Regional Differences, Dynamic Evolution and Driving Factors Analysis of PM_2.5 in the Yangtze River Economic Belt

Abstract

The proposal of a “dual-carbon” goal puts forward higher requirements for air pollution control. Identifying the spatial-temporal characteristics, regional differences, dynamic evolution, and driving factors of PM_2.5 are the keys to formulating targeted haze reduction measures and ameliorating air quality. Therefore, adopting the Dagum Gini Coefficient and its decomposition method, the Kernel Density Estimation model, and spatial quantile regression model, this study analyzes the regional differences, dynamic evolution, and driving factors of PM_2.5 concentrations (PM_2.5) in the Yangtze River Economic Belt (YREB) and the upstream, midstream, and downstream (the three regions) from 2003 to 2018. The study shows that: (1) PM_2.5 in the YREB was characterized by increasing first and then decreasing, with evident heterogeneity and spatial agglomeration characteristics. (2) Inter-regional differences and intensity of trans-variation were the primary sources of PM_2.5 differences. (3) The density curve of PM_2.5 shifted to the left in the YREB and the upstream, midstream, and midstream, suggesting that PM_2.5 has declined. (4) Industrial service level (IS) and financial expenditure scale (FES) exerted a significant and negative effect on PM_2.5 across the quantiles. On the contrary, population density (PD) showed a significant and positive influence. Except for the 75th quantile, the technology level (TEC) significantly inhibited PM_2.5. The remaining variables had a heterogeneous impact on PM_2.5 at different quantiles. The above results suggest that regional joint prevention and control mechanisms, collaborative governance mechanisms, and comprehensive policy mix mechanisms should be established to cope with PM_2.5 pollution and achieve green, sustainable economic development of the YREB.

1. Introduction

With the deepening of the “dual-carbon” goal, improving air quality and promoting green and low-carbon development become a broad consensus in the world. Among various air quality amelioration measures to achieve the “dual-carbon” goal, haze mitigation solutions have attracted extensive attention of scholars [1,2,3,4]. PM_2.5, particulate matter smaller than 2.5 μm, is the culprit of haze pollution [5,6]. The increase in PM_2.5 concentration will directly lead to a large increase in haze weather, toxic and harmful substances in the air. PM_2.5 may not only cause heart disease, and cardiovascular and cerebrovascular diseases, but also easily cause asthma, chronic bronchitis, and other respiratory diseases [7,8]. The World Health Organization identified air pollution, especially PM_2.5 fine particles, as carcinogens in 2013. Furthermore, PM_2.5 also adversely affects people’s production and life, urban image, and sustainable economic development [9,10,11].

China is still experiencing rapid industrialization and urbanization with a relatively high proportion of fossil energy consumption and a prominent air pollution problem characterized by PM_2.5 in China [12,13]. Due to the haze pollution characterized by PM_2.5, China has no choice but to bear huge economic costs every year, which can reach about 8% of the total GDP at the highest [14,15]. Therefore, the PM_2.5 pollution problem is still a major livelihood and environmental problem that cannot be ignored and needs to be solved urgently in China [16]. To this end, a series of policies, laws, and regulations have been promulgated by the government to tackle PM_2.5 pollution. In 2018, the Environmental Protection Tax Law of the People’s Republic of China, known as the “Green Tax Law”, came into effect [17]. The tax law imposes taxes on enterprises which directly discharge air pollutants, water pollutants, solid waste, and noise into the environment, forcing enterprises with high pollution and high energy consumption to transform and upgrade. In 2021, the “Opinions on In-depth Fighting the Battle of Pollution Prevention and Control” issued by the State Council formulated specific air pollution prevention and control objectives and implementation measures, which provided effective guidance for in-depth fighting the blue sky protection campaign. However, it is difficult for a single city to cope with PM_2.5 pollution [6]. PM_2.5 pollution control is a protracted war that requires long-term prevention and control, and multi-party efforts. Hence, investigating the spatiotemporal heterogeneity and driving factors of PM_2.5 is of critical significance to formulate differentiated PM_2.5 mitigation measures.

Given the urgent requirement for alleviating PM_2.5 pollution, many scholars conducted an intense discussion on the spatiotemporal distribution characteristics and driving factors of PM_2.5 [18,19]. Concerning the spatial-temporal heterogeneity characteristics, researchers mostly explore them based on a single city, a particular region, several urban agglomerations, and national scales [20]. Regarding driving factors, scholars mainly focused on exploring the influence of natural factors and socio-economic factors on PM_2.5. From the perspective of natural factors, some studies conclude that air temperature [21], air pressure [22], precipitation [23], wind speed [24], humidity [25], and other meteorological factors were the essential causes of PM_2.5 accumulation, diffusion, and sedimentation. For example, Pateraki et al. [26] simultaneously considered the effects of wind speed, temperature, relative humidity, and other meteorological factors on fine particles in the Mediterranean urban agglomeration. They found that temperature and humidity had a significant influence on PM_2.5. Similarly, when studying the formation mechanism of persistent extreme haze weather in Beijing in 2013, Yang et al. [27] also stated that meteorological factors played a crucial role in PM_2.5 pollution. Moreover, previous studies believed that the agglomeration and diffusion of PM_2.5 were also affected by terrain conditions [28], altitude [29], and other environmental factors. Besides natural factors, a large growing body of scholars confirmed that the fundamental factors of PM_2.5 pollution are socioeconomic factors [30,31], including population density [32,33], economic growth [34,35], industrial structure [36], urbanization level [37], energy intensity [38]. For instance, Ma et al. [35] verified the relationship between economic development level and pollutants based on the data selected from 152 cities in China and found an inverted U-shaped relationship between them. After comprehensively investigating the linkage between various socioeconomic factors and PM_2.5, Ding et al. [5] and Wu et al. [39] deemed that population density had the most significant impact on PM_2.5.

At the same time, a great number of econometric methods have been utilized by scholars to identify the driving factors of PM_2.5. For instance, Zhao et al. [40] applied the ordinary least square (OLS) model to explore the driving factors of PM_2.5 in 269 cities in China from 2015 to 2016. Their research results suggested that the Environmental Kuznets Curve (EKC) of air quality existed in China. A similar study was conducted by Wang and Komonpipat [41]. Furthermore, the Stochastic Impacts by Regression on Population, Affluence, and Technology model (STIRPAT) proposed by Dietz and Rosa [42] is often adopted to detect the impact of population, wealth, technology, and other variables on the environment [43,44]. Xu et al. [45] analyzed the nexus between economic growth, energy intensity, urbanization, and PM_2.5 by employing the STIRPAT model and nonparametric additive regression model. Nevertheless, due to the high mobility and diffusion of PM_2.5, ignoring the spatial effect may lead to biased regression results. Hence, it is essential to adopt spatial econometric models when studying the driving factors of PM_2.5. The commonly used spatial econometric models include the Geographically Weighted Regression Model (GWR) [46], Geographical Detector Model (GDM) [47], Spatial Lag Model (SLM) [48], Spatial Error Model (SEM) [49], Spatial Durbin Model (SDM) [50]. Among them, GWR could be used to measure local estimation values of the function in different locations by non-parameter estimation method and analyze the variation of these estimated values with space [51]. Using the GWR model, Wang et al. [52] found that natural and socio-economic factors had significant spatial differences on PM_2.5, thus providing a basis for formulating differentiated haze control measures in various regions [53,54]. Since PM_2.5 may be caused by the superposition of multiple factors, Yang et al. [55] detected the influence of natural and socio-economic factors on PM_2.5 by using GDM. They concluded that the interaction of industry and climate had the most significant impact on PM_2.5. Moreover, SLM, SEM, and SDM are the crucial tools to explore spatial correlation and the most widely used econometric methods for analyzing driving factors. Based on SLM and SDM, Zhu et al. [37] and Liu et al. [56] more intuitively decomposed the effects of various driving factors on PM_2.5 into direct effects and spatial spillover effects, thus effectively avoiding the errors caused by OLS, STIRPAT, and other panel models. Obviously, the above-mentioned studies conducted sufficient discussions on the driving factors of PM_2.5, providing crucial and valuable references for preventing PM_2.5 pollution. However, the econometric methods used in these studies are mostly mean regression models, which fail to reveal the local characteristics of each driving factor on PM_2.5 at different quantiles and are also vulnerable to the influence of sample data outliers [57]. To obtain more accurate and robust regression results, we adopted a spatial quantile regression model to analyze the driving factors of PM_2.5 empirically. Compared with the mean regression models, the spatial quantile regression model has some significant advantages. Firstly, the spatial quantile regression model combined the advantages of the spatial econometric models and the quantile regression model, so the direct and spatial spillover heterogeneity effects of various influencing factors on PM_2.5 at different quantiles could be detected. Secondly, since the stochastic error term of the spatial quantile regression model does not require satisfying classical econometric assumptions, more robust results than that OLS results can be obtained [7]. Therefore, the regression results obtained from the spatial quantile regression model can provide differentiated haze amelioration suggestions and references for governments in different regions and governance stages.

