Effects of Anthropogenic Emission Control and Meteorology Changes on the Inter-Annual Variations of PM 2.5 –AOD Relationship in China

: We identiﬁed controlling factors of the inter-annual variations of surface PM 2.5 –aerosol optical depth (AOD) relationship in China from 2006 to 2017 using a nested 3D chemical transport model—GEOS-Chem. We separated the contributions from anthropogenic emission control and meteorological changes by ﬁxing meteorology at the 2009 level and ﬁxing anthropogenic emissions at the 2006 level, respectively. Both observations and model show signiﬁcant downward trends of PM 2.5 /AOD ratio ( η , p < 0.01) in the North China Plain (NCP), the Yangtze River Delta (YRD) and the Pearl River Delta (PRD) in 2006–2017. The model suggests that the downward trends are mainly attributed to anthropogenic emission control. PM 2.5 concentration reduces faster at the surface than aloft due to the closeness of surface PM 2.5 to emission sources. The Pearson correlation coefﬁcient of surface PM 2.5 and AOD ( r PM-AOD ) shows strong inter-annual variations ( ± 27%) but no statistically signiﬁcant trends in the three regions. The inter-annual variations of r PM-AOD are mainly determined by meteorology changes. Except for the well-known effects from relative humidity, planetary boundary layer height and wind speed, we ﬁnd that temperature, tropopause pressure, surface pressure and atmospheric instability are also important meteorological elements that have a strong correlation with inter-annual variations of r PM-AOD in different seasons. This study suggests that as the PM 2.5 –AOD relationship weakens with reduction of anthropogenic emissions, validity of future retrieval of surface PM 2.5 using satellite AOD should be carefully evaluated.


Introduction
Long-term exposure to ambient fine particles (PM 2.5 ) in China causes more than 1 million early deaths every year [1,2]. To protect human health, it is critical to evaluate human exposure using high-resolution surface PM 2.5 data. However, nationwide surface in situ measurements of PM 2.5 were sparse and unavailable until 2013. Thus, studies usually retrieve surface PM 2.5 with horizontal resolution of 1-10 km using satellite aerosol optical depth (AOD) with large spatial and temporal coverage [3][4][5][6].
Accurate retrieval of surface PM 2.5 from satellite AOD requires a strong PM 2.5 -AOD relationship [7]. Studies use PM 2.5 /AOD ratio (η) and linear correlation coefficient of PM 2.5 and AOD (r PM-AOD ) to quantify the PM 2.5 -AOD relationship. Wang [8] explored the correlation between AOD from Moderate Resolution Imaging Spectroradiometer (MODIS) Most current studies focus on spatiotemporal variations of the PM 2.5 -AOD relationship in recent years, but studies on decadal trends are rare. In addition, most studies are observation-based, and thus it is difficult to separate contributions from different factors. Due to the tough clean air policies, anthropogenic emissions of SO 2 in China have declined markedly since 2006, and NOx emissions have reduced strongly after 2011, particularly after 2013 [18]. However, in 2006-2017, biomass burning emissions showed no statistically significant trends [19]. In addition, annual total biomass burning emissions of NMVOCs, NOx, NH 3 , SO 2 , BC, OC and primary PM 2.5 only account for 1-8% of the total emissions [18,19]. The objective of this work is to systematically quantify the relative contributions of anthropogenic emission control and meteorology changes to trends and the inter-annual variations of the PM 2.5 -AOD relationship in China in 2006-2017. We use a nested global 3D chemical transport model-GEOS-Chem-to simulate the PM 2.5 -AOD relationship in China. We separate the contribution from anthropogenic emissions and meteorology changes by fixing meteorology at the 2009 level and fixing anthropogenic emissions at the 2006 level, respectively. We investigate responses of the PM 2.5 -AOD relationship to anthropogenic emission changes and identify major meteorological elements that influence the inter-annual variations of the PM 2.5 -AOD link.

