Separating emission and meteorological contribution to PM2.5 trends over East China during 2000–2018

Abstract. The contribution of meteorology and emissions to long-term PM2.5 trends is critical for air quality management but has not yet been fully analyzed. Here, we used a combination of machine learning model, statistical model and chemical transport model to quantify the meteorological impacts on PM2.5 pollution during 2000–2018. Specifically, we first developed a two-stage machine learning PM2.5 prediction model with a synthetic minority oversampling technique to improve the satellite-based PM2.5 estimates over highly polluted days, thus allowing us to better characterize the meteorological effects on haze events. Then we used two methods, a generalized additive model (GAM) driven by the satellite-based full-coverage daily PM2.5 retrievals as well as the Weather Research and Forecasting/Community Multiscale Air Quality (WRF/CMAQ) modelling system, to examine the meteorological contribution to PM2.5. We found good agreements between the GAM model estimations and the CMAQ model estimations of meteorological contribution to PM2.5 on monthly scale (correlation coefficient between 0.53–0.72). Both methods revealed the dominant role of emission changes in the long-term trend of PM2.5 concentration in China during 2000–2018, with notable influence from the meteorological condition. The interannual trends in meteorology-associate PM2.5 were dominated by the fall and winter meteorological conditions, when regional stagnant and stable conditions were more likely to happen and haze events frequently occurred. From 2000 to 2018, the meteorological contribution became more unfavorable to PM2.5 pollution control across the North China Plain and central China, but were more beneficial to pollution control across the southern part, e.g., the Yangtze River Delta. The meteorology-adjusted PM2.5 over East China peaked at 2006 and 2011, mainly driven by the emission peaks in PM2.5 and PM2.5 precursors in these years. Although emissions dominated the long-term PM2.5 trends, the meteorology-driven anomalies also contributed −3.9 % to 2.8 % of the annual mean PM2.5 concentrations in East China estimated from the GAM model. The meteorological contributions were even higher regionally, e.g., −6.3 % to 4.9 % of the annual mean PM2.5 concentrations in the Beijing-Tianjin-Hebei region, −5.1 % to 4.3 % in the Fen-wei Plain, −4.8 % to 4.3 % in the Yangtze River Delta and −25.6 % to 12.3 % in the Pearl River Delta. Considering the remarkable meteorological effects on PM2.5 and the worsening meteorological conditions in the northern part of China where air pollution was severe and population was clustered, stricter clean air actions are needed to avoid haze events in the future.



