Effects of Anthropogenic Emissions from Different Sectors on PM2.5 Concentrations in Chinese Cities

PM2.5 pollution has gradually attracted people’s attention due to its important negative impact on public health in recent years. The influence of anthropogenic emission factors on PM2.5 concentrations is more complicated, but their relative individual impact on different emission sectors remains unclear. With the aid of the geographic detector model (GeoDetector), this study evaluated the impacts of anthropogenic emissions from different sectors on the PM2.5 concentrations of major cities in China. The results indicated that the influence of anthropogenic emissions factors with different emission sectors on PM2.5 concentrations exhibited significant changes at different spatial and temporal scales. Residential emissions were the dominant driver at the national annual scale, and the NOX of residential emissions explained 20% (q = 0.2) of the PM2.5 concentrations. In addition, residential emissions played the leading role at the regional annual scale and during most of the seasons in northern China, and ammonia emissions from residents were the dominant factor. Traffic emissions play a leading role in the four seasons for MUYR and EC in southern China, MYR and NC in northern China, and on a national scale. Compared with primary particulate matter, secondary anthropogenic precursors have a more important effect on PM2.5 concentrations at the national or regional annual scale. The results can help to strengthen our understanding of PM2.5 pollution, improve PM2.5 forecasting models, and formulate more precise government control policy.


Introduction
Particulate matter (including PM 2.5 and PM 10 , etc.), particularly fine particulate matter (PM 2.5 ), is an air pollutant that exerts an important negative impact on public health [1][2][3][4]. Worldwide, more than 3.2 million people die prematurely due to outdoor particulate matter exposure each year [5]. In recent years, with the acceleration of China's industrialization and urbanization and the continuous expansion of the urban scale, PM 2.5 pollution has gradually attracted the attention of the government and academic circles [6][7][8].
explored the impact of anthropogenic emissions between different sectors on PM 2.5 to clarify the difference in the impact of sector emissions on PM 2.5 from a more detailed perspective.
Based on the emission inventory and site monitoring data, this study used the geographic detector method to explore the influence mechanism of the anthropogenic emission factors of different sectors on the PM 2.5 concentrations. The research results will help researchers improve the accuracy of air quality forecasting models and also provide a basis and reference for government departments to formulate more accurate particulate matter emission government control policy.

Study Area
This study takes China as the research area and selects 366 major cities as samples to analyze the impact of the emissions of different sectors on the PM 2.5 concentrations of major cities. Considering China's natural and economic and social conditions and referring to the research of [52], the China region is divided into ten zones ( Figure S1): the upper zone of the Yellow River (UYR); the middle and upper zone of the Pearl River (MUPR); the middle zone of the Yellow River (MYR); the northern coastal zone (NC); the northeast zone (NE); the middle and upper zone of the Yangtze River (MUYR); the southeast coastal zone (SC); the Xinjiang zone (XJ); the Qinghai-Tibetan Plateau (QTP); and the eastern coastal area (EC). Furthermore, part of the district is divided into two regions, the south and the north. NE, MYR, UYR, and NC belong to the northern region, while SC, MUPR, EC, and MUYR belong to the southern region.

Data
The data used in this study include daily PM 2.5 monitoring data from 366 cities ( Figure S2) from 2015 to 2017 (National Environmental Monitoring Center). Based on this data, we can obtain the annual average concentrations (Figure 1a) from 2015 to 2017, as well as the seasonal average concentrations ( Figures S2 and S3). The anthropogenic emission data of primary PM 2.5 , nitrogen oxides (NO X ), volatile organic compounds (VOC S ), sulfur dioxide (SO 2 ) and ammonia (NH 3 ) come from the 2016 Tsinghua University Emission Inventory (MEIC) (http://www.meicmodel.org/ (accessed on 10 December 2018)) (Figure 1b-f), the grid resolution is 0.25 • × 0.25 • , and the temporal resolution is month by month. The emission sector is divided into five sectors: industrial sources, agricultural sources, transportation sources, power sources and residential sources. The types of pollutants in each sector are as follows (Figures S4-S8): primary industry PM 2.5 , primary power PM 2.5 , primary residential PM 2.5 , primary transportation PM 2.5 , industrial NO X , power NO X , residential NO X , transportation NO X , industry VOC S , power VOC S , residential VOC S , transportation VOC S , agricultural NH 3 , industrial NH 3 , residential NH 3 , transportation NH 3 , industrial SO 2 , and power SO 2 .