The Yangtze River Economic Belt (YREB), with rich natural resources, and unique geographical and industrial advantages, is the most dynamic and potential region for China’s economic development. In 2014, the Guiding Opinions on Promoting the Development of the Yangtze River Economic Belt by Relying on Golden Waterway issued by the State Council further promoted the construction and development of the YREB to the national strategic level, providing an action guide for protecting the ecological environment of the YREB and turning it into a new economic support belt in China [58]. However, the ecological environment along the Yangtze River has been seriously affected by the development of heavy chemical industries, including chemical industry, machinery manufacturing, metallurgy and building materials. Consequently, a series of ecological environmental problems emerged, such as massive discharge of pollutants, a sharp decline in air quality, and frequent occurrence of haze pollution, which become crucial bottlenecks restricting the high-quality economic development of the YREB [17]. Furthermore, there are significant differences in economic development, industrial agglomeration, and environmental pollution among the upstream, midstream, and downstream of the YREB [59,60]. It is urgent to identify the spatiotemporal distribution characteristics, regional differences, and influencing factors of PM_2.5 in the YREB, and then formulate specific haze reduction measures. To date, a growing number of researchers have conducted in-depth research on the severe PM_2.5 pollution in the YREB. For example, Mao et al. [61] analyzed the concentrations of major pollutants in 29 key monitored cities in the YREB from December 2017 to February 2018 based on the website data of the China National Environmental Monitoring Center. Yan et al. [62] investigated the spatial-temporal distribution features and driving factors of PM_2.5 in three urban agglomerations of the YREB from 2015 to 2018 using exploratory data analysis and GWR. Zhang et al. [63] evaluated the impact of land use change and anthropogenic emissions caused by urban expansion on PM_2.5 in the Yangtze River Delta.

However, previous studies mainly focused on the monthly data of some key cities and the annual data of a single urban agglomeration in the YREB, lacking large-scale and long-term analysis, thus affecting the formulation of haze mitigation policies. In addition, most existing studies only took the spatial-temporal distribution characteristics and driving factors of PM_2.5 into account, ignoring the regional differences of PM_2.5 caused by differences in resource endowments, climate, and economic development between different regions. Therefore, taking 108 cities of YREB from 2003 to 2018 as the research object, we attempt to comprehensively grasp the spatiotemporal evolution characteristics of PM_2.5 from a broader range and a longer time scale, thereby filling the gap of existing research and provide a theoretical reference for the overall air quality improvement and high-quality economic development of the YREB.

Concerning previous studies, the contributions of this paper are reflected in the following three aspects. First, three regions (upstream, midstream, and downstream), including 108 cities, were targeted as research objects to compare and analyze the spatiotemporal heterogeneity of PM_2.5 in the YREB. Second, utilizing the Dagum Gini coefficient and its decomposition, Kernel Density Estimation, we evaluated the regional differences and spatiotemporal evolution features of PM_2.5 in the three major regions of the YREB. Third, the spatial quantile regression model was introduced to identify the direct and spatial spillover heterogeneity effects of various influencing factors on PM_2.5. To conclude, this study is not only significant for identifying and analyzing the spatiotemporal heterogeneity characteristics and driving factors of PM_2.5 in the three regions of the YREB, but also provides comprehensive and new insights into implementing effective haze mitigation measures and joint prevention and control measures in the three regions of the YREB.

The remainder of this paper is arranged as follows. The methods adopted in this study are introduced in Section 2, including the Dagum Gini coefficient, Kernel Density Estimation, and spatial quantile regression model. Section 3 presents the empirical results and discussion. Section 4 provides discussions. Section 5 summarizes our findings and concludes.

2. Materials and Methods

2.1. Study Areas

To achieve the “dual-carbon” goal and high-quality economic development, China actively promotes the construction of ecological civilization, accelerated technological transformation, and formulated environmental protection laws and regulations, thus significantly ameliorating air quality. However, according to “The Bulletin of Ecological Environment in China (2019)”, the annual average PM_2.5 concentration of 168 key cities across China, including Beijing, Tianjin, Shanghai, Ningbo, Hangzhou, etc., was 44 μg/m³ in 2019, far higher than the annual target value of 10 μg/m³ set by the World Health Organization in the Air Quality Guideline (2005). Therefore, many cities in China are still facing severe haze pollution.

The Yangtze River Economic Belt (YREB) is the largest inland economic belt in China. It stretches across the eastern, central, and western regions, encompassing nine provinces and two municipalities. Relying on the broad market, perfect industrial structure, and convenient transportation, the YREB has contributed more than 40% of China’s GDP and has become the new engine and necessary support belt for China’s economic development. Taking the YREB, an important strategic economic belt in China, as the research object, identifying its PM_2.5 pollution status and mitigation measures can provide some empirical evidence and path selection for haze pollution control in China. Furthermore, YREB can be divided into three regions based on the heterogeneity of regional economic development level, natural conditions, and geographical location: the Upstream (Chongqing, Sichuan, Guizhou, Yunnan), the Midstream (Jiangxi, Hubei, Hunan), the Downstream (Shanghai, Jiangsu, Zhejiang, Anhui). For comprehensive and effective identification of the spatial-temporal distribution features and regional differences of PM_2.5 pollution in different regions of the YREB, the upstream, the midstream and the downstream (three regions including 108 cities) in the YREB were selected for comparative analysis (as shown in Figure 1).

2.2. Methodology

2.2.1. Dagum Gini Coefficient and Its Decomposition

Following Dagum’s idea of decomposing the Gini coefficient into subgroups [64], this paper employed the Dagum Gini coefficient to examine the regional differences and sources of PM_2.5 in the three regions of the YREB. The Dagum Gini coefficient is a critical indicator for measuring intra-regional differences and inter-regional differences. It is superior to traditional Gini coefficients because of its advantages in identifying the contribution rate of trans-variation density and solving the overlap problem between sub-samples [65]. The calculation formula of the Dagum Gini coefficient is as follows:

$$G=\frac{{\displaystyle \sum _{j=1}^k{\displaystyle \sum _{h=1}^k{\displaystyle \sum _{i=1}^{n_j}{\displaystyle \sum _{r=1}^{n_h}|y_{ji}-y_{hr}|}}}}}{2n^2\overline{y}}$$

(1)

where $G$ is the overall Dagum Gini coefficient; $\overline{y}$ represents the average value of PM_2.5 concentration in the YREB; $n$ is the number of cities, $n$ = 108; $k$ is the number of regions, $k$ = 3; $y_{ji}(y_{hr})$ refers to the PM_2.5 concentration value of any city in the $j(h)$ region; $n_j(n_h)$ refers to the number of cities in the $j-th$$(h-th)$ region; $j$, $h$ indicate the number of regions; $i$, $r$ indicate the number of cities in the region.