Introduction 40
The air pollution, especially PM2.5 pollution, has become a serious problem in China in the past decades. Variations in air pollution are primarily driven by two factors: emissions and meteorology. Anthropogenic emissions dominate the long-term trend of air pollution Cheng et al., 2019a), and meteorological conditions also notably influence the daily, seasonal, interannual and interdecadal air pollution variations Chen et al., 2020;Wang et al., 2019a;Zhai et al., 2019). In China, changes in major air pollutant emissions attributable to the economic development and clean air policies 45 have been widely studied (Guan et al., 2014;Shen et al., 2017). For example, during the 11 th Five-Year Plan (2006 and the 12 th Five-Year Plan (2011, gas pollutant emissions, i.e., SO2 and NOx, have been remarkably reduced Geng et al., 2019). During "the Air Pollution Prevention and Control Action Plan" (The Action Plan, 2013-2017) and the Blue Sky Protection Campaign (2018-2020), PM2.5 emissions dropped significantly and the PM2.5 concentrations substantially decreased (Bian et al., 2019;Liu et al., 2015). Meanwhile, air pollution was also affected by the long-term trend 50 of meteorological systems and climate change, especially in the context of global warming (Ruijin et al., 2017;Wang and Chen, 2016;Yi et al., 2019;. Distinguishing the contributions of emission and meteorology is critical for the evaluation of clean air policies, projection of the future air quality and understanding pollution process. Various methods have been reported to separate the contributions of emissions and meteorology. For example, chemical transport models (CTMs) simulate the atmospheric process with emission inventory and meteorology fields as inputs, thus 55 allowing researchers to assess the changes in air pollution attributable to one factor when controlling another factor (Wang et al., 2019a;Zheng et al., 2017). CTM simulations have been widely used to separate the contributions of meteorology and anthropogenic emissions to air pollution variations. With appropriate study design, the CTM modelling system can reasonably assess the influence of a specific emission reduction measure or a specific meteorological condition on air pollution. However, these model simulations require considerable computation resources, and the quality of inputs 60 (e.g., emission inventory and meteorology) affects the quality of simulations. Moreover, due to the interactions between emissions and meteorology, the simulations in the fixed emission scenarios and the fixed meteorology scenarios may not fully reflect real-world conditions.
Other studies have applied statistical methods to assess the meteorology-associate changes in air pollution and to quantify the contribution of emissions. Multiple linear regression (MLR) has been adopted to describe the relationships between meteorology and air pollutant concentrations (Cheng et al., 2019a;Sá et al., 2015). For example, Zhai et al. (2019) constructed deseasonalized and deseasonalized-detrended time-series data and assessed the meteorological effects by MLR.
Some studies also employed machine learning algorithms to better describe the non-linear relationships between meteorology and air pollution (Grange et al., 2018;Vu et al., 2019;. However, as such methods requires continuous PM2.5 data as inputs, previous studies relied on PM2.5 ground measurements that were limited to certain locations 70 (e.g., ground monitoring stations) and times (e.g., after 2013 in China). The limited sample size not only affected the model quality and introducing sampling bias, but also hampered the analyses on spatial heterogeneity of meteorology contributions across China. The relatively short study period failed to show the long-term trend of meteorology-associate PM2.5. The analysis on the complete-coverage long-term trends of meteorology and emission contributions to air pollution is urgently needed to support further evaluation of clean air policies and region-specific air quality management within the context of 75 climate change.
In this study, we aimed to analyze the spatiotemporal trends in meteorology-and emission-associate PM2.5 variations across China during 2000-2018. The meteorological impacts on PM2.5 trends were assessed with data-fusion PM2.5 predictions and chemical transport model simulations, taking advantage of the complete spatiotemporal coverage and long data records of these two datasets. The data-fusion PM2.5 predictions were derived by combining satellite data, chemical transport model 80 simulations, ground measurements and ancillary data with an optimized two-stage machine model that improved the PM2.5 estimates during highly polluted days. Then we assessed the long-term variations in meteorology-associate PM2.5 using a generalized additive model (GAM) that better described the non-linear associations between PM2.5 and meteorology. We also estimated the meteorological impacts on PM2.5 trends with chemical transport model simulations under different scenarios coupled with a most recent emission inventory. We showed that the temporal trends of meteorology-associate PM2.5 estimated 85 from the GAM method and from the chemical transport model were highly consistent. The trend analysis of the meteorology and emission contributions to PM2.5 could support making of air quality management plans in the future.

Data and methods
This study employed simulations from the Weather Research and Forecasting/Community Multiscale Air Quality (WRF/CMAQ) modelling system as well as gridded PM2.5 predictions fused from multiple data sources to assess the 90 meteorological effects on PM2.5 (Fig. 1). The study domain covers East China (east of longitude 105°) and the PM2.5 concentrations during 2000-2018 were analyzed.

Satellite-based PM2.5 retrievals
Previously reported satellite-based PM2.5 data tended to underestimate high pollution events (Xiao et al., 2018;Xue et al., 2019) because these events rarely occurred in the model training dataset and were less characterized by the model. Since high 95 pollution events were largely affected by meteorological conditions , correctly capturing https://doi.org/10.5194/acp-2021-28 Preprint. Discussion started: 10 March 2021 c Author(s) 2021. CC BY 4.0 License. these events was critical for the assessment of meteorological contributions. Thus, we developed a two-stage model to improve the prediction accuracy of PM2.5 estimates, especially over highly polluted days, and obtained spatiotemporally continuous daily PM2.5 dataset during 2000-2018.

