E ﬀ ects of Meteorological Factors and Anthropogenic Precursors on PM 2.5 Concentrations in Cities in China

: Fine particulate matter smaller than 2.5 µ m (PM 2.5 ) in size can signiﬁcantly a ﬀ ect human health, atmospheric visibility, climate, and ecosystems. PM 2.5 has become the major air pollutant in most cities of China. However, inﬂuencing factors and their interactive e ﬀ ects on PM 2.5 concentrations remain unclear. This study used a geographic detector method to quantify the e ﬀ ects of anthropogenic precursors (AP) and meteorological factors on PM 2.5 concentrations in cities of China. Results showed that impacts of meteorological conditions and AP on PM 2.5 have signiﬁcant spatio-temporal disparities. Temperature was the main inﬂuencing factor throughout the whole year, which can explain 27% of PM 2.5 concentrations. Precipitation and temperature were primary impacting factors in southern and northern China, respectively, at the annual time scale. In winter, AP had stronger impacts on PM 2.5 in northern China than in other seasons. Ammonia had stronger impacts on PM 2.5 than other anthropogenic precursors in winter. The interaction between all factors enhanced the formation of PM 2.5 concentrations. The interaction between ammonia and temperature had strongest impacts at the national scale, explaining 46% ( q = 0.46) of PM 2.5 concentrations. The ﬁndings comprehensively elucidated the relative importance of driving factors in PM 2.5 formation, which can provide basic foundations for understanding the meteorological and anthropogenic inﬂuences on the concentration patterns of PM 2.5 .


Datasets
According to CAAQS, annual average PM 2.5 concentrations are limited to 15 µg m −3 (Grade I) and 35 µg m −3 (Grade II). The daily average concentrations are 35 µg m −3 (Grade I) and 75 µg m −3 (Grade II) [39]. Grade I refers to the concentration limit required for scenic spots, nature reserves, and other areas requiring special conservation in China. Grade II refers to the concentration limits required for rural areas, residential areas, industrial areas, cultural areas, and mixed-use residential areas. The daily PM 2.5 concentrations of 366 cities in China were obtained from the China Environmental Monitoring Center. Due to the availability of data, data for Hong Kong, Macau, and Taiwan were not included. Figures S3 and S4 show maps of PM 2.5 concentrations [29].
We acquired meteorological data (839 sites) from the China Meteorological Data Network throughout the whole year of 2016. The daily meteorological data included surface air pressure (PS, hPa), air temperature (TE, • C), relative humidity (RH, %), wind velocity (WI, m s −1 ), sunshine duration (SS, h), and accumulated precipitation (PE, mm) (Figures S5-S11) [29]. The monthly anthropogenic emissions of VOCs ammonia (NH 3 ), sulfur dioxide (SO 2 ), and nitrogen oxide (NOx) were monitored in 2016. They were usually considered as AP of PM 2.5 and were collected from MEIC (multi-resolution emission inventory for China, http://www.meicmodel.org/). The MEIC includes emission data of the four sub-sectors of transportation, power, industry, and residential [52], and has been widely used in the research of air pollution [53][54][55][56]. Figure S12 indicates that the highest anthropogenic precursor emissions are mainly distributed in EC, MYR, SC, NC, and MUYR.

GeoDetector
In this study, the q statistics of GeoDetector were used to quantitatively analyze the impacts of AP and MCs on PM 2.5 in China. GeoDetector supports a series of statistical methods that can explore spatial difference and identify the driving factors. The main idea is based on the assumption that if an independent variable (X) causes a dependent variable (Y), then the spatial distribution of the independent variable and the dependent variable should be consistent [57][58][59][60]. GeoDetector can detect both qualitative data and numerical data. Compared with traditional linear statistical methods, this is a major advantage of GeoDetector. Another unique advantage of GeoDetector is the ability to detect the interaction between two factors acting on the dependent variable. The GeoDetector includes four detectors, which are factor detection, risk area detection, ecological detection, and interaction detection. In this study, factor detection and interaction detection were used.
Factor detector uses q statistic to detect the influence of X (e.g., MCs and AP) on Y (e.g., the PM 2.5 concentrations). The expression is: In the formula, h = 1, . . . , L, which classifies X or Y; N and N h are the numbers of the whole region and categories in classification h; σ 2 and σ h 2 are the Y value of the whole region and the variance of strata h, respectively. SST and SSW are the total variance of the whole region and the sum of variance within the strata, respectively. Greater values of q (0-1) indicate more spatial variation in Y. If the classification is based on X, a higher q value explains the influence of X on Y (i.e., explaining power: 100 × q%). Interaction detection can identify the impact of the interaction between potential driving factors. Based on that, we can assess whether the interaction between X 1 and X 2 will strengthen or weaken the explaining power of Y. Additionally, the influences of these factors on the dependent variable Y would be independent of each other. There are five types of interactions; please refer to [59] for more information. In addition, in order to identify the positive or negative correlations between PM 2.5 concentrations and influencing factors, this study calculated their Pearson correlation coefficients at different temporal and spatial scales.

