In order to analyze air quality changes in the THB, the observational data of hourly NO2, SO2 and PM2.5 concentrations from 2015 to 2019 were collected from the national air quality monitoring network (http://www.mee.gov.cn/, last access: 17 January 2022). The air quality observation data are under quality control, based on China's national standard of air quality observation.

The data of meteorological observations in the THB were sourced from the weather monitoring network of China Meteorological Administration (http://data.cma.cn/, last access: 17 January 2022), including air temperature (T), relative humidity (RH), sea level pressure (SLP), wind speed (WS) and precipitation (Pre), with temporal resolutions of 3 h.

To better understand the multiple timescale variations of PM2.5 and the relations to air pollutant emissions and meteorological drivers, a KZ filter (Rao and Zurbenko, 1994; Seo et al., 2018) is used to separate the daily data into multi-scale components, based on an iterative moving average that removes high-frequency variations in the data with applications in the study of air pollutants, especially O3 and PM2.5 variations (Chen et al., 2019; Ma et al., 2016; Seo et al., 2014; Zheng et al., 2020).

The KZ filter KZm,p with a length of moving average window m and number of iterations p can remove the high-frequency component of periods smaller than the effective filter width N (≥m×p1/2). The KZ filter is applicable to the time series with missing data owing to the iterative moving average process, which provides high accuracy when compared with the wavelet transform method (Eskridge et al., 1997). By comparing different sets of moving average m and number of iterations p, it was found that the decomposed time series using the KZ15,5 filter exhibited no white noise (short-term component), and the trend of the long-term component derived with the KZ365,3 filter corresponded approximately to the interannual trend of the original data (Rao and Zurbenko, 1994; Eskridge et al., 1997). Based on the spectral decompositions of the daily observational data and three components, the power spectra of daily observational data in periods of less than 33 d and longer than 632 d (1.7 years) have been well reproduced by short-term and long-term components, and the seasonal component well represents the seasonal variations, i.e., periods between 33 d and 1.7 years (Seo et al., 2018). Thus we applied KZ15,5 and KZ365,3 filters to remove the variations with periods shorter than 33 d and 1.7 years in this study.

A meteorological or environmental variable Xt observed in time series t can be decomposed into the short-term component XSTt and the baseline component XBLt presenting as 1Xt=XSTt+XBLt.

The baseline component XBL(t) is obtained by applying the KZ(15,5) filter to Xt, removing the short-term component XSTt with the temporal period shorter than 33 d from the observed data, expressed as 2XBLt=KZ15,5Xt.

The baseline component XBL(t) also can be separated into the daily climatic averages XBLclm over the study period, occupying most of the seasonality in XBL(t) and the residual ε(t): 3εt=XBLt-XBLclm.

To obtain the long-term component XLT(t) by removing the variations with the temporal period shorter than 1.7 years, the KZ365,3 filter is applied to ε(t) expressed as follows: 4XLTt=KZ365,3εt, with the short-term component 5XSTt=Xt-XBLt and the seasonal component 6XSNt=XBLt-XLTt.

The KZ filter was used to separate the daily surface PM2.5, NO2 and SO2 concentrations into short-term, seasonal and long-term components in this study. The short-term component presents a synoptic-scale variation of meteorological influences, which could control local accumulation and regional transport of air pollutants (Seo et al., 2017), partly associated with short-term fluctuations in air pollutant emissions (Russell et al., 2010). The seasonal and long-term components are attributable to the variations in air pollutant emissions related to human activities as well as the seasonal and interannual changes in meteorological conditions (Kim et al., 2018).