GeoDetector Model
Geographic detectors are a set of statistical methods that use spatial similarity to detect the influence of independent variables on dependent variables and reveal the driving force behind the phenomenon. This method has no linear assumptions and has clear physical meaning. It includes 4 modules: the factor detection module, interaction function detection module, risk area detection module, and ecological detection module [53]. This study uses the factor detection module in GeoDetector to evaluate the contribution of pollutants emitted by different sectors to the PM 2.5 concentrations of major cities in China. The factor detection module can express the degree of interpretation of the independent variable to the dependent variable through the q value. For specific methods, please refer to related research [54,55].

GeoDetector Model
Geographic detectors are a set of statistical methods that use spatial similarity tect the influence of independent variables on dependent variables and reveal the d force behind the phenomenon. This method has no linear assumptions and has clear ical meaning. It includes 4 modules: the factor detection module, interaction functi tection module, risk area detection module, and ecological detection module [53 study uses the factor detection module in GeoDetector to evaluate the contribution lutants emitted by different sectors to the PM2.5 concentrations of major cities in The factor detection module can express the degree of interpretation of the indepe variable to the dependent variable through the q value. For specific methods, pleas to related research [54,55].
The factor detection module used in this article is used to detect the spatial di tiation of the dependent variable and quantify the degree of interpretation of a independent variable to the dependent variable, measured by the q value. Assum the PM2.5 concentrations in the area in which the city is located are composed o emission data of 366 cities across the country. Assuming that there may be a fact The factor detection module used in this article is used to detect the spatial differentiation of the dependent variable and quantify the degree of interpretation of a certain independent variable to the dependent variable, measured by the q value. Assume that the PM 2.5 concentrations in the area in which the city is located are composed of daily emission data of 366 cities across the country. Assuming that there may be a factor that affects the concentrations of PM 2.5 , expressed as A = {A h }, h = 1, 2, . . . , L, L is the number of influencing factors, A h is the different types of influencing Factors A, and A-type h corresponds to one or more regional subsets of the city in space. To detect the spatial correlation between influencing Factor A and urban PM 2.5 concentrations, the layers of influencing Factor A and the urban PM 2.5 concentrations map are superimposed. In the h type of influencing Factor A, the discrete variance of urban PM 2.5 concentrations is denoted as σ h 2 , and the degree of interpretation of PM 2.5 concentrations by influencing Factor A can be expressed as: In the formula, q is the explanatory power of urban PM 2.5 concentrations influencing factors, and its value range is [0, 1]; L is the number of urban subregions divided by influencing factors, which this article uses equal interval classification to divide into 6 levels; h is a certain subarea; N h is the number of samples in a given subarea; N is the number of samples in the entire study area; σ h 2 is the discrete variance of the PM 2.5 concentrations in a given subregional city; σ 2 is the variance of the PM 2.5 concentrations in the entire regional city. If the q value of a certain influencing factor is higher, its explanatory power is stronger, and the influence of this factor on urban PM 2.5 concentrations is stronger. Conversely, if the q value of a certain influencing factor is smaller, its explanatory power is weaker, and the influence of this factor on urban PM 2.5 concentrations is weaker. In extreme cases, a q value of 0 indicates that the influencing factor and urban PM 2.5 concentrations score have no relationship. A q value of 1 indicates that the influencing factors completely affect the distribution of urban PM 2.5 concentrations.

