Impact of Fire Emissions on U.S. Air Quality from 1997 to 2016–A Modeling Study in the Satellite Era

: A regional modeling system that integrates the state-of-the-art emissions processing (SMOKE), climate (CWRF), and air quality (CMAQ) models has been combined with satellite measurements of ﬁre activities to assess the impact of ﬁre emissions on the contiguous United States (CONUS) air quality during 1997–2016. The system realistically reproduced the spatiotemporal distributions of the observed meteorology and surface air quality, with a slight overestimate of surface ozone (O 3 ) by ~4% and underestimate of surface PM 2.5 by ~10%. The system simulation showed that the ﬁre impacts on primary pollutants such as CO were generally conﬁned to the ﬁre source areas but its e ﬀ ects on secondary pollutants like O 3 spread more broadly. The ﬁre contribution to air quality varied greatly during 1997-2016 and occasionally accounted for more than 100 ppbv of monthly mean surface CO and over 20 µ g m − 3 of monthly mean PM 2.5 in the Northwest U.S. and Northern California, two regions susceptible to frequent ﬁres. Fire emissions also had implications on air quality compliance. From 1997 to 2016, ﬁre emissions increased surface 8-hour O 3 standard exceedances by 10% and 24-hour PM 2.5 exceedances by 33% over CONUS. ppbv, and 44 µ g m − 3 , respectively, in the northwestern U.S.; and 150 ppbv, 5.1 ppbv, and 48 µ g m − 3 in Northern California. The 20-year average ﬁre contributions to surface CO, O 3 , and PM 2.5 levels were 7.07 ppbv, 0.26 ppbv, and 1.68 µ g m − 3 , respectively, in Northern California, which were 3.2, 1.7, and 3.9 times of the domain-mean contributions. In the northwestern U.S., the ﬁre emissions accounted for 6.86 ppbv, 0.24 ppbv, and 1.40 µ g m − 3 of surface CO, O 3 , and PM 2.5 concentrations, respectively.


Introduction
In the past few decades, wildfires have occurred more frequently with longer durations in the western U.S. [1,2]. Wildfires not only wreak property damages and threaten human lives on their pathway, but also deteriorate local/regional air quality and pose imminent risks to both human and ecosystem health [3][4][5][6]. For example, case studies have shown that wildfires have caused large exceedances of ambient PM 2.5 (particulate matter with aerodynamic diameter equal to or less than 2.5 µm) standard in downwind areas [7][8][9][10][11] and contributed substantially to ambient ozone (O 3 ) pollution over the U.S. [12][13][14][15][16][17]. Under the Intergovernmental Panel on Climate Change's (IPCC) A1B and A2 scenarios, the researchers [18,19] project that the U.S. wildfire activities would increase toward the middle of the 21st century. In anticipation of continuing decreases in anthropogenic emissions in the U.S., Jacob and Winner [20] suggest that fire emissions would be more important to regulate future U.S. air quality. Meanwhile, aerosols emitted by fires and other sources can exert impacts on local/regional climate through complex aerosol-cloud-radiation interactions [21][22][23] that in turn affects U.S. air quality [24][25][26].
Investigation of the effects of fire emissions on U.S. air quality with air quality models (AQM) has emerged recently [27][28][29][30]. However, these studies either focused on isolated fire events [28,29]

Emissions Preparation
The National Emission Inventory 2011 (NEI2011) was adopted to create the baseline (2014) anthropogenic emissions with modifications using measurements from the Ozone Monitoring Instrument (OMI) onboard Aura satellite, the ground-based national Air Quality System (AQS), and the in-situ continuous emissions monitoring in power plants [48]. The Air Pollutant Emissions Trends Data compiled by the U.S. Environmental Protection Agency (EPA) (https://www.epa.gov/air-emissionsinventories/air-pollutant-emissions-trends-data) were then used to adjust the modified NEI2011 data into each simulation year with the factors scaled to the baseline emissions. In areas with no NEI2011 data, the Emissions Database for Global Atmospheric Research (EDGAR v3, http://edgar.jrc.ec.europa.eu/) at a 1 by 1 degree resolution developed by the Joint Research Centre of European Commission was projected to the domain and merged with the NEI2011 data.