By altering the local accumulation, regional transport, chemical conversion and wet and dry depositions of air pollutants, the meteorological factors such as wind, RH, T, air pressure and Pre could exert significant impacts on PM2.5 changes (Sun et al., 2013; Li et al., 2018; Z. Y. Chen et al., 2020). Therefore, with the multiple factors of the baseline components of 10 m WS, 2 m RH, 2 m T, SLP and Pre calculated by Eq. (2), a multiple linear regression equation was stepwise established for the baseline component of PM2.5 as follows: 7PM2.5BLMLRt=a0+∑iaiMETBLit, where METBLit(iϵ1,5) is the baseline component of the meteorological variable i with i=1,2,3,4,5, respectively, for WSBLt, RHBLt, TBLt, SLPBLt, PreBL(t). We fit the regression coefficient ai for each meteorological variable and the intercept a0. The residual εPM2.5 between PM2.5BL and PM2.5BLMLR regressed with the multiple linear Eq. (7) is given as 8εPM2.5t=PM2.5BLt-PM2.5BLMLRt. εPM2.5 contains not only the variability of PM2.5 related to long-term changes in air pollutant emissions but also the minor seasonal change of PM2.5 attributable to unconsidered meteorological influences in the multiple linear regression. By removing the minor seasonal change from εPM2.5 with the KZ365,3 filter, the emissions-related long-term component PM2.5LTemiss(t) can be isolated as follows: 9PM2.5LTemisst=KZ365,3εPM2.5t. Here the long-term component of surface PM2.5 concentrations can be further separated into the emission- and meteorology-related long-term components with Eqs. (9) and (4) (Seo et al., 2018). Similarly, the multiple timescale variations in SO2 and NO2 with long-term variations related to changes in air pollutant emissions and meteorological drivers are decomposed by the KZ filter with multiple linear regression. Seo et al. (2018) described the details of this method.

The daily PM2.5 concentrations observed at 14 sites over the THB (Fig. 1) were decomposed into short-term, seasonal and long-term components with Eqs. (4), (5) and (6) of the KZ filter. To verify the decomposition results, the spatial distribution of the total contributions of the short-term, seasonal and long-term PM2.5 components to the total variances of observed daily changes in PM2.5 concentrations over 2015–2019 are shown in Fig. 2a. The larger the total variance, the more independent the three components are of each other (Chen et al., 2019). The sum of the long-term, seasonal and short-term components contributed 91.4 %–94.4 % of the total variance, with regional averages of 92.7 % (Fig. 2), reflecting a satisfactory verification of the KZ filtering results.

Based on the PM2.5 decomposition results of the KZ filter, the short-term, seasonal and long-term components, respectively, accounted for 34.8 %–53.8 %, 29.2 %–56.3 % and 0.2 %–9.8 % of the total variances of daily PM2.5 changes in the THB over recent years (Fig. 2b, c and d), reflecting the different patterns of multiple timescale variations of PM2.5 over this region in central China with diverse effects of emissions and meteorology. The regional contributions of the averaged short-term, seasonal and long-term components were, respectively, with 47.5 %, 41.4 % and 3.7 % of the daily PM2.5 changes over the THB (Fig. 2), and thus it could be reasonably verified that the daily variation in atmospheric pollutant was generally dominated by the short-term and seasonal components, with the long-term component determining the change trend (Ma et al., 2016; Yin et al., 2019a).

The short-term, seasonal and long-term PM2.5 components were averaged for 14 sites of the THB to characterize the temporal variations of three components in the THB for 2015–2019 (Fig. 3). The correlation coefficients of 0.05, 0.01 and 0.04 among the decomposed short-term, seasonal and long-term components were near zero, indicating the orthogonal decomposition of multiple timescale components (Eskridge et al., 1997). According to the decomposed long-term, seasonal and short-term components demonstrated in Fig. 3, the notable peaks of decomposed seasonal and short-term components were highly consistent with the peaks of PM2.5 concentrations in the original observed data, which further proved the reasonable decomposition of the multi-scale components of PM2.5 change over 2015–2019.

The observed daily PM2.5 exhibited a distinct daily variation, with an overlapping of high-frequency variations, which could be caused by mesoscale and synoptic-scale meteorological processes (Ma et al., 2016). The short-term component of PM2.5 fluctuated frequently with a significantly positive correlation to the daily change of PM2.5 (r=0.68, p<0.05), indicating the important role of the short-term component with the temporal period <33 d in the day-to-day variations of PM2.5 concentrations in the THB (Fig. 3a).

