or S12 schemes. The B69 plume rise scheme has a higher downwind impact and slightly lower near-source impact for PM2.5 species that contain secondary aerosols (e.g., organic matter, nitrate, ammonium, and sulfate) than primary aerosols (e.g., elemental carbon and potassium), due to the time required to form secondary aerosols.

Among the three simulations, the largest differences in PM2.5 are found from the surface to 8 km over the source region. Over the downwind region, the bulk of PM2.5 differences is found in the middle and upper troposphere. In addition, we noticed that the simulations with F07 and S12 produce more PM2.5 than that with B69 between 6–8 km during the analysis period. This is because, in the cases of a strong fire, the plume injection height simulated by F07 or S12 could be higher than B69 (e.g., Fig. 3e). However, the difference in PM2.5 above 6 km is very small compared to those particles below 6 km. The total PM2.5 difference caused by different plume rise schemes is about 30 % near the source and 5 % in the downwind region. The difference in surface PM2.5 has a large impact on surface pollution levels and human health. More discussion on the impact of plume height on surface air quality is presented in Sect. 3.4. Although the upper-level PM2.5 difference is expected to have a smaller impact on human health, it may affect cloud formation, photochemical reactions, and the radiative budget in the Earth system.

Wildfire smoke increases the aerosol loading in the atmosphere and consequently the AOD over both the source region and downwind regions. According to our previous study (Li et al., 2021), from 14–17 September 2020 smoke from the West Coast was transported to the northeastern part of the USA. The downwind transport of wildfire smoke is highly dependent on plume rise estimation. Figure 8a shows the 2-month-averaged AOD from VIIRS compared with model simulations (Fig. 7b–d). The CMAQ-predicted AOD was interpolated to the VIIRS pixels that passed quality control using the nearest-neighbor approach. When comparing the CMAQ AOD to VIIRS AOD (Fig. 7b–d), we applied VIIRS AOD saturation level (AOD ≤5) to CMAQ AOD results (any CMAQ AOD values higher than 5 were changed to 5). In the West Coast high-peak region, all three runs capture the observed AOD high peak near the San Francisco region, but the simulated AOD peak is lower than VIIRS observed. The average AOD from VIIRS observation is higher than 2. However, among the three CMAQ runs, only F07 simulated an average AOD higher than 2. In the downwind region, all three CMAQ runs reproduce the general downwind transport pattern, but the simulated smoke-affected region (AOD >0.5) is smaller than the observations.

Figure 7e–h show the AOD differences and the difference ratio (percentage of the difference relative to B69) between the different plume rise scheme simulations. When comparing different model simulations, the AOD saturation level is removed. Near the source region, F07 and S12 simulate more AOD than B69, a pattern that is the opposite of that for plume rise estimation (lower plume height than that with B69). In the downwind region, B69 simulates more AOD than F07 and S12. The difference is approximately 20 %–30 % over the source region and 5 %–10 % over the downwind region. One possible reason that B69 predicts lower AOD near the source region and higher AOD in the downwind region compared to F07 and S12 is that a higher plume height will inject more aerosol into the free troposphere where the wind speed is stronger, accelerating the dispersion of the fire pollution. Therefore, the higher plume height will lead to lower AOD near the source region but higher AOD in the downwind region. The result is consistent with previous studies (Mallia et al., 2018; Vernon et al., 2018; Li et al., 2020).

The difference in the dispersion of fire pollution caused by the various estimated plume injection heights leads to further differences in the chemistry and photolysis reactions. Previous studies found that the thicker smoke, indicated by higher AOD, may absorb and/or scatter a larger fraction of sunlight, hence affecting photolysis reactions (Dickerson et al., 1997; Castro et al., 2001; Kumar et al., 2014; Baylon et al., 2018). Here, we simply examine how the plume rise differences affects photochemistry by comparing the photolysis rate of NO2 (NO2+hv→ NO + O) from the three runs, which is a key reaction that leads to the formation of tropospheric ozone. The differences in the NO2 photolysis rate (NO2_IUPAC10; in min-1) are shown in Fig. 9. Figure 9a and d show the photolysis rate difference and difference ratio between B69 and the NoFire experiments. The photolysis rate results in B69 were lower than the NoFire simulation, which proves that fire smoke led to the reduction in the photolysis rate, consistent with the findings of previous studies. The photolysis rate reduction caused by the fire smoke was found in the whole domain, both in the near-source region and the downwind region. Near the fire source, the photolysis rate reduction was more than 50 %. Figure 9b, c, e, and f show the photolysis rate difference and difference ratio between the three experiments with different plume rise schemes. Near the source region, where F07 and S12 simulate a higher AOD than B69 (Fig. 8), the NO2_IUPAC10 is reduced. Meanwhile, in the downwind region, where F07 and S12 simulate a lower AOD, the photolysis rate is higher than B69. Therefore, the difference in the plume injection height would affect the fire-induced photolysis rate reduction.

