Introduction

Fine particulate matter (PM2.5) has become one of the most serious and urgent environmental pollutants globally (Bari and Kindzierski 2016; Pant et al. 2017). PM2.5 represents fine particulate matter smaller than 2.5 mm. It is a pressing environmental pollution that is destructive owing to its prolonged and severe socioeconomic influence (Guo et al. 2009). It is regarded as an air pollutant that not only causes gray skies but also affects human health to a great extent (WHO 2018). There is plenty of evidence that exposure to PM2.5 significantly increases the occurrence of stroke, cardiovascular, and respiratory diseases, in particular (Fan et al. 2016; Pant et al. 2017). Given the negative effects of PM2.5 on human life, an increasing number of studies have explored the possible influence factors of PM2.5 (Chen et al. 2018; Wang et al. 2020). Exposure to PM2.5 has also been identified as an OECD Green Growth indicator because of the harmful features and long-term socioeconomic impacts of PM2.5 (OECD 2018).

From a social-ecological perspective, human society is rooted in the natural environment and integrated in the complex and synthetical social-ecological system (Ng et al. 2019). PM2.5 is not just a pollutant; it is also a social challenge with complicated social reasons (Bari and Kindzierski 2016). Air pollution is caused differently in varying geographical and social-economic contexts (Jiang et al. 2020). Recent literature has shown that the spatial distribution of PM2.5 was influenced by social economic indicators (Chen et al. 2018; Fu and Li 2020). To date, only limited topics such as population expansion, urbanization, and economic development have been empirically analyzed, and an elaborate and detailed exploration of other social structure factors on PM2.5 is lacking (Wang et al. 2020). As a useful theoretical perspective, social vulnerability has emerged as a meaningful analytic approach to understand the social conditions of a society and its predisposition to be influenced by pollutions (Lee 2014). Research on natural disasters has described environmental disasters as the production of the social vulnerability of the local population (Svoboda et al. 2015). As indicated in the literature, social vulnerability may be the root cause of socio-economic factors affecting environmental issues (Fekete et al. 2014). Social vulnerability perspective has also been widely used in environmental disasters and hazards (e.g., earthquakes, floods, and droughts), differences between environmental pollutants and natural disasters necessitate careful consideration regarding the factors that appropriately associate social vulnerability to PM2.5 pollution. With a view to the differences and similarities between PM2.5 pollution and natural disaster, social vulnerability perspective to the generation of PM2.5 pollution is reasonable and essential. On one hand, PM2.5 pollution has the special characteristics of diffusivity and chronicity compared with natural disasters (Brooks et al. 2005). But on the other hand, the generation mechanism of PM2.5 pollution and natural disasters by human society is similar. And these natural disaster studies also provide some basic and primary indicators that constitute social vulnerability to PM2.5 pollution. In addition, the social vulnerability perspective is also regarded as the preferred approach to illustrate the reasons of social phenomena with the use of social factors (Vommaro et al. 2020). Therefore, the construction of social vulnerability to the generation of PM2.5 pollution is reasonable and essential.

Over the previous decades, vulnerability research has analyzed environmental pollution and been widely implemented in engineering-based and biophysical-based vulnerability paradigms, with emphasis on natural characteristic analysis of environmental pollution (Guo et al. 2020). To some extent, social vulnerability represents the internal structure and conditions of the social system that exist before the occurrence of environmental pollution (Lee 2014). The literature has documented that social vulnerability to environmental issues originates in social and geographic characteristics, such as geographic distribution and the social resources configuration (Fu and Li 2020). These studies provide some basic and primary indicators that constitute social vulnerability. Less focus being placed on social vulnerability to environmental pollution has limited our understanding of environmental pollution governance. The relevance of social vulnerability in the creation of environmental problems is high, but the literature specializing in PM2.5 is scarce and focused mainly on environmental hazards and crises. So far, social vulnerability to PM2.5 has been limited to qualitative application and is yet to be applied quantitatively at the country scale. Overall, understanding social vulnerability to PM2.5 is a fundamental and crucial prerequisite for PM2.5 reduction. Consequently, examining the determinants of social vulnerability to PM2.5 globally can contribute to the formulation of effective mitigation and adaptation policies for PM2.5 governance. In response, the present study contributes to the literature in three aspects: (1) constructing the theoretical framework of social vulnerability to PM2.5; (2) examining the cause-effect relationship of social vulnerability indicators to PM2.5 globally; and (3) combining spatial econometric modeling and geographically weighted regression (GWR) modeling to test the global and local relationships in the theoretical model, considering the spatial autocorrelation and spatial heterogeneity of regional PM2.5 exposure to the extent possible.

