EN中文

Journal of Geographical Sciences >

2022 , Vol. 32 >Issue 5: 853 - 872

DOI: https://doi.org/10.1007/s11442-022-1975-8

Research Articles

Regional variation of urban air quality in China and its dominant factors

  • ZHAO Yanyan , 1 ,
  • ZHANG Xiaoping , 1, * ,
  • CHEN Mingxing 2, 1 ,
  • GAO Shanshan 1 ,
  • LI Runkui 1
Expand
  • 1. College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 100049, China
  • 2. Key Laboratory of Regional Sustainable Development Modeling, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
* Zhang Xiaoping (1972-), PhD and Professor, specialized in economic geography and regional sustainable development. E-mail:

Zhao Yanyan (1995-), Master, specialized in regional sustainable development.E-mail:

Received date: 2021-11-29

  Accepted date: 2022-01-15

  Online published: 2022-07-25

Supported by

National Natural Science Foundation of China(41771133)

National Natural Science Foundation of China(41822104)

The Strategic Priority Research Program of the Chinese Academy of Sciences(XDA19040403)

Fold

Abstract

It is of great theoretical and practical importance to carry out research on the spatio-temporal evolution of urban air pollution and its driving forces, which helps to facilitate a deeper understanding of the mutual feedback mechanisms between the urban environment and socio-economic systems. Comprehension of these mechanisms will contribute to the design and implementation of efficient environmental policies that ultimately will improve the quality of urbanization development. This paper illustrates the spatio-temporal evolutionary characteristics of six urban ambient air pollutant concentrations, namely, CO, NO2, O3, PM10, PM2.5, SO2, in 286 sample cities above the prefecture level in China from 2014 to 2019. The interactions between the pollutant concentrations are analyzed based on panel regression models. A random forest model is then employed to explore the correlations between the concentrations of these six pollutants and 13 natural and socio-economic impact factors to isolate the most crucial ones. The results reveal three aspects. First, within the research period, the average annual concentration of O3 increased while that of other pollutants decreased year by year. Second, there were significant interactions between concentrations of the six pollutants, leading to obvious compound air pollution in urban areas. Third, the impact of natural and socio-economic factors on urban air quality varied greatly among different air pollutants, with air temperature, vegetation coverage, urbanization level and traffic factors ranking high and the different response thresholds to the dominant influencing factors. In light of the limited ability of humans to control the natural environment and meteorological conditions, it is recommended that urban air quality be further improved by optimizing urban density, controlling anthropogenic emission sources, and implementing strict air pollution prevention and control measures.

Key words： urban air quality; spatio-temporal evolution; random forest model; impact factor

Cite this article

ZHAO Yanyan , ZHANG Xiaoping , CHEN Mingxing , GAO Shanshan , LI Runkui . Regional variation of urban air quality in China and its dominant factors[J]. Journal of Geographical Sciences, 2022 , 32(5) : 853 -872 . DOI: 10.1007/s11442-022-1975-8

1 Introduction

With China’s ongoing industrialization and urbanization processes, urban economic development has achieved remarkable successes. Yet, the metropolitan areas face severe environmental challenges, especially in the densely populated urban areas where haze, PM2.5, PM10, and other air pollutants are prominent, seriously affecting the health of residents and quality of their life in general. China has been continuously expanding its efforts to monitor and control atmospheric pollution. In this spirit, China has implemented the “Air Pollution Prevention and Control Action Plan,” which has outlined time-dependent air pollution regulations and controls for key industries in the Beijing-Tianjin-Hebei region, the Yangtze River Delta, and other key regions since 2014. Therefore, it is helpful to evaluate the effects of this policy implementation on urban air quality during this transition period. Further research on the impact factors of urban air quality could provide better targeted policy recommendations and decision-making references for the prevention and control of air pollution to improve the quality of urbanization development and promote refinement of urban governance.
Academia has carried out extensive research on the pressing issue of regional air quality and achieved rich results in which its perspective has mainly focused on two aspects. The first is related to the spatio-temporal variation of regional air quality. The characterization indexes of air quality usually have included a single pollutant, such as SO2, NO2, O3, PM10, and PM2.5 (Liu and Du, 2016; Xu and Li, 2016; Li et al., 2020) and a comprehensive air quality index, such as air pollution index (API) (Deng et al., 2013) and air quality index (AQI ) (Lin and Wang, 2016; Zhang et al., 2020) among others. The time scales involved in the research include daily, monthly, seasonal, and inter-annual evolutionary characteristics (Chen et al., 2015; Dai et al., 2020). The regional scale and spatial scope are also diverse, and include countries and provinces (Guo et al., 2019; Wang et al., 2019), urban agglomerations (such as the Beijing-Tianjin-Hebei urban agglomeration and the surrounding areas (Cheng et al., 2019), and the Yangtze River Delta (Chen et al., 2017)). Such features as the spatial agglomeration and spatial spillover can affect air pollution (Li et al., 2019). The second aspect is concerned with the impact factors of air quality. Air pollution is formed under the dual effects of natural meteorological factors and socio-economic activities, so its overall mechanism is more complex. Research shows that climatic-meteorological factors such as dust (Liao et al., 2016), wind (Yu et al., 2018), precipitation (Yang et al., 2019), and temperature (Wang and Wang, 2018), together with physio-geographical factors such as topography (Long et al., 2016) and vegetation (Liu et al., 2017b), exert a direct impact on the detention time and concentration of pollutants to either strengthen or slow down the interaction between pollutants. Such interactions ultimately affect the regional air quality through regional transmission, diffusion of air pollutants, and self-purification of air. Meanwhile, socio-economic activities are the main source of air pollutants. Some scholars have discussed the influence of socio-economic factors such as economic growth, industrial structure (Liu et al., 2017c), urbanization (Jiang et al., 2019), industrialization (Wang et al., 2019b), land use (Xu et al., 2015), foreign direct investment (FDI) (Jiang et al., 2018), and population density (Liu et al., 2017a) on air quality. Their research methods include a spatial econometric model (Bai et al., 2018; Liu et al., 2018), a geographically weighted regression (GWR) model (Bai et al., 2019), a geographic detector model (Zhang and Gong, 2018), principal component analysis, and a multiple linear regression model (Cheng et al., 2019). Generally speaking, the existing research on regional air quality mainly focuses on a single pollutant or the spatio-temporal characteristics of multiple pollutants over a single year. Yet, research that focuses on comprehensive evaluations based on multiple pollutants over multiple years is limited. Research on impact factors has paid less attention to the interaction between different types of air pollutants. In addition, most models are affected by the collinearity between variables, thus restricting their interpretation and robustness, causing them to be potentially ineffective in identifying the dominant impact factors. Urbanization, a geographical process dominated by human activities, is subjected to mutual feedback with the environmental system (Chen, 2015). The urban atmospheric-environmental system results from interactions between humans and nature; therefore, research on the impact factors and driving mechanisms of urban air quality still needs to be deepened.
Expanding upon the progress of existing research, this paper further addresses two issues. First, what are the regional differences in the patterns and evolutionary trends of the urban concentrations of six air pollutants in China since the adoption of the new “Ambient Air Quality Standard” (GB3095-2012) and the full implementation of the “Air Pollution Prevention and Control Action Plan” in 2014? Second, what are the differences in the leading factors affecting the urban concentrations of six air pollutants, and to what extent do socio-economic activities impact these concentrations? The answers to the above questions are expected to clarify the causes and evolutionary mechanisms related to urban air pollution and formulate more accurate air pollution prevention and control measures and strategies, thus to achieve high-quality development in urban areas by improving urban management and governing efficiency. In light of this, this paper performs a comprehensive assessment of urban air pollution in China from 2014 to 2019 regarding six air pollutants (CO, NO2, O3, PM10, PM2.5, and SO2) in 286 sample cities above prefecture level in China. The panel regression model is used to analyze the interaction between pollutants. The random forest model is employed to explore the correlation strength between the concentrations of six kinds of air pollutants and thirteen natural and socio-economic impact factors, from which the key influencing factors are extracted. Given that the random forest model can quickly process a large amount of data with robust results (Chen et al., 2018; Liu et al., 2019), it is expected to overcome the restriction of multi-collinearity in traditional research to achieve good fitting and provide a more in-depth exploration regarding the analysis of the influencing factors of urban air quality and selection of its leading factors.