Observations
We used MODIS Collection 6.1 Level-3 daily mean Dark Target and Deep Blue combined AOD data at 550 nm (https://modis-atmos.gsfc.nasa.gov/MOD08_M3/index.html, accessed on 16 June 2021). Collection 6.1 modified aerosol retrieval over the land surface when urban percentage is larger than 20% using a revised surface characterization and improved surface modeling in elevated terrain (Collection 6.1 Change Document). On a global scale, the expected errors are ± (0.05 + 15%) over land for Dark Target retrievals at the 10-km spatial resolution, ± (0.03 + 21%) for arid path retrievals and ± (0.03 + 18%) for vegetated path retrievals for Deep Blue retrievals. On regional scale, 60-83% of MODIS  C6.1 AOD data are within range of ± (0.05 + 15%) in NCP [20] and 90% of MODIS C5 data fall in the range of ± (0.05 + 20%) in YRD [21]. See details in reference [22].
We used surface in situ measurements of PM 2.5 from the China Ministry of Ecology and Environment network (https://www.mee.gov.cn, accessed on 16 June 2021) with 484 sites in 2013, 670 sites in 2014 and 1498 sites in 2015-2017 ( Figure 1). PM 2.5 concentrations were determined by two methods: Thermo Scientific Continuous Ambient Particle Monitor TEOM-FDMS (Waltham, MA, USA) (about 60% of the sites) and β-gauge (the remaining 40% of the sites) with quality control (National Ambient Air Quality Standards, GB3095-2012; available at: http://english.mee.gov.cn/Resources/standards/Air_Environment/ quality_standard1/201605/t20160511_337502.shtml, accessed on 16 June 2021). PM 2.5 concentrations determined by the two methods are highly correlated (r 2 = 0.95), but the concentrations measured by TEOM equipment are 15-23% lower than those measured by β-gauge [23]. Since the measurement method used at each site was unavailable, we used available data from all sites by the two methods, bringing uncertainties to the analysis.
We used surface in situ measurements of PM2.5 from the China Ministry of Ecology and Environment network (https://www.mee.gov.cn, accessed on 16 June 2021) with 484 sites in 2013, 670 sites in 2014 and 1498 sites in 2015-2017 ( Figure 1). PM2.5 concentrations were determined by two methods: Thermo Scientific Continuous Ambient Particle Monitor TEOM-FDMS (Waltham, MA, USA) (about 60% of the sites) and β-gauge (the remaining 40% of the sites) with quality control (National Ambient Air Quality Standards, GB3095-2012; available at: http://english.mee.gov.cn/Resources/standards/Air_Environment/quality_standard1/201 605/t20160511_337502.shtml, accessed on 16 June 2021). PM2.5 concentrations determined by the two methods are highly correlated (r 2 = 0.95), but the concentrations measured by TEOM equipment are 15-23% lower than those measured by β-gauge [23]. Since the measurement method used at each site was unavailable, we used available data from all sites by the two methods, bringing uncertainties to the analysis.

Model Description
We use the 3D chemical transport model, GEOS-Chem version 11.01, to simulate surface PM2.5 and AOD in China. We use a nested model with a horizontal resolution of 0.5° latitude × 0.667° longitude over Asia and the boundary conditions were archived from global simulations at 2° latitude × 2.5° longitude (see model grids at http://wiki.seas.harvard.edu/geos-chem/index.php/GEOS-Chem_horizontal_grids, accessed on 29 August 2022). Meteorological fields are from Modern-Era Retrospective analysis for Research and Application, Version 2 (MERRA-2). We ran the model with full