The Dagum decomposition theory can further decompose the Dagum Gini coefficient into three parts: the contribution of intra-regional differences $(G_w)$, the contribution of inter-regional differences $(G_{nb})$, the contribution of trans-variation density $(G_t)$, Which meets the requirements of $G=G_w+G_{nb}+G_t$. Equations (2)–(4) describe $G_w$, $G_{nb}$, $G_t$ in turn.

$$G_w={\displaystyle \sum _{j=1}^k{G}_{jj}}{p}_j{s}_j$$

(2)

$$G_{nb}={\displaystyle \sum _{j=2}^k{\displaystyle \sum _{h=1}^{j-1}{G}_{jh}}}(p_j{s}_h+p_h{s}_j)D_{jh}$$

(3)

$$G_t={\displaystyle \sum _{j=2}^k{\displaystyle \sum _{h=1}^{j-1}{G}_{jh}(p_j}}{s}_h+p_h{s}_j)(1-D_{jh})$$

(4)

In Equations (2)–(4), $p_j=n_j/n$, $s_j=n_j\overline{Y_j}/n\overline{Y}$; $D_{jh}=(d_{jh}-p_{jh})/(d_{jh}+p_{jh})$, $D_{jh}$ shows the relative impact of PM_2.5 between the $jth$ and the $tth$ regions; $d_{jh}$ refers to the difference in PM_2.5 between cities, which can be expressed as $d_{jh}={\displaystyle {\int }_0^{\infty }d{F}_j}(y){\displaystyle {\int }_0^y(y-x)d{F}_h}(x)$; $p_{jh}$ refers to the trans-variation first moment, which can be calculated by $p_{jh}={\displaystyle {\int }_0^{\infty }d{F}_h}(y){\displaystyle {\int }_0^y(y-x)d{F}_j}(x)$; $G_{jj}$ and $G_{jh}$, respectively, represent the intra-regional Gini coefficient, the inter-regional Gini coefficient. The $G_{jj}$ and $G_{jh}$ can be calculated by Equations (5) and (6).

$$G_{jj}=\frac{\frac{1}{2{\overline{Y}}_j}{\displaystyle \sum _{i=1}^{n_j}{\displaystyle \sum _{r=1}^{n_j}\left|y_{ji}-y_{jr}\right|}}}{n_j^2}$$

(5)

$$G_{jh}=\frac{{\displaystyle \sum _{i=1}^{n_j}{\displaystyle \sum _{r=1}^{n_h}\left|y_{ji}-y_{hr}\right|}}}{n_j{n}_h(\overline{Y_j}+\overline{Y_h})}$$

(6)

2.2.2. Kernel Density Estimation

Kernel density estimation, a nonlinear estimation method, uses a continuous density function to describe the distribution of the research object visually and then reveals the evolution trend of its morphological characteristics with time and space [66,67,68]. Instead of assuming specific distribution patterns in advance, it only estimates the probability density based on the data sample itself, which is widely used in economics, sociology, big data processing, and other fields [69,70]. The density function $f(x)$ of the random variable $x$ is defined as follows:

$$f(x)=\frac{1}{Nh}{\displaystyle \sum _{i=1}^{N}K(\frac{X_i-\overline{X}}{h}})$$

(7)

where $N$ is the number of observed values; $X_i$ stands for independent and identically distributed observations; $\overline{X}$ represents the mean value of observed values; $h$ refers to the bandwidth, and its small value means high estimation accuracy; $K(·)$ is the Kernel density function, which has many functions types, such as Gaussian kernel, trigonometric kernel, quadrangular kernel. Given the popularity and broad applicability, we employed the Gaussian kernel density function to estimate [71]. The Gaussian kernel density function is given by the Formula (8)

$$K(x)=\frac{1}{\sqrt{2\pi }}\exp (-\frac{x^2}{2})$$

(8)

2.2.3. Spatial Autocorrelation Analysis

According to the first law of geography [72], “everything is related to everything else, but near things are more related than distant things”. Therefore, spatial autocorrelation analysis measures the spatial correlation between things.

(1)

Global spatial autocorrelation

$TheGlobalMoran’sI$ is the most widely used to judge whether explanatory variables and explained variables exist in spatial agglomeration throughout the space. The function $GlobalMoran’sI$ is shown as follows [73,74]:

$$GlobalMoran’sI=\frac{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^m{\omega }_{ij}\left(Y_i-\overline{Y}\right)\left(Y_j-\overline{Y}\right)}}}{S^2{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^m{\omega }_{ij}}}}$$

(9)

where $S^2=\frac{1}{n}{\displaystyle \sum _{i=1}^n(Y_i}-\overline{Y}{)}^2$; $\overline{Y}=\frac{1}{n}{\displaystyle \sum _{i=1}^n{Y}_i}$; $Y_i$, $Y_j$ refer to the observed values of each city; $w_{ij}$ is weight matrix describing the spatial weight between the $i$ and $j$ regions; $n$ represents the total number of cities. In addition, $GlobalMoran’sI$ ranges from −1 to 1. When $Global$$Moran’sI$ > 0, a positive correlation exists among the index value of different cities, and the closer the index value is to 1, the stronger the spatial correlation. When $GlobalMoran’sI$ < 0, a negative correlation exists among the value of different cities, and the closer the index value is to −1, the stronger the spatial differentiation. When $GlobalMoran’sI$ = 0, complete random dispersion distribution exists among the value of different cities.

(2)

Local spatial autocorrelation

$Getis-Ord{G}_i^{*}$ statistics are adopted to measure the local spatial agglomeration characteristics of PM_2.5 in different cities [75]. $TheGetis-Ord{G}_i^{*}$ can be expressed by the following formula:

$$G_i^{*}(d)=\frac{{\displaystyle \sum _{i=1}^n{w}_{ij}(d)x_j}}{{\displaystyle \sum _{j=1}^n{x}_j}}$$

(10)

$$Z{(G_i^{*})}^2=\frac{G_i^{*}-E(G_i^{*})}{\sqrt{Var(G_i^{*})}}$$

(11)

where $n$ indicates the total number of cities; $d$ is the distance scale; $w_{ij}$ is the inverse distance matrix calculated by the Euclidean distance method. $x_i$ and $x_j$ represent the observed values of $i$ and $j$ area, respectively. The significance of $Z(G_i^{*})$ describes the spatial distribution of hot and cold spots in different regions. High z-scores and small p-values indicate hot spot areas, while high negative z-scores and small p-values represent cold spot areas.

2.2.4. Spatial Quantile Regression Model

Compared with the traditional OLS model, the quantile regression model does not require the random error term to satisfy the classical hypothesis of zero mean value and same variance and is not easily affected by extreme outliers. It can capture the whole picture of conditional distribution and comprehensively reflect the influence of explanatory variables on the explained variables at different quantiles [76]. Considering the advantages of the quantile regression model, the spatial aggregation and spillover characteristics of PM_2.5 concentration, this paper further constructed the spatial quantile regression model in the form of spatial Durbin for empirical analysis.