Effects on PM 2.5 Concentrations at the National Scale
The influence of each driving factor on PM 2.5 concentrations was acquired by calculating the corresponding q value (the power of determinant, Figure 1a), which indicated the contribution of each impacting factor on PM 2.5 concentrations. Figure 1 shows that there were obvious seasonal and annual difference in factors' impacts on PM 2.5 . Meteorological conditions were dominant impacting factors in PM 2.5 formation at the annual time scale. TE (q = 0.27) was the primary impacting factor, followed by PE (q = 0.22) and PS (q = 0.17).
Sustainability 2020, 12, x FOR PEER REVIEW 4 of 13 the explaining power of Y. Additionally, the influences of these factors on the dependent variable Y would be independent of each other. There are five types of interactions; please refer to [59] for more information. In addition, in order to identify the positive or negative correlations between PM2.5 concentrations and influencing factors, this study calculated their Pearson correlation coefficients at different temporal and spatial scales.

Effects on PM2.5 Concentrations at the National Scale
The influence of each driving factor on PM2.5 concentrations was acquired by calculating the corresponding q value (the power of determinant, Figure 1a), which indicated the contribution of each impacting factor on PM2.5 concentrations. Figure 1 shows that there were obvious seasonal and annual difference in factors' impacts on PM2.5. Meteorological conditions were dominant impacting factors in PM2.5 formation at the annual time scale. TE (q = 0.27) was the primary impacting factor, followed by PE (q = 0.22) and PS (q = 0.17). Meteorological factors were dominant driving forces in spring, such as PE (q = 0.12) and PS (q = 0.10). In summer, AP showed stronger impacts on PM2.5 concentrations than meteorological conditions, and the NOx, NH3, and VOCs were the three dominant factors (q > 0.10). In autumn, the meteorological factors and AP showed comparative influence on PM2.5. The dominant impacting factor in autumn was TE (q = 0.18), followed by PE (q = 0.13), NOx (q = 0.11), and NH3 (q = 0.10). Similar to the autumn, meteorological factors and AP had comparative impacts on PM2.5 concentrations in winter. TE (q = 0.25) was the dominant factor, followed by NH3 (q = 0.18), SSD (q = 0.17), PS (q = 0.16), and VOCs (q = 0.13). This indicated that AP were the dominant impacting factor in winter. Figure 2 shows the effects of AP and MCs on PM2.5 significantly varied at regional and seasonal scales in China. In general, meteorological factors were the major driving forces in China. PE and TE were primary driving forces in southern and northern China, respectively. Figure 2 shows that meteorological factors were primary drivers of PM2.5 formation in spring, which is similar to most regions at the annual time scale. The dominant meteorological factor was PE Meteorological factors were dominant driving forces in spring, such as PE (q = 0.12) and PS (q = 0.10). In summer, AP showed stronger impacts on PM 2.5 concentrations than meteorological conditions, and the NOx, NH 3 , and VOCs were the three dominant factors (q > 0.10). In autumn, the meteorological factors and AP showed comparative influence on PM 2.5 . The dominant impacting factor in autumn was TE (q = 0.18), followed by PE (q = 0.13), NOx (q = 0.11), and NH 3 (q = 0.10). Similar to the autumn, meteorological factors and AP had comparative impacts on PM 2.5 concentrations in winter. TE (q = 0.25) was the dominant factor, followed by NH 3 (q = 0.18), SSD (q = 0.17), PS (q = 0.16), and VOCs (q = 0.13). This indicated that AP were the dominant impacting factor in winter. Figure 2 shows the effects of AP and MCs on PM 2.5 significantly varied at regional and seasonal scales in China. In general, meteorological factors were the major driving forces in China. PE and TE were primary driving forces in southern and northern China, respectively. Sustainability 2020, 12, x FOR PEER REVIEW 5 of 13 and SS in southern and regions of MYR and NC, respectively. Meteorological factors and AP showed comparative impacts on PM2.5 concentrations in summer except for in XJ and QTP. TE was the dominant driving factor on PM2.5 concentrations in autumn in UYR, NE, and MYR, but PE and WI played the dominant role in regions of MUYR and SC, and of NC and EC, respectively. WI was the primary driving factor in NC, NE, EC, and UYR in winter, but PS was the major driving factor in MUYR and MUPR. However, in QTP, PS was the major impacting factor on PM2.5 concentrations throughout the whole year.