The two-stage prediction model
A two-stage prediction model was developed to estimate PM2.5 concentrations over China (Fig. 1). The first-stage model described high-pollution events that were underestimated in previous models, in which a synthetic minority oversampling 110 technique (SMOTE) was adopted (Torgo, 2010). The second-stage model predicted PM2.5 concentrations with the highpollution indicator from the first-stage model.
Since high-pollution events relatively rarely occur in the model training dataset, the model may not appropriately characterize the associations between high PM2.5 concentrations and the predictors, leading to underestimation of highpollution levels (Wei et al., 2020). To balance high-pollution samples and normal samples, we first defined a high-pollution 115 indicator, describing whether the daily PM2.5 observation was higher than the monthly average PM2.5 concentration plus two standard deviations at each location. A total of 3.9% of the daily data were assigned as high-pollution. Previous studies reported that balancing training data with SMOTE improved the classifiers' performance (Ghorbani and Ghousi, 2020;Saputra and Suharjito, 2019). Thus, we applied the SMOTE algorithm that oversampled the high-pollution data (the minority) by artificially generated new synthetic samples along the line between the high-pollution data and their selected 120 nearest neighbors (Chawla et al., 2002;Chawla et al., 2003). This method also under-sampled the normal data (the majority) to better balance the uneven proportion of the high-pollution and normal data. With SMOTE resampling, high-pollution data accounted for 23% in the new model training dataset.
The balanced model training dataset was adopted to train the first-stage extreme gradient boosting (XGBoost) model with all the predictors, excluding CMAQ simulations. The predicted high-pollution indicator from the first-stage model was passed 125 to the second-stage model as a predictor. We adopted the residual between the PM2.5 measurement and the CMAQ PM2. To fill any missing satellite data, in both the first-and second-stage model, we assigned the availability of satellite retrievals as a dichotomous predictor and constructed it as the cutoff point of the first layer of the decision tree to separate the training 130 data, thus mining the association between the availability of satellite retrievals and the PM2.5 concentration. This method that fills missing PM2.5 predictions with a decision tree outperformed other gap-filling methods in our previous evaluation study (Xiao et al., 2021a). The inclusion of CMAQ simulations also improved the accuracy of the gap-filled results.
The model's hyper parameter optimization and performance evaluation were conducted through five-fold CV, by-year CV and by-location CV (Appendix B). 135

Assessment of the meteorological effects on PM2.5 using GAM
Following the method described by Zhai et al. (2019), we constructed time-series data to distinguish the long-term, seasonal, and short-term trends of PM2.5 concentrations and meteorological conditions. Then the associations between PM2.5 and meteorology were fitted with a GAM model, using daily satellite-based PM2.5 predictions as dependent variable. GAM has been previously used to predict PM2.5 concentrations with meteorology and other predictors (Yanosky et al., 2014;Liu et al., 140 2009;Xiao et al., 2018). The meteorological predictors in the GAM included 10-meter wind speed, 2-meter specific humidity, 2-meter air temperature, total precipitation, 10-meter eastward wind (U wind), 10-meter northward wind (V wind), U wind at 500 hpa, and planetary boundary layer height, which have been reported to be strongly associated with PM2.5 concentrations in various regions in China (Chen et al., 2020).
Both the PM2.5 data and the meteorology data followed the same processing protocol. First, we calculated 10-day average 145 data, 50-day average data, and 19-year (2000-2018) average data based on the 50-day average data. We constructed deseasonalized-detrended data by removing the 50-day average data from the 10-day average data. We also constructed deseasonalized data by removing the 19-year average data from the 10-day average data. Assuming that the associations between PM2.5 and meteorological parameters remained constant, we estimated these associations by a grid-specific seasonal and year-round GAM model (Pearce et al., 2011) with the deseasonalized-detrended data. The GAM allows a nonlinear 150 response of PM2.5 levels to meteorological conditions, thus providing better fits to the training data (Table B1). We also fitted grid-specific seasonal stepwise MLR in a sensitivity analysis to examine whether the selection of model affects the assessment of meteorological effects. Additionally, normalized meteorological parameters were used to fit the linear regression. Hence, the estimated coefficients reflected the relative contribution of each meteorological parameter and supported the spatial analysis of the meteorological effects. Since the seasonal model attained a higher average model R 2 155 than did the year-round model (Table B1), the results obtained with the seasonal model are presented in this study.