Referring to [57,77], the econometric model of the Spatial Durbin quantile regression approach is given by the following equation:

$$y_{it}(\tau )=\rho (\tau )W{y}_{it}+\alpha (\tau )l_n+{\displaystyle \sum X_{it}}{\beta }_i(\tau )+W{\displaystyle \sum X_{it}}{\theta }_i(\tau )+v_t+{\epsilon }_{it}(\tau )$$

(12)

where $\tau$ is the quantile and refers to the percentage of data below the regression line in the total data; $i$ stands for the city, and $t$ stands for the period; $\rho$ represents the spatial autocorrelation coefficient of the explained variable; $W$ indicates the spatial weight matrix; $X_{it}$ and $y_{it}$ respectively denote the explanatory variable and explained variable of the $ith$ city in the $t$ year; $\alpha$, ${\beta }_i$ and ${\theta }_i$ are the parameters to be estimated in the model and vary with the quantile $\tau$; $v_t$ represents the time control variable, and ${\epsilon }_{it}$ represents the stochastic error term.

The estimators $\widehat{\rho }$, $\widehat{\alpha }$, $\widehat{\beta }$, $\widehat{\theta }$ need to meet the following minimization conditions:

$$\begin{array}{l}\underset{\rho ,\alpha ,\beta ,\theta }{\min }\{{\displaystyle \sum _{i:Y\ge \rho WY+\alpha +X\beta +WX\theta }^{n}q\left|Y-\rho WY-\alpha -X\beta -WX\theta \right|}+\\ {\displaystyle \sum _{i:Y<\rho WY+\alpha +X\beta +WX\theta }^n(1-q)\left|Y-\rho WY-\alpha -X\beta -WX\theta \right|}\}\end{array}$$

(13)

Due to the endogeneity of Wy, we cannot directly apply the panel quantile regression method to the estimation of spatial quantile regression models. Referring to the panel quantile processing method with endogeneity established by Chernozhukov and Hansen [78] and the quantile estimation method of instrumental variables proposed by Su and Yang [79] for the single-section spatial autoregressive model, the spatial lag variables of explanatory variables $\left[WX,W^{2}X\right]$ were selected as the instrumental variables of the regression equation.

2.3. Data and Data Source

Previous studies appreciated that socioeconomic factors were the fundamental causes of PM_2.5 pollution, which was mainly due to the fossil energy consumed by socio-economic departments in operation can cause a large number of PM_2.5 particulate emissions [55]. Thus, some possible socio-economic driving factors, including economic development level (PGDP), industrial service level (IS), population density (PD), technology level (TEC), financial expenditure scale (FES), greening level (GL), and intensity of public transportation (TN), were selected to analyze the direct and spatial spillover effects on PM_2.5. The original data of PM_2.5 were obtained from the Atmospheric Composition Analysis Group of Dalhousie University in Canada, and other data were obtained from the China City Statistical Yearbook (2004–2019). The selected influencing factors in detail is introduced below.

2.3.1. Data of Annual Average PM_2.5 Concentration

Since 2013, China has started to monitor PM_2.5 concentration. The monitoring areas are concentrated in Beijing-Tianjin-Hebei, Yangtze River Delta, Pearl River Delta urban agglomerations and some key cities, and the monitoring data are mostly daily and monthly. Due to the short monitoring time and imperfect monitoring area of PM_2.5 concentration in China, Chinese officials cannot provide long-term and continuous PM_2.5 concentration data at the city level, so it is difficult to analyze the temporal and spatial evolution trend of PM_2.5 concentration in cities in the Yangtze River Economic Belt for a long time.

Considering the availability of data, this paper uses the data of Atmospheric Composition Analysis Group of Dalhousie University in Canada. These data, the satellite grid data of the global average concentration of PM_2.5, are estimated based on the aerosol optical depth (AOD) data of two NASA satellites with the help of geographic information system. On this basis, this paper further uses ArcGIS 10.7 software for raster processing, matches the vector map of prefecture-level cities in China, and analyzes the annual average concentration data of 108 cities in the Yangtze River Economic Belt in China. This method can not only make up for the information gap and data missing caused by the insufficient construction of ground monitoring base stations, but also avoid the problem of inconsistent statistical caliber between ground monitoring and satellite monitoring data. Many scholars have confirmed the reliability of this satellite remote sensing data [80,81]. Based on this data, a large number of analyses have been carried out on the pollution situation and influencing factors of PM_2.5 in China [82,83,84], which provides a valuable reference for effectively identifying and controlling PM_2.5 pollution.

2.3.2. Data of Driving Factors

(1)

Economic development level (PGDP): Some studies believed that the energy consumption and pollutant emissions caused by economic development had aggravated the environmental pollution situation [85]; Nevertheless, some other studies confirmed that economic development indirectly alleviated environmental pollution by promoting the optimization and use of advanced environmental protection technologies [86]. To verify the impact of economic development on PM_2.5, the economic development level expressed by per capita GDP is included as an explanatory variable.

(2)

Industrial service level (IS): IS was defined as the proportion of tertiary industry in GDP. Generally, the higher proportion of the tertiary industry is conducive to reducing energy consumption per unit of GDP and alleviating haze pollution [87,88]. Thus, IS is beneficial to the environment.

(3)

Population density (PD): PD was measured by the population per square kilometer. It has a dual impact on the environment [33,40]. For one thing, higher population density would lead to more domestic garbage and sewage, which positively affects PM_2.5; For another, higher population density negatively affects PM_2.5 through scale effect and agglomeration effect, so the impact of PD on PM_2.5 is uncertain.

(4)

Technology level (TEC): TEC, expressed by science and technology expenditure, is a vital factor in mitigating haze pollution [89]. It can not only improve energy efficiency and productivity and prevent haze pollution from the source, but also strengthen pollution control and alleviate haze pollution from the terminal. Hence, the coefficient is expected to be positive.

(5)

Financial expenditure scale (FES): FES was denoted by the proportion of fiscal expenditure in GDP. Financial expenditure, especially environmental protection expenditure, provides special funds for preventing and controlling environmental pollution and improving environmental quality [90]. Therefore, FES can play a crucial role in haze pollution mitigation.

(6)

Greening level (GL): Greening level was defined as the green coverage rate in built-up. GL can effectively purify sewage, reduce noise, and absorb pollutants and radioactive materials [82]. Hence, high green coverage is a significant way to purify the air and mitigate PM_2.5 pollution.

(7)

Intensity of public transportation (TN): TN was measured by Annual public bus (tram) passenger traffic. Studies showed that TN hindered PM_2.5 pollution by alleviating the pressure of energy use and reducing air pollutant emissions [91]. Thus, the impact of TN on PM_2.5 is expected to be negative.

To sum up, the definition and descriptive statistics of dependent and independent variables were shown in

Table 1. The definitions of variables.

Variables	Meaning	Unit
PM2.5	The average concentration of PM2.5	mg/m3
The level of economic development (PGDP)	Per capita GDP	104 yuan
Industrial structure (IS)	The proportion of tertiary industry output value in GDP	%
Population density (PD)	Population per square kilometer	persons/square kilometer
Technology level (TEC)	Expenditure for science and technology	108 yuan
Financial expenditure scale (FES)	The proportion of fiscal expenditure in GDP	%
Greening level (GL)	Green coverage rate in a built-up area	%
The intensity of public transportation (TN)	Annual public bus (tram) passenger traffic	108 persons

and

Table 2. Descriptive statistics of variables.