The notable peaks of PM2.5 seasonal component that emerged in winters were highly consistent with the peaks of observed daily PM2.5 concentrations (Fig. 3b). A close linkage with the significant correlation coefficient of 0.75 (p<0.05) was found between the changes of PM2.5 seasonal component and daily PM2.5 concentrations, which could reflect a significant modulation of the PM2.5 seasonal oscillations in the day-to-day variations of PM2.5, driven by the seasonal shift of East Asian summer and winter monsoons, as well as the seasonal change of anthropogenic emissions (Zhu et al., 2012; Jeong and Park, 2017). The change of the long-term component of PM2.5 exhibited a steadily declining trend over 2015–2019 (Fig. 3c), which was consistent with the interannual trend of observed PM2.5 concentrations under the sustained impact of emission controls (Zhang et al., 2019; Xu et al., 2020). The correlation coefficient (r=0.24, p<0.05) of the long-term PM2.5 component with the observed daily PM2.5 change was much smaller than those of the short-term and seasonal PM2.5 components, implying less influence of emission reduction on the daily PM2.5 change and air pollution frequency, although the declining trend in PM2.5 was determined by anthropogenic emission reduction.

In previous studies, chemical transport models and statistical methods were both used to assess the changes in air pollution attributable to emissions and meteorology (Xiao et al., 2021). Significant declines in emissions-related PM2.5 concentrations occurred in central China (Wang et al., 2019; L. Chen et al., 2020), and the meteorology offset the impact of emission reduction in typical years of unfavorable meteorological conditions (Xu et al., 2020; Gong et al., 2021). The regional averaged emission- and meteorology-related long-term components and the long-term component over the THB are displayed in Fig. S1a in the Supplement, and suggest the steadily declining trend of PM2.5 and the dominant impact of emission reduction on long-term PM2.5 changes, which is consistent with previous studies using multiple linear regression models for central China (Fig. S1b). The meteorology-related long-term component has a positive value in certain periods, implying the significant modulating effect of meteorology on PM2.5 decline in the THB.

Since the short-term variations in meteorological variables were excluded, the correlations between baseline components of PM2.5 and meteorological variables were only related to their seasonal and long-term components, affected by the regional climate of East Asian monsoons rather than synoptic-scale meteorological processes. Based on our understanding of chemical and physical processes of diffusive transport, chemical transformation, emissions and depositions of PM2.5 in the atmosphere, the dominant meteorological factors for changing PM2.5 concentrations over china are wind speed, relative humidity, air temperature, atmospheric pressure and precipitation (Z. Y. Chen et al., 2020). We examined the significant correlations between baseline components of air pollutant concentrations and selected a set of meteorological factors, including air temperature, wind speed, precipitation, relative humidity and air pressure (Tables S1–S3 in the Supplement). The meteorological parameters selected in this study are consistent with previous studies (Z. Y. Chen et al., 2020).

Generally, the baseline components of air pollutants were negatively correlated with the baseline components of wind speed (WSBL) and positively correlated with the baseline components of sea level pressure (SLPBL; Tables S1–S3), which could be attributed to the ventilation effect of wind and the stagnant conditions of meteorology in high-pressure systems, restraining the horizontal and vertical dispersions of air pollutants (Hsu and Cheng, 2016; Wang et al., 2016; Miao et al., 2017). Although wind speed exerts a negative influence of on PM2.5 concentrations over the emission source region, increasing wind speed might cause the accumulation of PM2.5 concentrations over the downwind region of emission sources (Z. Y. Chen et al., 2020), which led to the inconsistent influence of WSBL in the region of central China (Tables S1–S3). Under high surface air temperature conditions, there are strong thermal activities such as turbulence, making an accelerated dispersion of air pollutants (Y. Yang et al., 2016). The negative influence of RHBL and TBL on PM2.5BL, SO2BL and NO2BL mainly reflected the effect of the seasonal cycle in East Asian winter and summer monsoons, whereas the influence of precipitation on air pollutants was more straightforward than other meteorological parameters, negatively influencing surface pollutant concentrations through the precipitation washout of air pollutants (Tables S1–S3).

