Long-Term (2005–2017) View of Atmospheric Pollutants in Central China Using Multiple Satellite Observations

: The air quality in China has experienced dramatic changes during the last few decades. To improve understanding of distribution, variations, and main inﬂuence factors of air pollution in central China, long-term multiple satellite observations from moderate resolution imaging spectroradiometer (MODIS) and ozone monitoring instrument (OMI) are used to characterize particle pollution and their primary gaseous precursors, sulfur dioxide (SO 2 ), and nitrogen dioxide (NO 2 ) in Hubei province during 2005–2017. Unlike other regions in eastern China, particle and gaseous pollutants exhibit distinct spatial and temporal patterns in central China due to di ﬀ erences in emission sources and control measures. OMI SO 2 of the whole Hubei region reached the highest value of ~0.2 Dobson unit (DU) in 2007 and then declined by more than 90% to near background levels. By contrast, OMI NO 2 grew from ~3.2 to 5.9 × 10 15 molecules cm − 2 during 2005–2011 and deceased to ~3.9 × 10 15 molecules cm − 2 in 2017. Unlike the steadily declining SO 2 , variations of OMI NO 2 ﬂattened out in 2016 and increased ~0.5 × 10 15 molecules cm − 2 during 2017. As result, MODIS AOD at 550 nm increased from 0.55 to the peak value of 0.7 during 2005–2011 and then decreased continuously to 0.38 by 2017. MODIS AOD and OMI SO 2 has a high correlation (R > 0.8), indicating that annual variations of SO 2 can explain most changes of AOD. The air pollution in central China has notable seasonal variations, which is heaviest in winter and light in summer. While air quality in eastern Hubei is dominated by gaseous pollution such as O 3 and NO x , particle pollutants are mainly concentrated in central Hubei. The high consistency with ground measurements demonstrates that satellite observation can well capture variations of air pollution in regional scales. The increasing ozone (O 3 ) and NO 2 since 2016 suggests that more control measures should be made to reduce O 3 -related emissions. To improve the air quality in regional scale, it is necessary to monitor the dynamic emission sources with satellite observations at a ﬁner resolution.


Introduction
Air pollution is a dynamic and complicated mixture of particulate matter and gaseous pollutants that emitted from anthropogenic activities and natural processes into the atmosphere. During the last decades, rapid industrialization and urbanization in China has caused serious air quality problems [1]. Massive anthropogenic emissions in populated regions of China lead to widespread air pollution [2][3][4][5], which is characterized by high concentration and diverse sources [6][7][8]. With a lifetime of several days to one-week, fine particle (PM 2.5 , fine particulate matter with aerodynamic diameter less than 2.5 µm) is the most-watched atmospheric pollutant in air quality monitoring. Moreover, as the main gaseous precursors of PM 2.5 , sulfur dioxide (SO 2 ), and nitrogen dioxide (NO 2 ) is usually measured to monitor emission levels for combustion of fossil fuels and vehicle exhaust. Ozone (O 3 ) pollution from photochemical reaction between NO x and volatile organic compounds (VOCs) is apt to occur under intense sunshine and high temperature conditions. Numerous epidemic studies have shown robust correlation between exposure level to atmospheric pollutants and increase in morbidity and mortality of respiratory and cardiovascular diseases [9].
Owing to adverse effects of air pollution on human health and daily life, the Chinese government has implemented continuous emission control measures. The anthropogenic emission sources and their amount in China have undergone dramatic changes during last decades due to both the rapid transition of economic structures and environmental policies [10]. There is also a large spatial difference in anthropogenic emissions sources [11], exerting considerable challenges for making timely and effective improvement measures. While ground measurements can accurately measure mass concentration of atmospheric pollutants, the national ground network for air quality monitoring in China since 2013 is only concentrated in major cities with very limited spatial coverage. Since late 1990s, advanced satellite sensors including moderate resolution imaging spectroradiometer (MODIS) and ozone monitoring instrument (OMI) provide an unprecedented opportunity to observe atmospheric particles and trace gases from regional to global scales [12,13].
The heavy air pollution in China has drawn the attention of numerous observational studies from various perspectives. Besides ground-based measurements and chemistry model simulations, satellite studies have greatly renewed knowledge concerning spatial distribution of air pollution hotspots and variations of their major emissions in northern China [14][15][16], the Yangtze River Delta (YRD) [17], and Pearl River Delta (PRD) [18]. However, previous studies mostly focused on economically developed regions with extensive urban agglomeration. As an emerging economic center located between polluted northern China and clean southern China, air quality in central China has been rarely concerned under the background of rapid development. Although recent air pollution such as PM 2.5 in central China (~40-60 µg/m 3 ) is obviously lower than that in northern China (~60-80 µg/m 3 ) [19,20], the magnitude of the atmospheric pollutants in central China is at a high level compared with health standard (e.g., PM 2.5 : 35 µg/m 3 ), which can further be seriously aggravated by [21]. By now, spatial patterns of major atmospheric pollutants in central China and how they change over space and time remain unclear due to few comprehensive observational studies.
In this study, we first provide comprehensive insight into spatial patterns and long-term trend of primary atmospheric pollutants including aerosols, SO 2 , and NO 2 in central China during 2005-2017 by utilizing multiple satellite products. Section 2 introduces satellite products and ground measurements used in this work. A comprehensive analysis regarding spatial and temporal variations of aerosol optical depth (AOD), SO 2 , and NO 2 in central China is given in Section 3. Further, its implication of on emission control is discussed in Section 4. At last, a brief summary is shown in Section 5. The main purpose of this work is to reveal spatial changes of pollution hotspot and temporal trends of the primary air pollutants in central China, giving a reference for future air quality improvements.

