China has experienced swift urbanization during the past two decades, with the urban ratio climbing from 36.22% in 2000 to 63.89% in 2020, accompanied by a huge population influx into cities (National Bureau of Statistics of China, 2001, 2021). This has heightened the environmental pressure on cities from anthropogenic production and consumption beyond the regional capacity. Air pollution has been recognized as one of the most critical threats to human health as no one is able to escape from the risks posed by severe haze episodes (Yin et al., 2017; Zhang et al., 2022). In 2019, it was the fourth-leading cause of early death throughout the world and strong attention has been focused on three pollutants in particular: PM_2.5, ground-level ozone and indoor fine particulate matter. In the same year, over 90% of the population was exposed to PM_2.5 concentrations greater than 10.00 μg/m³ annually, while 51.00% of the population experienced even worse conditions, with PM_2.5 concentrations exceeding 35.00 μg/m³, the standard and interim target-1 level in the World Health Organization (WHO) guidelines. Approximately 6.67 million deaths were ascribed to diseases linked to air pollution worldwide in 2019, of which 4.14 million were specifically related to PM_2.5 pollution. In China, around 1.85 million deaths were associated with air pollution and 1.42 million deaths with PM_2.5 pollution (Song et al., 2017; Health Effects Institute, 2020). The WHO tightened the standard from the previous 10.00 to 5.00 μg/m³ in 2021 (WHO, 2006; WHO, 2021). In China, Ambient Air Quality Standards (AAQS, GB3095-2012) were upgraded by including standards for PM_2.5: 15.00 μg/m³ (annual Grade-I), 35.00 μg/m³ (annual Grade-II and daily Grade-I) and 75.00 μg/m³ (daily Grade-II) in 2012. This came into implementation in 2016, but PM_2.5 monitoring data for major cities have been available publicly since 2015 (Ministry of Environmental Protection of China, 2012). Therefore, a study on PM_2.5 emissions and control in the context of the Sichuan Basin’s (SCB’s) rapid socioeconomic development is of great significance.

Normally PM_2.5 samples are taken across different seasons at a certain location, such as Chengdu, Lanzhou, Nanjing, Tianjin, Shenzhen, Beijing Chongqing, and Hangzhou. Further analysis found that inorganic ions and organic carbon dissolving in water, microelements, carbonaceous species, and certain aromatic hydrocarbons were the main components of PM_2.5 samples. A positive matrix factorization model was employed to explore whether road dust, coal combustion, power plants, secondary aerosol and sulfate, vehicle emissions, iron-steel industry, biomass burning and crustal dust were the sources. PM_2.5 emissions were extremely severe in winter (Gu et al., 2011; Liu G et al., 2015; Li L et al., 2017; Tan et al., 2017; Kong et al., 2020).

Regional studies of the main megacity cluster in China have highlighted similar topics, with consideration of transportation on PM_2.5 emission between cities. In the Yangtze River Delta region, the main components of PM_2.5 were sulfate, nitrate, organic matter, and ammonium, originating from biomass burning and vehicle emissions. PM_2.5 pollution was exacerbated by adverse meteorology (Hua et al., 2015). In the Beijing-Tianjing-Hebei megacity cluster, extreme haze events were driven by local emissions from Beijing and Hebei, whereas in Tianjin, external transport was the predominant source (Li J et al., 2017). In this region, the spatial association of PM_2.5 pollution was studied using Pearson’s correlation coefficient (An et al., 2020). In the Pearl River Delta region, ionic species were a major PM_2.5 component, accounting for 53.30% in winter and 39.40% in summer. Source apportionment identified industrial sources as the largest contributors while nitrate and ammonium ion mainly originated from agricultural emissions. Strong seasonal fluctuation in chemical composition was also found, with higher concentrations of ionic species occurring in the winter (Lai et al., 2007; Huang et al., 2018; Hou et al., 2019). The SCB is situated in southwest China; its unique topography as a basin surrounded by mountains causes frequent temperature inversion and traps air pollutants, inducing severe pollution in the winter (Zhang et al., 2012; Tao et al., 2013; Chen et al., 2014). Regional stagnant meteorology tended to prevent the particles from dispersing. Crustal elements were the predominant natural source of PM_2.5 whereas anthropogenic activities contributed heavily to other ion species (Feng et al., 2020; Qiao et al., 2019). In south SCB, high PM_2.5 concentrations were found (Ning et al., 2018; Fan et al., 2020). Previous studies have attempted to find out how PM_2.5 spatial variation is impacted by human activities and natural conditions in Chinese cities using global regression models, the drivers of air pollution in Sichuan province by structural decomposition method and the economic agglomeration impact on PM_2.5 pollution by output density model and spatial Durbin model in the Yellow River region (Liu et al., 2018; Wang et al., 2022).