Model Description
We use the 3D chemical transport model, GEOS-Chem version 11.01, to simulate surface PM 2.5 and AOD in China. We use a nested model with a horizontal resolution of 0.5 • latitude × 0.667 • longitude over Asia and the boundary conditions were archived from global simulations at 2 • latitude × 2.5 • longitude (see model grids at http://wiki.seas.harvard.edu/geos-chem/index.php/GEOS-Chem_horizontal_grids, accessed on 29 August 2022). Meteorological fields are from Modern-Era Retrospective analysis for Research and Application, Version 2 (MERRA-2). We ran the model with full gaseous chemistry and online aerosol calculations. GEOS-Chem simulates the thermodynamics of aerosols using the ISORROPIA II package [24]. The model couples aerosol and gas-phase chemistry through nitrate and ammonium partitioning [25], sulfur chemistry in clouds and aerosols [26], secondary organic aerosol formation [27,28] and uptake of acidic gases by sea salt and dust [29]. Monthly anthropogenic emissions of SO 2 , NOx, BC, OC, NMVOCs and NH 3 in Asia are from the multi-resolution emission inventory developed by Tsinghua University (Available at: http://meicmodel.org/, accessed on 29 August 2022) [18]. We updated anthropogenic emission inventories of these species in China in 2006-2017 [30]. Daily open biomass burning emissions are from the Global Fire Emissions Database, Version 4 (Available at: https://daac.ornl.gov/VEGETATION/guides/fire_emissions_v4_R1.html, accessed on 29 August 2022) [31] with horizontal resolution of 0.25 • latitude × 0.25 • longitude. Dry and wet removal of aerosols follow [32] and [33], respectively. The model simulation of AOD, surface PM 2.5 and its components are extensively validated against in situ station radiometer AOD measurements, MODIS AOD, surface in situ measurements of PM 2.5 and its components in previous studies [22,34].

Experimental Setup
We performed three experiments to quantify the contributions of anthropogenic emission control and meteorology changes to the PM 2. See details of the regions in reference [22]. PM 2.5 /AOD ratio η and Pearson correlation coefficient r PM-AOD were proved to be good parameters to quantify the PM 2.5 -AOD relationship [6,12,13,36]. The former is a conversion factor [37] and indicates the dry mass PM 2.5 concentration per unit aerosol optical thickness. The latter indicates the strength and direction of the linear relationship between surface dry mass PM 2.5 and AOD. A previous study showed that stronger PM 2.5 -AOD relationship produces better surface PM 2.5 retrieval [12]. We archived daily mean PM 2.5 concentration and AOD data from GEOS-Chem runs in the three experiments. We estimated the daily η (η = PM 2.5 _daily/AOD_daily) in each model grid first and then estimated the monthly, seasonal and annual mean in each region. We estimated r PM-AOD using daily mean PM 2.5 and AOD in each model grid in each month, season and year and then estimated the mean value in each region.