Study Area and Ground Measurments
We selected Hubei province as our study area (Figure 1), which is the main part of central China. Surrounded by mountains with evaluation at~600-1000 m, Hubei is relatively isolated from other regions. The mountains around Hubei provide a barrier for the transport of heavy pollution from northern China. In the other hand, the basin-like terrain in Hubei is not conductive for dispersion of atmospheric pollutants. Unlike the concentrated urban agglomeration in northern China, YRD, and PRD, only the megacity of Wuhan has a population of more than 10 million in Hubei, and other cities have much smaller scales.
Remote Sens. 2019, 11, x FOR PEER REVIEW  3 of 19 atmospheric pollutants. Unlike the concentrated urban agglomeration in northern China, YRD, and PRD, only the megacity of Wuhan has a population of more than 10 million in Hubei, and other cities have much smaller scales. The Environmental Protection Agency of China established a national air quality monitoring network in 2013, which releases hourly concentration of six main atmospheric pollutants (PM2.5, PM10, SO2, NO2, O3, and CO) in major cities with the same ambient air quality standard. Spatial distribution of air quality monitoring sites in Hubei is shown in Figure 1. There are 57 national air quality monitoring sites in Hubei, with nine sites in the megacity of Wuhan. In addition, the statistical emission inventory data of SO2 and NOx from the statistical yearbook of China is used for comparison and reference.  Figure 2, utilizing the National Centers for Environmental Prediction (NCEP) Final (FNL) and Operational Global Analysis data in the 1° × 1° grid. Seasonal characteristics regard the height of the planetary boundary layer (PBL) with 925 hPa winds, temperature, and relative humidity (RH). The PBL height of central China is around ~1000 m at similar levels as in northern China except a little lower (~800 m) in autumn. While northerly winds prevail in winter and spring within the PBL, southerly and easterly air flows are predominant in summer and autumn, respectively. In particular, wind speed over central China is obviously higher (~1-3 m/s) than adjacent areas. Central is located in the transition zone, with temperature and RH higher than in northern China and lower than in southern China. There is a high contrast in temperature near surface between the cold winter and hot summer, indicating distinct atmospheric stability. The Environmental Protection Agency of China established a national air quality monitoring network in 2013, which releases hourly concentration of six main atmospheric pollutants (PM 2.5 , PM 10 , SO 2 , NO 2 , O 3 , and CO) in major cities with the same ambient air quality standard. Spatial distribution of air quality monitoring sites in Hubei is shown in Figure 1. There are 57 national air quality monitoring sites in Hubei, with nine sites in the megacity of Wuhan. In addition, the statistical emission inventory data of SO 2 and NO x from the statistical yearbook of China is used for comparison and reference.
The meteorological background of central China is shown in Figure 2, utilizing the National Centers for Environmental Prediction (NCEP) Final (FNL) and Operational Global Analysis data in the 1 • × 1 • grid. Seasonal characteristics regard the height of the planetary boundary layer (PBL) with 925 hPa winds, temperature, and relative humidity (RH). The PBL height of central China is around 1000 m at similar levels as in northern China except a little lower (~800 m) in autumn. While northerly winds prevail in winter and spring within the PBL, southerly and easterly air flows are predominant in summer and autumn, respectively. In particular, wind speed over central China is obviously higher (~1-3 m/s) than adjacent areas. Central is located in the transition zone, with temperature and RH higher than in northern China and lower than in southern China. There is a high contrast in temperature near surface between the cold winter and hot summer, indicating distinct atmospheric stability.

MODIS Aerosol Dataset
The MODIS sensor onboard Terra and Aqua satellite since 2000 and 2002, respectively, can provide near daily global observations of aerosol and cloud properties with a wide swath of ~2330 km and broad spectrum between 0.4-14 μm at high spatial resolution of 250-1000 m. MODIS aerosol retrieval over land was firstly implemented at 10 km by the dart target (DT) algorithm utilizing spectral relationship between surface reflectance in visible and short infrared bands over dense vegetation [12]. Owing to the widely application of MODIS aerosol products in air quality research, DT aerosol products are at a much higher resolution of 3 km and are added in the Collection (C) 6 MODIS aerosol algorithm [22]. However, MODIS DT AOD is usually not available over urban areas and heavy pollution conditions [23]. To retrieve aerosols over bright surface such as deserts and urban regions, MODIS Deep Blue (DB) algorithm was developed to utilize pre-calculated surface reflectance database, and externed to all the cloud-free and snow/ice-free areas since C6 [24].
Both MODIS DT and DB AOD at 550 nm exhibits well accuracy compared with ground observations [23,25]. Ground validations show that retrieval error of MODIS AOD is generally better than 0.05 + 20% relative to Aerosol Robotic Network observations. The current version of MODIS DT and DB aerosol products is C6.1, which only conducts some slight improvements without changes in algorithm principle. Compared with C6 products, the C6.1 DB aerosol products have refined instrument calibration and aerosol models [26]. Quality assurance flags of MODIS DB AOD depend on number of retrieved pixels and their standard deviation. Usually, MODIS AOD of best estimate quality is suggested in application, but it is found that such selection can filter out many high-AOD (>1.0) cases in eastern China with only slight improvement in accuracy [23]. Considering the limitation of MODIS DT and MODIS DB products of best estimate, we selected C6.1 MODIS DB AOD of all quality at 10 km to investigate aerosol trends in central China.

OMI Products of Trace Gases
OMI onboard Aura since 2004 is a hyperspectral instrument that measures backscattered sunlight in ultraviolet-visible range (270 nm to 500 nm) with a spectral resolution of about 0.5 nm. As a sun-synchronous satellite instrument, OMI crosses the equator around local time of 13:30. With a

MODIS Aerosol Dataset
The MODIS sensor onboard Terra and Aqua satellite since 2000 and 2002, respectively, can provide near daily global observations of aerosol and cloud properties with a wide swath of~2330 km and broad spectrum between 0.4-14 µm at high spatial resolution of 250-1000 m. MODIS aerosol retrieval over land was firstly implemented at 10 km by the dart target (DT) algorithm utilizing spectral relationship between surface reflectance in visible and short infrared bands over dense vegetation [12]. Owing to the widely application of MODIS aerosol products in air quality research, DT aerosol products are at a much higher resolution of 3 km and are added in the Collection (C) 6 MODIS aerosol algorithm [22]. However, MODIS DT AOD is usually not available over urban areas and heavy pollution conditions [23]. To retrieve aerosols over bright surface such as deserts and urban regions, MODIS Deep Blue (DB) algorithm was developed to utilize pre-calculated surface reflectance database, and externed to all the cloud-free and snow/ice-free areas since C6 [24].