It is noteworthy the primary concerns regarding PM_2.5 study have shifted from individual cities to regional scales. But they mainly focused on the chemical characteristics and composition, source apportionment, and seasonal and regional fluctuations in PM_2.5 pollution. However few studies have investigated the overall PM_2.5 spatial association, relationship between cities and evolution as well as its drivers in SCB by social network analysis. An investigation in this field is essential to understand its formation and inform effective mitigation measures.

Here, the concentration of PM_2.5 from 2015 to 2020 and the statistical data for anthropogenic activities from 2015 to 2019 were collected to reveal trends in PM_2.5 concentrations and its spatial association network in SCB through social network analysis (SNA) using a gravity model with the following objectives: The PM_2.5 spatiotemporal evolution was visualized using ArcGIS. Its spatial association network was studied at the regional scale in SCB using a modified gravity model, with the characterization of overall network and each individual node by SNA. An intuitive spatial association network with a core-edge pattern was observed and the evolution displayed. Its driving factors were further investigated using the quadratic assignment procedure (QAP) method, which can generate realistic results due to its ability to deal with multicollinearity between independent variables. The root causes behind the driving factors are discussed and further regional-scale mitigation proposals are suggested herein; a clean industrial chain by life cycle assessment and an environmentally friendly ideology for lifestyle are expected from production and consumption in future development.

In SNA, the social agent, considered as a node in the social network, is an individual, company, organization, city or country and the connections between the nodes, edges, form the whole network structure. Originally used in sociology (Wasseman and Faust, 1994; Luo, 2005; Scott, J., 2007), SNA has been progressively applied to management, politics, geography, economics, physics (Liu, 2019), and environments (Song et al., 2019; An et al., 2020). The SCB includes eighteen cities with huge population and swift industrialization (Figure 1). It ranks as one of the most severely polluted regions in China due to its basin topography surrounded by high mountains and anthropogenic emission (Zhang et al., 2012). Therefore, the eighteen cities in SCB can be considered as nodes and the interactions between them as the edges in the study of the PM_2.5 spatial association, together with the driving factors.

The PM_2.5 spatiotemporal evolution in SCB from 2015 to 2020, including the annual average concentration and peak monthly average concentration in December and January, was visualized using ArcGIS. The PM_2.5 spatial association network was studied by considering each individual city as a node and inter-city PM_2.5 circulation as an edge. Firstly, a spatial correlation matrix should be established. In the gravity model the mutual interaction between cities is proportional to the target parameter but inversely proportional to the distance. It is widely used to construct the spatial correlation matrix and further to evaluate the relationship. This gravity model was shown in Eq. 1 and here the target parameter is PM_2.5 concentration in the city (Kuik et al., 2019; Bai et al., 2020; Wang et al., 2020). Y i j = K i j M i × M j D i j b (1) K i j = C i C i + C j (2) Y_ij: the gravitation between two cities; M_i: the mass (PM_2.5 annual average concentration) of city i; M_j: the mass (PM_2.5 annual average concentration) of city j; D_ij: the distance between two cities; C_i and C_j: PM_2.5 annual average concentration of city i and city j; K_ij: the empirical constant as Eq. 2; b: the distance attenuation coefficient and considered as 1 here.