Assessment of the meteorological effects on PM2.5 using WRF/CMAQ
We also used the WRF/CMAQ model to separate the contribution of emissions and meteorology on PM2.5 trends. The CMAQ model version 5.1 driven by the WRF model version v3.5.1 were utilized in this study, and the model configurations were following previous studies . The initial and boundary conditions for WRF were derived 160 from the National Centers for Environmental Prediction Final Analysis (NCEP-FNL) reanalysis data (NCEP, 2000). The boundary conditions for CMAQ were taken from the global GEOS-Chem model simulations. We used CB05 as the gas-phase mechanism, AERO6 as the aerosol module, and Regional Acid Deposition Model (RADM) as the aqueous-phase chemistry The evaluation of meteorological simulations of surface temperature, surface relative humidity, surface wind speed, and surface wind direction from WRF against ground-level observations from the National Climate Data Center (NCDC, ftp://ftp.ncdc.noaa.gov/pub/data/noaa/) were summarized in Fig B1. The WRF model well repoduced the near-surface temperature (r=0.98, normalized mean bias=-1.9%) and relative humidity (r=0.81, normalized mean bias=5.4%), but slightly overestimated surface wind speed (r=0.57, normalized mean bias=8.0%). The WRF simulation quality of temperature, 175 relative humidity and wind direction was consistent across years, but the simulation quality of wind speed showed slightly larger inter-annual variations. The validation results showed that the WRF simulations was acceptable to support further simulation of PM2.5 concentrations. The evaluation of PM2.5 simulations from CMAQ during the time period when ground measurements are available has been reported in our previous study . Compared to the measurements from ground monitoring stations, our model simulations well reproduced the spatial and temporal distributions of PM2.5 180 across China. Compared to the daily PM2.5 measurements in 74 cities, the CMAQ simulations obtained correlation coefficient r higher than 0.6 in 67 cities. The simulated PM2.5 decrease (30%) during 2013-2017 over China also well matched the observed PM2.5 decrease (33%).

Evaluation of the two-stage PM2.5 prediction model 185
The SMOTE resampling approach improved the prediction accuracy in the five-fold CV that the area under the curve (AUC) increased from 90.7 to 98.7 (Fig. B2). The two-stage model predictions in the five-fold CV matched the ground measurements https://doi.org/10.5194/acp-2021-28 Preprint. Discussion started: 10 March 2021 c Author(s) 2021. CC BY 4.0 License.
well with an R 2 of 0.80 and RMSE of 18.5 μg/m 3 (Fig. B2). The prediction accuracy in the by-location CV (R 2 of 0.71 and RMSE of 22.1 μg/m 3 ) and by-year CV (R 2 of 0.58 and RMSE of 27.5 μg/m 3 ) was lower than that in the five-fold CV, indicating unobserved temporal and spatial trends contributed to the PM2.5 prediction. The model performance was comparable to that 190 reported in previous studies (Xiao et al., 2018;He and Huang, 2018;Dong et al., 2020).
Specifically, compared to a benchmark model without SMOTE resampling and setting the PM2.5 concentration as the dependent variable, the two-stage model in this study better predicted high-pollution events (Fig. 2). The density distribution of the PM2.5 predictions from the two-stage model was very close to the density distribution of the PM2.5 measurements. The density distribution of the PM2.5 predictions from the benchmark model showed a higher percentage of low PM2.5 195 concentrations and a lower percentage of high PM2.5 concentrations than those revealed by the density distribution of the measurements. The greater ability of our two-stage model in capturing the daily variations in PM2.5 concentrations could better support our following analysis about meteorological impacts.  (Yin et al., 2017;Yin and Wang, 2017). We also showed that the meteorological conditions in 2014 and 2015 were more unfavorable to PM2.5 pollution control than those in 2013 over East China, as 235 previously reported Wang et al., 2019a).