Both MODIS DT and DB AOD at 550 nm exhibits well accuracy compared with ground observations [23,25]. Ground validations show that retrieval error of MODIS AOD is generally better than 0.05 + 20% relative to Aerosol Robotic Network observations. The current version of MODIS DT and DB aerosol products is C6.1, which only conducts some slight improvements without changes in algorithm principle. Compared with C6 products, the C6.1 DB aerosol products have refined instrument calibration and aerosol models [26]. Quality assurance flags of MODIS DB AOD depend on number of retrieved pixels and their standard deviation. Usually, MODIS AOD of best estimate quality is suggested in application, but it is found that such selection can filter out many high-AOD (>1.0) cases in eastern China with only slight improvement in accuracy [23]. Considering the limitation of MODIS DT and MODIS DB products of best estimate, we selected C6.1 MODIS DB AOD of all quality at 10 km to investigate aerosol trends in central China.

OMI Products of Trace Gases
OMI onboard Aura since 2004 is a hyperspectral instrument that measures backscattered sunlight in ultraviolet-visible range (270 nm to 500 nm) with a spectral resolution of about 0.5 nm. As a sun-synchronous satellite instrument, OMI crosses the equator around local time of 13:30. With a 2600 km swath, OMI has a daily global coverage with a pixel size of 13 × 24 km at nadir. By deriving abundance of various trace gases with their respective absorption characteristics, OMI products include vertical column density (VCD) of trace gases such as NO 2 and SO 2 . Considering the much shorter lifetime of gaseous pollutants of SO 2 and NO 2 (~one hour to half day) than aerosols', the continuous data record of OMI products has been widely used to infer anthropogenic pollution sources and verify emission inventory [27]. While NO 2 is the production from combustion of fossil fuels such as vehicle emissions in urban regions, SO 2 mainly originates from coal-burning sources.
The OMI NO 2 and SO 2 data was obtained from the NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC). OMI tropospheric NO 2 VCD was retrieved by the common differential optical absorption spectroscopy (DOAS) algorithm with uncertainties of <20% in polluted regions [28]. There are three key steps in NO 2 retrieval: spectral fitting of OMI backscattered radiances to laboratory absorption spectra, separating stratospheric contribution, and converting satellite observed slant column densities to VCDs. By contrast, the principal component analysis (PCA) technique was used in the current planetary boundary layer (PBL) SO 2 retrieval with priori profiles and a PBL height of 3 km (700 hPa) assumed [29]. The OMP PBL SO 2 derived using the PCA method registered a small bias (<0.1 DU) over background regions [30]. Unlike the common unit of OMI NO 2 products (10 15 molecules cm −2 ), Dobson units (1 DU = 2.69 × 10 16 molecule cm −2 ) are usually used for OMI SO 2 for convenience of retrieval calculation and error estimation. The pixel size of OMI has a large variation from 13 × 24 km at nadir to 26 × 120 km at the edge. Different from satellite AOD derived from aerosol scattering, OMI retrieval of trace gases by their absorption is not very sensitive to cloud contamination. To ensure representativity of OMI products, NO 2 retrievals with a cloud fraction >30% and for SO 2 with O 3 < 1500 DU in the data cell are filtered out. We selected a standard Level 3 OMI grid products at 0.25 × 0.25 • resolution.

The Characteristics of SO 2 and NO 2 Variation during 2005-2017
As the main water-soluble components, sulfate, and nitrate together accounted for a significant fraction (~20-50%) in mass concentration of PM 2.5 particles in China [4,11]. Other major particle components such as organic aerosol was generated via complicated chemistry reactions and mineral dust that originated from long-range transport and diverse local sources [4]. By contrast, emission sources of sulfate and nitrate's gaseous precursors, SO 2 and NO x , are mostly from coal-burning power plants and industry activities, which are the main control targets of the government due to the practicability for unified management. Since the short lifetime of SO 2 and NO 2 , satellite VCD products of SO 2 and NO 2 are usually used to map anthropogenic emission sources as well as their emission levels [13,17].  To have a long-term view of the changes in the overall SO2 level in central China, spatial means of OMI SO2 compared with statistical SO2 emission amount during 2005-2017 ( Figure 4). There is a generally consistent trend between OMI SO2 concentration and emission amount. Different from the peak value of SO2 concentration in 2007, statistical SO2 emissions in Figure 3 show the largest SO2 amount of ~760,000 tons in 2006. In 2007, the much higher OMI SO2 VCD in both Hubei and its adjacent part to the north demonstrates that the bottom-up statistical emission can miss some substantial emission sources (Figure 2), especially in the early stage of SO2 control. Considering the desulfurization policy in China from 2006 [31], the emission peak in 2006 can be partly caused by statistics of mainly large SO2 sources. On the other hand, OMI retrievals show a more rapid decline of SO2 amount (~50%) than statistical amount (~10%) since 2008. The increase of SO2 (~50%) in 2010-2011 can be connected with the recovery from the economic downturn in 2009 [32]. Considering that SO2 mainly originates from coal-burning combustion of fossil fuels, the low-level SO2 (~210,000 tons) after 2014 indicates effective effects of SO2 control policy. The striking deviation of the annual OMI SO2 shows large temporal variations of the SO2 level and the variances increase obviously with the mean concentration. Although there is no coal-burning heating during winter in central China as there is in northern China, the formation of sulfate from SO2 and their lifetime can have considerable seasonal differences due to variations of meteorological conditions such as illumination, temperature, and relative humidity ( Figure 2). To have a long-term view of the changes in the overall SO 2 level in central China, spatial means of OMI SO 2 compared with statistical SO 2 emission amount during 2005-2017 ( Figure 4). There is a generally consistent trend between OMI SO 2 concentration and emission amount. Different from the peak value of SO 2 concentration in 2007, statistical SO 2 emissions in Figure 3 show the largest SO 2 amount of~760,000 tons in 2006. In 2007, the much higher OMI SO 2 VCD in both Hubei and its adjacent part to the north demonstrates that the bottom-up statistical emission can miss some substantial emission sources (Figure 2), especially in the early stage of SO 2 control. Considering