The annual NH3 emissions of the YRD region for 2014 were estimated with a bottom-up method based on the constant emission factors and then allocated to the monthly level based on the previously investigated temporal profile of emissions. The inventory contained eight source categories, i.e., fertilizer application, livestock–poultry breeding, fuel combustion, biomass burning, transportation, sewage–waste treatment, industrial processes, and human metabolization (see Table 1 for the details). Note that the emissions from pets were not included in the current work, due to lack of detailed information. Given their relatively small fraction in total emissions, e.g., less than 2 % in the United Kingdom (Sutton et al., 1995, 2000), we believe that the uncertainty was limited. The annual emissions were calculated by prefectural city with Eq. (1): 1Ei=∑jALi,j×EFj×10-3, where E is the emissions (metric tons, t); i and j indicate the prefectural city and source type, respectively; AL is the activity level, which indicates the amount of livestock, the amount of used fertilizer, the fuel burned, or the industrial production, depending on the source type; and EF is the annual emission factor (kg-NH3 per unit of AL).

The activity data were mainly taken or estimated from official statistics at the prefectural city level (if available) or provincial level. For livestock–poultry breeding, the year-end stock and slaughter numbers were used for animals with a breeding cycle respectively more and less than 1 year. If the city-level stock was unavailable, the output of livestock products by prefectural city was applied as the scaling factor to calculate the number from the provincial data. Table S1 in the Supplement summarizes the annual numbers of livestock and poultry by prefectural city in the YRD. The amount of fertilizer used by prefectural city and type was calculated as the product of sown area of cropland and fertilizer rate per unit area of cropland. The sown area by crop type was taken from city-level statistics, and the application rate by fertilizer type was obtained at the provincial level from a national investigation by NDRC (2015). The detailed results of fertilizer activity data are summarized in Table S2. As can also be seen in the table, the aggregated amount of fertilizer used by province was close to the provincial-level statistics, and the deviation relevant to the official statistics was 2.3 % for the whole YRD. The methods and data sources for activity levels of other source categories were provided in our previous studies (Zhou et al., 2017; Zhao et al., 2017; Yang and Zhao, 2019).

The annual NH3 emission factors were obtained based on a thorough literature review and summarized by source category in Table S3. The results from domestic field measurements were preferentially selected. For sources without suitable domestic measurements, the emission factors were also obtained from previous inventories that shared a similar study period as this work. The values from the US and Europe, e.g., AP-42 database (USEPA, 2002) and the EMEP/EEA guidebook (EEA, 2013a, b), were adopted when the above information was lacking.

The monthly distribution of emissions by source was taken from domestic investigations in the YRD (Li, 2012; Zhao et al., 2015; Zhou et al., 2017). For the purpose of air quality modeling, the emissions by sector were allocated into a grid system with a horizontal resolution at 9 km × 9 km based on selected proxies. Those proxies included the distribution of land use (for fertilization), density of total population (for human metabolization and sewage–waste treatment) and rural population (for livestock–poultry breeding and residential solid fuel burning), gross domestic product (for industrial fuel combustion and processes), road network (for transportation), and satellite-derived fire points from the Moderate Resolution Imaging Spectroradiometer (MODIS, for open biomass burning; Davies et al., 2009).

The livestock–poultry breeding emissions were recalculated integrating the detailed regional information of soil, meteorology, and agricultural processes. The same annual activity data as E1 (e.g., livestock–poultry numbers in Table S1 and fertilizer use in Table S2) were applied.