The fire emissions were based on the Global Fire Emissions Database with small fires, version 4 (GFEDv4s), which was derived, in combination with the ancillary data of carbon pools, emissions factors, and combustion completeness, from the satellite measurements of burnt areas by the Moderate Resolution Imaging Spectroradiometer (MODIS) or the Visible Infrared Scanner (VIRS) and Along-Track Scanning Radiometer (ATSR) in the pre-MODIS era [49,50]. The major improvement of GFEDv4s over the previous versions was that it contained emissions from small fires that were derived by combining the measurements of 500 m resolution burnt area, 1 km resolution thermal anomalies for active fires, and 500 m resolution surface reflectance. The 0.25-degree resolution GFEDv4s data were projected onto the regional domain grids and split into the CB06 and AERO5 species. The GFEDv4s had a monthly resolution prior to year 2000 and daily one onward. A program was developed to merge the aforementioned anthropogenic and fire emissions into the temporalized, gridded, and speciated data ready for CMAQ.
The biogenic emissions were calculated online within CMAQ from the gridded normalized vegetation emissions under the standard condition (303 K and 1,000 µmol m −2 s −1 of photosynthetic active radiation) produced by SMOKE with the Biogenic Emissions Landuse Database version 3 (BELD3, https://www.epa.gov/air-emissions-modeling/biogenic-emissions-landuse-database-version-3-beld3). BELD3 contained 230 vegetation classes at a 1-km resolution covering most of North America, which were processed through the spatial allocator developed by Community Modeling and Analysis System (CMAS) (https://www.cmascenter.org/sa-tools/) to generate the gridded vegetation distribution map ready for SMOKE. An ancillary file was also developed to facilitate the CMAQ online calculation of sea salt emissions.

Experiment Set-Up
Two sets of 20-year (1997-2016) simulations were carried out, with one set including and the other excluding fire emissions. All other model configurations were kept identical for both experiments. The study domain covered the contiguous United States (CONUS), southern Canada, northern Mexico, and the adjacent oceans with a 30-km horizontal grid. Due to the computational constrain, a higher than 30-km resolution modeling was not possible for this study. However, the 30-km resolution simulation has been proven to realistically capture the spatial-temporal distributions of temperature and precipitation, including extreme events [32,51], which is important to an accurate air quality simulation. In addition, the 30-km grid is comparable to the 0.25-degree resolution of GFEDv4s fire emissions data. Therefore, the 30-km resolution modeling was appropriate for the scope of this study.
The CWRF control physics configuration was utilized in this investigation. This included the longand short-wave radiation scheme of Chou and Suarez [52] and Chou et al. [53], cloud microphysics scheme of Tao et al. [54], the ensemble cumulus parameterization of Qiao and Liang [35][36][37], the planetary boundary layer scheme of Holtslag and Boville [55], and the Conjunctive Surface-Subsurface Process model (CSSP) predicting detailed terrestrial hydrology and land-atmosphere interaction developed by Liang's group over years [56][57][58][59][60][61][62]. The large-scale meteorology conditions driving CWRF were derived from the European Centre for Medium-Range Weather Forecast Interim reanalysis (ERI) [63]. The chemical initial and boundary conditions were obtained from the climatological profiles built in CMAQ.    1b displays the ratios of major air pollutants from fires to those from anthropogenic emissions over 1997-2016. In general, fire emissions only accounted for a small portion (less than 5%) of the respective anthropogenic emissions of SO 2 , NOx, and NH 3 . The fire contributions to CO and black carbon (BC) became more prominent, frequently accounting for more than 10% of the annual man-made emissions. Fire was a major source of organic carbon (OC) with over 50% of anthropogenic emissions in 11 of 20 years between 1997 and 2016. In the strong fire years of 2002 and 2013, OC emissions from fires were 14% and 19% higher than those from man-made sources, respectively.