the desulfurization policy in China from 2006 [31], the emission peak in 2006 can be partly caused by statistics of mainly large SO 2 sources. On the other hand, OMI retrievals show a more rapid decline of SO 2 amount (~50%) than statistical amount (~10%) since 2008. The increase of SO 2 (~50%) in 2010-2011 can be connected with the recovery from the economic downturn in 2009 [32]. Considering that SO 2 mainly originates from coal-burning combustion of fossil fuels, the low-level SO 2 (~210,000 tons) after 2014 indicates effective effects of SO 2 control policy. The striking deviation of the annual OMI SO 2 shows large temporal variations of the SO 2 level and the variances increase obviously with the mean concentration. Although there is no coal-burning heating during winter in central China as there is in northern China, the formation of sulfate from SO 2 and their lifetime can have considerable seasonal differences due to variations of meteorological conditions such as illumination, temperature, and relative humidity (Figure 2).  Different from the spatial shift of SO 2 hotspot, NO 2 in the megacity of Wuhan was highest in Hubei during that time, where the number of vehicles exceeded three million by 2017 with rapid increases. Figure 6 gives comparison of variations of OMI NO 2 VCD and statistical NO x emissions. It can be seen that the overall NO 2 in central China increased by almost~90% to 100% from~3.2 to 5.9 × 10 15 molecules cm −2 in 2005-2011. Although the overall level of NO 2 amount since 2011 had a general downward trend as SO 2 , there were two notable increases in NO 2 during 2013 and 2017, respectively, which were not reflected in statistical NO x emissions. The large anthropogenic sources such as power and steel plants not only emitted SO 2 , but also lots of NO x . By 2011, NO x emissions of~690,000 tons were a little higher than SO 2 . Different from SO 2 emissions mainly concentrated in coal-burning sources, NO x also originated from dispersed vehicle emissions. Since 2011, the Chinese EPA began to control NO x emissions from large sources such as coal-burning plants, but the number of vehicles kept rapidly increasing. OMI NO 2 in 2017 (~3.9 × 10 15 molecules cm −2 ) indicated that NO x emissions tended to increase under the current background. It should be noted that NO x emission of 380,000 tons in 2017 was nearly twice of SO 2 (~210,000 tons), demonstrating the more important role of NO x emission in the future control. Similar as SO 2 , the yearly OMI NO 2 exhibited a large deviation from the mean. Different from the spatial shift of SO2 hotspot, NO2 in the megacity of Wuhan was highest in Hubei during that time, where the number of vehicles exceeded three million by 2017 with rapid increases. Figure 6 gives comparison of variations of OMI NO2 VCD and statistical NOx emissions. It can be seen that the overall NO2 in central China increased by almost ~90% to 100% from ~3.2 to 5.9 × 10 15 molecules cm −2 in 2005-2011. Although the overall level of NO2 amount since 2011 had a general downward trend as SO2, there were two notable increases in NO2 during 2013 and 2017, respectively, which were not reflected in statistical NOx emissions. The large anthropogenic sources such as power and steel plants not only emitted SO2, but also lots of NOx. By 2011, NOx emissions of ~690,000 tons were a little higher than SO2. Different from SO2 emissions mainly concentrated in coal-burning sources, NOx also originated from dispersed vehicle emissions. Since 2011, the Chinese EPA began to control NOx emissions from large sources such as coal-burning plants, but the number of vehicles kept rapidly increasing. OMI NO2 in 2017 (~3.9 × 10 15 molecules cm −2 ) indicated that NOx emissions tended to increase under the current background. It should be noted that NOx emission of ~380,000 tons in 2017 was nearly twice of SO2 (~210,000 tons), demonstrating the more important role of NOx emission in the future control. Similar as SO2, the yearly OMI NO2 exhibited a large deviation from the mean. To further examine spatial patterns of NO2 pollution, seasonal distribution of OMI tropospheric NO2 VCD is shown in Figure 7. The NO2 concentration was highest in winter and then in Autumn, with the lowest NO2 in summer. Since chemical reactions of NOx highly depend on ambient temperature [10], the hot and cold weather in summer and winter were the main drivers of this great seasonal contrast in NO2 levels. Moreover, the more stagnant weather in winter favored accumulation To further examine spatial patterns of NO 2 pollution, seasonal distribution of OMI tropospheric NO 2 VCD is shown in Figure 7. The NO 2 concentration was highest in winter and then in Autumn, with the lowest NO 2 in summer. Since chemical reactions of NO x highly depend on ambient temperature [10], the hot and cold weather in summer and winter were the main drivers of this great seasonal contrast in NO 2 levels. Moreover, the more stagnant weather in winter favored accumulation of NO x emissions (Figure 2). It can be seen that NO 2 hotspots (~12-13 × 10 15 molecules cm −2 ) existed over several smaller cities of Hubei during winter and exceeded~14 × 10 15 molecules cm −2 in the entire Wuhan region. There is a notable connection for high NO 2 between Hubei and adjacent areas, indicating potential transport from northern China [14,21]. By contrast, the NO 2 hotspots in central China were independent in other seasons. Compared with very low SO 2 concentration that was close to the background level in region scale by 2017, the considerable NO x emission with a slight increase can have a much larger contribution to PM 2.5 . To further examine spatial patterns of NO2 pollution, seasonal distribution of OMI tropospheric NO2 VCD is shown in Figure 7. The NO2 concentration was highest in winter and then in Autumn, with the lowest NO2 in summer. Since chemical reactions of NOx highly depend on ambient temperature [10], the hot and cold weather in summer and winter were the main drivers of this great seasonal contrast in NO2 levels. Moreover, the more stagnant weather in winter favored accumulation of NOx emissions (Figure 2). It can be seen that NO2 hotspots (~12-13 × 10 15 molecules cm −2 ) existed over several smaller cities of Hubei during winter and exceeded ~14 × 10 15 molecules cm −2 in the entire Wuhan region. There is a notable connection for high NO2 between Hubei and adjacent areas, indicating potential transport from northern China [14,21]. By contrast, the NO2 hotspots in central China were independent in other seasons. Compared with very low SO2 concentration that was close to the background level in region scale by 2017, the considerable NOx emission with a slight increase can have a much larger contribution to PM2.5.