The Models-3/Community Multiscale Air Quality (CMAQ) model system version 4.7.1 was applied to evaluate the NH3 emission inventories for the YRD. CMAQ is a three-dimensional Eulerian model designed for understanding the complex interactions of atmospheric chemistry and physics (UNC, 2010). The model has been widely applied and tested in China (Qin et al., 2015; Zhou et al., 2017; Zheng et al., 2019). As shown in Fig. 1, two nested domains were applied with the spatial resolutions of 27 and 9 km, on a Lambert conformal conic projection centered at (34∘ N, 110∘ E). The mother domain (D1, 177×127 cells) covered most parts of China, and the second domain (D2, 118×121 cells) covered the whole YRD region. The two inventories of YRD NH3 emissions developed in this work were applied in D2. Emissions from other pollutants of anthropogenic origin in D1 and D2 outside Jiangsu were obtained from the Multi-resolution Emission Inventory for China (MEIC, http://www.meicmodel.org/, last access: 4 April 2020) with an original spatial resolution of 0.25∘×0.25∘. Population density was applied to relocate MEIC to each modeling domain. A high-resolution inventory that incorporates more information of local emission sources was applied for Jiangsu (JS; Zhou et al., 2017). Both MEIC and JS inventories are for 2012. The emissions for 2014 were obtained using a simple scaling method based mainly on changes in activity levels (e.g., energy consumption and industrial production) between the 3 years. The biogenic emission inventory was from the Model of Emissions of Gases and Aerosols from Nature 2.1 (MEGAN2.1; Guenther et al., 2012; Sindelarova et al., 2014), and the emission inventories of Cl, HCl, and lightning NOx were from the Global Emissions Initiative (GEIA; Price et al., 1997). Meteorological fields were provided by the Weather Research and Forecasting Model (WRF) version 3.4, a state-of-the-art atmospheric modeling system designed for both meteorological research and numerical weather prediction (Skamarock et al., 2008), and the gas-phase Carbon Bond (CB05) mechanism and AERO5 aerosol module were adopted. Other details on model configuration and parameters were given in Zhou et al. (2017). The simulations were conducted for January, April, July, and October to represent the four typical seasons in 2014. A 5 d spin-up period of each month was used to minimize the influences of initial conditions in the simulations.

Using the observation data of the US National Climate Data Center (NCDC) at 43 stations in the YRD (see Fig. 1 for the locations of the stations), the WRF modeling performance was evaluated with statistical indicators including averages of simulations and observations, bias, normalized mean bias (NMB), normalized mean error (NME), root-mean-squared error (RMSE), and index of agreement (IOA). As can be found in Table S7, discrepancies between simulation and observation met the criteria by Emery et al. (2001) for most cases, implying the reliability of meteorological simulation. However, bigger errors were found for the simulation of wind direction.

There were very limited continuous ground measurement data available for ambient NH3 and NH4+ aerosol in the YRD region in 2014, particularly at rural–remote sites that are more representative for the regional atmospheric environment. We conducted online hourly measurements using MARGA (Monitor for AeRosols and Gases in ambient Air, ADI2080) at an urban site in the western downtown of Nanjing (32.03∘ N, 118.44∘ E) from August 2014. MARGA is a state-of-the-art instrument which monitors near-real-time water-soluble ions in aerosols and their gaseous precursors (Lanciki, 2018), and it was able to capture rapid compositional changes in PM2.5 (Chen et al., 2017). The site was on the roof of the building of the Jiangsu Provincial Academy of Environmental Science (30 m above the ground) surrounded by residential and commercial buildings and heavy traffic (JSPAES; Li et al., 2015; Chen et al., 2019). The data of October 2014 were applied in this work to evaluate the NH3 inventories through air quality simulation. In addition, the hourly data of online measurement with MARGA were available at a suburban site in Pudong, Shanghai (SHPD), for April, July, and October 2014 (unpublished data from Shanghai Environmental Monitoring Center).

Regarding satellite observation, the daily NH3 vertical column densities (VCDs) measured through the Infrared Atmospheric Sounding Interferometer (IASI) were downloaded from the ESPRI data center (http://cds-espri.ipsl.upmc.fr/etherTypo/index.php?id=1700&L=1, last access: 4 April 2020). We used the data in the domain (26.1–35.4∘ N, 114.2–124.1∘ E) with a 9:30 local time Equator crossing time to evaluate the NH3 emissions. Only pixels with radiative cloud fraction < 25 %, relative error < 100 %, and absolute error < 5×1015 molec. cm-2 were used following the criteria of previous studies (Van Damme et al., 2014, 2015). The monthly average VCDs for January, April, July, and October 2014 were calculated and allocated into a grid system of 0.5∘ (longitude) ×0.25∘ (latitude) using the Kriging interpolation method, as shown in Fig. 3.

