Paradigm shift in aerosol chemical composition over regions downwind of China

A rapid decrease in PM2.5 concentrations in China has been observed in response to the enactment of strong emission control policies. From 2012 to 2017, total emissions of SO2 and NOx from China decreased by approximately 63% and 24%, respectively. Simultaneously, decreases in the PM2.5 concentration in Japan have been observed since 2014, and the proportion of stations that satisfy the PM2.5 environmental standard (daily, 35 µg/m3; annual average, 15 µg/m3) increased from 37.8% in fiscal year (FY) 2014 (April 2014 to March 2015) to 89.9% in FY 2017. However, the quantitative relationship between the PM2.5 improvement in China and the PM2.5 concentration in downwind regions is not well understood. Here, we (1) quantitatively evaluate the impacts of Chinese environmental improvements on downwind areas using source/receptor analysis with a chemical transport model, and (2) show that these rapid emissions reductions improved PM2.5 concentrations both in China and its downwind regions, but the difference between SO2 and NOx reduction rates led to greater production of nitrates (e.g., NH4NO3) due to a chemical imbalance in the ammonia–nitric acid–sulfuric acid–water system. Observations from a clean remote island in western Japan and numerical modeling confirmed this paradigm shift.


Results
We analyzed hourly surface-level PM 2.5 observational data from Japan and China and calculated the annual average PM 2.5 concentration for each region. For Korea, we used annual average values from the Air Korea website. The observational data used in this study are described in the Methods section.
Satellite observations of NO 2 and SO 2 from the Ozone Monitoring Instrument (OMI) were also used for the analysis of emission trends between 2005 and 2019. Gridded (0.25 × 0.25 degree) Level 3 data from NASA were used for this study and annually-averaged data were used to examine year-to-year trends over China, Korea, and Japan (NASA OMI website) 24 . Figure 1 shows the annual average PM 2.5 concentrations over Fukuoka, Japan and Beijing (at the U.S. Embassy), China (Supplementary Figure S1 shows the locations of the observation sites). This figure includes estimated SO 2 , NO x , and NH 3 emissions over China 1 , and tropospheric vertical column densities (VCDs) of SO 2 and NO 2 over central eastern China (CEC) (OMI satellite data from 2011 to 2019) 24 . This figure also shows the PM 2.5 achievement ratio for Japan. Similar plots for the average of 74 cities in China and Korea (including two background sites), and several remote Japanese sites are provided in Supplementary Figure S2. Supplementary Figure S3 shows the annual average VCDs of SO 2 and NO 2 levels in East Asian regions between 2011 and 2019 based on satellite retrieval data. Supplementary Figure S4 shows the year-to-year average trends in three regions (CEC, Korea, and Japan) for SO 2 and NO 2 levels based on satellite observations. The PM 2.5 trends in Beijing and the average trend of 74 Chinese cities were very strongly correlated (Supplementary Figure S2(a)), although the PM 2.5 concentration in Beijing was 40% higher than the average concentration of 74 cities in 2013 (this difference became negligible in 2018, as the rate of decrease in Beijing was greater). This trend was quite consistent with that in Japan (Supplementary Figure S2(c)). The correlation coefficient (R) between Beijing and Fukuoka was 0.98. The average PM 2.5 trend for Korea (Supplementary Figure S2(b)) also showed a generally decreasing trend but with a slight difference from those in China and Japan up to 2017.