To isolate emissions-related long-term components from long-term components of PM2.5, NO2 and SO2, the stepwise multiple linear regressions of PM2.5BL, SO2BL and NO2BL, respectively, with the baseline components of the meteorological parameters (TBL, WSBL, RHBL, SLPBL and PreBL) were conducted with Eq. (7) for 14 sites, by adding and deleting meteorological variables based on their independent statistical significance to obtain the best model fit (Draper, 1998). We evaluated PM2.5BL, SO2BL and NO2BL fitted by multiple linear regression models with KZ decomposition (Table 1). The multiple linear regressions explained PM2.5BL, SO2BL and NO2BL with the adjusted determination coefficients (Adj. R2) of 0.5695–0.8093, 0.0630–0.4592 and 0.6304–0.8669, passing the confidence level of 99 % in all the THB sites, confirming the reasonable construction of the multiple linear regressions. The Adj. R2 of multiple linear regression for SO2BL were lower than those of PM2.5BL and NO2BL, which might be attributed to the larger impact of SO2 emission controls on the seasonal and long-term SO2 variations. In general, the variations of meteorological drivers can well reproduce the meteorology-related seasonal and long-term variations of PM2.5, SO2 and NO2 in the THB (Table 1).

PM2.5 consists of chemical components generated in the complex physical and chemical processes (S. Li et al., 2015). Primary particles are emitted directly from anthropogenic (e.g., industry, power plants and vehicles) and natural (e.g., outdoor biomass burning and dust storms) sources. Secondary particles (e.g., sulfate and nitrate) are converted in the chemical reactions of the precursor gases (e.g., SO2 and NOx), which are mainly produced by human activities (S. Li et al., 2015; H. Yang et al., 2016). Therefore, in addition to the reductions in primary particulate emissions, control of the secondary aerosol precursor emissions is of great importance in mitigating air pollution.

The interannual variations of the ratios in annual mean PM2.5, SO2 and NO2 concentrations relative to the annual averages in 2015 over the THB are displayed in Fig. 4. The declines of PM2.5 and SO2 in 2019 averaged over the THB were -26 % and -68 % relative to 2015, while the decrease ratio in NO2 was only -8 % over this region. The observed SO2 concentrations had a steeper decrease than PM2.5 and NO2, possibly because the dominant source sectors (i.e., power and industry) of SO2 significantly reduced their emissions (Zheng et al., 2018). The power sector was the major contributor to emission reduction but only accounted for one-third of NOx emissions and the contribution of transportation to NOx emissions was estimated to have increased over recent years (Zheng et al., 2018). The interannual variations in emissions for China were calculated from MEIC (Zheng et al., 2018), the annual total emissions of SO2 and NOx and the PM in the THB region reported by the National Bureau of Statistic of China (http://www.stats.gov.cn/tjsj/ndsj/, last access: 17 January 2022), presenting a more rapid decline of SO2 emissions in the THB than the changes of PM2.5 and NOx emissions (Fig. S2 in the Supplement). The declining trend of anthropogenic emissions estimated from emission inventories can support the explanation of the changes in air pollutant concentrations.

Figure 5 shows the spatial distributions of 5-year averaged concentrations, the linear trends and the change rates in interannual variations of PM2.5, SO2 and NO2 observed in the THB over 2015–2019. The change rates (% yr-1) were calculated with the linear trends by dividing with temporal-mean concentrations of air pollutants at the observation sites for the analysis period in Fig. 5. The 5-year averaged PM2.5 concentrations over the THB exceeded the Chinese National secondary air quality standard of 35 µg m-3 for annual mean PM2.5 concentrations (Fig. 5a), while SO2 and NO2 concentrations reached the secondary standards of 60 and 40 µg m-3 in annual mean SO2 and NO2 concentrations at most sites over the THB (Fig. 5d and g). Specifically, the 5-year averaged NO2 concentrations exceeded 40 µg m-3 in WH (Wuhan), a megacity in central China, which might be attributable to the large amounts of traffic transportation. From 2015 to 2019, both PM2.5 and SO2 decreased at all sites over the THB (Fig. 5b and e), whereas NO2 trends changed from mostly negative to positive in some sites (Fig. 5h), possibly due to the spatial disparity of NOx emissions in traffic sectors (Zheng et al., 2018). The comparison among the change rates of PM2.5, SO2 and NO2 in the THB presented the largest decreases of SO2 with -40 %–-20 % yr-1 over the 5 years (Fig. 5c, f and i), reflecting the effective control of SO2 emissions in terms of primary gaseous pollutants.