Table 2 summarizes the NH3 emissions estimated with E1 and E2 by source category and province for the YRD region in 2014. Agricultural activities (livestock farming and fertilizer) were identified as the most important sources of NH3, with the fraction of total emissions ranging from 74 % to 84 % in the two methods. Applying the constant emission factors, E1 derived a total NH3 emission estimate 60 % larger than that by E2 that characterized the agricultural processes. In particular, emissions from agricultural activities in E1 were calculated as twice those in E2. At the national scale, similarly, Dong et al. (2010) applied the constant emission factors and estimated the total NH3 emissions to be 16.1 Tg for China, 64 % larger than 9.8 Tg by Huang et al. (2012) with the agricultural processes characterized. The clearly larger estimation by constant emission factors was due partly to the fact that most domestic measurements on the emission factors of NH3 from fertilizer application were conducted in hot seasons (late spring and summer), when the basal dressing of single-season rice and maize and top dressing of wheat were usually conducted (Cai et al., 2002; Huo et al., 2015; Su et al., 2006). However, the crop rotation varied a lot in China, and part of the nitrogen fertilizer was actually not applied in hot seasons. Emission estimation based on those emission factors may thus overestimate the NH3 emission intensity (Huo et al., 2015; Wang et al., 2011; Zhang et al., 2010). Among the provinces, the fraction from Jiangsu of YRD emissions ranged from 45 % to 47 % in the two methods, followed by Anhui with around 37 %. Agricultural activities were relatively intensive in the two provinces: Jiangsu and Anhui contributed 46 % and 33 % of the economic output of agriculture and livestock–poultry farming in the YRD, and the collective fraction of fertilizer use by the two provinces reached 84 %. In contrast, agricultural activities were limited in Shanghai and Zhejiang, with smaller emissions estimated in both inventories.

The monthly distribution of NH3 emissions in the two inventories is illustrated in Fig. 4. Both inventories indicated relatively large emissions in summer (from June to August), and elevated emissions were also found in March and September in E2. The difference comes mainly from the effect of farming season on fertilization process. For example, the top dressing fertilization for winter wheat was conducted mostly during the seedling establishment and elongation stage in the following spring, resulting in enhanced use of nitrogen fertilizer in March. Moreover, September was the month with the highest temperature following summer in the YRD in 2014, and the elevated NH3 volatilization led to large emissions in E2. Compared to the fertilizer use, less variation in monthly emissions was found for livestock–poultry breeding, as very limited change in livestock amount was detected in both inventories.

Illustrated in Fig. 5 are the spatial distributions of emissions from fertilizer use, livestock–poultry breeding, and all categories in the two inventories. Both inventories indicated large emission intensities in northern Jiangsu (Xuzhou and Yancheng) and northern Anhui (Fuyang, Bozhou, and Suzhou) with abundant agricultural production. Xuzhou and Yancheng collectively contributed 36 %, 31 %, and 41 % of the provincial fertilizer use, agricultural economic product, and livestock–poultry farming product in Jiangsu, respectively. Similarly, Fuyang, Bozhou, and Suzhou collectively contributed 36 %, 36 %, and 35 % of the provincial sown area, agricultural economic product, and livestock–poultry farming product in Anhui, respectively.

The differences in spatial pattern between the two inventories were further investigated for total and fertilizer use emissions by month, through the indicator RD calculated with Eq. (4). As shown in Fig. 6, larger RD was found in northern Jiangsu, northern Anhui, and eastern Zhejiang, while smaller RD was found in western Zhejiang. The emissions in E1 were commonly larger than those in E2 across the YRD region for January and April. In contrast, larger emissions in E2 were found in northern Jiangsu (e.g., Xuzhou and Yancheng) and northern Anhui for July and October. The discrepancy resulted from the combined effect of varied activity data and emission factors as described in Sect. 2.2: top dressing fertilization and high temperature led to enhanced volatilization rate and thereby emissions of NH3 in E2, and the abundant fertilizer use in the cropland in the northern YRD was the main reason for the high emissions in October.

Figure 7 compares the NH3 emissions by province and source category in this work and other available downscaled national (MEIC) or provincial inventories in the YRD region. Results from other studies commonly ranged between E1 and E2 for agriculture, the most important NH3 source. With constant emission factors applied, the MEIC estimates were similar to those in E1. Most current provincial inventories made some corrections for emissions from fertilizer use or livestock–poultry breeding, but the local geographical and meteorological information was seldom applied in the emission estimation. For example, Liu and Yao (2016) calculated the emissions from livestock–poultry breeding for Jiangsu based on TAN but did not consider the impacts of varied monthly temperatures on the emissions. Zheng et al. (2016) calculated the agricultural NH3 emissions for Anhui based on a national guideline of NH3 emission inventory development (MEP, 2014) and ignored the impact of soil condition (e.g., pH) on NH3 volatilization from fertilizer use.

