Observations from space and ground stations were used in this study. Tropospheric vertical column densities (VCDs) of NH3 were derived from the measurements of Infrared Atmospheric Sounding Interferometer (IASI) on board MetOp-A (Van Damme et al., 2015, 2017; Clarisse et al., 2009). We determined the annual averages of NH3 column concentrations over the North China Plain on a 0.25∘ × 0.25∘ grid during 2008–2016, based on the relative error weighting mean method (Van Damme et al., 2014). The monthly NH3 concentrations were measured using passive NH3 diffusive samplers (Analyst, CNR Institute of Atmospheric Pollution, Rome, Italy), from September 2015 to August 2016 at 11 sites over northern China (Pan et al., 2018). The SO2 VCDs were provided by the Ozone Monitoring Instrument (OMI) measurements to test the trend of SO2 concentrations. They were derived from the Level-3 Aura/OMI Global SO2 Data Products (OMSO2e), released by the NASA Goddard Earth Sciences Data and Information Services Center (Krotkov et al., 2015). Besides, daily PM2.5 was sampled by quartz-fiber filters at an urban atmosphere environment monitoring station in Peking University (39.99∘ N, 116.3∘ E) of Beijing, China, since 2013. The major water-soluble inorganic compounds (e.g., NH4+, NO3-, and SO42-) were analyzed by ion chromatography.

In this study, the simulations with the Weather Research and Forecasting Model coupled with Chemistry (Grell et al., 2005) version 3.6.1 (WRF-Chem) were conducted for the domain of the North China Plain for the years 2008, 2010, 2012, 2014, 2015, and 2016 (referred to as Run_08–16). We ran the model with a horizontal resolution of 30 × 30 km and 24 vertical layers, extending from the surface to 50 hPa. The initial and boundary meteorological condition was derived from 6 h National Centers for Environmental Prediction reanalysis data. The detailed model configurations were described in our previous study (Huang et al., 2014). The anthropogenic emissions from power plant, industrial, residential, and vehicle sectors were taken from the MEIC database. The MEIC data show that the annual SO2 emissions in the North China Plain were reduced by about 60 %, from 9.9 Tg in 2008 to 4.2 Tg in 2016, while NOx emissions first increased from 8.0 to 8.8 Tg during 2008–2012, and then decreased to 6.7 Tg in 2016.

A high-resolution NH3 emission inventory (1 km × 1 km, month) was developed based on the bottom-up method. The emission factors were parameterized with regional farming practices, ambient temperature, soil pH, and wind speeds. The full details can be found in studies by Kang et al. (2016), Huang et al. (2012), and Huo et al. (2015). The inventory has similar spatial features with recent satellite observations (Van Damme et al., 2014), and its amount is close to the emission estimated by the inversion model using ammonium wet deposition data (Paulot et al., 2014). Recent modeling results also showed its good performance by comparing with ammonium observations in China (Huang et al., 2015). The inventory has covered the period from 1980 to 2016 and considered the inter-annual variability in activity levels and agricultural practices. It shows a distinct seasonal feature in NH3 emissions over the North China Plain. A total of 75 % of annual NH3 emissions are released in the spring and summer months (March–September), during which intensive agricultural fertilization and elevated ambient temperature facilitate the volatilization rates of NH3. Moreover, to integrate this inventory into WRF-Chem simulations, we adopted a diurnal profile with 80 % of NH3 emissions in the daytime, following previous studies (Zhu et al., 2015; Asman, 2001; Paulot et al., 2016).

According to the measurements by IASI, the North China Plain showed the highest VCDs of NH3 in China, which mostly ranged from 15 to 30 × 1015 molecules cm-2 during 2008–2014 and increased to above 30 × 1015 molecules cm-2 in 2015 and 2016 (Fig. S1 in the Supplement). We found the annual NH3 column concentrations increased significantly (p value < 0.05) over the North China Plain between 2008 and 2016 (Fig. 1a). The average tropospheric NH3 columns first fluctuated between 2008 and 2013, and then rapidly increased from 21 × 1015 molecules cm-2 in 2013 to 27 × 1015 molecules cm-2 in 2016. It showed an overall increase of 30 %, or an average annual increase of 0.9 × 1015 molecules cm-2 yr-1. Seasonally, the increase in NH3 columns was more pronounced in summertime (June–August, JJA), with an annual increase rate of 1.8 × 1015 molecules cm-2 yr-1 between 2008 and 2016, which was much higher than in other seasons (< 1 × 1015 molecules cm-2 yr-1).