Analysis on the National Annual Scale
The driving factors of PM 2.5 concentrations in all Chinese cities have significant annual changes on an annual scale. If the emission sectors are not considered, the primary two factors influencing the concentrations of PM 2.5 on the national annual scale are NH and P_PM ( Figure S9). If the emission sectors are considered, the dominant factors that affect PM 2.5 concentrations on the national annual scale come from residential sources. Among them, the dominant factor is R_NO (q = 0.20), followed by R_VO (q = 0.18), R_NH (q = 0.16) and R_SO (q = 0.14) ( Figure 2).
Factor A and the urban PM2.5 concentrations map are superimposed. In the h type of influencing Factor A, the discrete variance of urban PM2.5 concentrations is denoted as σh 2 , and the degree of interpretation of PM2.5 concentrations by influencing Factor A can be expressed as: In the formula, q is the explanatory power of urban PM2.5 concentrations influencing factors, and its value range is [0, 1]; L is the number of urban subregions divided by influencing factors, which this article uses equal interval classification to divide into 6 levels; h is a certain subarea; Nh is the number of samples in a given subarea; N is the number of samples in the entire study area; σh 2 is the discrete variance of the PM2.5 concentrations in a given subregional city; σ 2 is the variance of the PM2.5 concentrations in the entire regional city. If the q value of a certain influencing factor is higher, its explanatory power is stronger, and the influence of this factor on urban PM2.5 concentrations is stronger. Conversely, if the q value of a certain influencing factor is smaller, its explanatory power is weaker, and the influence of this factor on urban PM2.5 concentrations is weaker. In extreme cases, a q value of 0 indicates that the influencing factor and urban PM2.5 concentrations score have no relationship. A q value of 1 indicates that the influencing factors completely affect the distribution of urban PM2.5 concentrations.

Analysis on the National Annual Scale
The driving factors of PM2.5 concentrations in all Chinese cities have significant annual changes on an annual scale. If the emission sectors are not considered, the primary two factors influencing the concentrations of PM2.5 on the national annual scale are NH and P_PM ( Figure S9). If the emission sectors are considered, the dominant factors that affect PM2.5 concentrations on the national annual scale come from residential sources. Among them, the dominant factor is R_NO (q = 0.20), followed by R_VO (q = 0.18), R_NH (q = 0.16) and R_SO (q = 0.14) (Figure 2).  . The q value of the driving factor on the national annual and seasonal scales. (P_I_PM denotes primary industry PM 2.5 ; P_P_PM denotes primary power PM 2.5 ; P_R_PM denotes primary residential PM 2.5 ; P_T_PM denotes primary transportation PM 2.5 ; I_NO denotes industrial NO X ; P_NO denotes power NO X ; R_NO denotes residential NO X ; T_NO denotes transportation NO X ; I_VO denotes industry VOC S ; P_VO denotes power VOC S ; R_VO denotes residential VOC S ; T_VO denotes transportation VOC S ; A_NH denotes agricultural NH 3 ; I_NH denotes industrial NH 3 ; R_NH denotes residential NH 3 ; T_NH denotes transportation NH 3 ; I_SO denotes industrial SO 2 ; and P_SO denotes power SO 2 ).