Variables	Mean	Std. Dev	Min	Max	Obs.
PM2.5	48.1816	15.7575	8.0915	101.1900	1728
PGDP	3.5348	3.0292	0.0099	19.9017	1728
IS	38.0672	8.1988	20.6600	77.4900	1728
PD	481.6207	294.8146	52.7300	2305.6300	1728
TEC	80,839.2100	258,354.8000	77.0000	4,263,655.0000	1728
FES	16.4643	8.3178	4.9095	68.7608	1728
GL	37.5647	8.6913	0.3600	93.8100	1728
TN	2.0277	3.8147	0.0000	28.3800	1728

.

3. Results and Discussion

3.1. Temporal Variation Characteristics

Figure 2 presented the temporal change trend of the annual average PM_2.5 concentration in the YREB and three regions from 2003 to 2018. From the overall perspective of the YREB, the average PM_2.5 concentration from 2003 to 2018 was 48.18 μg/m³. It presented an inverted V-shape trend from 2003 to 2018, which rose from 45.09 μg/m³ in 2003 to 57.38 μg/m³ in 2011, and then declined to 34.77 μg/m³ in 2018. By comparing regions, we find that the PM_2.5 concentration in the midstream was the highest and demonstrated the same variation trend as the whole YREB, rising to the highest point in 2011 with a concentration of 61.96 μg/m³, and then declining to 38.23 μg/m³ in 2018. The PM_2.5 concentration downstream was the second highest and presented a wave shape, while the upstream was the lowest and presented an M-shape. In a word, the PM_2.5 in the upstream, midstream, and downstream exhibited a general decreasing trend during the research period. Especially from 2011 to 2018, the declined rate of the three regions reached 47.77%, 40.11%, and 32.13%, indicating that PM_2.5 pollution in the YREB has been effectively improved through multi-party efforts. However, we should also recognize that there is still considerable room for air quality improvement in the YREB. Strengthening the supervision and management of PM_2.5 and promoting environmental protection technologies will remain the focus of air pollution prevention and control in the future.

3.2. Spatial Distribution Characteristics

The spatial variation characteristics of PM_2.5 in the YREB in 2003, 2011, and 2018 were shown in Figure 3. In 2003, three clusters with high PM_2.5 concentration in the YREB were formed, including Chengyu urban agglomeration, Wuhan urban agglomeration and urban agglomeration along Huaihe River. By 2011, the PM_2.5 pollution in the midstream was aggravated, and the high concentration areas further expanded from Wuhan urban agglomeration to the Triangle of Central China, indicating that PM_2.5 pollution has not attracted enough attention from the public in the past eight years. By 2018, the high-concentration areas of PM_2.5 significantly decreased, and the situation of the “three pillars” in the high-value concentration areas of PM_2.5 had disappeared. The reason for this phenomenon is that after 2011, the implementation of various national environmental protection policies, laws, and regulations achieved remarkable results in PM_2.5 pollution control. Further analysis showed that the high PM_2.5 concentration area has moved to the vicinity of the urban agglomeration along the Huaihe River. Xuzhou, Suzhou, Huaibei, Bozhou, Fuyang, and Bengbu became the “hardest hit areas” of PM_2.5 pollution. This indicates that these cities, with a single economic structure, have not yet got rid of resource and factor-driven dependence. Therefore, the key for these cities to improve air quality lies in reducing their dependence on resource industries, accelerating the transformation and upgrading of energy-intensive industries, and constantly cultivating new industries, so that the balanced development of industries adapts to the sustainable development of the economy.

3.3. Spatial Correlation Analysis

To verify the spatial autocorrelation of PM_2.5 concentration in YREB, the global Moran index values of PM_2.5 calculated by geographical matrix were reported in

Table 3. The value of Moran’s I of PM_2.5 in YREB from 2003 to 2018.

Year	Moran’s I	Z-Score	Year	Moran’s I	Z-Score
2003	0.176 ***	14.659	2011	0.169 ***	14.084
2004	0.173 ***	14.492	2012	0.162 ***	13.635
2005	0.169 ***	14.167	2013	0.177 ***	14.714
2006	0.17 2 ***	14.454	2014	0.211 ***	17.453
2007	0.196 ***	16.285	2015	0.240 ***	19.677
2008	0.175 ***	14.618	2016	0.205 ***	16.960
2009	0.206 ***	17.063	2017	0.240 ***	19.769
2010	0.178 ***	14.789	2018	0.254 ***	20.837

Note: *** indicates the significance at 1% level.

. The table shows that the global Moran’s I of PM_2.5 from 2003 to 2018 was positive and passed the 1% significance level test. This indicates that the PM_2.5 in the YREB has prominent spatial agglomeration characteristics, high PM_2.5 cities are often surrounded by cities with high PM_2.5, and so are low PM_2.5 cities. Regarding the changing trend, the global Moran’s I showed an upward trend, rising from 0.176 in 2003 to 0.254 in 2018. This signifies that the spatial agglomeration of PM_2.5 is gradually increasing, and establishing the regional collaborative governance mechanism is essential and urgent.

To further analyze the local spatial agglomeration characteristics of PM_2.5 in the YREB, the local G* statistics from 2003 to 2018 were calculated by ArcGIS. These results were divided into four categories by natural breakpoint method and transformed into the evolution map of cold and hot spots. Due to space limitations, only three years (Figure 4) were reported in this paper. It shows that in 2003, the hot spot areas of PM_2.5 in the YREB were only Zigong and Neijiang in the upstream, while the sub-hot spot areas accounted for the majority, coving most cities in the upstream, all cities in the midstream and downstream; Lijiang was the only city in the cold spot area, Panzhihua, Baoshan, Lincang, Pu’er and other seven cities belonged to the sub-cold spot areas. The above results suggested that the PM_2.5 pollution in most areas of YREB was severe and formed high-value agglomeration in space. In 2011, the cold and hot spot areas of the YREB showed apparent changes. During this period, the PM_2.5 in Wenzhou decreased and retreated from the sub-hot spot area to the sub-cold spot area; the PM_2.5 in Kunming and Qujing increased and shifted from the sub-cold spot areas to the sub-hot spot areas; in Baoshan, Lincang, and Pu’er, PM_2.5 was relatively low and entered the cold spot area, presenting a low-value agglomeration state. In 2018, the high-value agglomeration areas of PM_2.5 have further increased and shifted from the upstream areas to Xuzhou, Suzhou, Huaibei, Bozhou, Bengbu, and Fuyang in the downstream areas, which is consistent with the conclusion of Section 3.2. The range of the low-value agglomeration areas of PM_2.5 gradually narrowed, and Baoshan, Lincang, and Pu’er entered the sub-cold spot areas from the cold spot areas. In conclusion, the above results indicate that PM_2.5 in most regions is in a medium-high agglomeration state except for some regions upstream. Therefore, the critical measure mitigating PM_2.5 pollution is to strengthen joint prevention and control between upstream cold spots and other regions and control PM_2.5 cooperatively.

3.4. Regional Difference and Its Decomposition

Figure 5 depicts the overall and intra-regional Dagum Gini coefficients of PM_2.5 concentration in the YREB. As described in Figure 5, the average Gini coefficient of the YREB was 0.1669, indicating that the overall difference of PM_2.5 in the YREB was slight during the study period. In terms of changing trends, its variation was not significant and can be roughly divided into two stages, with a slight decline from 0.1741 in 2003 to 0.1604 in 2009 and then an increase in fluctuation to 0.1745 in 2018. Furthermore, the comparative analysis showed that the Gini coefficient in the upstream was the largest, followed by the downstream and the midstream was the smallest, with average values of 0.2546, 0.1323, and 0.1074, respectively. Specifically, the Gini coefficient in the upstream declined by 38.97% from 0.3043 in 2003 to 0.1857 in 2018. The Gini coefficients in the midstream and downstream exhibited the same characteristics and variation trends, which were relatively small and showed an expanding trend, with increasing rates of 36.36% and 36.50%, respectively. Based on the above analysis, we can find that the intra-regional difference of PM_2.5 in the upstream was gradually narrowing, and positive results have been achieved in the collaborative control of PM_2.5 pollution. However, the intra-regional differences in the midstream and downstream were gradually expanding, which needed further attention.