Fire Emissions During 1997-2016
It is worth noting that the steady decreases in anthropogenic emissions, especially of CO, SO 2 , and NOx, have made fire more important an emission source over the years. For example, annual fire emissions of CO, SO 2 , and NOx were similar (approximately 5% in difference) between 1998 and 2011. Meanwhile, the 2011 anthropogenic emissions of CO, SO 2 , and NOx were 44%, 66%, and 41% lower than the respective 1998 ones. This increased the ratio of fire to anthropogenic emissions from 9.9% in 1998 to 16.9% in 2011, a nearly 70% jump for CO. The ratios for SO 2 and NOx increased from 0.9% and 2.6% in 1998 to 2.6% and 4.0% in 2011, respectively. If the current decreasing trend of anthropogenic emissions continues, fire emissions are anticipated to account for more percentage of total emissions and their role in regulating air quality will become more important in the future.
Though fire occurred at varying locations in different years, some regions were more susceptible to fires as illustrated in Figure 2. The southeastern U.S. (boxed on panel (a) of Figure 2), Northern California (boxed on panel (c)), and the northwestern U.S. (boxed on panel (d)) repeatedly experienced fire hazards. Therefore, these three regions were selected for detailed analysis of fire impact on regional air quality.

Evaluation of CWRF and CMAQ
The CWRF simulated 2-meter temperature (T2) and precipitation were compared to the respective observations to evaluate the model ability in reproducing the meteorology conditions. The T2 and precipitation observations were derived from the quality-controlled records of 8,516 stations in the National Weather Service (NWS) Cooperative Observer network [64,65]. The daily precipitation data were adjusted for topographic dependence using monthly mean data from the Parameter elevation Regression on Independent Slopes Model (PRISM) by Daly et al. [66] to correct the observation bias for mountain sites [67]. The station data were then mapped onto the CWRF grid following the mass-conservation Cressman [68] objective analysis method [67]. Figure 3 illustrates the biases of T2 (top panels) and precipitation (bottom panels) over CONUS for winter (DJF), spring (MAM), summer (JJA), and fall (SON) based on the 20-year average. In winter, CWRF underestimated T2 in the northwestern U.S. and overestimated it in the Great Plains with typical biases within ±2 K. In spring, the underestimate was generally weakened while the overestimate was enhanced by 0.5 -1 K. In summer, CWRF generally overestimated T2 across CONUS except in the Gulf States, with the largest biases of 3-4 K occurring along the coastal California as well as large portions of Iowa and Nebraska. In autumn, such overestimates were reduced to below 3 K and limited in the central to northern Great Plains. Overall, the biases in CONUS were -0.06, 0.68, 1.62, and 0.92 K for winter, spring, summer, and autumn, respectively. The corresponding root mean square error (RMSE) were 1.51, 1.33, 1.96, and 1.34 K, and the spatial pattern correlations with observations were high (> 0.97) for all seasons.  CWRF also performed well in reproducing precipitation, having the spatial pattern correlations with observations larger than 0.89 for all seasons. Its biases were typically within ±0.5 mm day -1 in most regions of CONUS and for all seasons. Major exceptions included overestimates of 0.5-1 mm day -1 in the states around the Great Lakes in winter, which were expanded to cover the Midwest and northern Great Plains in spring and turned to underestimates of a same magnitude in the central U.S. in summer. Larger biases occurred in some small regions, including the western mountainous areas (where observations were scarce) in winter, spring and autumn, as well as southern Texas and Florida in summer and autumn (when tropical storms prevailed). Overall, the biases were 0.16, 0.26, -0.19, and -0.06 mm day -1 for winter, spring, summer, and autumn, respectively. The corresponding RMSEs were 0.75, 0.66, 0.52, and 0.55 mm day -1 . Therefore, CWRF realistically reproduced the T2 and precipitation distributions.