Spatial and Temporal Distribution of Particle Pollution in 2005-2017
The spatial variations of annual MODIS AOD at 550 nm over Hubei during 2005-2017 are displayed in Figure 8. There were dramatic changes in aerosol loading over Hubei at both temporal and spatial scales. The aerosol particles were concentrated over the non-mountain areas with large spatial differences. Unlike gaseous precursors of SO 2 and NO 2 concentrated around Wuhan, particle pollution was mainly in the central part of Hubei and communicated with polluted regions in other provinces to the north and south. The high-AOD area (>0.8) grew and extended to the eastern part of Hubei in 2005, and in 2011 nearly covered the entire non-mountain area, including Wuhan. Then, the high-AOD area retreated to the central part of Hubei, which implies large changes in emission sources as well as control policies. It is worth noting that the high AOD in central part of Hubei have rapidly decreased since 2014. By 2017, the aerosol loading in non-mountain areas became homogeneous with an AOD around~0.6.
provinces to the north and south. The high-AOD area (>0.8) grew and extended to the eastern part of Hubei in 2005, and in 2011 nearly covered the entire non-mountain area, including Wuhan. Then, the high-AOD area retreated to the central part of Hubei, which implies large changes in emission sources as well as control policies. It is worth noting that the high AOD in central part of Hubei have rapidly decreased since 2014. By 2017, the aerosol loading in non-mountain areas became homogeneous with an AOD around ~0.6. Moreover, there were drastic temporal changes in aerosol loading over all the non-mountain areas of Hubei ( Figure 8). Compared with the persistent heavy aerosol loading (>0.8) in central Hubei during 2005-2011, there was an annual AOD increase from ~0.6-0.7 to ~1.0. In contrast, AOD in the eastern part (i.e., Wuhan) exhibited large yearly variations (>0.3), which indicates the obvious influence of regional transport. In addition, consistent or even larger increases in the magnitude of aerosol loading in adjacent areas of other provinces contributed to part of the heavy particle pollution Moreover, there were drastic temporal changes in aerosol loading over all the non-mountain areas of Hubei (Figure 8). Compared with the persistent heavy aerosol loading (>0.8) in central Hubei during 2005-2011, there was an annual AOD increase from~0.6-0.7 to~1.0. In contrast, AOD in the eastern part (i.e., Wuhan) exhibited large yearly variations (>0.3), which indicates the obvious influence of regional transport. In addition, consistent or even larger increases in the magnitude of aerosol loading in adjacent areas of other provinces contributed to part of the heavy particle pollution in central Hubei. The overall decrease of AOD in Hubei began in 2011 wherein AOD declined from >1.0 to~0.5-0.7. In 2017, there was more than a~30-50% decrease in AOD over most non-mountain areas.
Considering that there was no aerosol retrieval when land surface was covered by clouds, sampling frequency of available satellite AOD within one year is significant for whether its annual mean value can well represent the true variations. As shown in the available frequency of MODIS AOD (Figure 9), the annual sampling frequency of MODIS DB AOD of all quality is~120-150 days/per year, which is statistically adequate to capture the AOD gradient [33], even with some yearly difference. Moreover, long-term observations show that there are no significant changes for clouds during the last decade, except in 2013 [20]. Thus, the long-term trend of aerosol loading derived from MODIS AOD is generally representative of central China.
As shown in numerous studies, annual or seasonal mean maps of satellite AOD is usually used to denote particle pollution [19,33]. Different from the consistent distribution between gaseous precursors (NO 2 and SO 2 ) and AOD in northern China [15], the hotspots of gaseous pollutants and particles in central China are concentrated in completely different spatial locations of central China.
Although sulfate and nitrate only account for at most about half of PM 2.5 in major cities of China [11], anthropogenic emissions of SO 2 and NO x have a much larger amount than organic Carbon (OC) and black Carbon (BC) [31]. In particular, as the major control targets of the government, the overall reduction of SO 2 and NO x emission throughout China is approximately~5-10 times of that for OC and BC [4]. Thus, despite the spatial inconsistency, the substantial variability of SO 2 and NO x in central China can cause a considerable contribution to overall variations of overall aerosol loading.
Remote Sens. 2019, 11, x FOR PEER REVIEW 11 of 19 in central Hubei. The overall decrease of AOD in Hubei began in 2011 wherein AOD declined from >1.0 to ~0.5-0.7. In 2017, there was more than a ~30%-50% decrease in AOD over most non-mountain areas.