Interactive Effects on PM2.5
This study explored interactive effects on PM2.5 by using the interaction detector with a total of 45 pairs of interactions. The interaction of any two factors was analyzed by comparing their combined contribution with their individual contributions to PM2.5 concentrations. Figure 3 shows the q values of each pair of impact factors and their interaction through the whole year at the national scale. Interactions of PE ∩ TE, PS ∩ TE, PS ∩ VO, PS ∩ NO, VO ∩ NO, VO ∩ NH, and NO ∩ NH belong to bivariate enhancements and other interactions belong to nonlinear enhancements ( Figure 3). Generally, the interaction between NH and TE (q value = 0.46) was the strongest interaction among all impacting factors. Figure S14 indicates that there were obvious seasonal disparities in the interactive influence. In spring, fall, and winter (but not summer), the interactions between meteorological factors played major roles in PM2.5 concentrations. The interaction WI ∩ RH (q value = 0.38) had the strongest effect on PM2.5 in spring. In autumn and winter, the interaction between SS and TE (autumn: q value = 0.54; winter: q value = 0.42) played the strongest role in PM2.5 concentrations. However, NH ∩ RH (q value = 0.35) was the highest in summer.  Figure 2 shows that meteorological factors were primary drivers of PM 2.5 formation in spring, which is similar to most regions at the annual time scale. The dominant meteorological factor was PE and SS in southern and regions of MYR and NC, respectively. Meteorological factors and AP showed comparative impacts on PM 2.5 concentrations in summer except for in XJ and QTP. TE was the dominant driving factor on PM 2.5 concentrations in autumn in UYR, NE, and MYR, but PE and WI played the dominant role in regions of MUYR and SC, and of NC and EC, respectively. WI was the primary driving factor in NC, NE, EC, and UYR in winter, but PS was the major driving factor in MUYR and MUPR. However, in QTP, PS was the major impacting factor on PM 2.5 concentrations throughout the whole year.

Interactive Effects on PM 2.5
This study explored interactive effects on PM 2.5 by using the interaction detector with a total of 45 pairs of interactions. The interaction of any two factors was analyzed by comparing their combined contribution with their individual contributions to PM 2.5 concentrations. Figure 3 shows the q values of each pair of impact factors and their interaction through the whole year at the national scale. Interactions of PE ∩ TE, PS ∩ TE, PS ∩ VO, PS ∩ NO, VO ∩ NO, VO ∩ NH, and NO ∩ NH belong to bivariate enhancements and other interactions belong to nonlinear enhancements ( Figure 3). Generally, the interaction between NH and TE (q value = 0.46) was the strongest interaction among all impacting factors. Figure S14 indicates that there were obvious seasonal disparities in the interactive influence. In spring, fall, and winter (but not summer), the interactions between meteorological factors played There were obvious regional disparities in all the interactions in China (Figure 4). The interactions between meteorological factors were strongest in all regions. In the interactions between AP and meteorological factors, the interaction of AP∩TE played a primary role in EC, MUYR, and northern China, and the interaction of AP∩PE played a dominant role in MUYR, MUPR, and SC. Figures S15-S18 show the seasonal interactive effects of each pair of driving factors at the regional scale in China, indicating that there were also obvious regional and seasonal differences in all the interactions.