Long-term trends of PM2.5 concentrations over East China
Since haze events that greatly affects public health mainly occur in fall and winter (Zhao et al., 2013), we further analyzed the meteorological effects during fall-winter (September, October, November, December, January, and February) and springsummer. The meteorological conditions in fall-winter dominated the annual meteorological effects on PM2.5. We observed typical unfavorable meteorological conditions in the fall-winter of year 2006 (2.8 μg/m 3 ) and 2016 (2.5 μg/m 3 ). In certain 240 years, e.g., 2018, the spring-summer meteorological conditions were unfavorable to pollution control, but since the fallwinter meteorological conditions were favorable, the annual meteorological effect was beneficial. The significant fall-winter meteorological effects indicated the critical contribution of meteorology to haze event formation. The fall-winter weather conditions in 2017 were substantially better than the fall-winter weather conditions in 2013, leading to a 3.3 μg/m 3 decrease in the meteorology-associate PM2.5, thereby contributed to the achievement of pollution control targets of the Action Plan 245 Yi et al., 2019). Since the current evaluation of clean air policies focuses on changes in pollution levels over short periods, e.g., three or five years, policy performance can be largely affected by meteorological changes.
favorable year (-4.9 μg/m 3 ). The shape of the interannual trend of the meteorology-associate PM2.5 during wintertime in BTH was consistent with that in previous studies. For example, the 2014 and 2017 winter meteorological conditions were greatly favourable and the 2016 winter meteorological conditions were considerably unfavorable (Yi et al., 2019;Wang and Zhang, 2020). The meteorological effects showed a regional consistency with varying magnitudes. We observed notable regional heterogeneity in the long-term trends as well as seasonal trends of the meteorological effects on PM2.5 (Fig. 5, Fig. B3). In the northern part of China, especially in the North China Plain and central East China, the meteorological conditions worsened and were adverse to pollution control (Yin and Wang, 2018;Zhang et al., 2018). 260 Multiple climate systems could be associated with the long-term trend of meteorological effects. For example, greenhouse gas-induced warming may result in a decrease in light-precipitation days and surface wind speed, which are unfavorable to pollution control . In contrast, in the southern part of China, especially in the YRD and surrounding regions, the estimated meteorological conditions were improving and were beneficial to pollution control .
Regarding the seasonal trend of the meteorological effects, in spring-summer, we observed improving meteorological effects 265 in the southern part of China and worsening meteorological effects in the northern part of China. This spatially heterogeneous trend may result from the strengthening of the East Asia summer monsoon, which enhances the transportation of aerosols from the south to the north of China (Zhu et al., 2012;Liu et al., 2017a). In fall-winter, the East Asia winter monsoon significantly affects air pollution levels that benefits the air quality in North China but is unfavourable to air quality in the South China due to the southward transport of pollutant from north to south (Jeong and Park, 2017;Yin et al., 2015). The large-scale atmospheric circulations in some specific years also showed notably distinct effects on PM2.5 concentrations 275 over the north and south of East China, due to the opposite effects on meteorology parameters. For example, in 2015 and 2016 with strong El Niño, the fall-winter meteorology in the northern part of East China was significantly unfavorable for pollution control but that in the southern part of East China was considerably favorable. One reason is that the El Niño leads to excessive precipitation over southern China that in favour of wet deposition, but weakened the East Asia winter monsoon and led to south wind anomaly, weaker surface wind, and high humidity that were favorable to pollution events in the 280 northern region of East China He et al., 2019;Chang et al., 2016). On the country, during the year with La Niña , e.g. 2007 and 2010, we estimated beneficial winter-fall meteorology in northern regions but unfavourable meteorology in the southern region . Consistent with previous studies, we also observed spatially varying associations between PM2.5 and meteorological parameters that reflect the varying PM2.5 responses to meteorological changes (Fig. B4). Temperature was positively associated 285 with PM2.5 in spring, summer and fall across East China; however, in winter, the temperature was negatively associated with PM2.5 in northern China Qiu et al., 2015) due to the low-temperature-related stable atmosphere and decreased evaporation loss of PM2.5. Humidity yielded positive effects in northern China and negative effects in southern China in all seasons, especially in winter . The spatial difference in the effects of humidity on PM2.5 may occur due to a threshold of the humidity altering the direction of the humidity influence, from hygroscopic increase to wet deposition. 290 The boundary height and precipitation were negatively associated with PM2.5 across East China in all seasons, and the effect of precipitation was greater in northern China than that in southern China (Wang and Chen, 2016). Regarding the relative contribution of the different meteorology parameters, we found that over the south coast region, temperature and humidity showed greater effects than did the boundary layer height and precipitation. In winter, humidity, boundary layer height and precipitation were critical for the PM2.5 variations in the middle and north of China. In summer and fall, the temperature and 295 humidity were critical for the PM2.5 variations across southern China. In spring, the temperature showed notable effects in the south coast region, and the precipitation exhibited large effects in the North China Plain.