2 Data and methodology

2.1 Research area and data sources

Considering the changes in China’s administrative divisions and the lack of data in some cities, this paper selects 286 cities above prefecture level as the research samples. The data of urban air pollutants are derived from the monitoring data of CO, NO2, SO2, PM10, PM2.5, and O3 of China’s environmental monitoring stations from 2014 to 2019 (the number of stations in each year is shown in Table 1). Invalid data are eliminated, and the remaining quality-checked data are finally sorted into annual data according to the provisions on data validity in the “Ambient Air Quality Standard (GB3095-2012)” (Ministry of Environmental Protection of the People’s Republic of China, 2012). The evaluation index of O3 is the average annual concentration of the 8-hour maximum moving average of O3 (expressed hereafter as O3_8h), while the other five pollutants are given as annual average concentrations. The annual average data of urban pollutants come from the arithmetic average of the data from each monitoring station (excluding the control point) in the city (Cheng et al., 2017). When there is too much missing data in some cities to adequately summarize, these missing data are substituted with the data from the city’s environmental quality bulletin or by linear interpolation of the relevant time series. Due to a large amount of missing data in O3_8h from prefecture-level cities in 2014, the research period of O3_8h in this paper spans the five years from 2015 to 2019.
Table 1 The number of urban air quality monitoring stations in China from 2014 to 2019
Year 2014 2015 2016 2017 2018 2019
Number of sites 902 1446 1446 1545 1553 1589

Note: The number of stations in each year ends on December 31, and only valid stations with coordinate data are counted.

This paper divides the impact factors on urban air pollutant concentration into natural and socio-economic factors (Figure 1). Natural factors consist of four meteorological factors (annual precipitation, annual daily mean wind speed, annual daily mean relative humidity, annual daily mean temperature) and a vegetation index, affecting processes such as air pollutant transport, diffusion, secondary pollution, sedimentation, and digestion. The meteorological data come from the statistical yearbook and bulletin of various provinces and cities, and the missing data are collected from daily data sets of China’s surface climate data (V3.0) which is from the China Meteorological Data Service Centre (http://data.cma.cn) and computed using the Kriging interpolation method. The vegetation index data are based on the spatial distribution of the NDVI (normalized difference vegetation index) in China, collecting from the Resource and Environmental Science and Data Center of Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences (http://www.resdc.cn). Scholars have pointed out the influence of terrain, landform, slope, and other landform factors on air quality (Long et al., 2016; Bai et al., 2018; Wang et al., 2019), but their values are too stable over the study period to introduce into the panel analysis model and are thus not considered in this paper. Socio-economic factors are mainly related to the intensity of urban socio-economic activities as well as production and lifestyle, including eight indicators, such as economic development level, population agglomeration degree, power consumption, industrial structure, transportation, the intensity of social activities, urbanization level, science and technological investment, with the specific quantitative methods shown in Figure 1. The socio-economic factor data mainly come from the “China City Statistical Yearbook” and “China Urban Construction Statistical Yearbook” from 2016 to 2019, yet some missing data are supplemented by statistical yearbooks and bulletins of various provinces and cities, trend extrapolation, and linear interpolation methods. Research has shown a close relationship between the night light image data and population density, GDP, energy consumption, and lifestyle of residents (Su et al., 2013; Wu et al., 2014; Xiao et al., 2018). Therefore, this paper uses the night light image data as a proxy to reflect the comprehensive intensity of human socio-economic activities in the city, and these data come from the earth nighttime light data set from the Chinese Academy of Sciences in 2018 (https://www.jianshu.com/p/5fde55a4d267). These related brightness values are applied to the prefecture-level cities. Since academia has already mentioned the role of environmental control factors in optimizing urban air quality, this study does not model these factors due to the restriction of the availability and continuity of indicators. Instead, when it comes to the analysis of policy factors, this paper divides the 286 cities into cities with key environmental protections from those other 173 cities to compare the varied differences of air pollutant concentrations in these two kinds of cities with the intent of clarifying the regulatory role of environmental control factors on urban air quality.
Figure 1 Explanatory variables for urban air quality in China and their descriptions