Discussion
PM2.5 concentrations were found to be driven by natural conditions and human activities. Previous studies had reported that PM2.5 concentrations correlate with both MCs and AP emissions There were obvious regional disparities in all the interactions in China (Figure 4). The interactions between meteorological factors were strongest in all regions. In the interactions between AP and meteorological factors, the interaction of AP∩TE played a primary role in EC, MUYR, and northern China, and the interaction of AP∩PE played a dominant role in MUYR, MUPR, and SC. Figures S15-S18 show the seasonal interactive effects of each pair of driving factors at the regional scale in China, indicating that there were also obvious regional and seasonal differences in all the interactions. There were obvious regional disparities in all the interactions in China (Figure 4). The interactions between meteorological factors were strongest in all regions. In the interactions between AP and meteorological factors, the interaction of AP∩TE played a primary role in EC, MUYR, and northern China, and the interaction of AP∩PE played a dominant role in MUYR, MUPR, and SC. Figures S15-S18 show the seasonal interactive effects of each pair of driving factors at the regional scale in China, indicating that there were also obvious regional and seasonal differences in all the interactions.

Discussion
PM 2.5 concentrations were found to be driven by natural conditions and human activities. Previous studies had reported that PM 2.5 concentrations correlate with both MCs and AP emissions [16,[61][62][63][64][65]. We analyzed influences of AP and meteorological factors and their interactions on PM 2.5 in Chinese cities. Results showed that effects of AP and MCs and their interactions on PM 2.5 had obvious seasonal and regional variations across China.
In this study, we found that meteorological factors were the leading driving factors at the annual time scale and the national scale, indicating that the meteorological conditions had dominant influences on PM 2.5 concentrations. TE and PE were the two leading factors affecting PM 2.5 concentrations. TE was closely related to PM 2.5 concentrations by affecting atmospheric perturbation and chemical reactions [24]. PE could scavenge PM 2.5 from the air and had a moisture removal effect [11,66,67]. However, the major influencing factors on PM 2.5 concentration had significantly seasonal variations. The influences of meteorological conditions on PM 2.5 concentrations were the strongest in winter, and TE was the leading factor affecting PM 2.5 concentrations. This is because temperature inversion can weaken the scattering and dispersion of PM 2.5 , resulting in higher local PM 2.5 pollution [68,69]. In summer, influences of meteorological conditions on PM 2.5 concentrations were weakest. RH was the leading factor affecting PM 2.5 concentrations in summer. This was because RH was higher in summer than in the other three seasons, which had a suppression effect on PM 2.5 under moist air conditions [29].
At the annual time scale, TE was the dominant factor in northern and southern China (except MUPR), which is consistent with previous studies [70,71]. PE was the primary factor affecting PM 2.5 concentrations in southern China in most seasons throughout the whole year. This was because PE in southern China was higher than in other regions [29]. The increasing PE had scavenging effects on PM 2.5 by wet deposition, and could lower the PM 2.5 concentration [66,67]. WI was the main influencing factor in EC and northern China in winter, which is similar to some reports that WI was the most important and negative impacting factor on PM concentrations [72,73]. This was because weaker East Asia winter monsoons could slow wind speeds and increased the frequency of static wind, which had made it more difficult for PM to disperse [74].
Previous studies have indicated that AP were also crucial driving factors on PM 2.5 [48,63,75]. We found that the influence of AP on PM 2.5 in winter were higher than in other seasons in XJ and northern China (MYR, UYR, NC). This might be due to the high anthropogenic emissions from winter carbon-fired heating and less surface vegetation cover in winter in northern China, which significantly increased pollutant emissions in the atmosphere [76,77]. In addition, NH 3 had important impacts on PM 2.5 among AP factors in winter at the national scale and in some individual regions (NC, SC, MUYR, and MYR). Ammonia is an important precursor, and emissions of ammonia had stronger associations with PM 2.5 concentrations than other anthropogenic precursors; this is similar to results of previous studies [78]. Ammonia participates in photochemical reactions as an atmospheric alkaline gas, which is important in the SIA (secondary inorganic aerosol) formation of compounds such as ammonium salts, sulfate, and nitrate [79][80][81][82][83]. In China, SIA was an important driving factor of PM 2.5 pollution, especially during severe smog events [80,84]. SIA accounted for 32% of PM 2.5 mass concentration in China (e.g., Beijing, Guangzhou, Shanghai, and Xi'an) during the 2013 haze pollution events [84]. There was a 5.7% reduction in the annual concentration of PM 2.5 concentrations when ammonia emissions were cut by 47% in the Beijing-Tianjin-Hebei region of China [79]. PM 2.5 concentrations are affected by complex interactions between AP and meteorological conditions. This study found that interactions between any driving factors at all time and space scales had significant enhancement effects on PM 2.5 concentrations. The leading interactive effect between AP and MCs was between AP with RH and PE at the national scale in summer. This might be due to the fact that PE and RH have scavenging and suppression effects on PM 2.5 by wet deposition in summertime. However, the primary interactive effect between AP and MCs was AP ∩ TE at the national scale in winter. This was due to increased temperature inversion under the lower winter temperature, which weakened the diffusion and dispersion of pollution. The interaction between AP had the dominant effect on PM 2.5 concentrations over most regions in summer, which indicated that accelerated photochemical reactions between AP occurred under high-temperature conditions. Interactions between AP and factors of PE and PS were important at annual and seasonal scales in southern China. This might be due to the high precipitation and surface pressure in southern China [29].