There were obvious decreases in regional mean PM2.5, SO2 and NO2 concentrations over the THB (Fig. 4), while the declining degree of PM2.5 and SO2 varied from site to site over the THB and the change trends in NO2 were weak negative and even positive in certain sites (Fig. 5c, f and i). These interannual changes of air pollutants in the THB over recent years were investigated using the emission- and meteorology-related long-term components of the air pollutants in the next sections.

The declining trend of PM2.5 in China could be partly attributed to the reduced NOx and SO2 concentrations for producing secondary aerosols (Zhang et al., 2018). The reduction rates of anthropogenic emissions markedly accelerated after 2013, decreasing by 59 % for SO2, 21 % for NOx and 33 % for PM2.5 during 2013–2017 over China (Zheng et al., 2018). In order to assess the effect of changing precursor pollutant emissions on PM2.5 reductions, we compared the linear trends of emissions-related long-term components of PM2.5, NO2 and SO2 decomposed based on Eq. (9) over the THB for 2015–2019 (Fig. 6). The distinct declining trends of emissions-related long-term PM2.5 and SO2 components, as well as the variable trends of emissions-related long-term NO2 components were distributed basically consistently with the positive and negative trends in the interannual variations of air pollutant concentrations in the THB (middle column of Fig. 5; Fig. 6), demonstrating that the local emissions of air pollutants could spatially dominate the long-term variations of air pollutants in central China, especially the increasing trends in NO2 at some THB sites.

PM2.5 concentrations are changed by emissions of both primary PM2.5 and its gaseous precursors. As major gaseous precursors, SO2 and NO2 can be oxidized to convert nitrate and sulfate for secondary PM2.5 (S. Li et al., 2015). To investigate the effects of emission reductions on the interannual variations of PM2.5, NO2 and SO2 over recent years, the ratios of change trends in long-term (kLT) and emissions-related long-term (kemiss) components of PM2.5, SO2 and NO2, in the THB over 2015–2019 were demonstrated in Fig. 7, where the long-term and emissions-related long-term components of PM2.5, SO2 and NO2 were calculated with Eqs. (4) and (9). The trend ratios kLT/kemiss<1 indicated the more obvious downward trend of emissions-related long-term variations than the long-term trend of air pollutant concentrations, which might be attributed to the offsetting effect of meteorological conditions on emission reduction in the air quality change, whereas the long-term trend of air pollutant concentrations was more significant than the emissions-related long-term trend as kLT/kemiss>1, reflecting the synchronous impacts of anthropogenic emissions and meteorology on the long-term trend in air pollutant change. In addition, the trend ratios kLT/kemiss>1 and kLT/kemiss<1 of the gaseous precursors of PM2.5, SO2 and NO2, could reflect the high and weak efficiencies of SO2 and NO2, converting to sulfate and nitrate in the production of secondary PM2.5 during air pollutant emission reduction. The notable differences in Fig. 7 were spatially distributed with the trend ratios kLT/kemiss>1 and kLT/kemiss<1 in PM2.5, SO2 and NO2 concentrations under the same meteorological conditions, indicating the different influences of emissions on the long-term variations of PM2.5, SO2 and NO2 in the THB during recent years. The reduction in PM2.5 emissions was a primary cause of the long-term declines in PM2.5 concentrations in the THB, even though the meteorological changes might offset the effects of emission reduction on air quality improvement over the southern THB (Figs. 6 and 7). It is noteworthy that the trend ratios kLT/kemiss<1 of PM2.5 were accompanied with kLT/kemiss>1 of SO2 and NO2 at the downwind southern THB sites with both negative kLT and kemiss (Fig. 7, Table S4 in the Supplement), which could imply the increasing conversion efficiency of SO2 and NO2 to sulfate and nitrate for secondary PM2.5 during the reductions of air pollutant emissions over recent years. In the upwind northern THB sites, the kLT/kemiss>1 of PM2.5 were accompanied with kLT/kemiss>1 of SO2 and NO2 with an obvious facilitating effect of meteorology on the PM2.5 decline (Fig. 7, Table S4), revealing the underlying effect of regional transport of air pollutants on the spatial distribution of conversion efficiency of gaseous precursor to secondary PM2.5.