2.2 Research methods

2.2.1 Kernel density estimation

As a non-parametric estimation method to analyze the equilibrium degree of data distribution, the kernel density estimation model is used to describe the overall distribution and regional concentration differences of urban air pollutants, which is shown by the following calculation formula,
$f(x)=\frac{1}{nh}\sum\nolimits_{i=1}^{n}{k\left( \frac{x-{{X}_{i}}}{h} \right)}$
where n is the number of prefecture-level cities, $k\left( \frac{x-{{X}_{i}}}{h} \right)$ is a kernel function, and h is bandwidth,the use of which follows the idea of the minimum integral mean square error, and the value of which is related to the smoothness of kernel density distribution (Bai et al., 2015).

2.2.2 Panel regression model

Different air pollutants in cities have varied significant interactions, potentially producing secondary pollution, yet existing research on this is limited. This paper uses six kinds of air pollutants in 286 cities from 2015 to 2019 to build a panel regression model to explore further the interaction between air pollutants, which is expressed as follows,
${{Y}_{it}}={{\alpha }_{i}}+\sum\nolimits_{m=1}^{k}{{{\beta }_{m}}{{X}_{mit}}+{{\theta }_{it}}}$
where Yit, the explained variable, is a certain pollutant selected from the six kinds of air pollutants (CO, NO2, O3_8h, PM10, PM2.5, SO2) in the i-th city (i = 1, 2,..., 286) in the t-th year. Parameter Xmit indicates the value of the m-th explanatory variable of the i-th city in the t-th year, m=1, 2, …, 5 explanatory variables, which correspond to the concentrations of the other five pollutants, except for the explained variable Yit. The parameter θit is the random interference error term and αi represents the individual effect and varies with different cities. If the model test result shows that αi=α, it indicates that there is no difference between individual effects; however, if β is significantly not zero, there is significant interaction between all these pollutants.

2.2.3 Random forest regression model

As a machine learning method with multiple decision tree classifiers, random forest regression models generally use a bootstrap sampling method to extract multiple samples from the original samples to model them as decision trees and then combine these decision trees to get the final result through voting scoring rules (Breiman, 2001). In light of its convenience for the processing of large data sets, insensitivity to multiple collinearities, and robustness in the presence of missing data and unbalanced data, the random forest model can provide relatively high-precision prediction results (Liu et al., 2019; Xia et al., 2020), which can be used both for classification and regression analyses, with the intent of exploring the relative importance of impact factors. We use R language to construct the random forest models for each pollutant and evaluate their related influencing factors, and finally select the leading influencing factors of different pollutants based on the increase in the mean square error and the increase of node purity.

3 Results and analysis

3.1 Spatio-temporal characteristics of urban air quality in China

3.1.1 Temporal evolution characteristics

Considering the six urban air pollutants in China from 2014 to 2019, the concentrations of the five air pollutants showed a downward trend, except for O3_8h (Figure 2). Among these decreasing air pollutants, the annual average concentration of SO2 decreased most significantly by 70%, and that of PM2.5, PM10, and CO decreased by more than 30%, and that of NO2 decreased by about 19%. However, the annual average concentration of O3_8h increased yearly from 2015 to 2018, and it was only in 2019 that O3_8h was slightly lower than that in 2018. Overall, China has achieved phased results in controlling air pollution, evidenced by the obvious decreases in the concentration of traditional pollutants such as SO2 and CO, but its O3 pollution is relatively severe.
Figure 2 Annual average concentrations of six urban ambient air pollutants from 2014 to 2019
The concentration trend and change of urban ambient air pollutants in China are compared with kernel density curves (Figure 3). Apart from O3_8h in 2019, the peaks of the kernel density curves of the other five air pollutants converged to the left compared with those of 2014, thus indicating the concentration of O3_8h was higher than that in 2014 while the other five kinds of air pollutants decreased over the study period, which was consistent with the results in Figure 2. The slow decay of the kernel density curve in 2019 is significantly shorter than that in 2014, which implies an increase in the number of cities with lower pollutant concentrations, while those with higher pollutant concentrations decreased. The temporal characteristics are clearly shown by the shape of the curve forms of SO2 and CO, which transform from the “short and fat type” in 2014 to the “thin and tall type” in 2019, indicating that there are significant differences in the traditional air pollution caused by smoke from burning coal among cities. Most cities were mainly polluted by atmospheric particulate matter, nitrogen oxides, and O3.
Figure 3 Comparison of kernel density curves of six urban ambient air pollutants

