Machine Learning Explains Long-Term Trend and Health Risk of Air Pollution during 2015–2022 in a Coastal City in Eastern China

Exposure to air pollution is one of the greatest environmental risks for human health. Air pollution level is significantly driven by anthropogenic emissions and meteorological conditions. To protect people from air pollutants, China has implemented clean air actions to reduce anthropogenic emissions, which has led to rapid improvement in air quality over China. Here, we evaluated the impact of anthropogenic emissions and meteorological conditions on trends in air pollutants in a coastal city (Lianyungang) in eastern China from 2015 to 2022 based on a random forest model. The annual mean concentration of observed air pollutants, including fine particles, inhalable particles, sulfur dioxide, nitrogen dioxide, and carbon monoxide, presented significant decreasing trends during 2015–2022, with dominant contributions (55–75%) by anthropogenic emission reduction. An increasing trend in ozone was observed with an important contribution (28%) by anthropogenic emissions. The impact of meteorological conditions on air pollution showed significant seasonality. For instance, the negative impact on aerosol pollution occurred during cold months, while the positive impact was in warm months. Health-risk-based air quality decreased by approximately 40% in 8 years, for which anthropogenic emission made a major contribution (93%).


Introduction
Air pollution is one of the most important global environmental problems, which can significantly affect human health [1]. Air pollutants can be generally classified into two phases, i.e., gases including ozone (O 3 ), nitrogen oxide (NOx), carbon monoxide (CO), sulfur dioxide (SO 2 ), and volatile organic compounds (VOCs), and particles, such as fine particulate matter (PM 2.5 ) and inhalable particulate matter (PM 10 ) [2,3]. Exposure to ambient particulate matter is one of the most important health risk factors [4]. It was estimated that approximately 4.2 million deaths were attributed to particulate matter air pollution all over the world in 2015 [1].These air pollutants can be generated from natural [5] and/or anthropogenic [6,7] sources [2]. Concentrations of air pollutants in ambient air are affected by two major factors, including emissions and meteorological conditions [8][9][10][11]. It is essential to evaluate the impact of these two factors on the variability of air pollution in the atmosphere for making emission control policies.
Relatively serious air pollution is mainly distributed in developing and populous countries, such as China. Air pollution is generally characterized by complicated mixtures along In addition to emissions, meteorological conditions also play an important role in affecting air quality, especially in the typical coastal regions. However, the impact of meteorological conditions on air quality at a typical coastal region was not well quantified. In this study, we applied a machine learning approach to evaluate the impact of anthropogenic emissions and meteorology on trends in air quality during 2015-2022 in a coastal city of eastern China (i.e., Lianyungang). Additionally, drivers of health-risk air quality index (HAQI) and the corresponding premature mortality from 2015 to 2022 were also investigated.

Data Source
In the present study, the hourly mass concentration of air pollutants, including PM 2.5 , PM 10 , CO, NO 2 , O 3 , and SO 2 , in Lianyungang, which is a typical coastal city in eastern China, was obtained from the China National Environmental Monitoring Center network (https://quotsoft.net/air, last access 10 March 2023). The meteorological parameters were taken from ERA5 reanalysis data of the European Centre for Medium-Range Numerical Prediction (ECMWF), which is accessible at https://cds.climate.copernicus.eu/ (last access 10 January 2023). The meteorological data include zonal wind speed (U10), meridional wind speed (V10), temperature (T), boundary layer height (BLH), solar radiation (SR), sea level pressure (SP), total cloud coverage (TCC), total precipitation (TP), relative humidity (RH), zonal wind speed of 500 hpa (U500), meridional wind speed of 500 hpa (V500), 500 hpa vertical wind speed (W500), 850 hpa zonal wind speed (U_850), 850 hpa meridional wind speed (V850), and 850 hpa vertical wind speed (W850). The horizontal and temporal resolutions of these meteorological data were 0.25 degree and 1 h, respectively. The geographic location of Lianyungang is shown in Figure 1.
Toxics 2023, 11, x FOR PEER REVIEW 4 of 14 deweathered air pollutant, respectively, where i refers to individual month. The model configuration used in this study is consistent with previous studies [35]. In this work, the RF modeling analysis was performed using a random forest R package based on the R language [43].