Analysis on the National Seasonal Scale
If the emission sectors are not taken into account, the dominant factors that affect the PM 2.5 concentrations on the national scale in spring are NO and NH ( Figure S9). In summer, these types of anthropogenic emissions have similar effects on PM 2.5 concentrations on the national scale. Similar to summer, these types of anthropogenic emissions have similar effects on PM 2.5 concentrations in autumn (except SO). However, the influence of NH and P_PM on PM 2.5 concentrations are stronger than other factors in winter on the national scale.
If the emission sectors are considered, traffic sources play a leading role in the influence of PM 2.5 concentrations compared with several other emission sources in spring on the national scale ( Figure 2). The contribution of the influencing factors is ranked as follows: T_NO (q = 0.13), T_SO (q = 0.12), T_VO (q = 0.12). Similar to spring, the main driving factors come from traffic sources in summer. T_NO (q = 0.17) ranks first, followed by T_SO (q = 0.16) and T_NH (q = 0.15). In autumn, residential sources and traffic emissions play a leading role in PM 2.5 concentrations and have similar effects on the national scale. R_NO (q = 0.18) and R_NO (q = 0.17) are the top two factors. Similar to autumn, residential sources and traffic sources in winter play a leading role in the PM 2.5 concentrations and have similar effects, and the dominant factors are T_NO and T_SO.   . The q value of the driving factor on the regional annual and seasonal scales. (P_I_PM denotes primary industry PM2.5; P_P_PM denotes primary power PM2.5; P_R_PM denotes primary residential PM2.5; P_T_PM denotes primary transportation PM2.5; I_NO denotes industrial NOX; P_NO denotes power NOX; R_NO denotes residential NOX; T_NO denotes transportation NOX; I_VO denotes industry VOCS; P_VO denotes power VOCS; R_VO denotes residential VOCS; T_VO denotes transportation VOCS; A_NH denotes agricultural NH3; I_NH denotes industrial NH3; R_NH denotes residential NH3; T_NH denotes transportation NH3; I_SO denotes industrial SO2; and P_SO denotes power SO2).

Analysis on the Regional Seasonal Scale
If the emission sectors are not considered, NH is the dominant factor in northe China and SC in spring, and NO is the dominant factor in southern EC and MUYR (Figu S10). In summer, NH is the dominant factor of PM2.5 concentrations in most northern gions (except UYR) and in southern MUYR, VO is the dominant factor in URY and S and SO is the dominant factor in MUPR. In autumn, P_PM is the dominant factor of PM concentrations in NE, MYR, and SC. NH is the dominant factor of PM2.5 concentrations NC and MUYR, and the dominant factor of PM2.5 concentrations in MUPR and UYR VO. In winter, however, NH is the dominant factor of PM2.5 concentrations in NC, MY XJ, EC, SC and MUYR. VO is the dominant factor of PM2.5 concentrations in MUPR a NE, and SO is the dominant factor of PM2.5 concentrations in UYR in winter.
If the emission sectors are considered, emissions from residential sources and traf sources are the dominant factors affecting the change in PM2.5 concentrations in spring most regions (Figure 3). NE, NC, MYR in the northern region and SC in the southern gion are dominated by residential source emissions, while traffic emissions dominate southern EC and MUYR. Similar to spring, summer residential and traffic sources rema the basic influencing factors for PM2.5 concentrations changes in most parts of the count Among them, NC, MYR, and XJ in the north and EC and MUYR in the south are main due to traffic emissions, whereas residential source emissions play a leading role in t Figure 3. The q value of the driving factor on the regional annual and seasonal scales. (P_I_PM denotes primary industry PM 2.5 ; P_P_PM denotes primary power PM 2.5 ; P_R_PM denotes primary residential PM 2.5 ; P_T_PM denotes primary transportation PM 2.5 ; I_NO denotes industrial NO X ; P_NO denotes power NO X ; R_NO denotes residential NO X ; T_NO denotes transportation NO X ; I_VO denotes industry VOC S ; P_VO denotes power VOC S ; R_VO denotes residential VOC S ; T_VO denotes transportation VOC S ; A_NH denotes agricultural NH 3 ; I_NH denotes industrial NH 3 ; R_NH denotes residential NH 3 ; T_NH denotes transportation NH 3 ; I_SO denotes industrial SO 2 ; and P_SO denotes power SO 2 ).