In order to further assess the effect of gaseous precursor emissions on PM2.5 declines during recent 5-year air pollution mitigation, we selected seven and nine sites in the THB with decreasing trends of emissions-related long-term SO2 and NO2 components below -0.5 and 0.0 µg m-3 100 d-1, respectively, (Table S4) to compare the trend ratios kLT/kemiss of PM2.5, NO2 and SO2 for 2015–2019 (Fig. 8). The significantly negative linear correlations between the changes in kLT/kemiss of gaseous precursors (SO2, NO2) and PM2.5 could present the connection of kLT/kemiss>1 of NO2 and SO2 with kLT/kemiss<1 of PM2.5, which confirms the fact that the high conversion efficiency of SO2 and NO2 to sulfate and nitrate could have have offset the role of PM2.5 emission reduction in controlling PM2.5 pollution. The decreasing emissions of gaseous precursors drove faster oxidation of NO2 and SO2 to nitrate and sulfate components of PM2.5 in the source regions of air pollution in China (Zhai et al., 2021; Huang et al., 2021). This study identified the enhancing contribution of gaseous precursors to PM2.5 concentrations with reduced anthropogenic emissions of air pollutants over receptor regions in regional PM2.5 transport.

There are a few sources of uncertainty in the discussion of chemical transformation, for example in the separation of emission- and meteorology-related long-term components and in the selection of observational sites. Our results point to a better understanding of the offsetting effect of SO2 and NO2 oxidized to secondary particles in PM2.5 mitigation during the emission reduction in the THB. The long-term changes in PM2.5 are also caused by the emission variations of primary components like black and organic carbon, in addition to the chemical transformation of gaseous precursors. The difference in the emission of different primary pollutants may also lead to modifications in kLTkLTkemisskemiss of PM2.5. However, due to the current lack of long-term observations of PM2.5 components in the THB, the influence of emission variations of primary components on long-term changes in PM2.5 concentrations is not assessed in our study. Further work with long-term observational data of PM2.5 components like black and organic carbon could be conducted to quantify the influence of emissions of primary components and chemical transformation of gaseous precursors on PM2.5 changes.

As the air pollutant change trend is assumed to generally consist of emission- and meteorology-related changes (Seo et al., 2018; Yin et al., 2019b), the meteorological contribution rate Conmet to the long-term PM2.5 change trend is calculated with the following equation: 10Conmet=kLT-kemisskLT×100%. Here, Conmet (in %) is estimated with the linear trends kLT of the long-term component PM2.5LT(t) and kemiss of the emissions-related long-term component PM2.5LTemiss(t). PM2.5LT(t) and PM2.5LTemiss(t) are, respectively, calculated with Eqs. (4) and (9).

To quantitatively assess the meteorological contribution to the declining PM2.5 trends, the linear trends kLT and kemiss with the meteorological contribution rate Conmet in Eq. (10) were presented in Table S5 in the Supplement for 14 sites over the THB during 2015–2019. All the trends kLT and kemiss, respectively, in PM2.5LT(t) and PM2.5LTemiss(t) were negative over the THB (Table S5), indicating the significant effect of emission reductions on declining PM2.5 trends for improving regional air quality in central China. By comparing the declining PM2.5 trends kemiss and kLT from site to site (Table S5), the positive and negative contributions of meteorological variations to PM2.5 change trends over recent years were determined with the positive and negative differences between kemiss and kLT with the distinct meteorological influences on the change of the THB regional environment.