Figure 8 illustrates the observed and simulated hourly concentrations for gaseous NH3 and inorganic aerosol species (NH4+, SO42-, and NO3-) in PM2.5 for April, July, and October at SHPD and October at JSPAES. The normalized mean biases (NMBs) and normalized mean errors (NMEs) between observed and simulated concentrations and the monthly average concentrations from observation and simulation are summarized in Table 3. The simulated monthly average concentrations were close to the observed ones at both sites. The biggest discrepancy was found at SHPD for April, where the monthly average NH3 was simulated to be 56 % larger than observations with E1, and the smallest was at JSPAES for October, where the simulation was 1.7 % smaller than observations with E1. The simulated temporal variation, however, was much larger than the observed variation, leading to relatively large NME, particularly at SHPD for April. A clear difference was found for the simulation under two NH3 inventories. In general, the average of simulated NH3 concentrations at the two sites for available months was 27 % smaller in E2 than that in E1 (note the total NH3 emissions in E2 were 38 % smaller than those in E1 for the whole YRD region). At the SHPD site, application of E1 in CMAQ overestimated the NH3 concentration, indicated by the positive NMB values and the larger simulated concentrations than observed concentrations. Such overestimation was corrected when E2 was applied, and the NMEs with E2 were substantially reduced as well, as shown in Table 3. The better modeling performance implies improved estimation and spatiotemporal distribution of emissions. At JSPAES, the air quality modeling with both inventories underestimated the NH3 concentrations, and the simulated monthly average concentration with E1 was much closer to observation than that with E2. The close NMEs between the two inventories indicated very limited improvement at the site, in contrast to SHPD. Located in an urban area, JSPAES might be largely affected by the local sources like transportation and residential activities. NH3 emissions of such source categories, however, were not improved in E2.

To reduce the impact of the highly uncertain hourly meteorology simulation and emission data on air quality modeling, the daily NH3 concentrations derived from simulation and observation were further compared for October at JSPAES. As illustrated in Fig. S3, better agreement between observation and simulation was achieved for the daily concentrations than the hourly concentrations, and the NMEs for E1 and E2 were reduced from 56.9 % and 53.7 % to 37.0 % and 32.5 %, respectively. Besides the emission data, uncertainty in meteorology simulation also contributed to the discrepancy between simulation and observation. For example, both inventories overestimated the concentration on 7 October but underestimated that on 21–22 October. In contrast to the southeasterly wind observed at the ground meteorology station in Nanjing, the simulated wind direction on 7 October was from the north, enhancing the NH3 transport from Yancheng and Xuzhou in northern Jiangsu with intensive agricultural activities. On 21–22 October, the underestimation of NH3 concentration resulted largely from the overestimation in wind speed by WRF.

Compared to NH3, the modeling performance for inorganic aerosols (NH4+, SO42-, and NO3-) is better for most cases, indicated by the smaller NMEs and larger correlation coefficients (r) in Table 3. Some exceptions exist at SHPD for NH4+ and SO42 in October and NO3- in January. Application of E2 reduced the NMEs and improved the simulation of NH4+ and SO42 moderately, but there were no significant changes between the modeling results with E1 and E2. The averages of simulated concentrations at the two sites for available months were 7 %, 3 %, and 12 % smaller in E2 than those in E1 for NH4+, SO42-, and NO3-, respectively, and the differences were clearly smaller than for NH3 at 27 %. As a large fraction of inorganic aerosols come from secondary chemistry reaction, they are more representative for the regional atmosphere condition than the local environment around the measurement site. Therefore, the air quality modeling at a horizontal resolution of 9 km × 9 km is expected to be able to better simulate the concentrations for SIA than the primary gaseous pollutants, particularly when emissions from some local sources are not sufficiently quantified. The simulated concentrations were commonly larger than observations for NH4+ and SO42-, particularly at SHPD in July and October. The uncertainty of the model could be an important source of the discrepancy, as the recent reported mechanisms of gas-to-particle conversion were not sufficiently applied in the CMAQ version we used (Wang et al., 2016; Cheng et al., 2016). In addition, positive or negative artifacts also existed in ground observations with MARGA, resulting from the unexpected reaction between acid gaseous pollutants and nitrate aerosol (Chen et al., 2017; Schaap et al., 2011; Stieger et al., 2018; Wei et al., 2015). From an emissions perspective, the overestimation was partly corrected when smaller NH3 emissions in E2 were applied instead of E1 in the model. Moreover, due to missing information on individual industrial plants, the inventory we used in CMAQ failed to fully capture the progress of emission control in the YRD region and probably overestimated the SO2 emissions (Zhang et al., 2019). The formation of sulfate ammonium aerosols could then be enhanced through the irreversible reaction between SO2 and NH3. The process simultaneously reduced the amount of NH3 reacted with HNO3, further leading to the underestimation of nitrate aerosols. As shown in Table 3, application of E2 with less NH3 emissions than E1 could not improve the modeling performance of nitrate aerosols. The impact of SO2 and NOx emissions on SIA modeling will be further discussed in Sect. 3.4.