Discussion
The concentrations of PM 2.5 is affected by both natural and human factors, and anthropogenic emissions play an important role in its formation. Numerous studies have analyzed the formation of PM 2.5 and the influencing factors of its concentrations [56,57], but few studies have focused on the impacts of these factors on PM 2.5 concentrations from the perspective of sector emissions. In addition, the research results can provide a more accurate reference for formulating PM 2.5 control policies. In this paper, five types of pollutants emitted by five sectors, including residential sources, power, transportation, industry, and agriculture, were selected as driving factors. Based on the GeoDetector model, the impact of emission factors on the PM 2.5 concentrations at different temporal and spatial scales was analyzed. The study found obvious seasonal and regional changes in the impact of pollutant emissions from different sectors on PM 2.5 concentrations in Chinese cities. These variations are due mainly to the temporal and spatial differences in the emission of different precursors and photochemical reactions. Of course, anthropogenic emissions are also the main factors influencing the temporal and spatial differences in PM 2.5 concentrations in China and even at the global scale [58][59][60].
At the national scale, residential emissions are the main influencing factor of the PM 2.5 concentrations changes through the whole year. This result also verifies the proposal by other researchers that residential emissions play an important and non-negligible role in air pollution control [61,62]. However, at the national seasonal scale, the main driving factors of PM 2.5 concentrations show more significant changes. The impact of residential emissions on PM 2.5 concentrations in autumn and winter is stronger than that in spring and summer. These differences may be due to the unfavorable meteorological conditions of the planetary boundary layer in autumn and winter, such as temperature inversion and relatively frequent weak surface wind speed; moreover, the heating of living spaces and biomass burning in autumn and winter may easily cause an increase in PM 2.5 concentrations [10,11,63,64]. It is worth noting that ammonia emissions exert a stronger impact on PM 2.5 in winter compared to the other three seasons regardless of whether sectoral emissions are considered. Since the emission of NH 3 in winter is lower than in other seasons, its impact is also higher than in other seasons, indicating that NH 3 plays a very important role in the production of PM 2.5 in winter, which is consistent with previous studies [65,66]. Therefore, measures to limit NH 3 emissions should be considered when reducing PM 2.5 pollution in winter to achieve better pollution control effects. At the regional scale, we found that the impact of residential emissions on PM 2.5 concentrations is dominant in the northern region and part of the southern region (MUYR and EC) through the whole year. Moreover, residential emissions also play a leading role in most seasons in the northern region. This impact may be because coal combustion and biomass combustion accounted for a large proportion of residential sources in the northern region, and these sources exert a greater impact on PM 2.5 concentrations, especially in spring, autumn and winter, which is consistent with existing studies [67][68][69]. In addition, the importance of residential source NH 3 emissions has been neglected. Studies have shown that residential NH 3 emissions exert a significant impact on PM 2.5 concentrations at the regional annual scale and in most seasons in northern China. This finding is consistent with related studies. For example, NH 3 emissions from urban residential sewage account for a relatively high proportion of total NH 3 emissions and have an important impact on PM 2.5 concentrations [39]. It is worth noting that traffic emissions play a leading role in the four seasons for MUYR and EC in southern China, MYR and NC in northern China, and on a national scale. China's national statistics [70] show that from 1990 to 2012, the number of motor vehicles in China has increased 22 times in 22 years, and traffic emissions have become one of the largest sources of air pollution in China. On the other hand, this may be due to the fact that these areas are located in the central and eastern parts of China, with higher road network density, higher car ownership, and accelerated photochemical reactions under suitable weather conditions, which have an important impact on PM 2.5 concentrations [7,71,72].
The effects of primary emissions and secondary precursors on PM 2.5 concentrations have obvious differences at different temporal and spatial scales. At the annual scale, whether national or regional, primary PM 2.5 emissions contribute less to the PM 2.5 concentrations than secondary emissions. This finding is consistent with previous research, indicating that secondary emissions make an important contribution to PM 2.5 concentrations [73]. At the national seasonal scale, the contribution of primary PM 2.5 to the PM 2.5 concentrations is equivalent to that of secondary emissions, and the dominant factor comes from the primary emission of PM 2.5 from transport sources. However, on the regional seasonal scale, the contribution of primary PM 2.5 to PM 2.5 concentrations is significantly different from that of secondary precursor emissions. Therefore, when formulating PM 2.5 control policies, it is necessary to consider the actual conditions of the region and use differentiated emission control policies in different regions and seasons.
This study comprehensively quantifies the impact of different sectors' emissions on PM 2.5 concentrations. The results of the study can help us understand the anthropogenic emission mechanism of PM 2.5 pollution changes and provide a reference for follow-up research and the formulation of policies and measures for related sectors. However, there are some limitations. (1) Due to the uncertainties in the emission inventory adopted, our research results will also have corresponding uncertainties. Moreover, given data availability limitations, the current inventory data used is only for one year (2016); data from other years need to be analyzed in future work. (2) Given length restrictions and the previous research results that have been obtained, this article does not consider meteorological factors in its focus. In the future, it will be necessary to comprehensively consider the impact of land use/cover change, topography, and socioeconomic factors on PM 2.5 concentrations. (3) Related research shows that the concentrations of PM 2.5 and O 3 sometimes fluctuate, which also makes it difficult to formulate pollution control policies. Future research will combine the two pollutants of PM 2.5 and O 3 for analysis, thereby clarifying the impact mechanisms and formulating a more accurate policy reference for the government to comprehensively control air pollution.