Observed and Simulated Long-Term Trends of PM 2.5 -AOD Relationship
Observations show that the ratios of η observed at the in situ sites in 2013-2017 vary with seasons. The largest η is in winter (114-212 µg m −3 ) and the smallest in summer (44-61 µg m −3 , Figure 2). This is possibly explained by several reasons. First, anthropogenic emissions in winter are 30% larger than those in summer in NCP, while the differences in YRD and PRD are within 8%. Thus, the ratio in NCP in winter is higher than those in other seasons and regions. Second, stable stratification in winter confined surface emissions to the boundary layer and enhances surface PM 2.5 concentration. Simulated surface PM 2.5 concentration in winter is consistently 16-54% larger than those in summer in the three regions. Third, aerosol loading in NCP in summer is 25-43% larger than that in winter. In YRD and PRD, aerosol loading in summer is also smaller than that in winter, but the difference is smaller than those of surface PM 2.5 concentrations. Fourth, the simulated hygroscopic factors of different species in summer are 1-45% larger than those in winter, enhancing AOD in summer. fastest decline in summer (−9.1, −10.3 and −2.6% year ) and followed by those in fall (−7.0, −5.9 and −4.6% year −1 ). In 2006-2017, the simulated η show significant decreasing rates of −1.2, −0.7 and −1.4% year −1 in NCP, YRD and PRD (p-value < 0.01), respectively. Different from trends of AOD and surface PM2.5 [22], the difference of reduction rates of η before and after 2013 are much smaller ( Figure 3). Specifically, the simulated reduction rates of η in 2013-2017 are smaller than those in 2006-2013 by 16% in NCP, but the rates in YRD and PRD in 2013-2017 are 11% and 100% larger than those in 2006-2013.  Observations show that rPM-AOD in 2013-2017 is decreasing in the three key regions, but the trends are statistically insignificant (p-value > 0.76). These trends are in general agreement with recent studies [12]. GEOS-Chem reproduces the inter-annual variations of rPM-AOD with a bias of −33-222% (Figure 4). The model overestimates rPM-AOD for the annual mean and in spring-fall. The overestimate is possibly because the model does not resolve AOD from coarse particles. In contrast, the model underestimates rPM-AOD in winter, possibly due to the overestimated isolation of the boundary layer by the model [22]. rPM-AOD shows no significant trends in the three key regions in 2006-2017, but the inter-annual variations are substantial ( Figure 5). The rPM-AOD values vary by ±27% in the 12 years in spring-fall. In winter, rPM-AOD varies between −0.47 and 0.38. Observations show that r PM-AOD in 2013-2017 is decreasing in the three key regions, but the trends are statistically insignificant (p-value > 0.76). These trends are in general agreement with recent studies [12]. GEOS-Chem reproduces the inter-annual variations of r PM-AOD with a bias of −33-222% (Figure 4). The model overestimates r PM-AOD for the annual mean and in spring-fall. The overestimate is possibly because the model does not resolve AOD from coarse particles. In contrast, the model underestimates r PM-AOD in winter, possibly due to the overestimated isolation of the boundary layer by the model [22].

Contributions of Anthropogenic Emission Control and Meteorology Changes to PM 2.5 -AOD Relationship
The decrease of η in 2006-2017 is mainly attributed to anthropogenic emission changes (Table 1). Specifically, meteorology changes tend to increase η in NCP and YRD, but contribute 10% to the reduction of η in PRD in FIXMET. In addition, η in BASE correlate stronger with η in FIXMET (0.70 < r < 0.87) than those in FIXEMISS (0.26 < r < 0.64). The downward trends of surface PM 2.5 and AOD in recent years are also attributed to anthropogenic emission changes [38,39]. GEOS-Chem suggests that the annual mean surface PM 2.5 decreases faster than AOD by 68% in NCP, 59% in YRD and 72% in PRD in 2006-2017 in FIXMET.

Contributions of Anthropogenic Emission Control and Meteorology Changes to PM2.5-AOD Relationship
The decrease of η in 2006-2017 is mainly attributed to anthropogenic emission changes (Table 1). Specifically, meteorology changes tend to increase η in NCP and YRD, but contribute 10% to the reduction of η in PRD in FIXMET. In addition, η in BASE correlate stronger with η in FIXMET (0.70 < r < 0.87) than those in FIXEMISS (0.26 < r < 0.64). The downward trends of surface PM2.5 and AOD in recent years are also attributed to anthropogenic emission changes [38,39]. GEOS-Chem suggests that the annual mean surface PM2.5 decreases faster than AOD by 68% in NCP, 59% in YRD and 72% in PRD in 2006-2017 in FIXMET. On a seasonal scale, the downward trends of η (BASE) are also attributed to anthropogenic emission reductions ( Table 1). The downward trends of η in FIXMET are statistically significant in the four seasons. Meteorology changes increase η in spring and winter, but decrease η in summer and fall in NCP, and show limited effects on trends of η in other regions. In FIXMET in NCP, AOD decreases at the rate of 0.8% year −1 in spring, but surface PM 2.5 shows no trends; thus, η increases. The inter-annual variations of AOD in this region are controlled by temperature and vertical air movement at 850 hPa and surface RH [22]. However, none of these meteorological elements showed statistically significant trends over the 12 years. The weakening of the East Asian summer monsoon enhances aerosol concentrations but AOD increase (0.9% year −1 ) faster than surface PM 2.5 (0.1% year −1 ), producing a negative trend of η. Similar upward trends are observed for fall (AOD: 1.4% year −1 (p-value < 0.1); surface PM 2.5 : 0.3% year −1 ). The strong enhancement of Remote Sens. 2022, 14, 4683 9 of 16 AOD is related to the decreased potential vorticity (−0.02 PVU year −1 , p-value < 0.05) and the increased RH at 850 hPa (0.002 year −1 ). In winter, AOD decreases significantly over the 12 years (−1.2% year −1 , p-value < 0.01), but surface PM 2.5 increases; thus, η increases. The strong decrease of AOD is attributed to the significant increase of northerly wind speed at 850 hPa (0.15 m s −1 year −1 , p-value < 0.1). The inter-annual variations of η are also strongly affected by meteorology changes on the seasonal scale. H in BASE correlate stronger with η in FIXEMISS than those in FIXMET in fall and winter in the three regions (Table 2).    Table 3). Moreover, the inter-annual variations of annual r PM-AOD caused by meteorology changes (−14%-+7%) are much larger than those caused by anthropogenic emission changes (−5%-+3%). On the seasonal scale, r PM-AOD in BASE also correlate stronger to r PM-AOD in FIXEMISS than those in FIXMET, similar to the comparison on the annual scale (Table 3).