Bottom-up inventory results and satellite data also exhibit good correlations (Fig. 1). OMI SO 2 observations (Supplementary Figure S3) indicated that a rapid decrease in SO 2 was achieved over the CEC area, and the color www.nature.com/scientificreports www.nature.com/scientificreports/ representing SO 2 in the image has been nearly absent since 2017 (using the same color scale). We found that the SO 2 VCDs in Korea and Japan exhibited small decreases or constant levels after 2010 (see Supplementary Figure S4), whereas NO 2 VCDs in Korea and Japan remained almost constant and increased slightly, respectively, since 2017. NO 2 VCDs exhibited an increasing trend over CEC after 2016. Figure 2 shows the horizontal distribution of annually-averaged PM 2.5 in Japan from 2013 to 2018. The numbers in parentheses indicate the achievement ratios. PM 2.5 values were higher in the western part of Japan compared to the Tokyo metropolitan area. This pattern of high values in the west and low values in the east (i.e., a strong west-east gradient) indicates that the PM 2.5 concentration was strongly influenced by trans-boundary pollution from the west of Japan. From the year-to-year changes in PM 2.5 , we found that the PM 2.5 concentration over large areas of Japan decreased rapidly from 2014 to 2015, with the achievement ratio increasing from 37.8% to 74.5% within one year. It is important to note that there was a high rate of decrease in PM 2.5 from 2014 to 2017, but the rate slowed down after 2017. This is because the recent decrease in the trans-boundary fraction is significant, and the improvement was dramatic and rapid across the rural/remote sites impacted by the trans-boundary fractions. The domestic emissions from large urban and industrial areas contribute greatly to the PM 2.5 concentration, at levels near or sometimes exceeding the criteria. In 2018, a few PM 2.5 hotspots could be observed in a very limited area with strong local effects from volcanoes and industrial emissions.   The detailed changes over Japan were discussed in the context of Fig. 2. Figure 3 shows observed annual average rates of decrease (regionally averaged over 130°-142° E, 33°-37° N) of 0.104 ± 0.046, 0.151 ± 0.062, and 0.172 ± 0.064 since 2015. ΔPM yyyy values for both the Korean average and individual stations are shown. A complicated variation in ΔPM yyyy was observed, with some stations exhibiting positive changes or different variation patterns between years, except at the upwind background stations in Baengyeongdo and Jeju, which will be discussed later.
We used the 3-D Goddard Earth Observing System chemical transport model (GEOS-Chem) 25 for emission sensitivity analysis, including that of the S/R relationship for PM 2.5 . Details of the GEOS-Chem settings and S/R analysis are described in the Methods section.
The model results were analyzed to obtain S/R values. We confirmed that the annual average contribution of Japanese domestic emissions to Fukuoka PM 2.5 was approximately 20%, and the Chinese contribution was approximately 60% based on the meteorological conditions in 2014 26 . S/R results are very useful for evaluating possible strategies for improving PM 2.5 levels over downwind regions after enacting appropriate emission controls www.nature.com/scientificreports www.nature.com/scientificreports/ in one region. For example, the contribution of PM 2.5 from China (mainly from northern China) to Fukuoka was approximately 60%, and thus if the PM 2.5 concentration in China decreases by 40% (e.g., from 100 to 60 µg/ m 3 in Beijing between 2014 and 2017), the decrease in PM 2.5 concentration in Fukuoka can be calculated as follows: 60% × 40% = 24% (assuming that all emissions except those from China remain constant). The observed decrease in PM 2.5 concentration in Fukuoka (18.5 to 14.5 µg/m 3 ) was 22%, which is in good agreement with the model-based S/R estimate.
The contour lines in Fig. 3 represent S/R responses and were calculated by multiplying the fraction of the PM 2.5 contribution from China at each point by the rate of decrease in Chinese PM 2.5 concentration. The contours in Fig. 3(a-c) represent rates of decrease over East Asia in response to different decreases in PM 2.5 concentration in China. The rate of decrease in PM 2.5 concentration in China was set to 12%, 19%, or 26% based on observations. The observed relationship between rates of decrease in China and Japan can be explained using these contour lines.
For Korea, trends at the upwind background sites of Baengyeongdo and Jeju showed a consistent decreasing signal, in agreement with the decrease based on the S/R relationship. However, trends at other Korean sites cannot be explained by the S/R contour lines, and some cities (e.g., Seoul) exhibited significant increases in PM 2.5 concentrations in 2016 and 2017 compared with 2015 and large year-to-year variations.
OMI SO 2 and NO 2 variations across Korea cannot explain the observed changes in the PM 2.5 concentration. In Korea, the SO 2 VCD exhibited a small decrease or no change after 2010, and the NO 2 VCD remained almost unchanged (see Supplementary Figure S4); thus, the trend of the average PM 2.5 was not clearly correlated with the local emissions pattern (particularly in urban areas). Elucidating these patterns and their drivers in Korea is a subject for future research.