In contrast to the trends in tropospheric NH3 concentrations, the annual NH3 emissions first experienced a decreasing tendency from 2008 to 2011 (3.0 Tg in 2009 to 2.8 Tg in 2011) and then remained constant at around 2.8 Tg yr-1 during 2011–2016 over the North China Plain (Fig. 1b). The overall trend of NH3 emissions demonstrated a decrease of about 7 %. It is because the changes in mineral fertilizer use and livestock rearing practices have lowered NH3 emission rates. The increasing use of urea fertilizer (from 4.7 to 5.2 Tg yr-1) and compound fertilizers (from 1.2 to 1.7 Tg yr-1) but decreased ammonium bicarbonate (from 1.5 to 0.4 Tg yr-1) led to a 20 % reduction in NH3 emissions from fertilizer application during 2008–2016 (Table S1 in the Supplement). On the other hand, the number of some major livestock increased (beef -20 %, dairy +39 %, goat -23 %, sheep +55 %, pig +18 %, and poultry +19 %; see Table S1 in the Supplement for details), while the proportion of intensive animal rearing systems rises to nearly half of the livestock industry in 2016, compared to only 28 % in 2008 (Table S1 in the Supplement). The intensive systems are characterized with more effective livestock manure management in favor of lower volatilization rates of NH3 (Kang et al., 2016). The transition from the free-range to the intensive system in livestock animal rearing offsets the effect of increased animals on the NH3 emissions, thereby resulting in the annual livestock emissions in the North China Plain being almost constant (around 1.2 Tg yr-1). Overall, the decreasing NH3 emissions cannot track the upward trend of tropospheric NH3 concentrations.

During 2008–2016, SO2 column concentrations were subject to a dramatic decline (p < 0.01) due to a 60 % decrease in SO2 emissions. The annual mean SO2 VCDs were reduced from 14 × 1015 molecules cm-2 (2008) to 4 × 1015 molecules cm-2 (2016), showing a percent reduction of nearly 70 %. Especially during 2012–2016, the decreases in SO2 emissions and VCDs accelerated owing to the implementation of the Action Plan for Air Pollution Prevention and Control by the Chinese government (Zheng et al., 2018). The ground measurements in a typical urban station in the North China Plain indicated that the annual average SO42- concentration in PM2.5 decreased by 35 % (2013–2016) along with rapid SO2 reductions, which was accompanied by a 33 % decrease in particulate NH4+ (Fig. 1b). Seasonally, the decrease in ground-level SO42- reached 60 % during summertime (JJA), which was much higher than in other seasons.

We performed numerical simulations with WRF-Chem to interpret the cause of the NH3 increase. We first evaluated model results against measurements of surface NH3 concentrations available in the North China Plain as well as the satellite-retrieved NH3 columns. The simulated monthly averaged surface NH3 concentrations at 11 stations (mean + standard deviation: 13.5 ± 6.8 µg m-3) generally agreed with corresponding observations (13.4 ± 9.7 µg m-3) with a correlation coefficient of 0.57. More than 70 % of the comparisons differed within a factor of 2 (Fig. 2). Both simulations and observations show high NH3 concentrations of about 30 µg m-3 in warm seasons (March–October) due to enhanced NH3 volatilization and frequent fertilization activities, as well as lower values (mostly < 15 µg m-3) in other months (Fig. 3). Spatially, the hotspot of NH3 was mainly concentrated in Hebei, Shandong, and Henan provinces, which have the most intensive agricultural production in China and thus emit considerable gas-phase NH3 into the atmosphere. We note that the simulated monthly NH3 concentrations were underestimated by 25 %–70 % in several stations in wintertime (January, February, and December). Recently, NH3 emissions from the residential coal and biomass combustion for heating are considered to be a potentially important source of NH3 in suburban and rural areas during wintertime (Li et al., 2016), but it has not been fully included in our bottom-up inventory, which was partially responsible for such deviation between the model and observations.

We calculated the NH3 VCDs from the simulations by integrating NH3 molecular concentrations from the surface level to the top of the troposphere. The results agreed well with the observed NH3 columns of 2016 on the magnitude and spatiotemporal patterns (Fig. S2 in the Supplement). Both IASI measurements and the WRF-Chem simulation showed high annual mean NH3 column concentrations in Hebei, Shandong, and Henan provinces, reaching above 30 × 1015 molecules cm-2. Moreover, we evaluated the modeled SNA concentrations using the filter-based PM2.5 samples at an urban atmospheric monitoring station in the North China Plain during 2014–2016 (Fig. S3 in the Supplement). The model generally reproduced the observed SNA concentrations, with a small annual mean bias for sulfate (-2 %) and ammonium (-13 %) and a relatively large bias for nitrate (-24 %). Overall, the model performed well in modeling the concentrations in tropospheric NH3 as well as secondary inorganic aerosols, which provides high confidence for the following interpretation of the NH3 increases.