Calculation of Health-Risk-Based AQI (HAQI)
In this paper, we also calculated the HAQI value to evaluate the health impact of air pollution. A detailed calculation method can be found in Hu et al., (2015) [44]. Meanwhile, we also evaluated the impact of anthropogenic emissions and meteorological conditions on changes in HAQI from 2015 to 2022. More discussion will be given in the next section. Relative risk (RR) was used to estimate the health effect for the six air pollutants (including PM2.5, PM10, NO2, SO2, CO, and O3) by using Equation (2). The RR based on the equivalent concentration of the pollutants (RR ) was calculated by Equation (3): where β refers to the exposure-response relationship coefficient, C represents the concentration of each pollutant, and C0 is the threshold concentration, below which the pollutant proves no obvious adverse health effects: where ER refers to the sum of the total excess risk for simultaneous exposure to the six air pollutants. Equation (4) introduces the equivalent pollutant concentration of the ith criteria pollutant (C ′): where β represents the exposure-response relationship coefficient of the ith pollutant, and C , is the threshold concentration of the ith pollutant. Then C can be used to calculate the equivalent concentration of the ith criteria pollutant (HAQI ), which is shown in Equations (5) and (6):

Meteorological Normalization Using RF Model
In the present study, a meteorological normalization approach based on the RF algorithm [35] was applied to separate the impact of anthropogenic emissions and meteorological conditions on trends in air pollutants, including PM 2.5 , PM 10 , CO, NO 2 , O 3 , and SO 2 , respectively. A more detailed description of such meteorological normalization approach can be found elsewhere [35][36][37]42]. The daily average data were used for the RF model analysis, which was consistent with Grange et al., (2018) [35]. Briefly, the prediction features for the RF model included time variables (i.e., proxy of emissions) and meteorological parameters. To explain long-term, seasonal, and weekly variations in anthropogenic emissions, three corresponding time variables were used here, which were Unix time, Julian day (day of year), and weekdays (day of week), respectively. The RF model prediction features for the meteorological field include U10, V10, T, BLH, SR, TCC, TP, RH, U500, V500, W500, U850, V850, and W850, which could explain the impact of these meteorological parameters on variations in air pollution. The day of the week variable was a categorical variable, while all the other variables were numeric variables. In practice, a specific RF model for an individual air pollutant (e.g., PM 2.5 , PM 10 , CO, NO 2 , O 3 , or SO 2 ) was first established. The 70% and 30% of the input data were divided as training and testing data sets, respectively. In this processing, the RF model's built-in importance indexes could be obtained. In addition, a fivefold cross validation was applied to evaluate the model performance. The normalized mean square error (NMSE), root mean square error (RMSE), and correlation coefficient between observation and prediction were calculated for evaluation metrics.
To obtain weather-normalized (so-called deweathered) prediction, the total 1000-time prediction was achieved. To do so, the model input data for daily meteorological variable features were randomly resampled from the historical weather data during 2001-2022, while time variables were resampled. This process was repeated in the corresponding 1000 predictions. For each processing, a specific model prediction result was then taken. The final deweathered concentration for the specific air pollutant was obtained by aggregating the 1000 predictions using arithmetic mean.
In addition, we applied a meteorological impact index method to evaluate the extent of meteorological influence on the concentrations of air pollution in different years and/or months. This method has been described in a previous study [34]. Briefly, the relative difference in the monthly mean concentration of an individually observed and deweathered air pollutant was due to the meteorological impact. The meteorological (MET) impact index could be calculated using Equation (1). In Equation (1), C i,observed and C i,deweathered represent the monthly mean concentrations of an observed and deweathered air pollutant, respectively, where i refers to individual month. The model configuration used in this study is consistent with previous studies [35]. In this work, the RF modeling analysis was performed using a random forest R package based on the R language [43].