The theoretical construction of social vulnerability to PM2.5

Social vulnerability theoretical framework to PM2.5

Conceptually, vulnerability describes a region’s predisposition to be affected by external adverse disturbances and the absence of adaptive capacity (Chakraborty et al. 2018). It is an abstract concept that cannot be measured easily through the use of a single variable. In addition, vulnerability is considered a multidimensional construct with differences in indicator selection related to the target object (Fatemi et al. 2017). The Intergovernmental Panel on Climatic Change described vulnerability as the degree to which a system was susceptible to influence from climate change (Mackay 2007). Similarity, the United Nations International Strategy for Disaster Reduction (UNISDR 2009) defined vulnerability as the features of the population that made it susceptible to the influence of a potential pollution. For the concept definition of social vulnerability, it is generated under the definition of general vulnerability. The core components of social vulnerability are also consisted of sensitivity and capacity dimensions. However, social vulnerability focuses on social factors and emphasizes the deep understanding of how social factors that influence social phenomena. Different from general vulnerability, social vulnerability is based on the fundamental assumption that environmental pollutions are socially constructed, which places social structure and human beings at the center (Gallopín 2006). In general, social vulnerability describes the extent to which a society is susceptible to and would be able to adapt to the threats of environmental pollutions (Carrão et al. 2016).

Social vulnerability to PM2.5 involves not only regional sensitivity to PM2.5 but also the capacity to cope with PM2.5 effects. Sensitivity and capacity are two dimensions of social vulnerability to PM2.5. From the conceptual composition of sensitivity dimension, it indicates the degree to which an area is influenced by different kinds of environmental disturbances, which is determined by the basic demographic characteristics and intrinsic social structure of the region (Gallopín 2006). For basic demographic characteristics, it has been widely demonstrated that demographic characteristics impinge on the distribution of environmental pollution among regions (Climent-Gil et al. 2018; Wu and Geng 2020). Despite this, social structural factors also affect environmental pollution with a deep and complex mechanism (Wang and Wang 2020; Yang et al. 2020). For a deeper understanding of social structure connections to social vulnerability, social inequality has been mentioned as a factor for various kinds of environmental pollution (Yang and Liu 2018). The issue of disparities in exposure to environmental burdens and their underlying causes among humans in different social conditions has been of interest in environmental justice literature (Boone et al. 2009). It has also been mentioned by scholars that the unequal distribution of social resources further increased the exposure to environmental pollution (Song et al. 2019). Social inequality is therefore constructed to explain the influence of social resource distribution to the generation of PM2.5 pollution. Capacity is the ability of a region to accommodate potential environmental impacts, take advantage of opportunities, and manage subsequent consequences, which mainly depends on its economic, ecological, and social capacities (Vommaro et al. 2020). With regard to economic capacity, economic development level has been found to be linked to environmental degradation governance (Dinda 2004). Ecological capacity is the capacity of a particular ecosystem to produce and regenerate what people demand from surfaces (Vačkář 2012). With respect to social governance capacity, the general social governance level is regarded as the guarantee for various regional development issues, such as environmental pollution adaption (Nguyen and Liou 2019). Governance ability has also been used as a component of social vulnerability (Brooks et al. 2005). Based on the fundamental definition of social vulnerability and the supporting literature discussed above, the sensitivity dimension of social vulnerability is constituted by demographic structure and social inequality, and the capacity dimension of social vulnerability consists of social economic, ecological, and social governance capacities.

Selection of social vulnerability assessment indicators

According to the theoretical framework of social vulnerability constructed above, a series of social vulnerability assessment indicators can be selected.

Sensitivity dimension of social vulnerability to PM2.5

Social demographic structure

In terms of demographic structure, population is considered to indicate vulnerability because it contributes to determining the human pressure on the natural environment (Dondo Bühler et al. 2013; Nguyen and Liou 2019). Population density reflects the population agglomeration level, and it is related to the increasing demand for public transportation and services that may cause more environmental pollution (Chen et al. 2020). Thus, both high population and population density tend to lead to higher PM2.5. With the rapid growth of urbanization in the recent decades, various environmental challenges, such as excessive consumption of natural resources and global warming, have appeared everywhere (Childers et al. 2015). Urbanization refers to the transformation of the rural population into an urban population; it means that population, wealth, and resources are highly concentrated in urban areas. It has boosted the regional economy but put pressure on environmental quality (Wang et al. 2020). Previous research has reported a link between urbanization and PM2.5 concentrations in a number of different regions (Zhu et al. 2019).