The model successfully reproduced the observed increasing trend in NH3 columns over the North China Plain during 2008–2016 (Fig. 4). The modeled NH3 columns were systemically lower than the measurements because the relative error weighting mean method always biased a high result due to the smaller relative error in a larger column (Van Damme et al., 2014; Whitburn et al., 2016). An overall increase of 39 % in NH3 columns with an average annual increase of 0.8 × 1015 molecules cm-2 yr-1 was found in the simulations between 2008 and 2016, and meanwhile the SO2 columns averaged over the North China Plain decreased by approximately 50 % in this period. These results were close to the measurements.

To verify our hypothesis, we replaced SO2 emissions during 2010–2016 by those in 2008 and repeated the simulations (referred to as Run_10_S08 to Run_16_S08). It was noticeable that, under these conditions, the increasing trend of NH3 column concentrations was not observed, and even a decrease of 13 % took place (Fig. 4). The largest differences were found in 2015 and 2016, when the annual NH3 columns in these sensitivity simulations were about 40 % (8–10 × 1015 molecules cm-2) lower than those in the baseline cases, corresponding to the 60 % reduction in SO2 emissions between 2008 and 2016. By comparing the results among Run_08, Run_16, and Run_16_S08, we found that the reduction in SO2 emissions increased the NH3 column concentrations by 52 % during 2008–2016, which was even higher than the overall increase (39 %) in the baseline cases. Therefore, we deduce that the rapid SO2 emission reductions are responsible for the increased NH3 levels during 2008–2016, while other mechanisms may be negative contributors. More details on these effects are shown in the following.

As we indicated above, SO42- was observed to be decreasing over recent years in response to the reductions of SO2 emissions. This was also reproduced by our simulations, which showed that the annual average sulfate concentrations decreased by almost 50 % in the lower troposphere. This decreasing trend was especially pronounced after 2013 owing to the very effective SO2 emission reductions. Given that the vapor pressure of H2SO4(g) is practically zero over atmospheric particles, atmospheric SO42- is predominately in the particle phase and can combine with NH3 available in air, forming sulfate salts (mostly ammonium sulfate/bisulfate) (Seinfeld and Pandis, 2006). Since the North China Plain is typically under rich NH3 regimes, SO42- is mainly in the form of ammonium sulfate (Meng et al., 2011; Huang et al., 2017); and the aforementioned SO42- reductions would therefore increase atmospheric NH3 concentrations by driving the phase state of NH3 from particulate to gaseous.

By assuming that a 1 mol decrease in simulated SO42- would lead to a 2 mol increase in ambient gaseous NH3 in this region, the average annual increase in the tropospheric NH3 columns due to the reductions of SO42- was estimated to be approximately 1.5 × 1015 molecules cm-2 yr-1 over the North China Plain during 2008–2016. This is comparable with or higher than the simulated results from Run_08 to Run_16, as well as the IASI observations (0.9 × 1015 molecules cm-2 yr-1). By neglecting the deposition processes, we found that the rapid SO2 emission reduction of 50 % from 2012 to 2016 resulted in a 55 % increase in the NH3 columns, compared to that of 30 % recorded by IASI observations. Overall, the estimation results confirmed that the increasing trend of NH3 was mainly determined by the SO2 emission reductions.

We compared the spatial patterns of decreased SO42- and increased NH3 between 2008 and 2016 (Run_08 vs. Run_16). Large reductions of 6–10 × 1015 molecules cm-2 in annual averages of sulfate columns were concentrated in Hebei, Shandong, and Henan provinces, the area subject to high SO2 loadings and stringent emission controls (Fig. 5a). Meanwhile, the simulated increases in NH3 columns reached more than 8 × 1015 molecules cm-2 in most parts of the North China Plain (Fig. 5b) and were comparable with those observed by the IASI (8–16 × 1015 molecules cm-2). In addition, we found that NH4+ concentrations have decreased with a similar magnitude to the increases in gas-phase NH3 levels between Run_08 and Run_16. The proportion of NH3 in the total (NH3+NH4+) increased on average from 26 % in 2008 to 37 % in 2016 over the North China Plain. Figure 5c and d illustrate that, without the large SO2 emission reductions between 2008 and 2016 (i.e., replacing SO2 emissions in 2016 by those in 2008, Run_08 vs. Run_16_S08), the sulfate columns partly increased. Correspondingly, the NH3 columns remained constant or decreased by about 5 × 1015 molecules cm-2 (-13 % relative to the 2008 level) in parts of the North China Plain. Thus, the increase in the tropospheric NH3 columns was the result of a transition in NH3 phase partitioning, which was strongly associated with the decreased formation of ammonium sulfate due to SO2 emission reductions.