3.1.2 Spatial evolution characteristics

Figure 4 depicts the regional differences of concentrations of six urban ambient air pollutants.
Figure 4 Spatio-temporal variation of concentrations of six urban ambient air pollutants at monitoring stations in China
Cities with high CO concentration are mainly resource-intensive industrial cities, such as Hegang, Anshan, Benxi, Baotou, Datong, Changzhi, Jincheng, Tangshan, Xingtai, Handan, Zibo, and Hanzhong. By 2019, the air quality of traditional typical CO pollution areas, such as Xianyang, Baoji, Zhengzhou, Luoyang, Zhuzhou, Chenzhou, Shaoguan, was significantly improved. Compared to CO, the concentration of SO2 has a similar spatial distribution pattern that it is distributed most especially in resource- based cities. Although the SO2 concentration in 2019 was significantly lower than that in 2014, some cities, such as Tangshan, Jinan, Zibo, Taiyuan, Datong, Shuozhou, and the like, still had relatively high SO2 concentrations.
Compared with 2014, the concentration of NO2 in the atmosphere decreased, but the area with higher NO2 concentration did not show an obvious spatial change in 2019. The areas with relatively higher NO2 concentration were mainly focused in the metropolitan areas and economic circles with dense populations and intense socio-economic activities, such as the Beijing-Tianjin-Hebei Urban Agglomeration, the Yangtze River Delta, the Pearl River Delta, the Central China Plains Urban Agglomeration, the Jinan Metropolitan Area, and the Chengdu Metropolitan Area, among others.
Relatively high concentrations of PM10 and PM2.5 occupied wide areas and appeared to be coupled to urban ambient air pollutants. The urban agglomerations of Beijing-Tianjin-Hebei, the Shandong Peninsula, and Central China Plains make large contributions, in addition to typical energy-rich cities as well as those with heavy industry in central and southern Shanxi province, have relatively high pollutant concentration. In 2019, the concentration of PM10 decreased significantly compared with that in 2014. The areas with higher pollutant concentration in the South also shrunk significantly so as to display a more remarkable spatial pattern described by “high in the north and low in the south.” The PM2.5 concentration in most urban stations also decreased, but the distribution range of the area with relatively high values was wider than that of PM10.
The concentration of O3_8h has a spatial distribution trend of high in the east and low in the west. At the beginning and end of the study period, the areas with relatively high O3_8h concentrations were concentrated in Beijing, the Yangtze River Delta Urban Agglomeration, the Shandong Provincial Capital Urban Agglomeration, Dongguan City, and others. In 2019, the area with relatively high ozone pollution expanded significantly compared with 2015, with further intensified ozone pollution observed in the Beijing-Tianjin-Hebei Urban Agglomeration, the Shandong Peninsula, the Yangtze River Delta Urban Agglomeration, and the Central China Plains Urban Agglomeration. Additionally, cities such as Handan, Zhengzhou, Kaifeng, Jiaozuo, Puyang, and Xuzhou also showed significant increases in ozone pollution. Thus, the major urban agglomerations in China face the dual pressure of significant PM2.5 and ozone pollution, resulting in arduous air pollution control measures.

3.2 Analysis of impact factors of urban air quality in China

3.2.1 The interrelationship between pollutants

Based on Equation 2, a panel regression model is constructed to quantitatively analyze the interactions between pollutants to understand further the spatio-temporal coupling effects among different air pollutants between cities. The values of R2 are all above 0.7 with good overall fitting results of each model (Table 2), indicating that the individual fixed effect model is all suitable for model calculations. The results show a significant positive relationship evidenced by strong correlations and intensifying effects between most pollutants, including the extraordinarily positive relationships among CO, NO2, SO2, PM2.5, and PM10. In contrast, O3_8h has significantly negative correlations with CO, PM2.5, and SO2 while air pollutants maintain positive ones with NO2 and PM10, which indicates a complex reaction relationship between nitrogen oxides and other pollutants, thus making the control of their pollution level fundamental for improving air quality. The concentrations of PM10 and PM2.5 show a positive relationship, significant at 99% confidence level, indicating intensified mutual interactions between PM10 and PM2.5. Besides, PM2.5 concentration also has a significant positive relationship with NO2 and SO2 concentration. Inorganic compounds such as sulfate, nitrate, and ammonium salts generated by atmospheric reactions can form PM2.5 under certain conditions with an obvious secondary conversion process between them (Wang et al., 2020), further indicating the complex evolution mechanism of PM2.5 pollution.
Table 2 Interactions between six urban ambient air pollutants from the results of panel data regression models
Variables (1) (2) (3) (4) (5) (6)
CO NO2 O3_8h PM10 PM2.5 SO2
CO - 2.397*** -11.662*** 3.228*** 0.808 15.944***
NO2 0.00470*** - 0.636*** 0.320*** 0.0787** 0.150***
O3_8h -0.00305*** 0.0848*** - 0.0586*** -0.0373*** -0.171***
PM10 0.000197*** 0.0997*** 0.137*** - 0.489*** 0.0458
PM2.5 0.00124 0.0616** -0.219*** 1.229*** - 0.349***
SO2 0.0100*** 0.0482*** -0.412*** 0.0473 0.144*** -
Constant 0.681*** 8.467*** 89.084*** 5.296** 3.576** -4.619**
Observations 1430 1430 1430 1430 1430 1430
R-squared 0.861 0.935 0.741 0.972 0.965 0.878
Number of ID 286 286 286 286 286 286

Note: *, **, and *** represent the significance level of 10%, 5%, and 1%, respectively.