Calculation of Health-Risk-Based AQI (HAQI)
In this paper, we also calculated the HAQI value to evaluate the health impact of air pollution. A detailed calculation method can be found in Hu et al., (2015) [44]. Meanwhile, we also evaluated the impact of anthropogenic emissions and meteorological conditions on changes in HAQI from 2015 to 2022. More discussion will be given in the next section. Relative risk (RR) was used to estimate the health effect for the six air pollutants (including PM 2.5 , PM 10 , NO 2 , SO 2 , CO, and O 3 ) by using Equation (2). The RR based on the equivalent concentration of the pollutants RR was calculated by Equation (3): where β refers to the exposure-response relationship coefficient, C represents the concentration of each pollutant, and C 0 is the threshold concentration, below which the pollutant proves no obvious adverse health effects: where ER Total refers to the sum of the total excess risk for simultaneous exposure to the six air pollutants. Equation (4) introduces the equivalent pollutant concentration of the ith criteria pollutant (C i ): where β i represents the exposure-response relationship coefficient of the ith pollutant, and C 0,i is the threshold concentration of the ith pollutant. Then C i can be used to calculate the equivalent concentration of the ith criteria pollutant (HAQI i ), which is shown in Equations (5) and (6): where j is the health category index; C i,j and C i,j−1 represent the upper-limit concentrations for the jth and j-1th health categories; and AQI i,j and AQI i,j−1 refer to the air quality index of the pollutant that corresponds to C i,j and C i,j−1 . Finally, Equation (7) shows that the overall HAQI is determined by the maximum of all HAQI i s. HAQI i = max(HAQI 1 , HAQI 2 , . . . , HAQI n ), n = 1, 2, . . . , 6.

Calculation of Premature Mortality (M)
The premature mortality (M) attributable to PM 2.5 and O 3 can reflect health effects of atmospheric pollutants. A detailed calculation method can be found in Apte et al., (2015) [45]. Briefly, the premature mortality is calculated in Equation (8): where Y o refers to the baseline mortality rate corresponding to a particular disease category in regions, which can be found in the Statistical Yearbook of Public Health and Family Planning in China. Pop represents the population for this region in 2015. RR has been calculated in Equation (2), and RR−1 RR here refers to the attribution fraction, which is attributed to respiratory disease linked to O 3 and stroke, ischemic heart disease (IHD), chronic obstructive pulmonary disease (COPD), and lung cancer (LC) linked to PM 2.5 . HAQI = max(HAQI , HAQI , … , HAQI ) , n = 1, 2, … , 6.