Social inequality

Social inequalities in education, economic level, and health have been commonly considered to indicate the imbalanced resource distribution within regions (Binelli et al. 2015). According to the literature, inequalities in education and income may be considered antecedents of environmental pollution, and inequality in health is the consequence of environmental pollution (Grunewald et al. 2017; Yang and Liu 2018). In empirical studies, the unequal distribution of social resources has been demonstrated to cause environmental pollution such as haze pollution and water pollution (Ezbakhe et al. 2019). Income inequality is linked to consumption competition and leads to subsequent emissions of air pollution from transportation and resource consumption (Demir et al. 2019). The relationship between inequality in education and PM2.5 is rarely studied in the existing literature. Taking income inequality as a reference, it can be deduced that inequality in education is related to occupational resource competition that may cause PM2.5 from mass labor-intensive industrial production. A few studies have linked social inequality with environmental pollution by correlation analysis. However, a detailed causal analysis is absent in the literature. Therefore, we examine whether inequality in education and income contribute to the distribution of PM2.5.

Capacity dimension of social vulnerability to PM2.5

Social economic capacity

The correlation of economic development and environmental pollution has been examined widely (Altıntaş and Kassouri 2020). As the environmental Kuznets curve theory implies, although economic development causes environmental pollution, it also provides feasible economic support to solve environmental pollution and improve environmental quality (Hao et al. 2018). Both gross domestic product (GDP) per capita and gross national income (GNI) per capita are common economic indicators to evaluate the economic development of a country (Mia et al. 2018). Compared with GDP per capita, GNI per capita is more popular as a social economic capacity indicator. It was developed by World Bank and classifies countries into four groups (low, lower-middle, upper-middle, and high economies).

Social-ecological capacity

Biocapacity is an ecosystem’s capacity to produce biological materials for humans and to absorb waste materials generated by humans (Global Footprint Network 2019). It is regarded as an important kind of ecological wealth that plays a fundamental role in social production (Niccolucci et al. 2012). Biocapacity is the carrying capacity of a country’s biologically productive land; it determines social sustainability (Nguyen and Liou 2019). Land has been used as a convincing social-ecological vulnerability indicator (Ng et al. 2019). Biocapacity is also the regional capacity to supply sufficient ecological resources and absorb human-generated waste and pollution (Marti and Puertas 2020). It is of strategic value in geopolitics and reflects the competitiveness and potential quality of ecological resources among countries (Danish et al. 2019).

Social governance capacity

As an institutional guarantee, social governance plays a basic role in effective governance to environmental pollutions. Environmental pollutions combined with poor governance have historically resulted in widespread socioeconomic loss, huge mortality, and intense social conflicts (Below et al. 2007). Resilience is the positive expression of vulnerability, which is viewed as a relatively proactive expression of a region’s engagement with the goals of reduction of impacts and damages. From the resilience perspective, the resilience-building approach aims at strengthening the ability to cope with potential damages through effective social management strategies (Juhola and Kruse 2013). Compared with a resilient society, a vulnerable society is predisposed to be severely affected by environmental threats (Mafi-Gholami et al. 2020). Governance has also been recognized as an important force in reducing national-level vulnerability (Nanda et al. 2019). In new urban governance concepts, governance has been recognized as an approach for the improvement of society and avoidance of potential environmental risks (Van der Heijden 2016). The World Governance Index (WGI) jointly produced by the World Bank Development Research Group, Natural Resource Governance Institute, and Brookings Institution is a comprehensive and synthetic assessment index combining more than 30 data sources, such as a variety of surveys covering institutes, enterprises, citizens, and experts. It has been regarded as a comparable and transparent country governance level indicator (Ward and Dorussen 2015). The WGI is selected as the social governance indicator in this study.

Data and methods

Data

PM2.5 for countries was obtained from OECD’s environmental statistics database (https://stats.oecd.org/). It estimates PM2.5 concentrations by integrating satellite observations, chemical transport models, and measurements from ground monitoring station networks. The social governance index derived from the WGI has six aggregate indicators: voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law, and control of corruption. They are expressed in standard normal units ranging from −2.5 to 2.5 (GWI 2019). In the National Footprint Accounts, biocapacity is summarized by six types of ecologically productive lands (cropland, fishing grounds, built-up land, forest products, carbon, and grazing land). The newest data for biocapacity in the Global Footprint Network database are updated for the year 2016 (Global Footprint Network 2019). Therefore, all the data in our study are for the year 2016. Table 1 provides a detailed description and lists the sources of all social vulnerability variables. The statistical description of the variables is presented in Table 2.