The seasonal variations in SO42- decreases and NH3 increases were consistent (Fig. 6). We can see that the reduction of sulfate column concentrations between the Run_08 and Run_16 reached 1.3 × 1015 molecules cm-2 in summer (JJA), which was about 3 times larger than in other seasons. The corresponding percent reductions ranged from 15 % in DJF to 36 % in JJA. As aforementioned, the observations of PM2.5 in Beijing also showed the highest decrease in sulfate in summer. Considering that the SO2 emission reductions were uniform throughout the year, this seasonal pattern was likely attributed to the conversion efficiency of SO2 to H2SO4. Our simulations showed that a 1 mol decrease in SO2 corresponded to an approximately 0.7 mol decrease in particulate sulfate in summer over the North China Plain, but the values dropped to below 0.4 in other seasons. It is known that the photochemical oxidation of SO2 by OH radical is most active in summertime due to high atmospheric oxidizing capacity, and it dominates the formation of SO42-, which makes the response of SO42- concentrations to SO2 emission reductions more sensitive (Paulot et al., 2017; Huang et al., 2014). The comparison of modeled NH3 columns also showed a markedly higher increase in summer months than during other seasons, driven by the variations in SO42-. Furthermore, by comparing the model results between the Run_16 and Run_16_S08 cases, we found that, without considering the SO2 emission reductions, the seasonal increases in NH3 columns and decreases in SO42- concentrations were not observed.

Since the chemical formation of particulate ammonium nitrate also affects the gas–particle partitioning of NH3, the role of NOx emissions should be discussed. We noted that, unlike the trend of particulate sulfate in PM2.5, the simulated concentrations of particulate nitrate in PM2.5 increased on average by 28 % over the North China Plain between 2008 and 2016, despite a 16 % reduction in NOx emissions (Fig. S4 in the Supplement). This trend can be partially explained by the increased NH3 in the atmosphere that would facilitate the formation of ammonium nitrate. To quantitatively understand the effect of NOx emission on the trend of NH3, we performed a sensitivity experiment by repeating the simulation of 2016 with the NOx emissions in 2008 (Run_16_08N). By comparing the results among Run_16, Run_16_08N, and Run_08, we found that the reduction in NOx emissions (16 % from 2008 to 2016) decreased the gaseous NH3 concentrations by about 3 % (Fig. S5 in the Supplement). Specifically, because the reduced NOx in this period promoted the transition of ozone (O3) photochemistry from a volatile organic compound (VOC)-limited to a transitional regime with high O3 production efficiency (Jin and Holloway, 2015), the simulated annual mean O3 concentrations were elevated by 3.7 ppb over the North China Plain between the Run_16_08N and Run_16 cases. The resultant enhancement in atmospheric oxidizing capacity would favor the conversion of NO2 to NO3- and therefore derive more NH3 partitioning from gaseous to particulate phases via aerosol thermodynamic equilibrium. Moreover, the measurements at an urban station of Beijing indicated a fluctuating trend of the annual mean NO3- concentrations during 2013–2016 (Fig. 1). Overall, the limited reduction in NOx emissions cannot be responsible for the increased NH3, because the concentrations of particulate nitrate has remained high over the North China Plain during recent years.

Besides, meteorological conditions are known to have an influence on NH3 concentrations. Both Warner et al. (2017) and Fu et al. (2017) found that elevated annual surface temperature has partially contributed to the increase in NH3 in eastern China over the past decade. In this work, we tested the effects of meteorological conditions on NH3 variations by a simulation with meteorological fields in 2016 and anthropogenic emissions in 2012 (Run_12_M16). We selected these 2 years because NH3 concentrations experienced a rapid increase during the period. This change in meteorological fields for the Run_12_M16 resulted in a decrease of about 3 % in annual mean NH3 concentrations relative to the Run_12 (Fig. S6 in the Supplement). Therefore, the inter-annual variability in meteorological conditions cannot explain the observed significant increase over the North China Plain.

Interestingly, increasing trends of gas-phase NH3 in the atmosphere have also been observed in the last 20 years in the Midwest of the United States and Western Europe by satellite retrievals and ground measurements (Saylor et al., 2015; Warner et al., 2017; Ferm and Hellsten, 2012). The marked decreases in SO2 and NOx emissions were largely responsible for these increases, as confirmed by the corresponding trends of particulate sulfate and nitrate concentrations. Warner et al. (2017) infer that SO2 emission reduction in China may be a leading cause of the increased NH3. More recently, Yu et al. (2018) quantified the contributions of the acid gases on the trends of NH3 and found that emissions of SO2 contributed to two-thirds and NOx to one-third of the change in NH3 over the United States from 2001 to 2016. In this work, we demonstrate that the rapid reduction in SO2 emissions was responsible for the increase in NH3 over the North China Plain during 2008–2016, while other potential pathways (NH3 emissions, NOx emissions, and meteorological conditions) decreased its concentrations by approximately 13 % for this period.