Normally, the gravity model was extended by including the relevant elements on the target parameter, which was helpful to make a comprehensive evaluation on the relationship. Therefore, the “mass” of PM_2.5 concentration was modified according to the source appointment of PM_2.5 in Chongqing and Chengdu. The major components are mainly from coal combustion, vehicle emissions, industrial pollution, and soil dust (Chen et al., 2017; Kong et al., 2020). Based on the modified gravity model in previous research and considering the availability of raw data for each individual city, the following parameters were integrated: urban population (P_i), land area of urbanized area (L_i); the number of civilian vehicle (V_i), gross domestic production from secondary industry (S_i), and gross domestic production from tertiary industry (T_i). These variables were converted to a comparable order of magnitude as PM_2.5 concentration by geometric mean; the square root of the obtained value and PM_2.5 concentration was taken as M_i in Eq. 1 producing the modified gravity model in Eq. 3. Y i j = K i j × C i × P i × L i × V i × S i × T i 5 × C j × P j × L j × V j × S j × T j 5 D i j (3)

A gravity matrix with i rows and j columns was obtained from Eq. 3. It was dichotomized by comparing individual gravity values with the average gravity value of the matrix; if greater than the average, a value of 1 was assigned, indicating strong association between the cities, otherwise a value of 0 was assigned (Liu H et al., 2015; Su and Yu, 2019; Wang et al., 2020). The spatial association was visualized, and the characteristic parameters of the overall network structure and each individual city were assessed using SNA and Ucinet. The QAP method was applied to study the driving factors, without considering the direction of the spatial association, as this paper studies its formation and evolution, following Bai et al. (2020).

The spatiotemporal evolution shows that PM_2.5 levels in most cities in the SCB exceeded the national standard in 2015 and dropped sharply from 2015 to 2020. Nevertheless, a slightly strengthened PM_2.5 spatial association network is observed by SNA analysis based on the modified gravity model from 2015 to 2019. Alongside the dual impacts of socioeconomic growth and strict environmental regulations discussed previously, the regional fusion policy in SCB plays an important role: developed as Chengdu-Chongqing Metropolitan Area in 2016 and updated as Chengdu-Chongqing Economic Circle in 2021 (National Development and Reform Commission, Ministry of Housing and Urban-Rural Development, 2016; State Council of the PRC, 2021). However, two cities, Guangyuan and Bazhong, excluded from the latest economic circle, also exhibited an increasingly stronger relationship with other cities in the PM_2.5 spatial association in 2019 as compared to their initial isolation in 2015. This phenomenon reminds us regardless of how cities are economically classified, PM_2.5 emissions and dispersion inevitably cause interdependent impacts across the whole region. Therefore, regionalism is critical to realize further PM_2.5 control and mitigation. Attempting to tackle the emissions of each city individually will likely fail to achieve the desired targets.

The evolution of the density, efficiency, and connectedness of the overall network shows that the closeness degree of spatial association is promoted by some fluctuations. Despite the rapid drop in PM_2.5 concentration due to strict environmental regulations, upgrades to the economic structure and smart urban planning, the rebound of the overall network parameters implies the collaborative governance effect of inter-city PM_2.5 emissions show a backward trend. The growth of all the socioeconomic parameters is a lasting motivation for local governments. Environmental targets have been outlined in national policy but occasionally fail to be fully addressed in practice. The degree, closeness, and betweenness reveal that Chengdu and Chongqing occupy the center of the network and their strong influence is intensified evidently, of which Chengdu is even more powerful. Based on this core-edge model, they should be prioritized to control PM_2.5 emissions.