3.2.2 Natural and socio-economic factors based on random forest model

This paper used the annual average concentrations of CO, NO2, O3_8h, PM10, PM2.5, SO2 as the explained variable separately and the variables listed in Figure 1 as explanatory variables to construct six regression models to analyze the impact of the main natural and socio-economic factors on urban air quality. Due to the lag of socio-economic statistical data among the explanatory variables in 2019, the period for modeling analysis is selected from 2015 to 2018. A random forest model is employed to analyze the impact factors of urban air pollutants, among which the number of parameter decision trees, denoted by ntree, is selected according to the principles of small out-of-band errors and small amount of calculations; while the number of features, denoted by mtry, is chosen according to the principle of minimum average errors. The results show that the overall goodness-of-fit of the six regression models for six urban ambient air pollutants are 63.05%, 77.59%, 56.63%, 82.50%, 76.78%, and 61.81%, respectively.
Figure 5 demonstrates the importance ranking of the 13 influencing factors as computed by the random forest method when six pollutants are used as the explained variables. In Figure 5, IncMSE represents the increase of the estimated error in the random forest model compared with the original error after the values of each variable are assigned randomly; IncNodePurity is the increment of node purity, indicating the degree of influence for variables on each decision tree node. For both methods, the larger the value, the more important the variable (Wang et al., 2019a). Factors that rank first in these two methods can be determined as the dominant ones. On the whole, meteorological factors have a relatively strong influence on pollutants. In terms of socio-economic factors, the level of science and technology investment has little influence on the concentration of pollutants, thought to be because technology transformation takes a certain time, causing a certain lag without exerting a significant influence in a short time.
Figure 5 Importance ranking of the impact factors on six urban ambient air pollutants
Due to length limitations, this paper only analyzes the top four factors that strongly impact each pollutant in detail, determined by its ranking in IncMSE. Among the meteorological factors, the annual average temperature has the greatest impact on the concentrations of the other five air pollutants, except for O3_8h. The vegetation index strongly affects the concentrations of O3_8h, PM10, and SO2. Among the socio-economic factors, land urbanization generally exerts a strong influence on pollutants, except for O3_8h and SO2, with the second and third rankings, respectively. The proportion of secondary production greatly affects the concentration of CO, NO2, PM2.5, SO2, and PM10, which ranks second only to the urbanization factor. Also, the influence of power consumption and traffic factors is greater on O3_8h. In terms of pollutants, the influencing factors of CO concentration include (in ranked order): annual average temperature, population density, land urbanization, and proportion of secondary industry (Figure 5a). Unlike other pollutants, NO2 concentration is strongly influenced by annual relative humidity, second only to annual average temperature within natural factors (Figure 5b). For O3_8h concentrations, the ranked influencing factors are vegetation index, electricity consumption, annual average temperature, and transportation, from top to bottom (Figure 5c). The PM10 concentration is most affected by the annual average temperature and vegetation index among the natural factors, and relatively less affected by the land urbanization and the proportion of secondary industry regarding the socioeconomic factors, which is similar to that for CO, NO2, and PM2.5 (Figure 5d). As with many of the other pollutants, the annual average temperature has the greatest impact on PM2.5 concentration; Figure 5e shows that the other three dominant factors, for PM2.5, all belong to socio-economic factors, namely, the level of land urbanization, the proportion of secondary industry, and the level of economic development. The SO2 concentration is largely impacted by population density and the proportion of secondary production, in addition to the temperature and vegetation index (Figure 5f).
To further explore the effects of the leading influencing factors, this paper uses the partial dependence diagram of the impact factors to show the differences. Figure 6 demonstrates the results, yet only the responses of the top four impact factors on the six kinds of air pollutants are listed.
Figure 6 Plots of partial dependence of the impact factors on six urban ambient air pollutants
Figure 6 indicates a nonlinear response relationship between the impact factors and the concentration of pollutants. From the perspective of natural factors, the impact of the annual average temperature on the concentration of pollutants shows a nonlinear trend of “first rising and then falling” together with various thresholds of each pollutant concentration responding to mutation. Taking temperature as an example, the concentrations of NO2, O3_8h, PM10, and PM2.5 increase along with increasing air temperatures when the annually-averaged air temperature is lower than 15℃, before gradually decreasing with the increase of temperature when it is above 15℃. However, the concentrations of CO and SO2 show a positive response when the annual air temperature is below 10℃, and a negative response when it is above 10℃. The impact of annual daily mean relative humidity on the concentration of NO2 reflects a response relationship of “first steady, then up, and then down” which means that the NO2 concentration increases significantly with the increase of relative humidity when it is about 54% to 64%. In addition, the concentrations of O3_8h, PM10, and SO2 vary in their response trends towards the vegetation index. That of O3_8h responds in the type of “first fluctuating and then sharply decreasing” with the increase of vegetation coverage. A plausible reason is that an increased vegetation index can reduce the wind speed and hinder the trans-regional transmission of pollutants, thus promoting the deposition of particulate matter. This would have the net effect of significantly decreasing the concentration of pollutants in the atmosphere, consistent with the conclusions from other scholars (Liu et al., 2018; Wang et al., 2013; Wang, 2014).
Regarding the socio-economic factors, the concentrations of CO, NO2, PM10, and PM2.5 in response to land urbanization have a trend of “first rising and then stabilizing,” which means that the air pollution is rising when the level of urbanization gradually increases from a relatively lower level. Yet, its impact stabilizes when the urbanization level is relatively high because the emission characteristics of processes related to further rapid urban expansion weaken. The concentrations of CO, NO2, PM10, PM2.5, and SO2 positively respond to the proportion of secondary production, which enters an accelerated period of influence when the proportion of secondary production exceeds 20%. This indicates that increases beyond 20% will aggravate the concentration of various air pollutants due to the increased emissions of air pollutants caused by the development of the heavy chemical industry in the acceleration period of industrialization. Additionally, the change in the concentrations of CO and SO2 in response to population density suggests a trend of “first decreasing while fluctuating, then rising while fluctuating and then stabilizing” with the mutation thresholds of population density as 3200 people/km2 and 1800 people/km2, respective to the two pollutants. Moderate increases in population density enhance the intensity of human activities; however, the atmospheric environment at this time has a certain carrying capacity without causing a synchronous aggravation of air pollution. It follows that when the population density exceeds a given threshold, the air carrying capacity reaches a certain upper limit, and the intensity of production and living activities continuously leads to increases in pollutant emission intensity, resulting in the aggravation of air pollution. The concentration of O3_8h positively responds to the “gradual increase” of total electricity consumption in society while demonstrating a trend of “first rising and then falling” in response to increased traffic. Its mechanisms may be related to the compound pollution of nitrogen oxides and secondary pollution of vehicle exhaust. The concentration of PM2.5 has a “first positive and then negative” response trend to the level of urban economic development, which shows that the coordination and unification of economic development and environmental protection could be achieved by the improvement of the quality of economic development, rather than the so-called antagonistic relationship between economic development and environmental protection. This result is consistent with the view of an inverted U-shape of the environmental Kuznets curve, but it still needs to be verified by subsequent studies.