To be consistent with the local crossing time of IASI at 09:30 local time, the average of simulated hourly NH3 concentrations at 09:00 and 10:00 local time was applied to calculate the NH3 VCDs, using the following equations: 5nNH3=∑k=123mk×ΔHk×100,6ΔHk=H×ln⁡pkpk+1, where nNH3 is the NH3 VCDs from the CMAQ model (molec. cm-2); mk is the simulated NH3 concentrations at the vertical layer k in the CMAQ (molec. cm-3); ΔH is the height of layer k (m); H represents the height when the pressure of atmosphere declines to 1/e of the original value; and p is the air pressure. Figure 9 illustrates the simulated NH3 VCDs with E1 and E2 for January, April, July, and October. Similar spatial patterns are found with the two inventories; i.e., relatively large NH3 VCDs were simulated mostly in northern Jiangsu and northern Anhui Province, consistent with the hot spot of NH3 emissions. The simulated NH3 VCDs with E1 were 53 % larger than those with E2 across the whole YRD region, with the maximum and minimum monthly differences calculated at 73 % and 31 % for April and October, respectively. The NMB, NME, and correlation coefficient (r) between the observed and simulated VCDs and the monthly average VCDs from observation and simulation are summarized in Table 4. Application of both inventories resulted in larger NH3 VCDs than those from satellite observation for January and October, while the simulated VCDs for April and July were smaller. Besides the uncertainty from monthly distribution of NH3 emissions, the bias from WRF modeling on temperature might also contribute to the discrepancy between simulated and observed VCDs. As shown in Table S7, WRF overestimated the monthly temperature in January and October, with the NMBs calculated at 26.6 % and 0.34 %, and underestimated it in April and July, with the NMBs calculated at -1.62 % and -2.51 %, respectively. Compared to E1, application of E2 significantly reduced the NMEs from 83.8 % to 37.5 % for January and largely corrected the overestimation in VCD simulation for January and October. The simulated VCDs were 4.3 % larger and 1.4 % smaller than observations for the 2 months, respectively. The results implied satisfying agreement between the simulated and observed VCDs over the YRD region. Improvement in NH3 VCD simulation was also found for April when E2 instead of E1 was applied in the air quality modeling, with the NME reduced from 65.8 % to 60.7 %. For July, however, application of E2 did not improve the model performance, implying that the current method in E2 could possibly underestimate the NH3 volatilization when the actual ambient temperature was high. Besides the emissions, the discrepancy could result from various factors including the uncertainty in chemical mechanisms in CMAQ and environmental conditions. Errors from satellite retrieval could also contribute to the inconsistency between simulation and observation. In their study, van Damme et al. (2014), for example, estimated an error of 19 % for the total NH3 columns in Asia. As the ESPRI product of NH3 VCDs we applied in the study does not provide the averaging kernel, however, uncertainty in NH3 column retrieval could result from the reduced sensitivity of satellite measurement towards the surface.