PM2.5 trends after adjusting the meteorological effects
In East China, after adjusting for the meteorological influence, PM2.5 started increasing in 2000 and peaked in 2006 with an increase of 9.6 μg/m 3 compared to the 2000 level (Fig. 6). Then, the PM2.5 varied, with the second highest PM2. primarily driven by SO2 emissions, and the second PM2.5 peak was driven by NO2 and PM2.5 emissions. The PM2.5 decreasing trend after 2013 in BTH was higher than that in the other regions (5.8 μg/m 3 per year), mainly driven by the 310 emission reduction in SO2 and PM2.5. The annual average meteorology-adjusted PM2.5 concentration in BTH from 2014-2018 was consistent with that in a previous study (Qu et al., 2020). We found that the observed high-pollution events in the fall-winter of year 2006, 2013, and 2016 were partly attributable to unfavorable meteorological conditions that led to a 5.9, 3.4, and 11.1 μg/m 3 PM2.5 increase, respectively. Since the meteorology contributed up to 25% of the observed PM2. The meteorology-adjusted PM2.5 trend from 2013-2018 showed varying spatial patterns. The highest decrease occurred in Beijing, Tianjin, south of Hebei and the capital cities, including Xi'an, Wuhan, Zhengzhou, and Changsha (Fig. 7), indicating the more efficient implementation of clean air policies in these regions. As described above, the effects of meteorology also showed spatial differences. Over the Northeast China Plain, North China Plain, and Sichuan Basin, the adjusted PM2.5 decreasing trend was weaker than the observed trend. Over the Shanxi, the intersection of Hubei-Henan-Anhui and south of 340 Jiangsu, the adjusted PM2.5 decreasing trend was stronger than the observed trend. The interquartile range of the meteorological effects on the PM2.5 trend varied between -17.2% and 1.8% across East China. From 2013-2018, the decreasing trend of the meteorology-adjusted PM2.5 level was lower than that of the observed PM2.5 level by 8.4% in East China, 7.9% in the BTH region, 3.3% in the YRD, and 7.5% in the PRD while the adjusted trend was greater than the observed trend by 2.01% in the FWP. 345 https://doi.org/10.5194/acp-2021-28 Preprint. Discussion started: 10 March 2021 c Author(s) 2021. CC BY 4.0 License.