The CMAQ simulation was evaluated against the ground observations archived in AQS managed by U.S. EPA (https://www.epa.gov/outdoor-air-quality-data). The measured hourly surface air quality data at the monitoring sites were processed using the EPA Remote Sensing Information Gateway (RSIG) processor (https://www.epa.gov/hesc/remote-sensing-information-gateway) and mapped onto the CMAQ grid. The detailed description of CMAQ model evaluations can be found in He et al. [69], who investigated the long-term trends of O3 pollution observed by the AQS and modeled by the CWRF-CMAQ system, focusing on the summer season. Here only an overview CMAQ evaluation was provided. Figure 4 displays the scatter density plots of the comparison of annual-mean surface O3 and PM2.5 between modeled and measured results paired in time and space from 2000 to 2016. CMAQ slightly overestimated surface O3, with the normalized mean bias (NMB) of 4.07% and RMSE of 5.31 ppbv. It well captured the observed surface O3 spatiotemporal variations, with a correlation CWRF also performed well in reproducing precipitation, having the spatial pattern correlations with observations larger than 0.89 for all seasons. Its biases were typically within ±0.5 mm day −1 in most regions of CONUS and for all seasons. Major exceptions included overestimates of 0.5-1 mm day −1 in the states around the Great Lakes in winter, which were expanded to cover the Midwest and northern Great Plains in spring and turned to underestimates of a same magnitude in the central U.S. in summer. Larger biases occurred in some small regions, including the western mountainous areas (where observations were scarce) in winter, spring and autumn, as well as southern Texas and Florida in summer and autumn (when tropical storms prevailed). Overall, the biases were 0.16, 0.26, −0.19, and −0.06 mm day −1 for winter, spring, summer, and autumn, respectively. The corresponding RMSEs were 0.75, 0.66, 0.52, and 0.55 mm day −1 . Therefore, CWRF realistically reproduced the T2 and precipitation distributions.
The CMAQ simulation was evaluated against the ground observations archived in AQS managed by U.S. EPA (https://www.epa.gov/outdoor-air-quality-data). The measured hourly surface air quality data at the monitoring sites were processed using the EPA Remote Sensing Information Gateway (RSIG) processor (https://www.epa.gov/hesc/remote-sensing-information-gateway) and mapped onto the CMAQ grid. The detailed description of CMAQ model evaluations can be found in He et al. [69], who investigated the long-term trends of O 3 pollution observed by the AQS and modeled by the CWRF-CMAQ system, focusing on the summer season. Here only an overview CMAQ evaluation was provided. It well captured the observed surface O 3 spatiotemporal variations, with a correlation coefficient of 0.52. CMAQ slightly underestimated surface PM 2.5 with NMB of approximately -9.8%. Its RMSE was 3.83 µg m −3 and correlation coefficient with observations was around 0.41. However, zooming into the fire susceptible sub-regions as highlighted in Figure 2, the NMB became larger in magnitudes. For example, the NMBs for PM2.5 were -39.6%, -47.7%, and -58.1% for the northwestern U.S., Northern California, and the southeastern U.S., respectively, approximately 4.0, 4.9, and 5.9 times of the NMB over CONUS. This might come from the uncertainty of fire emissions. Based on a recent study by Pan et al. (2020) who compared the six commonly used biomass burning emissions inventories, GFEDv4s had the smallest fire emissions over the temperate North America. The usage of other fire emissions inventories may improve CMAQ performance over CONUS. In addition, the potential misrepresentation of plume height might mistakenly allocate fire emissions at wrong altitudes. In the future, the fire emissions inventory with plume height information will greatly help in accurate simulations of fire-related air quality.

Average Air Quality
The fire contributions were obtained by the differences between the CMAQ results with and without fire emissions. Figure 5 (top panels) displays the 20-year average contributions of fire emissions to surface air quality of CO, O 3 , and PM 2.5 . In general, the impacts on primary pollutants such as CO were typically at or close to fire locations, including Canada, the northwestern U.S., northern California, southeastern U.S., west coastal areas, and Yucatan Peninsula of Mexico. The 20-year mean fire contribution to surface CO level reached more than 20 ppbv in certain regions, with the domain average as 2.21 ppbv. On the other hand, the impacts on secondary pollutants (chemically formed in the atmosphere from precursor species) such as O 3 and PM 2.5 (which has both primary and secondary origins) spread into broader areas and far away from fire source locations. For example, fires led to more than 0.2 ppbv increase in average surface O 3 concentrations across most of the domain with the largest enhancement reaching 1 ppbv. Averaged over the entire domain, the 20-year mean surface O 3 and PM 2.5 levels were enhanced by 0.15 ppbv and 0.43 µg m −3 , respectively, due to fire activities.  Three regions susceptible to fires, as shown on Figure 2, were selected for analyzing the time evolution of fire effects on regional air quality. Figure 6   The fire emissions from the southeastern U.S. were relatively lower but generally more persistent than those from the northwestern U.S. and Northern California. As a result, the peak fire contributions to surface CO, O3, and PM2.5 concentrations were approximately 20 ppbv, 1.6 ppbv, and 5 µg m -3 , respectively, while the corresponding 20-year averages were 4.25 ppbv, 0.24 ppbv, and 0.75 µg m -3 . The peak-to-average ratios for CO, O3, and PM2.5 were 4.7, 6.7, and 6.7, respectively, in the southeastern U.S., much lower than the corresponding ratios, which were at least 20, in the northwestern U.S. and Northern California.