Considering that there was no aerosol retrieval when land surface was covered by clouds, sampling frequency of available satellite AOD within one year is significant for whether its annual mean value can well represent the true variations. As shown in the available frequency of MODIS AOD (Figure 9), the annual sampling frequency of MODIS DB AOD of all quality is ~120-150 days/per year, which is statistically adequate to capture the AOD gradient [33], even with some yearly difference. Moreover, long-term observations show that there are no significant changes for clouds during the last decade, except in 2013 [20]. Thus, the long-term trend of aerosol loading derived from MODIS AOD is generally representative of central China. As shown in numerous studies, annual or seasonal mean maps of satellite AOD is usually used to denote particle pollution [19,33]. Different from the consistent distribution between gaseous precursors (NO2 and SO2) and AOD in northern China [15], the hotspots of gaseous pollutants and particles in central China are concentrated in completely different spatial locations of central China. Although sulfate and nitrate only account for at most about half of PM2.5 in major cities of China [11], anthropogenic emissions of SO2 and NOx have a much larger amount than organic Carbon (OC) and black Carbon (BC) [31]. In particular, as the major control targets of the government, the overall reduction of SO2 and NOx emission throughout China is approximately ~5-10 times of that for OC and BC [4]. Thus, despite the spatial inconsistency, the substantial variability of SO2 and NOx in central China can cause a considerable contribution to overall variations of overall aerosol loading. Spatial mean of aerosol loading is usually used to evaluate variations of regional emission levels [34]. The annual average of MODIS AOD values within Hubei province during 2005-2017 are shown in Figure 10. Consistent with spatial variations, there was a slow increasing trend (0.55-0.61) in AOD during 2005-2008 with an obvious decrease (0.56) in 2009. This indicates significant changes in anthropogenic emissions or regional meteorological conditions. The increase of AOD from 2010 (0.61) was much larger and annual AOD of the entire Hubei region reached a peak value of~0.7 in 2011. Unlike northern China's growing AOD in 2013 [30], aerosol loading in Hubei continuously decreased from~0.7 in 2011 to the lowest value of~0.38 in 201. This was largely due to spatial difference in emission trends. Although the annual mean of MODIS AODs was at similar levels around 0.5 in most years, their daily values had very large fluctuations from~0.1 to~1.3, which was consistent with the frequent heavy pollution events and their cleanups in eastern China [20,21]. The particle pollution in central China is obviously lower than in northern China [19]. However, it should be noted that regional AOD exceeds 0.8 for about more than one quarter of the days during 2005-2014. After the clean air actions taken by the Chinese government in 2013, there has been a rapid decline in particle pollution in eastern China [35]. While there has been a continuous decrease in SO 2 and particle pollutants in central China, the notable increase in NO x indicates a transition from air pollution, i.e., a larger fraction of OC and enhanced O 3 pollution. regional AOD exceeds 0.8 for about more than one quarter of the days during 2005-2014. After the clean air actions taken by the Chinese government in 2013, there has been a rapid decline in particle pollution in eastern China [35]. While there has been a continuous decrease in SO2 and particle pollutants in central China, the notable increase in NOx indicates a transition from air pollution, i.e., a larger fraction of OC and enhanced O3 pollution. Seasonal mean of MODIS AOD at 550 nm showed variations of particle pollution in different meteorological conditions (Figure 11). Similar to gaseous pollution, MODIS AOD had the highest AOD in winter and autumn. MODIS AOD in central Hubei exceeded ~0.8-1.0 and influenced the northern part by adjacent transport under the prevailing northerly winds in winter [2,21]. By contrast, MODIS AOD reached moderate levels (~0.6) in the spring and summer. Different from the regional particle hotspot in winter, high AOD values were only concentrated in certain parts of central Hubei during the other seasons. Moreover, these AOD hotspots exhibited notable spatial variations in different seasons, which implies a significant intra-transport due to higher wind speeds in central China (Figure 2). The mountains around Hubei have an obvious blocking effect on outside transport Seasonal mean of MODIS AOD at 550 nm showed variations of particle pollution in different meteorological conditions (Figure 11). Similar to gaseous pollution, MODIS AOD had the highest AOD in winter and autumn. MODIS AOD in central Hubei exceeded~0.8-1.0 and influenced the northern part by adjacent transport under the prevailing northerly winds in winter [2,21]. By contrast, MODIS AOD reached moderate levels (~0.6) in the spring and summer. Different from the regional particle hotspot in winter, high AOD values were only concentrated in certain parts of central Hubei during the other seasons. Moreover, these AOD hotspots exhibited notable spatial variations in different seasons, which implies a significant intra-transport due to higher wind speeds in central China (Figure 2). The mountains around Hubei have an obvious blocking effect on outside transport of particle pollutants, even in the polluted winter. On the other hand, the limited spatial coverage of aerosol hotspots (except in winter) can partly explain the lack of gaseous pollution hotspots in central Hubei from OMI products with a coarse resolution.
Remote Sens. 2019, 11, x FOR PEER REVIEW 13 of 19 of particle pollutants, even in the polluted winter. On the other hand, the limited spatial coverage of aerosol hotspots (except in winter) can partly explain the lack of gaseous pollution hotspots in central Hubei from OMI products with a coarse resolution.