3.2.3 The role of environmental regulation

Environmental regulation and planning guidance play an important role in the comprehensive management of urban air quality. In “The National Eleventh Five-year Plan for Environmental Protection (2006-2010)” issued by the State Council of China in 2007, 113 cities were identified as key cities for environmental protection and were required to carry out comprehensive measures to prevent and control air pollution. China’s government has successively issued the “National Plan on New Urbanization” and the “Integrated Reform Plan for Promoting Ecological Progress” as well as various other environmental plans since 2014. Through structural optimization and economic transformation and other pollution control measures, the urban environmental quality of China has been greatly improved. This paper compares the air pollutant concentrations of 113 key cities of environmental protection and 173 other key cities from 2015 to 2019 with the intent of reflecting upon the comprehensive impact of environmental regulation factors on air quality. Compared with 2015, the concentration of pollutants in 2019, except O3_8h, decreased more in 113 key environmental protection cities than other key cities, indicating that the regional environmental regulation measures put in place have had a significant effect on the control of most pollutants (Table 3). However, the relatively implicit effects of regional environmental regulation regarding the treatment of O3 pollution are related to the fact that as a typical secondary pollutant, O3 has wide precursor sources, restricting the control effect of current environmental policies on the synergistic effect of its precursors.
Table 3 Concentration changes of six ambient air pollutants in 2019 compared with 2015 (by city type)
Concentration of ambient air pollution CO
(mg/m3)		 NO2
(μg/m3)		 O3_8h
(μg/m3)		 PM10
(μg/m3)		 PM2.5
(μg/m3)		 SO2
(μg/m3)
Variation 113 key environmental protection cities -0.28 -3.62 10.23 -23.51 -14.38 -16.14
173 compared key cities -0.27 -1.93 9.46 -18.31 -12.30 -13.28
Rate of
change (%)		 113 key environmental protection cities -25.29 -10.3 12.13 -25.18 -26.56 -58.38
173 compared key cities -25.92 -7.10 11.54 -21.84 -24.84 -54.19
The above analysis shows that urban air quality is affected by natural and socio-economic factors, together with the interactions between different kinds of pollutants. Considering the variety of impact factors and pollutants, this paper explores the importance of influencing factors on the concentration of urban air pollutants, with the results shown in Figure 7. The main impact factors of air pollutant concentrations have similarities yet also have variations among them. In terms of natural factors, the annual average temperature is one of the leading factors of all pollutants, while the vegetation index, annual daily mean relative humidity, precipitation, wind speed are leading factors of specific pollutants. Regarding socio-economic factors, land urbanization and the proportion of secondary industry are the dominant factors, while population density, power consumption, and transportation have a greater impact on the concentration of certain pollutants.
Figure 7 Flow chart of core impact factors on urban ambient air pollutants

4 Conclusions and discussion

4.1 Conclusions

The urban atmospheric environment is a multi-scale and multi-element system with a complex driving relationship and mutual feedback mechanisms between elements. Based on the monitoring data of the China National Environmental Monitoring Centre from 2014 to 2019, this paper comprehensively portrays the air pollution characteristics of 286 cities above prefecture level in China in terms of concentrations of CO, NO2, O3, PM10, PM2.5, and SO2, to explore the impact factors and dominant mechanisms which appear to drive urban air pollution.
The conclusions are as follows.
First, urban air quality in China has generally improved during the research period. Out of six air pollutants, only O3_8h concentration increased, the concentrations of the remaining five air pollutants showed a downward trend, among which SO2 experienced the steepest decline. Because the O3 concentration is rising, its driving mechanism is worthy of further study. The traditional atmospheric pollution related to burning coal dramatically improved, but this type of pollution significantly varied among different cities. Some resource-based heavy industrial cities still have high concentrations of SO2 and CO. Most cities are polluted mainly by atmospheric particulate matter, nitrogen oxides, and ozone. Although there was a reduction in typical heavily polluted areas, air pollution issues in Beijing, Tianjin, Hebei, Shanxi, Shandong, Henan, and some metropolitan areas are still prominent.
Second, there are significant interactions between the six ambient air pollutants. Most pollutants are significantly correlated with one another. The secondary pollution caused by the complex interactions between air pollutants increases the difficulty of air pollution control, thus requiring further joint prevention, targeted controls, and comprehensive treatment.
Third, natural and socio-economic factors exert different influences on the concentration of different kinds of air pollutants, and some have a nonlinear relationship between them. Among the natural factors, the urban annual average temperature is most strongly related to air pollutant concentration, followed by vegetation index. For the socio-economic counterpart, the leading factors include the level of land urbanization, the proportion of secondary industry, and the total power consumption, with traffic factor ranking the second among other factors. Also, the partial dependence analysis further simulates the response to mutation thresholds for different pollutant concentrations and dominant influencing factors. The conclusions of this research further support the existence of complex mutual feedbacks between human socio-economic activities and regional ambient air quality under the constraints of physical geography, meteorology, and urban environmental carrying capacity. Therefore, the comprehensive integration of theoretical and empirical frameworks into the impacts of key socio-economic factors on the urban environment is suggested for future research.