Calculation of Premature Mortality (M)
The premature mortality (M) attributable to PM2.5 and O3 can reflect health effects atmospheric pollutants. A detailed calculation method can be found in Apte et al.,(201 [45]. Briefly, the premature mortality is calculated in Equation (8): where Y refers to the baseline mortality rate corresponding to a particular disea category in regions, which can be found in the Statistical Yearbook of Public Health an Family Planning in China. Pop represents the population for this region in 2015. RR h been calculated in Equation (2), and here refers to the attribution fraction, which attributed to respiratory disease linked to O3 and stroke, ischemic heart disease (IHD chronic obstructive pulmonary disease (COPD), and lung cancer (LC) linked to PM2.5.     Figure 3 shows the relative importance taken from the RF model for an individual air pollutant. For most air pollutants (i.e., PM 2.5 , PM 10 , NO 2 , SO 2 , and CO), the Unix time-as a long-term anthropogenic emission proxy-shows highly relative importance. This reveals the important impact of anthropogenic emission on trends in air pollutants (i.e., PM 2.5 , PM 10 , NO 2 , SO 2 , and CO) during 2015-2022. However, the Unix time shows a relatively low contribution to the relative importance, suggesting that the variability of atmospheric ozone in Lianyungang was not mainly driven by anthropogenic emissions. a long-term anthropogenic emission proxy-shows highly relative importance. This reveals the important impact of anthropogenic emission on trends in air pollutants (i.e., PM2.5, PM10, NO2, SO2, and CO) during 2015-2022. However, the Unix time shows a relatively low contribution to the relative importance, suggesting that the variability of atmospheric ozone in Lianyungang was not mainly driven by anthropogenic emissions.     The deweathered NO2 trend showed a similar trend w more rapid reduction during the latter period (−1.04 µg m −3 a −1 ) rather than the fo period (−0.06 µg m −3 a −1 ). These results could be attributed to more rapid reductio PM2.5 and NO2 from 2020 to 2022. The deweathered SO2 showed a larger reduction du 2015-2018 than that during 2019-2022, highlighting a more effective reduction in during the former period. This is also consistent with emission inventory studies where they also found substantial emission reduction of SO2 over eastern China. deweathered PM10 also presented a larger annual reduction ratio (−5.35 µg m −3 a −1 ) du 2015-2018 than that during 2019-2022, which could be partly explained by mankind emission control (such as urban road dust emissions) [48,49]. During these three per the deweathered O3 presented a comparable trend with a range of 0.21-0.37 µg m which reflects the continuity of the O3 pollution trend during the different periods in region. It should be noted that an unexpected short-term emission reduction due t COVID-19 pandemic lockdown in the spring of 2020 in eastern China has been w The deweathered NO 2 trend showed a similar trend with a more rapid reduction during the latter period (−1.04 µg m −3 a −1 ) rather than the former period (−0.06 µg m −3 a −1 ). These results could be attributed to more rapid reductions in PM 2.5 and NO 2 from 2020 to 2022. The deweathered SO 2 showed a larger reduction during 2015-2018 than that during 2019-2022, highlighting a more effective reduction in SO 2 during the former period. This is also consistent with emission inventory studies [18], where they also found substantial emission reduction of SO 2 over eastern China. The deweathered PM 10 also presented a larger annual reduction ratio (−5.35 µg m −3 a −1 ) during 2015-2018 than that during 2019-2022, which could be partly explained by mankind dust emission control (such as urban road dust emissions) [48,49]. During these three periods, the deweathered O 3 presented a comparable trend with a range of 0.21-0.37 µg m −3 a −1 , which reflects the continuity of the O 3 pollution trend during the different periods in this region. It should be noted that an unexpected short-term emission reduction due to the COVID-19 pandemic lockdown in the spring of 2020 in eastern China has been widely reported [38,[50][51][52][53][54], which might have influence on the long-term trend observed in the present study. Due to the methodology limitation, we here roughly evaluate such influence. To do so, the data in 2020 were replaced by that in 2019. The comparison between the original trends and the modified trends by using the replaced data could represent the upper limit of the COVID-19 impact. The modified trends in observed and deweathered concentrations of air pollutants during 2015-2022 were −3.00 and −1.87 µg m −3 a −1 , −5.03 and −3.24 µg m −3 a −1 , −1.07 and −0.63 µg m −3 a −1 , −2.71 and −1.93 µg m −3 a −1 , 0.91 and −0.26 µg m −3 a −1 , and −0.04 and −0.03 mg m −3 a −1 for PM 2.5 , PM 10 , NO 2 , SO 2 , O 3 , and CO, respectively. The difference between the original trends (see Figure 4) and the modified trends in observed and deweathered concentrations was very small or even negligible (1-7% and 0-2%, respectively). These results suggest that the short-term emission reduction during the COVID-19 pandemic lockdown could influence the long-term trend during 2015-2022, but the magnitude of the influence might be small. Moreover, it would be more helpful to evaluate such influence by using chemical transport modeling, which could be further performed in the future.  Figure 4) and the modified trends in observed and deweathered concentrations was very small or even negligible (1-7% and 0-2%, respectively). These results suggest that the short-term emission reduction during the COVID-19 pandemic lockdown could influence the long-term trend during 2015-2022, but the magnitude of the influence might be small. Moreover, it would be more helpful to evaluate such influence by using chemical transport modeling, which could be further performed in the future.

Impact of Meteorology on Air Pollution Trends
To understand the impact of meteorological conditions on trends in air quality, we further quantified an annual and seasonal mean meteorological impact index, a proxy for the potential contribution of the meteorological impact to changes in concentrations of air pollutants (see Figures 6 and 7). The positive value for the meteorological impact index could reflect negative meteorology, that is, adverse meteorological conditions for air pollution, while the negative values for that reflect those favorable meteorological conditions for improving air quality [36,37,42]. As shown in Figure 6, the meteorological impact index presented negative values for all six air pollutants during different time periods, suggesting favorable meteorological conditions that play an important positive impact on air quality in this coastal city region. Except for O3, the magnitudes of such values for most air pollutants were larger during 2019-2022 than that during 2015-2018,