The Implication of Changing Air Pollution in Central China
Spatial variations of gaseous precursors, SO2 and NOx, can provide a visible view on their major emission sources and the effects of control measures [13]. Moreover, comparison of MODIS AOD and OMI trace gases that exhibit a significant correlation in main global pollution hotspots [36]. Although

The Implication of Changing Air Pollution in Central China
Spatial variations of gaseous precursors, SO 2 and NO x , can provide a visible view on their major emission sources and the effects of control measures [13]. Moreover, comparison of MODIS AOD and OMI trace gases that exhibit a significant correlation in main global pollution hotspots [36]. Although other components such as OC and mineral dust account for a considerable fraction (~30-60%) in PM 2.5 of China [11], sulfate and nitrate had a much larger variability in the last decade due to specific and strict control of their emissions [4]. Variations of regional AOD in Hubei showed consistent trends with SO 2 and NO 2 ( Figure 12). In particular, there was almost a linear relationship between regional mean value of MODIS AOD and OMI SO 2 with a high correlation coefficient (R), 0.84. Despite a much lower R (0.366) between AOD and NO 2 , it can be seen that variations of AOD and NO 2 had a close relation. Conversely, the connection between MODIS AOD and OMI SO 2 and NO 2 was weaker in seasonal scales. The yearly variations of seasonal meteorological conditions as well as their influence on the amount of atmospheric pollutants and satellite sampling frequency had considerable magnitude compared with the actual trends of air pollution [20,21]. The MODIS AOD in Hubei had slow changes from 2005 to 2008, with a decrease in SO 2 and increase in NO 2 . While the reduction of SO 2 emission in Hubei was~60,000 tons and growth of NO x emission approximately 150,000 tons-inferred from comparison with OMI NO 2 -it is clear that the effect of desulfurization was exceeded by the increase of NO x and other emissions.  (Figures 4 and 6). There was a persistent and rapid decline in MODIS AOD with the decrease of both SO2 and NOx from 2012. Although OMI NO2 became slightly higher (~0.5 × 10 15 molecules cm −2 ) and SO2 did not change, MODIS AOD fell by ~0.03 during 2017 as SO2 and NOx emissions. It has to be stated that other anthropogenic emission sources as well as natural dust also make important contributions to aerosol loading [3]. Different from the notable gaseous hotspot in Wuhan, high-AOD (>0.8) that converge in central Hubei mainly originated from emissions of primary aerosols and small  (Figures 4 and 6). There was a persistent and rapid decline in MODIS AOD with the decrease of both SO 2 and NO x from 2012. Although OMI NO 2 became slightly higher (~0.5 × 10 15 molecules cm −2 ) and SO 2 did not change, MODIS AOD fell by~0.03 during 2017 as SO 2 and NO x emissions. It has to be stated that other anthropogenic emission sources as well as natural dust also make important contributions to aerosol loading [3]. Different from the notable gaseous hotspot in Wuhan, high-AOD (>0.8) that converge in central Hubei mainly originated from emissions of primary aerosols and small local sources.
To have an examination on reliability and representativity of satellite observations, annual variation of the main atmospheric pollutants during 2013-2017 was analyzed based on ground measurements in Hubei ( Figure 13). It can be seen that there was a rapid decline of PM 2.5 concentration from~100 µg/m 3 in 2013 to~52 µg/m 3  Despite the large overall decrease of PM2.5, SO2, and NOx during 2013-2015, O3 pollution became more serious in early 2014. The precursors of O3 included NOx and VOCs, but there was also a notable rise in O3, as NO2 decreased in 2014-2016. Considering the nonlinear relationship between O3 and its precursors, the decrease of NO2 in high NOx/VOC ratio conditions promoted O3 formation [37]. In addition, it was found that the large reduction of aerosol loading enhanced O3 production rate to stronger photolysis [38]. By now, NOx emission has become the main gaseous pollutant in central China and increases with continuous growth in the number of motor vehicles. Therefore, it is significant to strengthen the control measures for NOx emissions as well as VOCs in the eastern and central regions of Hubei in the future.

Discussion
The spatial and temporal variations of satellite AOD, SO2, and NO2 reveal dramatic changes of anthropogenic pollutants in central China and their emission levels. OMI NO2 (~15 × 10 15 molecules cm −2 ) and SO2 (~0.5 DU) in the hotspot of Wuhan in central China is much lower than the high concentrations of NO2 (~20 × 10 15 molecules cm −2 ) and SO2 (~1.0 DU) in northern China [15,16]. Unlike the concentrated air pollution in northern China, pollution hotspots in central China have smaller scales and larger spatial variations. Temporal changes (~30%-50%) in the amount of main atmospheric pollutants in central China is similar to northern China, demonstrating that the control measure of the government is the main driver for changes of air quality in China. In particular, the  [37]. In addition, it was found that the large reduction of aerosol loading enhanced O 3 production rate to stronger photolysis [38]. By now, NO x emission has become the main gaseous pollutant in central China and increases with continuous growth in the number of motor vehicles. Therefore, it is significant to strengthen the control measures for NO x emissions as well as VOCs in the eastern and central regions of Hubei in the future.

Discussion
The spatial and temporal variations of satellite AOD, SO 2 , and NO 2 reveal dramatic changes of anthropogenic pollutants in central China and their emission levels. OMI NO 2 (~15 × 10 15 molecules cm −2 ) and SO 2 (~0.5 DU) in the hotspot of Wuhan in central China is much lower than the high concentrations of NO 2 (~20 × 10 15 molecules cm −2 ) and SO 2 (~1.0 DU) in northern China [15,16]. Unlike the concentrated air pollution in northern China, pollution hotspots in central China have smaller scales and larger spatial variations. Temporal changes (~30-50%) in the amount of main atmospheric pollutants in central China is similar to northern China, demonstrating that the control measure of the government is the main driver for changes of air quality in China. In particular, the large spatial difference of MODIS AOD has reached a close level after 2016 in all the non-mountain areas (Figure 8). For instance, in 2017, MODIS AOD ranged between 0.5-0.6 in both central and eastern Hubei.