We examined recent PM 2.5 and aerosol composition changes over a clean and remote island, Fukue Island, which is located at the western edge of the Japanese mainland and eastern edge of the East China Sea (see Supplementary Figure S1). Details of the observations from Fukue Island and comparison with the GEOS-Chem simulation can be found in the Methods section. Figure 4 shows observation results from Fukue Island averaged between February and April. Figure 4(a) shows the aerosol composition ratios among chloride, NO 3 − , SO 4 2− , NH 4 + , and organic aerosols. Figure 4(b) shows a scatter diagram of SO 4 2− and NO 3 − for each year. As shown in Fig. 4(a), the observed sulfate concentration decreased significantly (by 40%) at Fukue Island. This result is consistent with the decrease in SO 2 emissions over China. Although NO x and NH 3 emissions were also reduced, the observed nitrate concentrations increased continuously. This result could be explained by the chemical balance of the ammonia-nitric acid-sulfuric acid-water system. This thermodynamic equilibrium process is included in the GEOS-Chem simulation described in the Methods section, which allows for detailed studies on chemical balance. Due to the extremely low vapor pressure of sulfuric acid, sulfuric acid produced in the atmosphere consumes ammonia and is neutralized, forming ammonium sulfate aerosol. Then, the leftover ammonia, referred to as free ammonia, is available for the potential formation of ammonium nitrate. As a result, the reduction of sulfuric acid causes more free ammonia to be available, leading to the formation of more ammonium nitrate. Seinfeld and Pandis (2016) 27 indicates that about half of the decrease in concentration of (NH 4 ) 2 SO 4 will be offset by the increase in NH 4 NO 3 . The relationship between the decrease in sulfate and increase in nitrate depends primarily on the concentrations of their precursors, relative humidity (RH), and temperature.
Although SO 2 , NO x , and NH 3 emissions over China have all been reduced, the decrease in NO x is significantly smaller than that in SO 2 , and the decrease in NH 3 is much smaller than that in either SO 2 or NO x (see Fig. 1). If the increase in nitrate due to SO 2 reduction is larger than the nitrate decreases due to decreases in NO x and NH 3 emissions, the overall effects of emission control will lead to increased nitrate concentration. These phenomena were actually observed on Fukue Island from 2012 to 2019 (as seen in Fig. 4b), where sulfate decreased by 1.7 μg/ m 3 , while nitrate increased by 1.7 μg/m 3 , causing the NO 3 − concentration to increase by almost four-fold compared to the 2012-2014 period. A more detailed analysis of these phenomena based on the GEOS-Chem model is described below.

Discussion
To quantitatively analyze the increase in NO 3 − , we modeled an additional four cases of sensitivity experiments, changing the SO 2 and NO x emission intensities based on the bottom-up Multi-resolution Emission Inventory for China (MEIC) 1 results (Table 1). Emission reduction was applied only in the China region, and emissions in all other regions were the same as in the control experiment. CNTL (S10N10) was the control experiment. Case S04N08 was designed based on the MEIC emission reduction rate and OMI satellite changes, and thus is suitable for examining recent emission changes. Case S07N09 was designed to examine the linearity of decreases in SO 2 (between the S10, S07, and S04 cases), and case S04N10 was designed to examine NO x sensitivity under a constant SO 2 condition (with S04N08). In this sensitivity study, emissions of NH 3 and non-methane volatile organic compounds were the same as in CNTL.
The results of the modeled sensitivity experiments are shown in Fig. 4(c) for SO 4 2− and NO 3 − in CEC, the centers of the Yellow Sea and East China Sea, and Fukue Island (model results were averaged between February and April for consistency with observations). Note that Fig. 4(c) demonstrates the typical response of the ammonia-nitric acid-sulfuric acid-water system to the emission sensitivity shown in Table 1. Thus, the absolute concentration level is different from the ACSM observation, but the fundamental changes observed can be explained by the model emission sensitivity experiment.
As shown in Fig. 4(c), the sensitivity experiment between CNTL (=S10N10) and S04N10 for CEC showed decreased   − ranges from 1:0.5 to 1:0.65, becoming larger as the transport distance increases. As noted above, the relationship between the decrease of sulfate and the increase of nitrate is strongly dependent on RH, temperature and a heterogeneous reaction with sea salt (for NaNO 3 formation) during transport from mainland China over the ocean. These responses are quite consistent with observations at Fukue Island.
The change in NH 3 concentration between CNTL and S04N08 is of great interest, and this result is shown in Supplementary Figure S5. The NH 3 concentration in the S04N08 experiment was more than double that over CEC (i.e., increase in free NH 3 ) 28 , and NH 3 concentration increases were simulated over western Japan, including Fukue Island. The changes in NH 3 concentration over CEC were also supported by Infrared Atmospheric Sounding Interferometer (IASI) satellite observations 29 . The conclusions of these sensitivity studies were reasonable, showing that reductions in SO 2 emissions change the balance of the ammonia-nitric acid-sulfuric acid-water system, creating free NH 3 that reacts with HNO 3 to form NH 4 NO 3 , which is transported to downwind regions, especially in the cold season. Figure 5 shows the horizontal distribution of scaled annual mean ΔSO 4 2− and ΔNO 3 − from the model sensitivity study, based on the CNTL and S04N08 experiments. These indices are calculated as follows: The SO 4 2− decrease (ΔSO 4 2− ) over mainland China exceeded −50%, consistent with the 60% decrease in SO 2 (Fig. 1a). Over western and eastern Japan, SO 4 2− decreased by 30% and 20%, respectively. We found that this rate of decrease was linearly proportional to the SO 2 reduction rate within China via a comparison with case S07N09. The impacts of the decrease in SO 2 in China clearly covered a large area downwind.