Impact of Meteorology on Air Pollution Trends
To understand the impact of meteorological conditions on trends in air quality, we further quantified an annual and seasonal mean meteorological impact index, a proxy for the potential contribution of the meteorological impact to changes in concentrations of air pollutants (see Figures 6 and 7). The positive value for the meteorological impact index could reflect negative meteorology, that is, adverse meteorological conditions for air pollution, while the negative values for that reflect those favorable meteorological conditions for improving air quality [36,37,42]. As shown in Figure 6, the meteorological impact index presented negative values for all six air pollutants during different time periods, suggesting favorable meteorological conditions that play an important positive impact on air quality in this coastal city region. Except for O 3 , the magnitudes of such values for most air pollutants were larger during 2019-2022 than that during 2015-2018, suggesting that meteorological conditions (such as atmospheric diffusion conditions and/or air temperature) during the latter period were more favorable for reducing concentrations of air pollutant than during the former period. Except for O 3 , the meteorological impact index of the five air pollutants presents relatively high values during warm months, while it presents relatively low values and even positive values during cold months in this coastal city region. These results could provide direct evidence to prove that meteorological conditions during cold months could promote aerosol (i.e., PM 2.5 and PM 10 ) and gas (i.e., SO 2 , NO 2 , and CO) pollution. The meteorological conditions during warm months play an important role in reducing the concentrations of these air pollutants. The meteorological impact index of O 3 shows an overall opposite variation compared with other air pollutants, which present relatively low and high values during cold months and warm months, respectively. This is in line with the fact that the most frequent O 3 pollution occurs in warm seasons due to unfavorable meteorological conditions, such as strong solar radiation and high air temperature. This is consistent with the RF built-in importance results with the most important prediction feature for solar radiation and U10 (Figure 3).
Toxics 2023, 11, x FOR PEER REVIEW 9 of 14 meteorological impact index of the five air pollutants presents relatively high values during warm months, while it presents relatively low values and even positive values during cold months in this coastal city region. These results could provide direct evidence to prove that meteorological conditions during cold months could promote aerosol (i.e., PM2.5 and PM10) and gas (i.e., SO2, NO2, and CO) pollution. The meteorological conditions during warm months play an important role in reducing the concentrations of these air pollutants. The meteorological impact index of O3 shows an overall opposite variation compared with other air pollutants, which present relatively low and high values during cold months and warm months, respectively. This is in line with the fact that the most frequent O3 pollution occurs in warm seasons due to unfavorable meteorological conditions, such as strong solar radiation and high air temperature. This is consistent with the RF built-in importance results with the most important prediction feature for solar radiation and U10 ( Figure 3).   contributed approximately 50% to this emission driver. This suggests that the decline i HAQI value from 2015 to 2022 was driven by anthropogenic emission reduction. A reported by a previous study, approximately 0.894 million premature deaths in 2017 wer estimated due to PM2.5 and O3 pollution across China [55]. To further evaluate health ris associated with long-term exposure to air pollution, we estimated premature mortalit from ambient PM2.5 and O3 pollution in the Lianyungang region. Figure 9 presents trend in cause-specific premature mortality related to observed and deweathered PM2.5 and O from 2015 to 2022, respectively. Among the five specific causes, PM2.5-IHD and PM2. stroke accounted for a major fraction (approximately 71-78%) of the total number o deaths, due to the fact that stroke and IHD dominate the total mortality [45]. The highes premature mortality was attributable to observed PM2.5 and O3 in 2015, which would caus a total of 11,208 deaths. The specific causes were PM2.5-IHD (3931), PM2.5-stroke (4764 PM2.5-LC (770), PM2.5-COPD (1446), and O3-respiratory (297). Overall, the observe mortality from PM2.5 pollution declined during 2015-2022. The results indicated by th present study were comparable to a previous study, which showed a significant reductio in premature deaths attributable to long-term PM2.5 exposures from 2013 to 2016 in Chin due to emission control [56]. The deweathered PM2.5-related mortality was decreased from 10,088 deaths to 6501 deaths in 8 years, while the deweathered O3-related mortality wa increased from 804 deaths to 1192 deaths. The deweathered PM2.5-related mortalit presented a decreasing trend (−531 deaths per year). The O3-related mortality presente   10 contributed approximately 50% to this emission driver. This suggests that the decline in HAQI value from 2015 to 2022 was driven by anthropogenic emission reduction. As reported by a previous study, approximately 0.894 million premature deaths in 2017 were estimated due to PM 2.5 and O 3 pollution across China [55]. To further evaluate health risk associated with long-term exposure to air pollution, we estimated premature mortality from ambient PM 2.5 and O 3 pollution in the Lianyungang region. Figure 9 presents trends in cause-specific premature mortality related to observed and deweathered PM 2.5 and O 3 from 2015 to 2022, respectively. Among the five specific causes, PM 2.5 -IHD and PM 2.5 -stroke accounted for a major fraction (approximately 71-78%) of the total number of deaths, due to the fact that stroke and IHD dominate the total mortality [45]. The highest premature mortality was attributable to observed PM 2.5 and O 3 in 2015, which would cause a total of 11,208 deaths. The specific causes were PM 2.5 -IHD (3931), PM 2.5 -stroke (4764), PM 2.5 -LC (770), PM 2.5 -COPD (1446), and O 3 -respiratory (297). Overall, the observed mortality from PM 2.5 pollution declined during 2015-2022. The results indicated by the present study were comparable to a previous study, which showed a significant reduction in premature deaths attributable to long-term PM 2.5 exposures from 2013 to 2016 in China due to emission control [56]. The deweathered PM 2.5 -related mortality was decreased from 10,088 deaths to 6501 deaths in 8 years, while the deweathered O 3 -related mortality was increased from 804 deaths to 1192 deaths. The deweathered PM 2.5 -related mortality presented a decreasing trend (−531 deaths per year).