Responses of PM 2.5 /AOD Ratios to Anthropogenic Emission Changes (FIXMET)
AOD is determined by both aerosol loading and the hygroscopic growth factor from the surface to the top of the atmosphere ( [22], Section 2.1). GEOS-Chem shows that in FIXMET the hygroscopic growth factors do not change over the years, and the decrease of η in 2006-2017 is mainly due to faster decrease of PM 2.5 at the surface than aloft ( Figure 6). We estimate the reduction rates of PM 2.5 in 2006-2017 ((PM 2.5_2017 -PM 2.5_2006 )/PM 2.5_2006 ) at various heights from the surface to 500 hPa. The reduction rates of PM 2.5 at 800 hPa (500 hPa) are 7% (48%), 5% (47%) and 20% (55%) smaller than those at the surface in NCP, YRD and PRD, respectively. The largest difference in reduction rates between the surface and aloft is from OA and the ratio decreases with increasing altitude monotonically. In contrast, the reduction rate of sulfate-nitrate-ammonium (SNA) increases slightly below 800 hPa in NCP and YRD ( Figure 6, see model validation of PM 2.5 components in [22,34]).
GEOS-Chem shows that reduction rates of OA in surface PM 2.5 are slightly larger than those of AOD OA in winter (<25%), and are markedly larger (48-81%) than those of AOD OA in summer. The reason is that OA reduction rates are decreasing with increasing height both in summer and winter ( Figure 6), but at a faster rate in winter due to stable stratification and lower PBLH. In contrast, reduction rates of PM 2.5_SNA are slightly larger than AOD SNA in summer (by up to 8%), but are 1-38% smaller in winter. We find that reduction rates of SNA are decreasing with increasing altitude in summer, but the trend is the opposite in winter below 850 hPa. The model shows that in winter, concentration of NO 3 − is increasing at a faster rate at the surface than aloft. In addition, the ratio of NO 3 − /SNA decreases quickly with increasing height (e.g., NCP in winter: surface: 57%; 750 hPa: 18%). Thus, the resulting total reduction rates of SNA increase with increasing height in winter. The unfavorable chemical processes that buffer NO 3 − reduction in winter have been widely observed and simulated [40,41]. Very few studies have investigated the vertical distribution of PM 2.5 components in China. The authors in [42] showed that NO 2 is the most important factor that determines the vertical profile of