For NO 3 − , ΔNO 3 − over China was not significant, which is consistent with recent observations 9 . Several areas downwind of China exhibited increased NO 3 − . Over the East China Sea, the rate of increase in NO 3 − exceeded 90%, as NO 3 − concentrations were low in this area in the CNTL experiment; thus, a small increase in NO 3 − results in a large rate of increase. In the Fukue Island region, this increase was approximately 60%. Figure 5 shows annually averaged values, and ratios increased when averaged over the cold season (February to April) because NH 4 NO 3 is more stable in cold weather, as discussed below. Figure 6 shows time-longitude trends of the NO 3 − increase from experiments CNTL to S04N08 along the latitude of 32.5° N. This latitude corresponds the typical transport route from Shanghai to Fukue Island. The increase in concentrations over the downwind regions of China between December and March was significant and was caused by cold weather increasing NH 4 NO 3 stability. During the warm season, the transport path changes and warm temperatures cause NH 4 NO 3 aerosols to enter the gas phase as HNO 3 . At Fukue Island, the increase reached 1 µg/m 3 in winter, consistent in magnitude with observations at Fukue Island. Notably, the eastern edge of area in which NO 3 − increased (on the order of 0.5 µg/m 3 ) approaches 134° E to 136° E, where large cities such as Osaka are located. Note that the changes in concentration were small and usually difficult to detect from observations www.nature.com/scientificreports www.nature.com/scientificreports/ over urban areas of mainland Japan due to large local NO x emissions. However, this small increase may contribute significantly to the presence of excess nitrogen over the downwind region in East Asia 23 .
Recent studies have described how a decrease in PM 2.5 can enhance the lifetime of OH radicals and increase the O 3 level 30 (followed by increases in the atmospheric oxidation capacity and NO 3 − formation). This is a reasonable mechanism that might increase the NO 3 − formation. However, our version of the GEOS-Chem model does not include the heterogeneous interaction between PM 2.5 and OH. Our results explain the observed SO 4 2− / NO 3 − changes exactly, and this indicates that a change in the atmospheric oxidation capacity is not the primary reason for the observed changes in SO 4 We analyzed the PM 2.5 observation data from 2014 to 2019 over Japan, Korea, and China, and found that there was a clear decreasing trend over Japan, which was strongly correlated with levels in China. An emission sensitivity study based on the GEOS-Chem chemical transport model was carried out to quantify the relationship between emission levels in China and PM 2.5 concentrations over downwind regions. The model results showed that the trend of an annual decrease in PM 2.5 in Japan was explained primarily by reduced PM 2.5 concentrations in China. We also used this model to quantitatively evaluate the impact of Chinese environmental improvements on downwind areas using S/R analysis. Rapid emission reductions played an important role in reducing PM 2.5 concentrations, but a chemical imbalance in the ammonia-nitric acid-sulfuric acid-water system caused an increase in long-range NO 3 − transport to downwind regions. Observations on a clean remote island and numerical modeling confirmed that this paradigm shift has occurred since 2014-2015. Concentrations of sulfate, a chemical that undergoes long-range transport, are decreasing, whereas those of nitrate, which is subject only to short-distance transport, are increasing. This increase in nitrate could lead to an excess nitrogen burden in East Asia and the surrounding oceanic regions 31,32 . We found that the most recent satellite NO 2 and SO 2 VCDs, for 2019 (see Supplementary Figure S4), revealed that this paradigm shift is accelerating (because SO 2 is still decreasing, whereas NO 2 is now increasing), indicating that there is a need for careful continuous observation of changes in aerosol chemical compositions, both in China and the downwind regions of Japan and Korea.

Methods
Surface pM 2.5 observation data. For Japan, hourly PM 2.5 observation data from the Atmospheric Environmental Regional Observation System 33 (AEROS; also referred to as 'Soramame-kun') were used for calculating annually averaged PM 2.5 concentrations from 2013 to 2018. A total of 662 AEROS sites were used in this study, and after quality control processing, the AEROS data were interpolated into a 0.375° longitude-latitude grid. Data averaged across Japan were obtained from the Ministry of Environment of Japan.