The spatial distribution of meteorological contribution rates Conmet regarding the long-term declining PM2.5 trend presented a unique pattern of northern positive and southern negative values over the THB (Fig. 9), with high positive contributions in the northern sites XY (61.92 %) and EZ (37.31 %), and as low negative contributions in the southern sites CD (-24.93 %) and CS (-23.03 %). It is worth mentioning that the contribution rates of meteorological variations show great spatial disparities at a small scale, i.e., EZ, HG and HS, which seems not be induced by the variation in synoptic weather or meteorological conditions. The underlying surface conditions dominate the near-surface meteorological conditions in the atmospheric boundary layer at a small scale (Wang et al., 2017). The topography and land use of HG, HS, EZ and the surrounding regions vary distinctly with the underlying surface conditions of plains, lakes and hilly areas. The underlying surface of observational sites with different near-surface meteorology effectively influences the local accumulation, chemical transformation, and dry and wet depositions of air pollutants (Bai et al., 2022). Therefore, the heterogeneity of meteorological contributions to PM2.5 at such a small spatial scale might be attributed to the local meteorological conditions in the atmospheric boundary layer, which are largely affected by the underlying surface changes.

Compared with the statistical studies using synthetic data of meteorological influence on regional PM2.5 changes in central China and the meteorological contribution from -45.5 % to 29.0 % over recent years (Gong et al., 2021; L. Chen et al., 2020), the PM2.5 pollution over the THB was affected contrarily by meteorological drivers with the northern positive and southern negative contributions from 2015 to 2019 (Fig. 9). The meteorological change could accelerate and offset the effects of emission reductions on declining PM2.5 trends in the northern and southern THB, which might be attributed to the regional transport of air pollutants conducive to the upwind diffusion and downward accumulation of air pollutants, respectively, over the northern and southern THB under the declining wind of East Asian monsoons over recent years (Hu et al., 2020; Zhong et al., 2019).

The above observational study investigated the meteorological influence on the changes in PM2.5 concentrations in the THB using the KZ filter, concluding on the large impact of meteorology on the PM2.5 changes over 2015–2019. To validate this conclusion of analyses with the KZ filter, we designed three sets of modeling experiments CTRL, SENS-MET and SENS-EMI (Table S6 in the Supplement) for December of 2015–2019, respectively, driven with the changing meteorology and anthropogenic emissions over 2015–2019, fixed meteorological conditions and anthropogenic emissions of 2015 with atmospheric chemical model WRF-Chem (Weather Research and Forecasting model with Chemistry). Air pollutant emission inventories, modeling configuration, experiment design and modeling verification are described in the Supplement. The modeling verification of experiments CTRL indicated that PM2.5 and meteorology were reasonably reproduced by the WRF-Chem simulation (Figs. S4–S5, Table S7 in the Supplement), and the three designed sets of modeling experiments CTRL, SENS-MET and SENS-EMI can be used in the further analyses of emission and meteorological impact on PM2.5 change over 2015–2019 to confirm the results of the KZ filter.

We derived the impact of meteorology by comparing the simulated PM2.5 concentrations in three sets of experiments CTRL, SENS-MET and SENS-EMI (Table S6). The relative contribution of meteorology to the interannual changes of PM2.5 concentrations was calculated with a linear additive relationship of contributions of meteorology and emission in the following equations: 11ConMET=kMETkCTRL,12ConEMI=kEMIkCTRL,13RConMET=ConMETConMET+ConEMI×100%. kCTRL,kMET and kEMI represent the trends in interannual changes of PM2.5 concentrations simulated by the experiments CTRL, SENS-MET and SENS-EMI, respectively. ConMET and ConEMI are the contributions of meteorology and emission, and RConMET is the contribution rate (%) of meteorology to interannual changes of PM2.5 concentrations (Zhang et al., 2020).

Based on WRF-Chem modeling experiments, we assessed the impact of meteorological changes on interannual PM2.5 variations from 2015 to 2019 with Eqs. (11)–(13). The relative contribution of meteorology to interannual PM2.5 variations displayed a regional pattern of northern positive and southern negative values over the THB (Fig. 10), confirming the impact of meteorological changes by accelerating and offsetting the effects of emission reductions on declining PM2.5 trends in the northern and southern THB, respectively. The general spatial distribution of meteorological contribution rates regarding declining PM2.5 trends from the WRF-Chem simulation was consistent with the results using the KZ filter (Figs. 9 and 10), validating these results and confirming that meteorological drivers exerted a contrary impact between the northern positive and southern negative contributions to long-term changes of PM2.5 concentrations in the THB.