Extreme Air Quality
Investigation of fire emission effect on extreme air quality has regulatory importance. Here the extreme air quality was defined as the surface pollutant concentration exceeding the corresponding National Ambient Air Quality Standards (NAAQS). According to U.S. EPA, two criteria pollutants, O3 and PM2.5, constantly exceed the respective NAAQS and plague public health in many counties in the U.S. Currently, the NAAQS is 70 ppbv for O3 averaged over 8 hours and 35 µg m -3 for PM2.5 averaged over 24 hours (https://www.epa.gov/criteria-air-pollutants/naaqs-table). As of 2019, there have been 197 counties designated as O3 nonattainment (exceeding NAAQS) with more than one third of total U.S. population residing in those counties. Meanwhile, there have been 16 counties designated as PM2.5 nonattainment with over 20 million people suffering from the heavy PM2.5 pollution (https://www.epa.gov/green-book). Therefore, the following analysis would focus on O3 and PM2.5. Figure 7 shows the probability distributions of surface 8-hr average O3 and 24-hr average PM2.5 over CONUS. It can be found that approximately 71% of surface 8-hr O3 fell between 35 and 50 ppbv over the years of 1997-2016. Although less than 1% of surface 8-hr O3 exceeded the NAAQS of 70 ppbv, fire emissions increased the occurrences of NAAQS exceedance by 10% over the same time period. Surface 24-hr average PM2.5 displayed two distinct peaks with approximately 10% occurring around 3 µg m -3 and 12% centering around 10 µg m -3 . Similar to surface 8-hr average O3, less than 1% The fire emissions from the southeastern U.S. were relatively lower but generally more persistent than those from the northwestern U.S. and Northern California. As a result, the peak fire contributions to surface CO, O 3 , and PM 2.5 concentrations were approximately 20 ppbv, 1.6 ppbv, and 5 µg m −3 , respectively, while the corresponding 20-year averages were 4.25 ppbv, 0.24 ppbv, and 0.75 µg m −3 . The peak-to-average ratios for CO, O 3 , and PM 2.5 were 4.7, 6.7, and 6.7, respectively, in the southeastern U.S., much lower than the corresponding ratios, which were at least 20, in the northwestern U.S. and Northern California.

Extreme Air Quality
Investigation of fire emission effect on extreme air quality has regulatory importance. Here the extreme air quality was defined as the surface pollutant concentration exceeding the corresponding National Ambient Air Quality Standards (NAAQS). According to U.S. EPA, two criteria pollutants, O 3 and PM 2.5 , constantly exceed the respective NAAQS and plague public health in many counties in the U.S. Currently, the NAAQS is 70 ppbv for O 3 averaged over 8 hours and 35 µg m −3 for PM 2.5 averaged over 24 hours (https://www.epa.gov/criteria-air-pollutants/naaqs-table). As of 2019, there have been 197 counties designated as O 3 nonattainment (exceeding NAAQS) with more than one third of total U.S. population residing in those counties. Meanwhile, there have been 16 counties designated as PM 2.5 nonattainment with over 20 million people suffering from the heavy PM 2.5 pollution (https://www.epa.gov/green-book). Therefore, the following analysis would focus on O 3 and PM 2.5 . Figure 7 shows   Figure 8, red color indicates 99% or more exceedance days were due to the addition of fire emissions). Fire also resulted in more than doubled O 3 exceedance days in South Missouri, a large portion of Arkansas, and South Georgia.  Figure 8, red color indicates 99% or more exceedance days were due to the addition of fire emissions). Fire also resulted in more than doubled O3 exceedance days in South Missouri, a large portion of Arkansas, and South Georgia. In CONUS, the surface PM2.5 exceedance days caused by additions of fire emissions mostly occurred in the source regions such as the northwestern U.S., Northern California, and scattered locations in the southeastern U.S. (panel (c) of Figure 8). The large fire contributions (over 175 days) to PM2.5 exceedance days occurred in the portions of Northern California and North Idaho over the years of 1997-2016. Percentagewise, Northern California, large portions of the northwestern states, and Florida saw that the PM2.5 exceedance days were almost exclusively induced by the addition of fire emissions (panel (d) of Figure 8). Large portions of the other southeastern U.S. experienced more than doubled PM2.5 exceedance days due to fires.