Sensitivity analysis
To evaluate whether the selection of statistical models affects the assessed associations between meteorology and PM2.5, we compared the meteorology-associate PM2.5 estimated by GAM and MLR. The estimated meteorology-associate PM2.5 levels from the MLR and GAM matched well, with correlation coefficients larger than 0.98 across East China (Fig. B5). Hence, the results of this study are robust and not affected by the selection of PM2.5-meteorology model. 350 To examine the effects of length of the time window when constructing the deseasonalized PM2.5, we conducted a sensitivity analysis with a 90-day averaging window in the BTH region, and the estimated PM2.5 concentrations after adjusting for meteorological effects were almost identical to the results using a 50-day time window (Fig. B5). Thus, this statistical method was not sensitive to the averaging time window.
Compared to previous studies, we employed the GAM to better describe the nonlinear associations between PM2.5 and 355 meteorology in this study. We observed consistent temporal trends of the meteorological effects and the meteorologically adjusted PM2.5 concentrations compared to previous studies, but the magnitude of the assessed meteorological effects and adjusted PM2.5 concentrations varied. Thus, when comparing the meteorological effects of a specific year, the conclusion may be inconsistent Zhai et al., 2019;. Assessing the meteorologyassociate PM2.5 with different methods may also lead to varying long-term trends . Several factors may 360 affect the uncertainty of the assessed meteorological contributions in this study. First, the satellite retrievals exhibited an increasing prediction error when hindcasting historical pollution levels. The satellite-driven PM2.5 prediction model used in this study is a state-of-the-art prediction model with improved prediction accuracy for high-pollution events, but the hindcast prediction quality needs to be further improved to better describe the historical PM2.5 spatiotemporal distribution. Second, we obtained meteorological information from the MERRA-2 reanalysis dataset with a spatial resolution lower than that of the 365 PM2.5 predictions. This resolution mismatch with smooth spatial variations in the meteorological fields may not fully describe the meteorological effects at the local scale.

Conclusions
In this study, we analyzed the meteorology-and emission-driven variations in the PM2.5 concentration during 2000-2018 across East China by the GAM-based method and CMAQ simulations. To support the GAM-based analysis, we combined satellite 370 data, CMAQ simulations and ground observations to predict complete-coverage PM2.5 concentrations with a two-stage machine learning model that attained improved prediction accuracy of high-pollution events. Both methods showed significant meteorological influences on PM2.5 dominated by the meteorological conditions in fall and winter. The greatly varying meteorological effects on PM2.5 concentration over a relatively short time period may remarkably affect the evaluation of clean air policies during a certain period. We also observed distinct regional differences in the long-term and seasonal trends of the 375 meteorological effects. The meteorology-associate PM2.5 tended to increase in the North China Plain and Central China, but https://doi.org/10.5194/acp-2021-28 Preprint. Discussion started: 10 March 2021 c Author(s) 2021. CC BY 4.0 License. decrease across southern China, e.g. in the YRD. After adjusting for the meteorological effects, the average PM2.5 concentration decreased 13.1 μg/m 3 from 2013-2018 over East China, and the BTH region showed the greatest decrease (28.5 μg/m 3 ) among the studied urban agglomeration regions. The decreasing trend of PM2.5 after adjusting for the meteorological effects was 8.4% weak than the observed PM2.5 decreasing trend in East China, 7.9% weak in the BTH region, 3.3% weak in the YRD, and 7.5% 380 weak in the PRD while the adjusted trend was 2.0% greater than the observed trend in the FWP. Considering the remarkable meteorological contributions to PM2.5, further emission reduction measures are required to prevent the occurrence of haze events under unfavourable meteorological conditions.

Appendix A. Data collection and processing
We collected hourly PM2.5 measurements from 2013-2018 from both the national air quality monitoring network (~1,593 385 stations) and local air quality monitoring stations (~ 1,700 stations) mainly located in East China. Continuous identical measurements over at least three hours were removed due to instrument malfunction. Daily average concentrations were calculated based on at least 12 hourly measurements.
We obtained Aqua and Terra MODIS Collection 6 level 2 aerosol products at a 0.1-degree resolution from https://ladsweb.modaps.eosdis.nasa.gov/. Since the aerosol optical depth (AOD) retrieved with the Deep Blue (DB) 390 algorithm and the Dart Target (DT) algorithm (Levy et al., 2013;Hsu et al., 2013) exhibit different coverage and retrieval accuracy , we fitted daily linear regressions to fill the missing retrievals when only DT or DB AOD was presented. Then, we calculated the average of the DT AOD and DB AOD separately for each sensor. Similarly, since the Aqua AOD and Terra AOD are observed at different pass over times, to improve the data coverage, we fitted daily linear regressions to fill the missing retrievals when only Aqua AOD or Terra AOD was presented. We calculated the average of 395 the Aqua and Terra AODs to characterize the daily aerosol loadings (Jinnagara Puttaswamy et al., 2014).
We also used daily PM2.5 simulations at a spatial resolution of 36 km during 2000-2018 from the WRF/CMAQ model as an important predictor. The inverse distance weighting (IDW) method was applied to interpolate the CMAQ simulations to match the grid of 0.1°. Detailed description of the WRF/CMAQ simulations could be found in Sect. 2.3.