Meteorological Elements That Influence the Correlation of PM 2.5 and AOD (FIXEMISS)
We estimated the correlation coefficient of r PM-AOD (correlation coefficient of daily mean PM 2.5 and AOD in each month in 2006-2017) and monthly mean meteorological elements in each season. The meteorological elements are from MERRA-2 reanalysis data and include temperature (T), an east-west wind component (U), a north-south wind component (V), vertical air movement (O), relative humidity (RH), potential vorticity (PV) at the surface, 850 hPa and 500 hPa, and tropopause pressure (TROPPT), pressure at the surface (PS) and sea level pressure (SLP).
Meteorological elements that have strong correlation with r PM-AOD vary with regions and seasons (Table 4). T at the surface, 850 hPa and 500 hPa are strongly correlated with r PM-AOD with correlation coefficients of 0.72-0.88 in NCP and YRD in spring. T surface is also positively related to r PM-AOD in YRD and PRD in fall, and to r PM-AOD in PRD in winter. Higher T is usually related to stronger vertical mixing, thus a larger correlation of the surface PM 2.5 and column AOD.  (Table 4). Zonal wind is positively related to r PM-AOD in NCP but negatively related to r PM-AOD in YRD and PRD. Meridional wind is positively related to r PM-AOD in YRD and negatively related to r PM-AOD in NCP. The positive or negative correlation coefficients are attributed to the wind direction. For example, U 500hPa is positive (from west to east) in the three regions and is consistently negatively related to r PM-AOD in YRD and PRD. Faster wind at 500 hPa blows aerosols away and decreases the correlation of surface PM 2.5 and AOD in the column. In contrast, U 850hPa in summer and U surface in winter in NCP are negative (from east to west) in 1/3 of the 36 months and are positively related to r PM-AOD in NCP. We find that O are positively related to r PM-AOD in NCP and YRD in winter, but negatively related to r PM-AOD in PRD in summer and winter. In the former two regions, the vertical air movement is upward, thus larger O means stronger mixing and larger r PM-AOD . In PRD, the vertical movement is downward, thus larger O means stronger isolation between the surface and aloft and smaller r PM-AOD .
PS and SLP are negatively related to r PM-AOD in the three regions in spring, summer and fall (Table 4). Air flows up and together in a low-pressure system, enhancing vertical mixing. Lower PS means stronger mixing and larger r PM-AOD . dT, dV and dU are indicators of atmospheric stability, thus, they are mostly positively related to r PM-AOD in the three regions.
We investigated the correlation of r PM-AOD and RH and PBLH in every season. In spring, RH 500hPa is positively related to r PM-AOD in PRD (r = 0.77), but shows weaker relation in NCP and YRD. RH surface and RH 850hPa have relatively weaker relations with r PM-AOD in the three regions. In summer, RH has relatively weaker correlation with r PM-AOD in NCP and YRD than in PRD. This is in general agreement with observations, which showed that RH correction increased the r PM-AOD in the three regions with the largest percentage increase in PRD [15]. PBLH is positively related to r PM-AOD in NCP (r = 0.61), but is relatively weakly related to r PM-AOD in other seasons and other regions (−0.33 < r < 0.33). This is in general agreement with a recent study, which showed that PBLH-PM correlations are stronger in polluted regions than in clean regions [43]. PBLH correction deteriorates the correlation in YRD and PRD in spring [15].