To further investigate the impact of soil pH on the emissions and thereby the modeling performance on NH3 VCDs, the soil in the YRD region was classified to three types, acidic soil (pH ≤6.5), neutral soil (6.5<pH≤7.5), and alkali soil (pH > 7.5), and the NMBs and NMEs between the simulated and observed NH3 VCDs were calculated by soil type and month, as summarized in Table 5. For neutral and acidic soil, application of E2 that considers the effect of farming season, geophysical condition, and manure management on NH3 emission rates resulted in clearly smaller NMEs than E1, implying the improvement in emission estimation. For acidic soil, however, the NMBs were negative for all the months when E2 was applied, and the NMEs were elevated compared to E1 except for January. Moreover, application of E2 resulted in negative NMBs for neutral and alkali soil in April and July as well. Those results implied that E2 possibly underestimated the NH3 emissions for acidic soil in particular for warm seasons. With the correction of pH and temperature, the NH3 volatilization rate from basal dressing fertilization was relatively low, indicating that the current linear assumption between the soil pH and NH3 volatilization rate might not be appropriate for soil with low pH values for eastern China. As shown in Fig. S4, the measured NH3 volatilization rates from urea and ABC fertilizer use under relatively high soil pH (Zhang et al., 2002; Zhong et al., 2006) were close to the estimated values in E2, but the measured results for acidic soil were clearly larger than those in E2.

Limitation should be acknowledged in the emission comparison and evaluation. Besides those we paid extra attention to in E2 (e.g., temperature, soil property, fertilizer application method, and manure management process), other factors could also be influential on air-surface exchange of NH3 and thereby NH3 emissions, including meteorology parameters (wind speed, precipitation, and leaf surface wetness), surface layer turbulence, air and surface heterogeneous-phase chemistry, and plant physiological conditions (Flechard et al., 2013; Gyldenkaerne et al., 2005, Skjøth et al., 2011). With those factors integrated in a bidirectional surface–atmosphere exchange module in air quality modeling, the NH3 emission inventories were improved and the biases in simulation of NH3 and NH4+ aerosol concentrations were reduced for both the US and Europe (Bash et al., 2013; Wichink Kruit et al., 2012). The fact that we ignored given parameters and processes in the current work could thus partly explain the discrepancy between the simulation and observations. Applying the bidirectional NH3 exchange module, for example, Wichink Kruit et al. (2012) found increased NH3 concentrations for agricultural source areas due to the elevated lifetime and transport distance of NH3 in the model. The result implied a possible correction on the underestimation in NH3 concentrations, as shown in Tables 3 and 4. Therefore, a more comprehensive evaluation and comparison in NH3 emissions is thus suggested for the future, including the bidirectional NH3 exchange and the top-down constraint with inverse modeling.

Besides the meteorology condition, NH3 emissions, and soil pH, the estimates of SO2 and NOx emissions could influence the NH3 and SIA simulation as well. SO2 can be transformed to S (IV) through liquid-phase reaction and then be oxidized to S (VI) by O3, or it can be directly oxidized to H2SO4 by H2O2 or the hydroxyl radical (⚫OH). HNO3 can be formed through NO2 oxidation by ⚫OH in daytime or through hydrolysis of N2O5 at the aerosol surface at night. Normally NH3 preferentially reacts with H2SO4 and relatively stable (NH4)2SO4 is produced, while NH4NO3 could easily be decomposed under high-temperature or low-humidity conditions. Therefore, the ambient NH3 concentrations and formation of NH4+ aerosols are influenced by the balance between acidic (SO2 and NOx) and alkaline component (NH3) emissions.

As described in Sect. 2.3, the SO2 and NOx emissions for 2014 used in this work were scaled from those for 2014 based on the changes in activity data. Ignorance of emission control progress during 2012–2014 would probably result in overestimation in emissions. The bias was evaluated through satellite observation. The daily planetary boundary layer (PBL) SO2 and tropospheric NO2 VCDs were obtained from the OMSO2 Level-3 product (http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/OMI/omso2e_v003.shtml, last access: 4 April 2020) and the POMINO Level-3 product from the Ozone Monitoring Instrument (OMI), respectively. As shown in Table S8, all the provinces in the YRD had their SO2 and NO2 VCDs substantially reduced during 2012–2014, and the VCDs declined by 48 % and 31 %, respectively, for the whole region. From a recent unpublished emission study, however, the SO2 and NOx emissions were estimated to decrease by only 16 % and 8 % in the YRD region for the 2 years (personal communication with Cheng Huang from Shanghai Research Academy of Environmental Science, 2019). It can be inferred that the overestimation of SO2 emissions might enhance their reaction with NH3 and thereby the formation of (NH4)2SO4 in the air quality modeling. The formation of NO3-, in contrast, might be suppressed accordingly.