Figure 6 reported the characteristics and evolution trends of inter-regional Gini coefficients of PM_2.5 in the upstream, midstream, and downstream of the YREB from 2003 to 2018. It shows that the average Gini coefficients of the upstream-midstream (UP-MI), upstream-downstream (UP-DO), and midstream-downstream (MI-DO) were 0.2148, 0.2204, and 0.1279, respectively. Moreover, for UP-MI and UP-DO, the Gini coefficients exhibited a fluctuating downward trend, with a decreased rate of 11.38% and 8.55%, respectively; the Gini coefficient of the MI-DO presented an upward trend, with an increased rate of 30.14%. These results indicated that the inter-differences in UP-MI and UP-DO show narrowing trends, but the difference between MI-DO was obviously on the rise. It may be that while PM_2.5 pollution has been effectively alleviated under the supervision and governance of various regions; the high pollution generated by resource-dependent cities in the lower reaches of the Huaihe River urban agglomeration has led to a large inter-regional difference in the MI-DO.

Figure 7 describes the sources and contribution rates of regional differences in PM_2.5 concentration in the YREB. As shown in Figure 7, the average contribution values of $G_w$, $G_{nb}$, and $G_t$ during the study period were 31.33%, 31.03%, and 37.64%, respectively. From the perspective of changing trends, the contribution rate of $G_w$ fluctuated between 29.69% and 33.07%, with a fluctuation range of only 3.38%; the contribution rate due to $G_{nb}$ increased from 27.86% in 2003 to 38.39% in 2018, with an increased rate of 37.80%; the contribution rate of $G_t$ decreased from 41.55% in 2003 to 29.59% in 2018, with a decreasing rate of 28.78%. From the evolution process, the contribution rate of $G_w$ was relatively stable. The contribution rate due to $G_{nb}$ experienced a process of “declining–rising–declining–rising”, while the contribution rate $G_t$ was contrary. It can be concluded that trans-variation density was the primary source of the PM_2.5 difference in the YREB. The key to narrowing the PM_2.5 difference is to reduce the impact of cross-overlap among the three regions on the overall difference.

3.5. Dynamic Evolution Characteristics

In this section, Kernel density estimation was employed to identify the dynamic evolution characteristics of PM_2.5 in the YREB and the three regions. Figure 8 exhibited the evolution of PM_2.5 in YREB from 2003 to 2018. According to Figure 8, the distribution dynamics of PM_2.5 showed the following characteristics: Concerning the distribution position, the PM_2.5 distribution curve in the YREB roughly moved to the left, indicating that the PM_2.5 concentration in the YREB decreased during the study period. Regarding the distribution pattern, the height of the main peak of the PM_2.5 distribution curve rose in fluctuation, and the width became slightly wider after going through the process of “narrowing first and then widening”. These results mean that the absolute difference of PM_2.5 in the YREB slightly expands. In addition, the PM_2.5 distribution curve had the characteristics of right ductility narrowing, with only one main peak and no regional polarization.

Based on the Kernel density estimation, this paper further presented the dynamic evolution diagram of PM_2.5 distribution in upstream, midstream, and downstream (as described in Figure 9a–c) to identify and compare the distribution dynamics and evolution features of PM_2.5. The Kernel density curve would be analyzed from the four aspects: distribution position, distribution pattern, distribution ductility, and polarization characteristics.

Regarding the distribution position, the kernel density curve of the upstream, midstream, and downstream in the research period shifted to the left, indicating that the PM_2.5 concentration in these three regions showed a downward trend, which coincided with the conclusion of Section 3.1.

Regarding the distribution pattern, the height of the main peak in the upstream increased, and the width tended to narrow, suggesting that the absolute difference of PM_2.5 decreased. Furthermore, the height of the main peak in the midstream and downstream showed a trend of first increasing and then decreasing, and the width gradually became wider, indicating that the absolute difference of PM_2.5 in these regions is in a state of expanding.

Regarding the distribution ductility, PM_2.5 kernel density curves in the upstream, midstream, and downstream had a right trailing phenomenon, which revealed that there were cities with high PM_2.5 concentration in each region, such as Chengdu, Ziyang, and Neijiang in the upstream, Wuhan, Jingzhou and Xiaogan in the midstream, Xuzhou, and Suzhou in the downstream. The high PM_2.5 concentration in these cities led to right-tailed distribution. In addition, the ductility of PM_2.5 kernel density curves in the three regions was generally narrowing.

Regarding the polarization characteristics, the kernel density curve of PM_2.5 in the upstream has experienced the evolution process of “multi-peak to single-peak”, with the side peak gradually disappearing. In contrast, the kernel density curve of PM_2.5 in the midstream and downstream had only one main peak. The above conclusions implied that the multi-polarization of PM_2.5 upstream was weakening, and there was no regional polarization in the midstream and downstream.

3.6. Driving Factors

Quantile regression reflects the heterogeneous influence of independent variables on the dependent variables. Meanwhile, it also ensures the regression results are not easily affected by extreme values. Therefore, spatial quantile regression was adopted to identify the heterogeneous influence of each factor on PM_2.5 at different quantiles and the five representative quantiles (i.e., 10th, 25th, 50th, 75th, 90th) were selected for estimation (as shown in

Table 4. The results of spatial quantile regression and spatial Durbin model.

Variables	Spatial Durbin Model	Spatial Quantile Regression
		τ = 10	τ = 25	τ = 50	τ = 75	τ = 90
WY	1.6962 ***	0.8839 ***	0.7802 ***	0.7635 ***	0.7834 ***	0.7055 ***
	(0.0219)	(0.0255)	(0.0198)	(0.0134)	(0.0142)	(0.0172)
PGDP	−0.8459 ***	−0.0632	0.0050	0.2144 **	0.0975	−0.1355
	(0.1524)	(0.1946)	(0.1335)	(0.1010)	(0.1670)	(0.1336)
IS	−0.1979 ***	−0.1269 **	−0.1343 ***	−0.1278 ***	−0.0763 *	−0.1226 **
	(0.0358)	(0.0582)	(0.0284)	(0.0226)	(0.0416)	(0.0573)
PD	0.0181 ***	0.0163 ***	0.0102 ***	0.0108 ***	0.0122 ***	0.0105 ***
	(0.0011)	(0.0019)	(0.0015)	(0.0011)	(0.0015)	(0.0011)
TEC	−0.0185	−0.0278 ***	−0.0457 **	−0.0366 ***	−0.0396	−0.0302 ***
	(0.0120)	(0.0101)	(0.0189)	(0.0076)	(0.0259)	(0.0079)
FES	−0.5255 ***	−0.3226 ***	−0.3394 ***	−0.2425 ***	−0.2475 ***	−0.3646 ***
	(0.0358)	(0.0520)	(0.0406)	(0.0309)	(0.0328)	(0.0371)
GL	−0.0115	0.0193	−0.0124	−0.0081	0.0001	0.0128
	(0.0268)	(0.0431)	(0.0258)	(0.0182)	(0.0217)	(0.0315)
TN	−0.4326 ***	−0.4603 ***	−0.0406	−0.0395	0.0520	0.1141
	(0.0957)	(0.1490)	(0.1171)	(0.0586)	(0.1149)	(0.1466)
Cons		54.1386 ***	41.7269 ***	44.1979 ***	54.1068 ***	55.1763 ***
		(0.1810)	(0.2582)	(0.1122)	(0.2103)	(0.2213)
Obs	1728	1728	1728	1728	1728	1728

Note: Figures in parentheses are t-statistics. ***, ** and * indicate the significance at 1%, 5% and 10% levels, respectively.