Ground measurements from the air quality monitoring network showed similar spatial patterns of atmospheric pollutants ( Figure 14). While PM 2.5 hotspots are concentrated in central China as high values of MODIS AOD, NO 2 is mainly distributed in Wuhan. Despite the low concentration <~35 µg/m 3 in central Hubei, small SO 2 and NO 2 hotspots indicate the existence of emission sources in these small cities, which cannot be detected in OMI trace gases due largely to the small scale and limited amount. On the other hand, new problems of air pollution begin to emerge in central China due to different reduction amounts of the anthropogenic emissions. There have been great improvements in particle pollution and its gaseous precursors. Recently, rapid increases in motor vehicles has led to more NO 2 in urban regions of central China, especially in Wuhan. Spatial distribution of O 3 pollution covers both central and eastern Hubei, implying that O 3 pollution is a regional air quality problem. The photochemical pollution has become the primary air pollution in most cities, which is characterized by high O 3 centration (>160 µg/m 3 ) in the daytime of summer. Besides NO x , VOCs are another type of complicated precursor for O 3 , as they consist of numerous volatile gases. By now, only few VOCs can be measured by satellite observations with much lower accuracy than SO 2 and NO 2 [27], exerting large challenges for constraining their spatial and temporal distributions. most cities, which is characterized by high O3 centration (>160 μg/m 3 ) in the daytime of summer. Besides NOx, VOCs are another type of complicated precursor for O3, as they consist of numerous volatile gases. By now, only few VOCs can be measured by satellite observations with much lower accuracy than SO2 and NO2 [27], exerting large challenges for constraining their spatial and temporal distributions. The high consistency with ground measurements indicates that satellite observations can provide reliable information regarding the overall trends of SO2, NO2, and particles in regional scales. The distinct seasonal patterns of satellite AOD, SO2, and NO2 give further insight into the main factors of air pollution in central China. Unlike the close connection between the annual mean of AOD and SO2 and NO2 (Figure 12), their lower correlation in seasonal scales indicates that other factors such as outside transport can also have important influence on air quality in central China during certain periods.
Our results demonstrate that different control measures should be taken in view of the large spatial discrepancy of gaseous and particle pollutants in central China. It should be stated that spatial resolution of MODIS AOD and OMI gaseous products we used is 10 × 10 km and 0.25 × 0.25°, respectively, which can be too coarse to reveal point sources such as steel plants and small cities in The high consistency with ground measurements indicates that satellite observations can provide reliable information regarding the overall trends of SO 2 , NO 2 , and particles in regional scales. The distinct seasonal patterns of satellite AOD, SO 2 , and NO 2 give further insight into the main factors of air pollution in central China. Unlike the close connection between the annual mean of AOD and SO 2 and NO 2 (Figure 12), their lower correlation in seasonal scales indicates that other factors such as outside transport can also have important influence on air quality in central China during certain periods.
Our results demonstrate that different control measures should be taken in view of the large spatial discrepancy of gaseous and particle pollutants in central China. It should be stated that spatial resolution of MODIS AOD and OMI gaseous products we used is 10 × 10 km and 0.25 × 0.25 • , respectively, which can be too coarse to reveal point sources such as steel plants and small cities in central Hubei. Other than SO 2 and NO x emissions, trace gases such as ammonia (NH 3 ) and formaldehyde (HCHO) also have important contribution to air pollution in central China. Although satellite products of HCHO and NH 3 are available, their accuracy is much lower compared with NO 2 and SO 2 [27,35]. Despite much smaller variations in emissions than sulfate and nitrate [4], the long-term changes of other major components in PM 2.5 including OC and mineral dust have to be considered in central China. They cannot by be quantified by current satellite observations and have no corresponding ground measurements.
As an enhanced version of OMI, the TROPOspheric monitoring instrument (TROPOMI) onboard Sentinel-5 satellite since 2017 has a higher spatial resolution of 7 × 3.5 km, which can capture much finer emission sources [39]. Additionally, satellite observations can provide an observational constraint in emission sources for chemistry transport models [40]. To improve air quality on the regional scale, it is necessary to monitor spatial and temporal variations of emission sources and make detailed control measures timely.

Conclusions
During the last few decades, air pollution in China has drawn wide attention due to high-level anthropogenic emissions and drastic changes. However, most studies focus on economically developed and densely populated regions such as northern China, the YRD, and PRD. As an emerging pollution hotspot with diverse emission sources, air pollution in central China has been of less concern. To have a long-term insight into the spatial pattern, variations, and influence factors of atmospheric pollutants in central China, we utilized ground measurements and long-term multiple satellite observations including MODIS AOD, OMI SO 2 , and NO 2 during 2005-2017.
Satellite observation showed large spatial and temporal variations in AOD, SO 2 , and NO 2 over central China. The air quality exhibited notable seasonal differences, which was heaviest in winter and lightest in summer. There was a large overall decrease in atmospheric pollutants in central China due to continuous control measures. Different from the common collocation of gaseous and particle pollution, eastern Hubei was mainly influenced by gaseous pollutants including NO x and O 3 , whereas particle pollution was predominant in central Hubei. Despite slight recovery in 2011, OMI SO 2 in Hubei declined by more than 90%, which was close to the background level. By contrast, with a later emission control, OMI NO 2 reached its peak value in 2011 and fell back to the level of 2005 with ã 50% reduction in 2017. Thus, before 2011, MODIS AOD slowly changed due to the offset of decreased SO 2 and increased NO 2 and then sharply declined as the combined reduction of both SO 2 and NO 2 . It should be noted that there was a slight rise in OMI NO 2 during 2017, which was consistent with an increase in vehicle usage.
High consistency with ground measurements shows that satellites can reliably reflect the overall trends of atmospheric pollutants in regional scales. Annual concentration of PM 2.5 , SO 2 , and NO 2 in ground network declines~45 (45%),~40 (75%), and~17 (29%) µg/m 3 during 2013-2017, respectively. There is a high correlation (R > 0.8) between satellite SO 2 and AOD, demonstrating that reduction of SO 2 emissions has a large contribution to decrease of PM 2.5 . Despite the large overall decrease of PM 2.5 and its gaseous precursors, increased O 3 has become the main air pollution problem in eastern Hubei with growth of NO x emissions. For air quality improvement in central Hubei, primary emission of particle pollutants should be controlled. To improve air quality on a regional scale, it is necessary to monitor the distinct spatial and temporal variations of atmospheric pollutants in central China via satellite observation at finer resolution in the future.