Meteorological parameters were extracted from the Modern-Era Retrospective analysis for Research and Applications 400
Version 2 (MERRA-2) dataset at a resolution of 0.5° latitude × 0.625° longitude . We extracted parameters including surface albedo, cloud area fraction for low clouds, total cloud area fraction, surface net downward longwave flux, surface incoming shortwave flux, surface net downward shortwave flux, total incoming shortwave flux, total net downward shortwave flux, surface pressure, 2-meter specific humidity, 2-meter air temperature, 2-m dew point temperature, total column ozone, total column odd oxygen, total precipitable ice water, total precipitable liquid water, total 405 precipitable water vapor, 2-meter eastward wind (U wind), 2-meter northward wind (V wind), 10-meter U wind, 10-meter wind speed, 10-meter V wind, U wind at 500 hPa, U wind at 850 hPa, V wind at 500 hPa, V wind at 850 hPa, total latent https://doi.org/10.5194/acp-2021-28 Preprint. Discussion started: 10 March 2021 c Author(s) 2021. CC BY 4.0 License. energy flux, evaporation from turbulence, planetary boundary layer height, snowfall, and bias-corrected total precipitation. These parameters have been reported to be strongly associated with the PM2.5 concentration in various regions in China (Chen et al., 2020). The inverse distance weighting method was applied to estimate the daily smooth surface of 410 meteorological data and to match with the modelling grid at a 0.1° spatial resolution.
Elevation data from the Global Digital Elevation Model (GDEM, https://earthexplorer.usgs.gov/) version 2 at a 30-m resolution were averaged to match the modelling grid. We calibrated the gridded population distribution data from the LandScan Global (https://www.worldpop.org/) at the county level with the total population reported in China City Yearbooks. These calibrated gridded population data were fused to better characterize the population distribution across China (Xiao et al., 2020).The land cover classification data of urban and rural regions at a 30-m resolution for 2000-2017 were downloaded from http://data.ess.tsinghua.edu.cn (Gong et al., 2019a;Gong et al., 2019b). The fraction of urban/rural region at the 30-m resolution was averaged according to the modelling grid. 420 https://doi.org/10.5194/acp-2021-28 Preprint. Discussion started: 10 March 2021 c Author(s) 2021. CC BY 4.0 License.

Appendix B. Model performance evaluation
The hyperparameters of XGBoost, including the maximum number of boosting iterations, the learning rate, the maximum depth of a tree, the minimum sum of the instance weight needed in a child, the subsampling ratio of a training instance, and the subsampling ratio of columns when constructing each tree, were optimized by grid search with the five-fold cross-425 validation (CV) root-mean-square error (RMSE) as a performance evaluation statistic.
The model performance was evaluated through five-fold CV, by-year CV and by-location CV. The five-fold CV approach randomly selects 20% of the data for model testing and train the model with the remaining data. This process is repeated five times, and each record is selected once as testing data. The by-year CV approach validates the model hindcast ability by sequentially selecting one year of data for testing and using the remaining yearly data for model training such that each year 430 is selected once for testing. The by-location CV approach validates the model ability for spatial prediction by using the data at 20% randomly selected locations for testing and uses the remaining data for model training. This process is repeated five times until each location has been selected once for model testing.

Data availability 465
All the data used to predict PM2.5 concentrations are openly available for download from the websites given in the supplement. with contributions from all co-authors.

Competing interests
The authors declare that they have no conflict of interest.

Acknowledgments
This work was supported by the National Natural Science Foundation of China (grant no. 42007189, 42005135, 41921005, 475 and 41625020).