). Moreover, we further presented the Spatial Durbin regression results (as described in column 2 of Table 4) for comparative analysis. The detailed analysis is shown below.

(1)

The level of economic development (PGDP) had a negative effect on PM_2.5 at the 10th and 90th quantiles while having a positive impact at other quantiles. This result indicates that in cities with low PM_2.5 concentration, economic development is easier to mitigate haze pollution. In cities with medium PM_2.5 concentration, the rapid industrialization process leads to excessive energy consumption and increasing industrial pollutants, thereby aggravating PM_2.5 pollution. For cities with high PM_2.5 concentration, the pressure of haze governance and the urgent requirement of ecological civilization construction force the local governments to take a series of PM_2.5 mitigation measures, including increasing investment in environmental protection, guiding technological innovation, strengthening environmental protection law enforcement, etc. This reduces the PM_2.5 concentration within the region and triggers the spillover and diffusion of pollution control technologies between regions, inhibiting PM_2.5 pollution in adjacent regions.

(2)

At different quantiles, the estimated coefficients of industrial service level (IS) were negative and passed the 1% significance level test, indicating that the tertiary industry has played a crucial role in alleviating PM_2.5 pollution in local and neighboring areas. Similar conclusions were drawn by Zhu et al. [92]: The rapid development of the tertiary industry has shifted the production factors originally attached to the industrial chain with high energy consumption and low efficiency to the industrial chain with high efficiency and low consumption, thereby improving the effective utilization rate of resources and environment and alleviating the PM_2.5 pollution. At the same time, developing an environmentally friendly tertiary industry in a particular area is conducive to promoting environmental protection technology cooperation and sharing advanced governance concepts with neighboring areas, suppressing PM_2.5 pollution in neighboring areas.

(3)

The coefficients of population density (PD) were significantly positive across different quantiles at the significance level of 1%, which implies that the positive impact of population density on PM_2.5 is more significant than the negative impact. Thus, PD exacerbates PM_2.5 pollution status. This result has been confirmed by previous studies [93,94]: A great demand for production, living, and infrastructure caused by high population density easily triggers a series of problems, such as excessive consumption of energy and resources, a massive discharge of pollutants, and insufficient carrying capacity of the urban environment, which aggravates the PM_2.5 pollution situation. Moreover, under atmospheric circulation, atmospheric chemistry, and other natural factors, PD easily worsens the PM_2.5 pollution level in adjacent areas while causing PM_2.5 pollution in a certain area.

(4)

The estimated coefficients of technology level (TEC) were negative and significant at a 1% significance level except for the 75th quantile level. These results suggest that the improvement of technology level is more conducive to promoting clean production at the source of enterprises, improving energy efficiency and resource utilization rate [95], thus inhibiting PM_2.5 pollution in local and adjacent areas. In addition, the negative effects of TEC were more significant at the low and high quantiles than at the middle quantile. The reason may be that the improvement of TEC can effectively ameliorate PM_2.5 pollution through innovative technology and equipment, and it may also promote the rapid development of the economy, leading to a large amount of pollutants emission and causing PM_2.5 pollution. Therefore, the dual effect of technology level may lead to the insignificant negative effect of technology level on PM_2.5 at the 75th quantile.

(5)

The financial expenditure scale (FES) adversely affected PM_2.5, and all estimated coefficients were significantly negative at the 1% significance level, which was consistent with our expectations. As mentioned previously, the exclusive funds provided by the fiscal expenditure play a substantial role in haze control and prevention in local and neighboring areas. On the one hand, it can directly control PM_2.5 pollution; On the other hand, it can induce enterprises to adopt environmentally friendly production technologies and pollution control technologies through scientific research and policy incentives, indirectly realizing the prevention and control of PM_2.5 pollution. In addition, FES also has a spillover effect. While promoting the development of special haze control activities in a certain area, it also provides a demonstration role for neighboring areas and promotes the improvement of air quality in neighboring areas. Hence, improving the financial expenditure scale is a crucial pathway to effectively mitigate PM_2.5 pollution. The cities in the YREB should pay more attention to the level and efficiency of fiscal expenditure.

(6)

The estimated coefficient of greening level (GL) was positive in the low and high quantiles, but negative in the middle quantiles. The reason may be that the areas in the middle quantiles attach more importance to the coordination of economic development and environmental protection, formulate a relatively complete set of environmental protection schemes, and actively promote the construction of urban greening levels. As a result, the GL can better enhance the regional ecological carrying capacity and environmental self-purification ability and exert its powerful functions in dust removal, toxic and harmful gas filtration, and air purification, thus suppressing PM_2.5 pollution. At the same time, the improvement of GL in a certain area can drive green level management and construction in neighboring areas and play a certain role in controlling PM_2.5 pollution in neighboring areas. However, at the low quantiles of PM_2.5 concentration, due to the public’s lack of attention to air quality and the problems of management and maintenance in the greening process, the suppression effect of GL on PM_2.5 pollution has not yet appeared in local and adjacent areas. At high quantiles, the ability of green plants to absorb PM_2.5 and purify the air is very limited. Under large-scale pollution emissions, relying solely on GL to control PM_2.5 pollution is only a drop in the bucket, so GL cannot effectively reduce PM_2.5 pollution in local and adjacent areas.

(7)

Intensity of public transportation (TN) has a negative correlation with PM_2.5 in the 10th–50th quantiles while a positive correlation with PM_2.5 in the 75th–90th quantiles. In the 10th–50th quantiles of PM_2.5, TN is more conducive to human capital agglomeration. Human capital agglomeration can drive the flow of knowledge and technology to central cities and promote the application of clean and environmental protection technologies, thus improving the air quality in central cities. Meanwhile, it can also mitigate the PM_2.5 pollution situation in neighboring areas through the spillover effect of knowledge and technology. However, the transportation infrastructure industry is not a smokeless industry, and the increase in public transportation intensity often leads to more emissions of carbon dioxide and nitrogen oxides. In addition, it is rather difficult to control PM_2.5 pollution at high quantiles, so TN cannot hinder the PM_2.5 pollution in the local and neighboring areas.

4. Discussion

In the process of dual-carbon goal and ecological civilization construction, it is extremely significant to reduce air pollutant emissions and improve air quality. As an important source of air pollution, PM_2.5 plays an adverse impact on health, climate change, and economic development. Identifying and analyzing the spatial-temporal distribution, regional differences and influencing factors of PM_2.5 is of critical significance for winning the blue sky defense war and realizing high-quality economic development. The Yangtze River Economic Belt, a key strategic development area of China, undertakes the critical task of national ecological civilization demonstration, and green and low-carbon development. Therefore, what is essential is that take the YREB as the research object, based on identifying the spatial-temporal differences and driving factors of PM_2.5, some policy suggestions are provided for formulating targeted and differentiated PM_2.5 reduction measures.