It was interesting to examine the effect of fire emissions on air quality over the metropolitan areas where a large population lives. Figure 9 illustrates the changes of surface O3 and PM2.5 exceedances due to the addition of fire emissions in the selected metropolitan areas from 1997 to 2016. It can be seen that except for San Francisco, fire generally caused higher percentage of more PM2.5 exceedances than that of O3. This was understandable since fire was the direct source of many PM2.5 components such as black carbon and organic carbon, whereas its impact on O3 was through emissions of its precursors such as CO and NOx in a complex nonlinear system. Fires led to 5.2% (Washington D.C.) ~ 29.9% (Houston) more PM2.5 exceedances while 1.5% (New York City) ~ 18.1% (San Francisco) more O3 exceedances. Moreover, it was not surprising that fire emissions affected air In CONUS, the surface PM 2.5 exceedance days caused by additions of fire emissions mostly occurred in the source regions such as the northwestern U.S., Northern California, and scattered locations in the southeastern U.S. (panel (c) of Figure 8). The large fire contributions (over 175 days) to PM 2.5 exceedance days occurred in the portions of Northern California and North Idaho over the years of 1997-2016. Percentagewise, Northern California, large portions of the northwestern states, and Florida saw that the PM 2.5 exceedance days were almost exclusively induced by the addition of fire emissions (panel (d) of Figure 8). Large portions of the other southeastern U.S. experienced more than doubled PM 2.5 exceedance days due to fires.
It was interesting to examine the effect of fire emissions on air quality over the metropolitan areas where a large population lives. Figure 9 illustrates the changes of surface O 3 and PM 2.5 exceedances due to the addition of fire emissions in the selected metropolitan areas from 1997 to 2016. It can be seen that except for San Francisco, fire generally caused higher percentage of more PM 2.5 exceedances than that of O 3 . This was understandable since fire was the direct source of many PM 2.5 components such as black carbon and organic carbon, whereas its impact on O 3 was through emissions of its precursors such as CO and NOx in a complex nonlinear system. Fires led to 5.2% (Washington D.C.)~29.9% (Houston) more PM 2.5 exceedances while 1.5% (New York City)~18.1% (San Francisco) more O 3 exceedances. Moreover, it was not surprising that fire emissions affected air quality more in areas close to fire sources, e.g., San Francisco, Los Angeles, Houston, and Atlanta. However, fire emissions can cause extra exceedances over regions not often threatened by fires, e.g., Washington D.C. and New York City. Seeing the large number of people live and work in those mega cities, fire-induced bad air quality and its associated health damage should be taken into account in the future environmental planning.

Discussion and Conclusions
Taking advantage of the satellite observations of fire activities from space over the past two decades (1997-2016), this study examined the contributions of fire emissions to regional air quality over CONUS using the state-of-the-art regional modeling system encompassing CWRF climate, CMAQ air quality, and SMOKE emission models. In comparison with the ground observations, CWRF reproduced precipitation well with the domain average biases of less than 0.3 mm day −1 across 4 seasons. It overestimated surface temperature, especially over the Great Plains and Midwest in summer, but underestimated it over the western U.S. inter-mountain regions in winter. The domain average biases for winter, spring, summer, and autumn were -0.06 K, 0.68 K, 1.62 K, and 0.92 K, respectively. Overall, CMAQ slightly overestimated surface O 3 by approximately 4% but underestimated surface PM 2.5 by around 10%, as compared with the AQS observations.