For China, PM 2.5 concentration data were obtained from the China National Environmental Monitoring Center 34 . In China, PM 2.5 concentrations have been monitored in 74 major cities since the end of 2012, including cities in the Beijing-Tianjin-Hebei region, Yangtze River Delta, and Pearl River Delta, as well as Chongqing municipalities and all provincial capitals. Data from these 74 cities were collected between 2013 and 2018, averaged, and used for analysis in this study. We also analyzed PM 2.5 observations taken at the U.S. Embassy in Beijing from 2011 to 2019.
For Korea, the annual average PM 2.5 values were obtained from the official Air Korea website of the Ministry of Environment of Korea 35 . We selected nine sites, including the cities of Seoul, Busan, Gwangju, Gangneung, Deagu, Daejeon and Mokpo, as well as Jeju and the background site of Baengnyeongdo (the locations of the latter two sites are shown in Supplementary Figure S1) for analysis between 2015 and 2018. Data collected from multiple points within large cities were averaged.
We used observation sites with more than 250 days of qualified observations.
Chemical transport model and S/R analysis. We used the GEOS-Chem model for analysis 25 . The model was run using the full GEOS-Chem NOx-Ox-VOC-HOx-CO chemistry option to simulate the formation of aerosols, including mineral dust, sea salt, black carbon (BC), organic carbon (OC), and secondary inorganic aerosols. The GEOS-Chem model used ISORROPIA-II 36 to calculate the detailed thermodynamic equilibrium www.nature.com/scientificreports www.nature.com/scientificreports/ processes for the H + -NH 4 + -K + -Ca 2+ -Mg 2+ -Na + -OH − -SO 4 2− -NO 3 − -Cl − -H 2 O aerosol system. The model used the assimilated meteorological fields from GEOS of the NASA Global Modeling and Assimilation Office. The model has a horizontal resolution of 2° × 2.5° for global runs, and 0.5° × 0.667° for Asian one-way nesting runs (11° S−55° N, 70−150° E), both containing 47 vertical levels from the surface to 0.01 hPa. We used anthropogenic emissions data from the Emission Database for Global Atmospheric Research 37 for the global domain and from the Regional Emission Inventory in Asia (REAS; ver. 2.1) for the Asian domain 38 . REAS NH 3 emissions data were modified to include seasonal variations in China 39 . PM 2.5 concentrations from the model were calculated by summing the concentrations of relevant aerosols (BC, OC, SO 4 2− , NO 3 − , NH 4 + , dust, and sea salt). Model simulation was conducted from the beginning of December 2013 to the end of July 2019, and the results from the first 8 months were used for model training. We primarily used the S/R model results for 2014 (when the pollution level was high) and assumed that the model results would be similar for the meteorology of different years. Other basic numerical settings were as reported in Uno et al. 18 . We set 19 source regions (including Japan, Korea, northern China, and central China) for S/R analysis 26 . We used a 20% reduction of emission (not zero emission) to avoid undesirable non-linearity of the chemical reactions. The source contribution from region A is calculated as follows: Aerosol observations using the ACSM at Fukue Island. The chemical compositions and mass concentrations of atmospheric fine aerosols, i.e., fine particulate matter (PM 1 ), were observed at the remote island of Fukue, Nagasaki Prefecture, Japan (32.75° N, 128.68° E; see Supplementary Figure S1). The population on this island is approximately 40,000 and it is generally considered to have few emission sources. Aerosol chemical composition was measured using a quadrupole-type ACSM (Q-ACSM; Aerodyne Research Inc., Billerica, MA, USA). The mass concentrations of PM 2.5 were obtained from an air-pollution monitoring station at Goto (located on Fukue Island), which is the site of municipal government offices for Nagasaki prefecture. The chemical compositions of ammonium, nitrate, sulfate, chloride, and organic compounds were analyzed. Because our main interest was trans-boundary air pollution from mainland China, measurements were taken only from January to May on Fukue Island. The details of Q-ACSM and calibration procedures for Fukue Island have been described previously 22,40,41 . To confirm the model simulation, the ACSM observations and GEOS-Chem model results were compared over a 4-month period (Supplementary Figure S6). The GEOS-Chem experiment (CNTL) used emissions from 2010, and thus SO 4 2− and NO 3 − concentrations from GOES-Chem were over-and underestimated, respectively, but we confirmed that the model results reproduce the observed variations well.

Data availability
The datasets generated for the present study are available from the corresponding authors upon reasonable request.