Discussion
We use model experiments to separate contributions from anthropogenic emission control and meteorology changes to the PM 2.5 -AOD relationship. We find that η decreased significantly in 2006-2017, due mainly to anthropogenic emission control. With further reduction of anthropogenic emissions in the future, the PM 2.5 -AOD relation is predicted to become weaker. Previous observation-based studies also detected weakening trends of the PM 2.5 -AOD relationship in the last five years [12]. However, it was difficult to investigate reasons for the trends based only on observations. GEOS-Chem simulation showed that r PM-AOD showed no statistically significant trends but large inter-annual variations. Meteorological elements are critical in explaining the inter-annual variations of r PM-AOD , such as T, U, V, O, PS, atmospheric instability, RH and PBLH. Among these elements, RH and PBLH were well discussed in previous observation-based studies. Using correction of RH and PBLH improves the correlation of monthly PM 2.5 and AOD in Beijing in 2011-2015 from 0.63 to 0.76 [9]. The authors in [17] suggested correcting surface PM 2.5 retrieval using PBLH in northwest China. RH tends to weaken the r PM-AOD regardless of geographical location [13]. Corrected by RH and PBLH, r PM-AOD increased in most regions but decreased in a few of the 368 cities in China [15]. r PM-AOD decreases with increasing surface wind speed [15]. Other meteorological elements were rarely discussed. However, this study shows that T, PS and atmospheric instability are also important to the variations of r PM-AOD , and should be considered in future research.
Despite the strong relation between surface PM 2.5 and AOD, they show a lot of differences. First, surface PM 2.5 and AOD show completely different seasonality [22,36]. Second, surface PM 2.5 and AOD respond differently to emission changes. With the anthropogenic emission changes in 2006-2017, fractional reduction rates of surface PM 2.5 are larger than AOD. Third, influences of meteorology changes on the inter-annual variation of AOD are larger than that of surface PM 2.5 [22]. Fourth, despite steady improvement of data quality, uncertainties of AOD values obtained by space-borne remote sensors are so large that they can hardly be used to detect the long-term variations [44]. Even for a global mean quantity, the discrepancies among different products exceed the signal of inter-annual variability [44]. On regional scales, the uncertainties are much larger and more complex. MODIS Terra and Aqua show opposite trends (Terra: −0.009 yr −1 ; Aqua: +0.0012 yr −1 ) in China in 2001-2011, and both are statistically significant at 95% confidence level [45]. Lastly, studies showed that weaker PM-AOD relationship deteriorate PM 2.5 retrieval. The authors in [12] showed that adjusted R 2 of PM 2.5 retrieval decreased from 0.87 in 2013 to 0.69 in 2017 owing to the weakening of the PM-AOD relationship. With the strong reduction of surface PM 2.5 in recent years and in the future, the PM 2.5 -AOD relationship becomes weaker and the retrieval of PM 2.5 becomes worse [12]. The predictability of surface PM 2.5 using space-borne AOD needs further validation.

Conclusions
We studied the PM 2.5 -AOD relationship in NCP, YRD and PRD in China using a nested 3D chemical transport model-GEOS-Chem. We separated the contributions from anthropogenic emission control and meteorology changes by fixing meteorology at the 2009 level and fixing anthropogenic emissions at the 2006 level, respectively. We found that η was decreasing in 2006-2017, but r PM-AOD showed no statistically significant trends. The decrease of η was determined to be caused by anthropogenic emission changes. The vertical distribution of reduction rates varies with seasons and PM 2.5 components. In summer, all components reduce slower with increasing height, while in winter the reduction rate of SNA increases first and then decreases. The overall effect of the different trends of different components is that PM 2.5 concentration decreases slower at higher altitude than at the surface. The inter-annual variations of r PM-AOD were mainly determined by meteorology changes. We found that major meteorological elements that have strong correlation with r PM-AOD vary with regions and seasons. T was positively related to r PM-AOD in the three regions and was particularly important in spring and fall. Horizontal wind speed and vertical air movement show a strong correlation with r PM-AOD . PS is mostly negatively related to r PM-AOD , while atmospheric instability is positively related to r PM-AOD . RH is negatively related to r PM-AOD in NCP and YRD in fall and winter, but is positively related to r PM-AOD in PRD in spring and summer. PBLH is positively related to NCP in fall and negatively related to YRD in winter and PRD in spring. This study suggests using other meteorological elements mentioned above when analyzing the PM 2.5 -AOD relationship or retrieving surface PM 2.5 using satellite AOD. In addition, as the PM 2.5 -AOD relationship weakens with decreasing anthropogenic emissions, validity of remote-sensing surface PM 2.5 retrieval should be regularly evaluated.