Surface or ambient PM2.5 is the common measure used to link exposure to wildfire smoke to health endpoints such as asthma and chronic obstructive pulmonary disease (Reid et al., 2016). To protect human health and the environment, the National Ambient Air Quality Standards (NAAQS) have been established for PM2.5 and other criteria air pollutants (NO2, O3, SO2, CO, PM10, and lead). The daily PM2.5 NAAQS level is 35 µgm-3 for the 24 h mean PM2.5 concentration (United States Environmental Protection Agency, 2020b). The simulated surface PM2.5 differences caused by different plume rise schemes are shown in Fig. 10. The F07 and S12 simulations, which have averaged lower initial plume heights, yield higher surface PM2.5 concentrations than the B69 simulation over the West Coast, whereas the opposite patterns are found in the central and the eastern USA. The surface PM2.5 difference caused by different plume rise schemes reaches 70 % over the West Coast, which is much higher than the AOD differences. In the downwind regions, the surface PM2.5 difference caused by different plume rise schemes is less than 15 %, meaning that the effects of the plume rise estimation on surface PM2.5 occur mainly near the source region.

Next, we examine how the plume rise estimation affects the prediction of PM2.5 exceedances. Figure 11a shows the daily mean surface PM2.5 difference between the F07 and B69 runs for 20 August 2020 (the first fire peak during the study period). The simulated PM2.5 exceedance regions (PM2.5>35 µgm-3, as defined by NAAQS, and at same level as U.S. EPA-defined unhealthy regions for sensitive groups), unhealthy regions (PM2.5>55 µgm-3, as defined by U.S. EPA), and very unhealthy regions (PM2.5>150 µgm-3, as defined by U.S. EPA) as seen by different plume rise schemes overlaid by the AirNow-observed exceedance for the same day are shown in Fig. 10b, c, and d. According to Fig. 11b and c, on 20 August 2020, B69 and S12 simulated more PM2.5 exceedance and a larger unhealthy region in the downwind regions (Wyoming (WY) and Idaho (ID); magenta and blue region), whereas F07 and S12 simulated more exceedance and a larger unhealthy region in the southeastern USA (yellow and orange region), where prescribed fires were the major biomass burning sources. In WY and ID, where F07 did not simulate the PM2.5 exceedance, whereas B69 and S12 did, the difference between F07 and B69 reached 15 µgm-3 (Fig. 11a). Furthermore, B69 and S12 simulate some very unhealthy regions in Nevada, whereas F07 simulates more very unhealthy regions in central and southern California. Although these schemes agree on the PM2.5 exceedance forecast in the majority of the region, the disagreements in the downwind areas (i.e., ID and WY for this case) may affect key decision-making on early warnings of extreme air pollution episodes at local levels during large wildfire events.

The total number of predicted exceedance days from the B69 simulation and the differences between B69, F07, and S12 are shown in Fig. 12. All the states on the West Coast and in the mountain region experienced at least 1 d of PM2.5 exceedance (Fig. 12a). Most of the region in northern California experienced more than 20 exceedance days, with a maximum of more than 35 d. F07 and S12 simulate more exceedance days on the West Coast near the source region and in the southeast. The difference in the exceedance days could be as large as 20 d in northern California. B69 simulates more exceedance days in downwind regions such as Nevada, Idaho, Montana, and Wyoming. The difference could reach 4 d in the downwind regions.