Conclusions
The effects of AP and MCs on PM 2.5 in Chinese cities were systematically analyzed in this study. The findings revealed significant seasonal and regional disparities in the impacts of examined factors and how they interacted on PM 2.5 . The results can help us to better understanding the relative importance of the driving factors in the formation of PM 2.5 . The study indicated that local AP and meteorological factors had important impacts on PM 2.5 in China, and had obvious regional and seasonal variations. Meteorological conditions played a leading role in determining PM 2.5 concentrations at the regional and national scales throughout the whole year. At the seasonal time scale, WI was the primary factor on PM 2.5 concentrations in winter in northern China and XJ, but PE was the major driver on PM 2.5 concentrations during most seasons in southern China. However, AP had stronger impacts on PM 2.5 during winter than in other seasons in XJ and most regions of northern China. NH 3 had a stronger effect on PM 2.5 concentrations during winter than other anthropogenic precursors. Interactions between all influencing factors have enhanced effects on PM 2.5 concentrations. In addition, the interaction between MCs and AP played a leading role at the national scale throughout the whole year and in summer and winter. The results could provide a basis for the government to develop more precise air pollution control strategies.
Some limitations to the study should be clarified to assist future studies. First, land use and land cover, socioeconomic conditions, elevation, and topography were not considered to assess their influence on PM 2.5 concentrations. Second, the uncertainty of MEIC emission inventory may lead to some uncertainties in the research results. When processing emissions inventory data, because the inventory only has monthly emissions data, so as to match the daily PM 2.5 concentration and daily weather monitoring data, we took the arithmetic average of the monthly inventory emissions data, which also increased the uncertainty of the analysis results. Third, the data used in this study is limited to 2016, and does not include data analysis of other years. This is because the emission inventory data that we could obtain were for 2008, 2010, 2012, 2014, and 2016, but the PM 2.5 data in 366 cities were available only from 2015 to 2017. In order to maintain consistency between the data, we selected 2016 as the research period in this study. There was no comparative analysis of inter-annual variability. Therefore, we should comprehensively consider other factors on PM 2.5 concentrations including socioeconomic, land use, terrain, and elevation in the future. In addition, inter-annual change analysis based on multi-year data needs to be added.   Figure S14: The seasonal interactive q values and the original q value of each pair of factors. Figure S15: The interactions between impacting factors in spring at the Sustainability 2020, 12, 3550 9 of 13 regional scale in China. Figure S16: Interactions between impacting factors in summer at the regional scale in China. Figure S17: Interactions between impacting factors in autumn at the regional scale in China. Figure S18: Interactions between impacting factors in winter at the regional scale in China. Table S1: Effect of various factors on PM 2.5 in China in 2016. Table S2: Effect of various factors on PM 2.5 throughout the whole year at the regional scale. Table S3: Effect of various factors on PM 2.5 in spring at the regional scale. Table S4: Effect of various factors on PM 2.5 in summer at the regional scale. Table S5: Effect of various factors on PM 2.5 in autumn at the regional scale. Table S6: Effect of various factors on PM 2.5 in winter at the regional scale.