Health Risk and Premature Mortality Assessment
The O 3 -related mortality presented an increasing trend (38 deaths per year). These results suggest that reduction in exposure risk to ambient PM 2.5 was driven by anthropogenic emission control in this region. This is consistent with the results obtained from chemical transport modeling, which showed that air pollution control avoided 0.39 million deaths caused by PM 2.5 pollution in China from 2012 to 2017 [57]. However, health risk linked to ambient O 3 pollution was increased. The difference of the total deweathered deaths from 2015 to 2022 was 3199, suggesting that China's clean air actions might avoid airpollution-induced deaths in the Lianyungang region. The relative change in observed and deweathered values reflected the magnitude of the meteorological impact (see Equation (1)). The highest meteorology-driven impact could lead to 2050 deaths, which occurred in 2021. This suggests that changes in meteorological conditions could significantly affect air pollution deaths in this coastal region. Although we have estimated cause-specific premature mortality in this study, a further comparison between estimation and observation in the real world would be more useful to understand the uncertainty of our estimates and more helpful to further understand the health risk of air pollution in a future study. risk to ambient PM2.5 was driven by anthropogenic emission control in this region. This is consistent with the results obtained from chemical transport modeling, which showed that air pollution control avoided 0.39 million deaths caused by PM2.5 pollution in China from 2012 to 2017 [57]. However, health risk linked to ambient O3 pollution was increased. The difference of the total deweathered deaths from 2015 to 2022 was 3199, suggesting that China's clean air actions might avoid air-pollution-induced deaths in the Lianyungang region. The relative change in observed and deweathered values reflected the magnitude of the meteorological impact (see Equation (1)). The highest meteorologydriven impact could lead to 2050 deaths, which occurred in 2021. This suggests that changes in meteorological conditions could significantly affect air pollution deaths in this coastal region. Although we have estimated cause-specific premature mortality in this study, a further comparison between estimation and observation in the real world would be more useful to understand the uncertainty of our estimates and more helpful to further understand the health risk of air pollution in a future study.

Conclusions
In this study, the impact of anthropogenic emissions and meteorology on trends in air quality during 2015-2022 in a coastal city was investigated using an RF modeling approach. The annual mean PM2. 5  risk to ambient PM2.5 was driven by anthropogenic emission control in this region consistent with the results obtained from chemical transport modeling, which that air pollution control avoided 0.39 million deaths caused by PM2.5 pollution i from 2012 to 2017 [57]. However, health risk linked to ambient O3 pollution was in The difference of the total deweathered deaths from 2015 to 2022 was 3199, sug that China's clean air actions might avoid air-pollution-induced deaths Lianyungang region. The relative change in observed and deweathered values r the magnitude of the meteorological impact (see Equation (1)). The highest meteo driven impact could lead to 2050 deaths, which occurred in 2021. This sugge changes in meteorological conditions could significantly affect air pollution death coastal region. Although we have estimated cause-specific premature mortality study, a further comparison between estimation and observation in the real world be more useful to understand the uncertainty of our estimates and more helpful to understand the health risk of air pollution in a future study.

Conclusions
In this study, the impact of anthropogenic emissions and meteorology on tr air quality during 2015-2022 in a coastal city was investigated using an RF m approach. The annual mean PM2.5, PM10, NO2, SO2, and CO decreased from 53.

Conclusions
In this study, the impact of anthropogenic emissions and meteorology on trends in air quality during 2015-2022 in a coastal city was investigated using an RF modeling approach. The annual mean PM 2.5 , PM 10  conditions to the changes in HAQI were 93% and 7%, respectively. These results suggest that the substantial improvement in overall air quality was driven by anthropogenic emission reduction, which proved the effectiveness of clean air actions on air pollution control in the coastal city. Moreover, the clean air action could avoid 3199 deaths from 2015 to 2022 in the Lianyungang region, highlighting the health benefit of air pollution control policies.