4.2 Discussion

First, the multi-dimensional urbanization process needs to be deconstructed to realize an urban economy-environment system’s coordination and sustainable development. Only by taking into account the multi-dimensional scientific measures of population, land, economy, and society can we achieve the goal of sustainable development. Caused by both natural and human-induced factors, the formation and evolutionary mechanisms of urban air pollution are very complicated and need to be discussed from a multi-disciplinary perspective. This is especially true after China put forward the lofty goals of achieving “peak carbon” and “carbon neutrality.” Urban areas would be tasked with multiple goals within these endeavors, such as carbon emission reduction and air pollution prevention and controls. Future research should further strengthen the cross-integration of the natural and social sciences to build an interdisciplinary perspective to plan and design a feasible path to achieve comprehensive and sustainable urban development.
Second, the results show that air temperature and vegetation coverage rank as the two most important natural factors due to their strong influence upon pollutant concentrations, while wind speed, precipitation, and other factors follow behind in terms of their comprehensive influence after taking an annual average, though they may have obvious short-term impacts. In the future, the effects of climate change and global warming will affect the large-scale background of urban air pollution controls. People can favorably alter the urban natural environmental factors by constructing sponge cities, afforestation, improving upon urban microcirculation, and slowing down the urban heat island effect.
Third, the processes of urbanization, industrialization, transportation, and energy consumption are still the dominant socio-economic factors affecting the concentration of urban air pollutants. The pathways to achieving high-quality economic development should include considering the mechanisms of policy regulation and planning guidance to optimize the socio-economic spatial layout and upgrade industrial infrastructure. Identifying the threshold limits of key impact factors, such as the inflection point effect of population density at 1800 people/km2 and 3200 people/km2, directly help future planning regulation. Future studies could focus on exploring the differences among impact factors of air pollutants in different types of cities, in order to make more educated and accurate suggestions for policy-making.
Fourth, since the regional economy-environment system is a multi-scale coupled complex, multi-regional scale analysis and interdisplinary study can be carried out in the future. Adding explainary variables and research samples, extending the research period, and employing new analysis methods (e.g., machine learning) could also be considered. Taking these steps should help to further clarify the mechanisms and pathways needed to achieve high-quality urban development as well as providing a valuable reference for more accurate and effective governance countermeasures.

References

Publishing order | Descend order by publishing year | Descend order by cited within
[1]
Bai C Q, Wang X M, Xie Y Q et al., 2015. Evolutionary tendency and income gap of county residents in Yangtze River Delta: Based on kernel density estimation method. Acta Agriculturae Zhejiangensis, 27(3): 498-503. (in Chinese)

[2]
Bai L, Jiang L, Zhou H F et al., 2018. Spatio-temporal characteristics of air quality index and its driving factors in the Yangtze River Economic Belt: An empirical study based on Bayesian spatial econometric model. Scientia Geographica Sinica, 38(12): 2100-2108. (in Chinese)

[3]
Bai L, Jiang L, Zhou H F et al., 2019. Spatiotemporal heterogeneity of air quality index and its socio-economic factors in the Yangtze River Economic Belt. Research of Soil and Water Conservation, 26(2): 312-319. (in Chinese)

[4]
Breiman L, 2001. Random forests. Machine Learning, 45(1): 5-32.

DOI

[5]
Chen D L, Lu X H, Kuang B, 2018. Dynamic evolution and spatial convergence of urban land use efficiency in the middle reaches of the Yangtze River. China Population, Resources and Environment, 28(12): 106-114. (in Chinese)

[6]
Chen M X, 2015. Research progress and scientific issues in the field of urbanization. Geographical Research, 34(4): 614-630. (in Chinese)

[7]
Chen T, Deng S L, Gao Y et al., 2017. Characterization of air pollution in urban areas of Yangtze River Delta, China. Chinese Geographical Science, 27(5): 836-846.

DOI

[8]
Chen W, Tang H Z, Zhao H M, 2015. Diurnal, weekly and monthly spatial variations of air pollutants and air quality of Beijing. Atmospheric Environment, 119: 21-34.

DOI

[9]
Cheng L J, Wang S, Gong Z Y et al., 2017. Pollution trends of ozone and its characteristics of temporal and spatial distribution in Beijing-Tianjin-Hebei region. Environmental Monitoring in China, 33(1): 14-21. (in Chinese)

[10]
Cheng Y, Liu T T, Zhao Y L et al., 2019. Spatiotemporal evolution and socio-economic driving mechanism of air quality in Beijing-Tianjin-Hebei and surrounding areas (“2+26” cities). Economic Geography, 39(10): 183-192. (in Chinese)

DOI

[11]
Dai F, Chen M, Yang B, 2020. Spatiotemporal variations of PM2.5 concentration at the neighborhood level in five Chinese megacities. Atmospheric Pollution Research, 11(6): 190-202.

DOI

[12]
Deng X J, Liao L Q, Hu G P, 2013. Air pollution index and their correlation with meteorological data in major cities of China during the last decades. Environmental Science & Technology, 36(9): 70-75, 80. (in Chinese)

[13]
Guo Y M, Lin X Q, Bian Y, 2019. The spatial-temporal characteristics and influencing factors of air quality in China’s urban agglomerations. Ecological Economy, 35(11): 167-175. (in Chinese)

[14]
Jiang L, Zhou H F, Bai L, 2018. Spatial heterogeneity analysis of impacts of foreign direct investment on air pollution: Empirical evidence from 150 cities in China based on AQI. Scientia Geographica Sinica, 38(3): 351-360. (in Chinese)

DOI

[15]
Jiang L, Zhou H F, Bai L, 2019. Impact of urbanization on urban air quality based on spatial econometric models. Tropical Geography, 39(3): 461-471. (in Chinese)

[16]
Li H J, Qi Y J, Li C et al., 2019. Routes and clustering features of PM2.5 spillover within the Jing-Jin-Ji region at multiple timescales identified using complex network-based methods. Journal of Cleaner Production, 209: 1195-1205.