Conclusions
The results of this study show that the effects of emission factors among various sectors on PM 2.5 concentrations present large regional and seasonal differences. Residential emissions were the dominant driver at the national annual scale, and also played the leading role at the regional annual scale and during most of the seasons in northern China. Residential NH 3 emissions exert a significant impact on PM 2.5 concentrations at the regional annual scale and in most seasons in northern China. Transportation emissions play a leading role in the four seasons for MUYR and EC in southern China, MYR and NC in northern China, and on a national scale. Compared with primary particulate matter, secondary anthropogenic precursors have a more important effect on PM 2.5 concentrations at the national or regional annual scale.
The influence mechanism of pollutants discharged by different sectors on PM 2.5 concentrations is more complicated, which also makes it difficult for us to clarify the contributions of various factors to PM 2.5 concentrations. This study comprehensively and quantitatively assessed the impact of anthropogenic emissions from various sectors and their interactions on PM 2.5 concentrations. Given the complex nonlinear relationship between PM 2.5 concentrations and its influencing factors, this study will help us better understand the impact mechanism of PM 2.5 pollution. Moreover, it will help researchers improve the accuracy of PM 2.5 prediction models and help governments formulate more precise PM 2.5 pollution control policies.  Figure S9: Annual and seasonal q values of driving factors for PM 2.5 concentrations between PM 2.5 concentrations and impacting factors at the national scale (China). P_PM denotes primary PM 2.5 ; NO denotes NO X ; VO denotes VOCs; NH denotes NH 3 ; SO denotes SO 2 . Figure S10. Annual and seasonal q values of driving factors for PM 2.5 concentrations between PM 2.5 concentrations and impacting factors at the regional scale (China). P_PM denotes primary PM 2.5 ; NO denotes NO X ; VOCs VO denotes VOCs; NH denotes NH 3 ; SO denotes SO 2 . Table S1: Effect of various anthropogenic factors with different emission sectors on PM 2.5 concentrations in China in 2016. Table S2: Effect of various anthropogenic factors with different emission sectors on PM 2.5 concentrations throughout the whole year at the regional scale. Table S3: Effect of various anthropogenic factors with different emission sectors on PM 2.5 concentrations in spring at the regional scale. Table S4: Effect of various anthropogenic factors with different emission sectors on PM 2.5 concentrations in summer at the regional scale. Table S5: Effect of various anthropogenic factors with different emission sectors on PM 2.5 concentrations in autumn at the regional scale. Table S6: Effect of various anthropogenic factors with different emission sectors on PM 2.5 concentrations in winter at the regional scale.

Institutional Review Board Statement:
The studies not involving humans or animals. So ethical review and approval were not applicable for this study.

Informed Consent Statement:
Informed consent obtained from all subjects or patient consent was waived are not applicable for studies not involving humans.

Conflicts of Interest:
The authors declare no conflict of interest.