Ultimately, the driving factors identified by QAP regression call for regional collaboration concentrating on the cities with short distances, small differences in the number of civilian vehicle and urban ratio, but with a large difference in urban population and GDP gap between secondary industry and tertiary industry. These driving factors exhibit a significant impact on the evolution of PM_2.5 overall network and the relationship between the nodes in the SCB. Direct distance is mainly involved in the PM_2.5 transportation and ties between nodes and the difference of urban ratio is relative small, both with coefficient around 0.33-0.34 from QAP regression. Therefore, GDP gap, urban population and the number of civilian vehicle with the coefficient above 0.50 in the QAP regression were considered to explore the relationship of spatiotemporal evaluation for PM_2.5 in individual city in Figure 2. They were illustrated in Figure 9 for sixteen cities except Chengdu and Chongqing whose data were quite huge and unsuitable for the scale herein. Moreover, the GDP sum for secondary industry and tertiary industry were also included. All the data increased from 2015 to 2019 except the GDP gap which were all positive in 2015 but negative for most cities in 2019. This negative GDP gap implies the extent to tertiary industry greater than secondary industry in the economic structure. Its shift partly explains the PM_2.5 annual concentration decrease from 2015 to 2019 observed in Figures 2A, E although the other data climbs heavily. The cities with PM_2.5 < 35.00 μg/m³ in Figures 2A, E correspond to those with a combination of relatively small value for driving factors in Figure 9. However, Ya’an and Bazhong, with quite small driving factors in Figure 9A is not observed in Figure 2A PM_2.5 < 35.00 μg/m³ as their PM_2.5 annual concentrations were 36.00 μg/m³, exceeding the standard slightly. Moreover, a synergetic effect among these factors plays an important role as some cities with quite low value in one or two driving factors fails in reaching the PM_2.5 < 35.00 μg/m³. These findings demonstrate the dominant factors impacting the spatial and temporal differences are the GDP gap between secondary industry and tertiary industry, urban population and the number of civilian vehicle identified by QAP regression. In Figure 10, Chengdu and Chongqing shows the same trend for all the driving factors which accounts for their PM_2.5 mitigation from 2015 to 2019. The negative GDP gap indicates their tertiary industry exceeded second industry since 2015. It is inadequate to achieve PM_2.5 < 35.00 μg/m³ in 2015 and 2019 without the synergetic effect of all the factors mentioned above.

Although the study findings provide some interesting and valuable insights, they do not reveal the fundamental influencing mechanism of these factors; hence, further study is required. A knowledge of the mechanism could help propose more accurate and precise policy and measures for PM_2.5 mitigation. Moreover, the data for the subdivision of tertiary industry are unavailable publicly for each individual city, preventing a comprehensive in-depth analysis of its impact. Further research could be continued with more specific information in the future.

Herein, we discuss the spatiotemporal evolution of PM_2.5 pollution by visualizing PM_2.5 concentrations from 2015 to 2020 in SCB. Overall network structure and node behavior was investigated using SNA based on the modified gravity model of the PM_2.5 spatial association from 2015 to 2019. In addition, the driving factors were explored by QAP analysis. The following conclusions were made:

(1) PM_2.5 levels dropped significantly from 2015 to 2020 in SCB. The annual PM_2.5 concentration decreased from 62.92 to 40.83 μg/m³ in Chengdu and from 55.08 to 32.67 μg/m³ in Chongqing. The number of cities with an annual PM_2.5 concentration above 35.00 μg/m³ fell from seventeen to six. The most heavily polluted season was typically winter, particularly in December and January, which had a monthly concentration around 1.50–2.00 times the annual average. The maximum annual and monthly concentration globally decreased. The number of days where pollution exceeded 75.00 μg/m³ dropped dramatically from more than 600 days to approximately 200 days. Nevertheless, ensuring the continuous improvement of PM_2.5 in SCB remains a great challenge in the face of the swift socioeconomic growth, as no city reached less than 15.00 μg/m³ annually and PM_2.5 levels rebounded, particularly in peak months occasionally but significantly.

These conclusions direct us to the following proposals:

(1) A comprehensive regional strategy should be adopted to control PM_2.5 emissions in SCB; its effectiveness should be continuously evaluated and adjusted accordingly, particularly in the peak emissions months in winter. The Construction outline of Chengdu-Chongqing Economic Circle (State Council of the PRC, 2021) stipulates that the air quality excellent rate should reach 88%, excluding two cities in SCB. Integration of the economic and environmental spheres is heavily recommended. PM_2.5 emissions rebounds should be prevented by strictly and persistently implementing environmental measures.