DOI

[17]
Li K, Liu M, Mei R B, 2020. Pollution characteristics and sensitivity analysis of atmospheric ozone in Taian city. Environmental Science, 41(8): 3539-3546. (in Chinese)

[18]
Liao Q Y, Luo B, Du Y S et al., 2016. Effects and characteristics of floating dust weather on air quality in Sichuan Basin. Environmental Monitoring in China, 32(5): 51-55. (in Chinese)

[19]
Lin X Q, Wang D, 2016. Spatio-temporal variations and socio-economic driving forces of air quality in Chinese cities. Acta Geographica Sinica, 71(8): 1357-1371. (in Chinese)

[20]
Liu H J, Du G J, 2016. Spatial pattern and distributional dynamics of urban air pollution in China: An empirical study based on AQI and six sub-pollutants of 161 cities. Economic Geography, 36(10): 33-38. (in Chinese)

[21]
Liu H J, Sun Y N, Chen M H, 2017a. Dynamic correlation and causes of urban haze pollution. China Population, Resources and Environment, 27(3): 74-81. (in Chinese)

[22]
Liu H M, Fang C L, Huang J J et al., 2018. The spatial-temporal characteristics and influencing factors of air pollution in Beijing-Tianjin-Hebei urban agglomeration. Acta Geographica Sinica, 73(1): 177-191. (in Chinese)

[23]
Liu H M, Fang C L, Zhang X L et al., 2017b. The effect of natural and anthropogenic factors on haze pollution in Chinese cities: A spatial econometrics approach. Journal of Cleaner Production, 165: 323-333.

DOI

[24]
Liu W D, Tang Z P, Xia Y et al., 2019. Identifying the key factors influencing Chinese carbon intensity using machine learning, the random forest algorithm, and evolutionary analysis. Acta Geographica Sinica, 74(12): 2592-2603. (in Chinese)

[25]
Liu Y Q, Ma Y X, Wu Z X, 2017c. The impact of industrial structure upgrade in Beijing-Tianjin-Hebei region on air pollution. Urban Problems, (12): 65-71. (in Chinese)

[26]
Long X, Tie X X, Cao J J et al., 2016. Impact of crop field burning and mountains on heavy haze in the North China Plain: A case study. Atmospheric Chemistry and Physics, 16(15): 9675-9691.

[27]
Ministry of Environmental Protection of the People’s Republic of China. Ambient Air Quality Standards. National Environmental Protection Standards of the People’s Republic of China (GB3095-2012), 2012-02-29. (in Chinese)

[28]
Su Y X, Chen X Z, Ye Y Y et al., 2013. The characteristics and mechanisms of carbon emissions from energy consumption in China: Using DMSP/OLS night light imageries. Acta Geographica Sinica, 68(11): 1513-1526. (in Chinese)

[29]
Wang C, Kan A K, Zeng Y L et al., 2019a. Population distribution pattern and influencing factors in Tibet based on random forest model. Acta Geographica Sinica, 74(4): 664-680. (in Chinese)

[30]
Wang F Y, Zheng J, Wang Z S, 2019b. Analysis of the influence of urbanization and industrialization level on air quality: Based on the space-time model of data from 16 cities in Hubei Province in 2005-2017. Resources and Environment in the Yangtze Basin, 28(6): 1411-1421. (in Chinese)

[31]
Wang H, Lu S W, Li S N et al., 2013. Inhalable particulate matter and fine particulate matter: Their basic characteristics, monitoring methods, and forest regulation functions. Chinese Journal of Applied Ecology, 24(3): 869-877. (in Chinese)

PMID

[32]
Wang J J, Xia X S, Cheng X F et al., 2020. Temporal and spatial distribution characteristics and influencing factors of PM2.5 concentration in Hefei city. Resources and Environment in the Yangtze Basin, 29(6): 1413-1421. (in Chinese)

[33]
Wang W, 2014. Characteristics of negative air ion concentration and its relationships with environmental factors. Ecology and Environmental Sciences, 23(6): 979-984. (in Chinese)

[34]
Wang Y D, Wang Z X, 2018. Characteristics of PM2.5 concentration variability and its meteorological factors in Shanghai. Arid Land Geography, 41(5): 1088-1096. (in Chinese)

[35]
Wang Z B, Liang L W, Wang X J, 2019. Spatio-temporal evolution patterns and influencing factors of PM2.5 in Chinese urban agglomerations. Acta Geographica Sinica, 74(12): 2614-2630. (in Chinese)

[36]
Wu J S, Niu Y, Peng J et al., 2014. Research on energy consumption dynamic among prefecture-level cities in China based on DMSP/OLS nighttime light. Geographical Research, 33(4): 625-634. (in Chinese)

[37]
Xia X S, Chen J J, Wang J J et al., 2020. PM2.5 concentration influencing factors in China based on the random forest model. Environmental Science, 41(5): 2057-2065. (in Chinese)

[38]
Xiao H W, Ma Z Y, Mi Z F et al., 2018. Spatio-temporal simulation of energy consumption in China’s provinces based on satellite night-time light data. Applied Energy, 231: 1070-1078.

DOI

[39]
Xu Q L, Li H, 2016. Study on the pollution characteristics and correlation of PM2.5 and NO2 in the urban area of Chenzhou City. China Population, Resources and Environment, 26(Suppl.2): 80-82. (in Chinese)

[40]
Xu S, Zou B, Pu Q et al., 2015. Impact analysis of land use/cover on air pollution. Journal of Geo-information Science, 17(3): 290-299. (in Chinese)

[41]
Yang Q, Gao Y H, Chen G C, 2019. Influence analysis of the precipitation on atmospheric pollutant concentration in Chongqing. Meteorological and Environmental Sciences, 42(2): 68-73. (in Chinese)

[42]
Yu C X, Deng X L, Shi C E et al., 2018. The scavenging effect of precipitation and wind on PM2.5 and PM10. Acta Scientiae Circumstantiae, 38(12): 4620-4629. (in Chinese)

[43]
Zhang X M, Luo S, Li X M et al. 2020. Spatio-temporal variation features of air quality in China. Scientia Geographica Sinica, 40(2): 190-199. (in Chinese)

[44]
Zhang X P, Gong Z Z, 2018. Spatiotemporal characteristics of urban air quality in China and geographic detection of their determinants. Journal of Geographical Sciences, 28(5): 563-578.

DOI

Options
Outlines

/

〈 〉