4.1. Suggestions for PM_2.5 Mitigation Measures

Thanks to the government’s leading role in haze pollution mitigation, PM_2.5 concentration in the YREB has dropped significantly in recent years. This provides new ideas and empirical evidence for ecological environment management in China and many other developing countries similar to China. That is to say, in haze pollution amelioration and ecological environment improvement, the dominant role of the central government should be brought into full play, and the haze control system of central and local linkage and local cross-regional cooperation should be established. Furthermore, given the obvious temporal and spatial differences of PM_2.5 in the YREB, local governments should avoid “one size fits all” policy formulation and pay more attention to differentiated regional PM_2.5 mitigation measures. The high-value clusters have gradually decreased in the YREB and turned into many resource-dependent cities along the Huaihe City Agglomeration, which requires sufficient attention from local governments. It is urgent to update the production process of resource-dependent enterprises through technological innovation, drive the development of high-tech industries, and then promote the transformation of resource-dependent enterprises, thus reducing PM_2.5 pollution.

According to the results of the Dagum Gini Coefficient and Kernel Density Estimation, we found that trans-variation density was the primary source of PM_2.5 difference in the YREB, and solving the problem of over-lapping among regions is beneficial to narrow the PM_2.5 difference. To this end, cities with low PM_2.5 concentration should further play their advantages in haze control, develop energy conservation and environmental protection industries, and increase financial input to boost the environmental protection industry into a new pillar industry. In comparison, cities with high PM_2.5 concentrations should reduce their dependence on coal, optimize the urban energy structure and increase the use of clean energy. Moreover, industrial pollution control should be strengthened, and production processes and equipment with greater pollution in industrial production should be phased out to reduce fine particle emissions.

The global and local spatial autocorrelation results show that PM_2.5 has apparent spatial agglomeration and spatial spillover characteristics, and the high PM_2.5 concentration areas tend to be adjacent to the high-value areas. Therefore, regional joint prevention and control mechanisms and regional cooperative governance mechanisms should be established to effectively govern haze pollution. Local governments should make efforts to mitigate haze pollution and ameliorate air quality by strengthening environmental protection supervision, improving pollution prevention and control technologies, and establishing air quality prevention and early warning mechanisms. Meanwhile, it is essential for local governments to strengthen cooperation and exchange of pollution prevention and control technologies with neighboring regions, establish strict emission standards, enhance joint air pollution law enforcement, and jointly win the battle of the blue sky.

Based on the results of spatial quantile regression, it is found that the selected seven influencing factors have spatial heterogeneity on PM_2.5 in the YREB, which provides a new insight for local governments to improve the air quality in the YREB. First, since the level of economic development (PGDP) has a negative impact on the high quantile of PM_2.5, promoting high-quality economic development should become the key path to improving air quality in areas with severe PM_2.5 pollution. Second, according to the negative effects of the industrial service level (IS), the technology level (TEC), and the financial expenditure scale (FES) in all quantiles, local governments should appropriately promote the transformation of industrial services, improve the level of science and technology, increase the scale of financial expenditure, strive to achieve clean production at the source and end pollution control, so that significantly ameliorate PM_2.5 pollution. Third, it is verified that population density (PD) is not favorable to PM_2.5 mitigation in all quantiles, and so is the intensity of public transportation (TN) in high quantiles. Therefore, in addition to increasing the level of human capital investment, improving the utilization intensity of public transport infrastructure and giving full play to the scale effect and agglomeration effect brought by population agglomeration are critical. Fourth, given the fact that greening level (GL) only inhibits PM_2.5 in the middle quantiles, local governments should clarify the limited effect of greening level on PM_2.5 pollution. Emission reduction and greening should be taken into consideration at the same time to avoid the waste of resources caused by blindly following the trend of greening construction.

4.2. Future Research Direction

Exploring the spatial-temporal evolution characteristics, regional differences and driving factors of PM_2.5 is essential in achieving the dual-carbon goal and promoting the high-quality development of China’s economy. However, there are still some limitations in our study. Firstly, in terms of the sample time, this paper is based on the city panel data from 2003 to 2018. In the future, the latest PM_2.5 concentration data, and natural and socio-economic factors data will be included in our study. Secondly, this paper only analyzes the direct and indirect effects of social and economic factors on PM_2.5. More detailed natural factors, such as temperature, relative humidity, and rainfall, need to be further involved in our future research. Finally, considering that the spatial quantile regression model in our paper only analyzes the total impact of each driving factor, we will try to improve this model in the future to capture the net direct and indirect effects of each driving factor.

5. Conclusions

Based on the panel data of 108 cities in China’s YREB from 2003 to 2018, this paper analyzed the spatial-temporal characteristics of PM_2.5. Subsequently, the Dagum Gini coefficient and kernel density estimation were used to explore the regional differences and dynamic evolution of PM_2.5 in the three regions of the YREB. Finally, based on verifying the global and local spatial autocorrelation of PM_2.5 by using the Global Moran index and The Getis -Ord G_i^*, we also adopted the spatial quantile regression model to explore the nexus between the influencing factors and PM_2.5 in the YREB. The main conclusions are as follows:

The annual average PM_2.5 concentration in the YREB increased from 45.09 μg/m³ in 2003 to 57.38 μg/m³ in 2011 and then decreased to 34.77 μg/m³ in 2018, indicating that remarkable results have been achieved in controlling haze pollution in recent years. These remarkable achievements are closely related to the central and local governments, which have played an indispensable leading role in improving the air quality of the YREB and promoting its ecological civilization construction.

In comparing regions, PM_2.5 concentration in the midstream was the highest, followed by the downstream and the upstream was the lowest. Furthermore, the results of the spatial distribution pattern showed that PM_2.5 in the YREB had obvious spatial agglomeration characteristics, and the high-value agglomeration area gradually decreased over time and transferred to resource-dependent cities along the Huaihe River. Therefore, we can infer that reducing the dependence on coal resources, adjusting the energy structure, and improving the energy utilization rate are significantly critical to effectively ameliorating PM_2.5 pollution.

According to the Dagum Gini coefficient and its decomposition, the overall difference of PM_2.5 in the YREB was relatively small and the variation was not significant during the study period; the order of intra-regional differences in the three regions from large to small is: upstream, downstream, and midstream; the inter-regional differences in the UP-MI and UP-DO were large and in a decreasing trend, while the inter-regional difference in the MI-DO was small and presented an expanding trend. Moreover, trans-variation density was the primary source of the PM_2.5 difference in the YREB and solving the cross-overlap problems is of great necessity to narrow the PM_2.5 difference in the YREB. According to the Kernel Density Estimation, the absolute difference of PM_2.5 decreased upstream and increased in the midstream and downstream during the study period. In addition, the PM_2.5 distribution curves also suggested that the multi-polarization trend of PM_2.5 upstream was weakening, and there was no multi-polarization phenomenon in the midstream and downstream.

The spatial autocorrelation test showed that PM_2.5 had obvious spatial autocorrelation and spatial dependence characteristics. Furthermore, the results of spatial quantile regression implied that the selected factors had spatial heterogeneity on PM_2.5 at different quantiles. Specifically, the level of economic development (PGDP) inhibited PM_2.5 at the 10th and 90th quantiles, while aggravating PM_2.5 at other quantiles. Industrial service level (IS) had a significant negative impact on PM_2.5 at all quantiles, and so did the financial expenditure scale (FES). On the contrary, population density (PD) had a significant positive effect on PM_2.5 at all quantiles. The technology level (TEC) imposed significantly negative effects on PM_2.5 except for the 75th quantile. Greening level (GL) only exerted a negative impact on PM_2.5 at the middle quantiles, while the intensity of public transportation (TN) exerted a negative impact on PM_2.5 at the 10th–50th quantiles. The above results suggest that it is essential for different cities to make clear the heterogeneous effects of various influencing factors and formulate different haze mitigation measures.