Fire emissions increased the domain-average 20-year mean surface pollutant concentration slightly. The impacts on primary pollutant such as CO were generally confined to the fire source areas but the effects on secondary pollutant like O 3 spread much broadly. The inter-annual variations in fire emissions impact on air quality were large with the maximum effect occurring during fire events. Fire occasionally accounted for more than 100 ppbv of monthly mean surface CO and over 20 µg m −3 of monthly mean PM 2.5 from 1997 to 2016 in the northwestern U.S. and Northern California, two regions susceptible to frequent fires. Fire emissions did pose air quality compliance implications. From 1997 to 2016, fire emissions increased surface O 3 exceedances (over NAAQS) by 10% and PM 2.5 exceedances by 33%. In large portions of Northern California, Idaho, Montana, Wyoming, and New Mexico, there would have been no O 3 exceedance days should no fire occur, suggested by the model simulation. Meanwhile, the PM 2.5 exceedance days in large areas of California, the northwestern states, and Florida were almost exclusively caused by the addition of fire emissions.
It should be noted that the results out of the current investigation bear uncertainties. Although the quantification of model uncertainty is beyond the scope of this study, several uncertainty sources and their potential impacts on the research results would be discussed. First, the base year (2014) anthropogenic emissions was derived from the NEI2011 instead of NEI2014. This was mainly due to the logistic issue. When the project was initiated, the NEI2014 was not available. Therefore, the NEI2011 based emission data used by NOAA NAQFC was the best choice and adopted by this study. However, the usage of the NAQFC emissions data was not expected to make a large difference in the results for two reasons. (1) The NAQFC team has updated the NEI2011 to reflect changes in major sources, such as using EPA's Continuous Emissions Monitors (CEMs) to update NO x and SO 2 emissions of electricity generation units (EGUs) from the base year (2011) to model year [48]; (2) The focus of this study was to investigate the effects of fire emissions on air quality through examining the difference between two experiments with and without fire emissions. Therefore, the uncertainty introduced by the anthropogenic emissions tended to cancel out each other from two experiments.
Second, the uncertainty in any fire emissions inventory is large. Large negative biases of surface PM 2.5 simulation have been identified in Section 3.2 when compared to the AQS observations in the fire susceptible regions. As pointed out by Pan et al. [70], the GFEDv4s dataset used in this study bears the smallest fire emissions over the temperate North America that includes CONUS. Moreover, the potential misrepresentation of plume height might mistakenly allocate fire emission at the wrong altitude. This indicates that the contribution of fire emissions to air quality may be at the low end out of this study. Therefore, an ensemble approach that considers multiple fire emissions inventories including plume height information may assist in reducing uncertainty in air quality modeling in the future.
Last but not the least, it is a known issue that CMAQ bears large uncertainty in modeling the secondary organic aerosol (SOA). Recent implementation of the volatility basis set (VBS) in CMAQ (CMAQ-VBS) [71] has significantly improved SOA prediction. However, the SOA formation efficiency is still two times too low [72]. Recent lab study [73] shows that more than half of the biomass burning emitted non-methane VOC can be converted to SOA after 6 days. Therefore, the underestimate of SOA in CMAQ likely affects the predicted PM 2.5 resulting from fire emissions, especially in the downwind regions.
Although it is typically not factored into the designation of air quality non-compliance areas, wildfire induced bad air quality cannot be ignored for its adverse health effects. Based on the projections by Spracklen et al. [18] and Hawbaker and Zhu [19], the U.S. wildfire activities would increase toward the middle of the 21 st century. This indicates that, with the continuing effort of reducing anthropogenic emissions and the increasing likelihood of more fire incidents in the future, air quality managers must take fire into account when formulating strategies on air quality compliances and human health protection. In addition, according to the National Interagency Fire Center, the prescribed fires account for between 14% and 42% of the total burnt areas over U.S. from 1998 to 2016. The inter-annual variation is large, but the prescribed fire is an important source of emissions in the U.S. Therefore, besides taking necessary measures to mitigate impacts of wildfires, the environmental policy makers should also better manage the prescribed fires to minimize their impact on air quality.