Table 1 Description and data sources of social vulnerability variables
Full size table
Table 2 Statistical summary of all variables
Full size table

Spatial autocorrelation analysis

Environmental phenomena are usually spatially autocorrelated because of their geographical distributions (Qin et al. 2019). Spatial autocorrelation analysis is a powerful spatial statistical method for exploring spatial disparity and relativity. There are two types, namely, global and local spatial autocorrelation analysis. Global spatial autocorrelation identifies the general pattern of spatial aggregation and heterogeneity but fails to reflect the specific aggregation regions. Local spatial autocorrelation illustrates the specific association and variation among adjacent positions (Anselin 1995).

Global Moran’s I is the most generally used statistical index to measure the spatial dependence in observations. It determines whether the observations are dispersed, clustered, or randomly distributed based on locations. Its value ranges from −1 to 1 (Moran 1950). A positive value indicates positive spatial correlation, and a negative value corresponds to negative spatial correlation. There is no spatial correlation when the value is 0. The closer the index is to 1, the stronger is the spatial correlation. The global Moran’s I is defined as follows:

$$ I=\frac{\frac{N}{S_0}{\sum}_i{\sum}_j{W}_{ij}{Z}_i{Z}_j}{\sum_i{Z}_i^2},\kern0.5em {S}_0=\sum \limits_{\mathrm{i}}\sum \limits_{\mathrm{j}}{\mathrm{W}}_{\mathrm{i}\mathrm{j}}, $$

where N is the number of countries; Z is the deviation from the mean of variables; S0 is the summary of spatial weights; and Wij is the spatial weight matrix based on distance.

Local spatial autocorrelation (also called local indicator of spatial autocorrelation (LISA)) examines the specific spatial accumulation features and finds the degree of local spatial heterogeneity among areas and their surrounding positions (Anselin 1995). Local Moran’s I demonstrates the particular characteristics of location-specific information by local indication of spatial clusters. The local Moran’s I is given as

$$ I={Z}_{\mathrm{i}}\sum \limits_{\mathrm{j}}{W}_{\mathrm{i}\mathrm{j}}{Z}_{\mathrm{j}}, $$

where the parameter meaning is consistent with the global Moran’s I index.

Spatial econometric model analysis

A spatial econometric model analyzes variables in space and incorporates spatial autocorrelation and spatial dependence efficiently. This regression technique has been broadly used in fields such as economics, demography, environmental studies, and social sciences (Gilbert and Chakraborty 2011). Depending on the correlation of variables, both the spatial lag model (SLM) and the spatial error model (SEM) are commonly used spatial econometric models in the literature (Zhang et al. 2018). Specifically, the SLM adds a spatial lagged dependent variable to the model, which examines the spatial endogenous interaction effects of the explanatory variable. In this case, it would suggest that PM2.5 nearby would affect PM2.5 in the country. The spatial lag model is given by the following equation:

$$ Y=\rho \mathrm{WY}+ X\beta +\mu, $$

where WY denotes the spatial lag term; ρ denotes the spatial coefficient; and μ is the error term.

In contrast to the SLM, in the SEM, the spatial autocorrelation appears in the error term over space. The SEM assumes spatial autocorrelation is derived from the error term and considers variables with spatial heterogeneity that are ignored in traditional non-spatial models. The spatial error model is calculated by the following formula:

$$ Y= X\beta +\varepsilon, \varepsilon =\lambda W\varepsilon +\mu, $$

where λ is the spatial coefficient of the error term; ε represents the spatial error term; Wε is the spatial weight matrix of ε; and μ is a random error.

Geographically weighted regression model analysis

Unlike traditional global regression models that assume that the parameters of the variables examined are constant over space, a geographically weighted regression model allows for the parameters to vary across space (Wang et al. 2019). GWR is superior to ordinary least squares (OLS) regression modeling because it supports the calculation of parameters’ estimation of local variations between independent and dependent variables rather than implement a global regression (Xie et al. 2018). The GWR model is useful for examining spatial non-stationarity and spatial heterogeneity, and it has been widely employed in environmental research (Fotheringham et al. 2002). The detailed mathematical formula of the GWR model is as follows:

$$ Y\mathrm{i}=\beta 0\left(u\mathrm{i},v\mathrm{i}\right)+\sum \limits_{k=1}^p\beta k\left(u\mathrm{i},v\mathrm{i}\right)X\mathrm{i}k+\varepsilon \mathrm{i}, $$

where Yi is the PM2.5 of each country; ui and vi are the geographical coordinates for country i; β0 is the intercept of the model; βk is the coefficient of Xik; Xik represents the independent variable Xk for country i; p is the number of the independent variables; and εi is the residual error.

The estimated coefficient βk can be calculated by the following matrix equation:

$$ \beta k={\left[{X}^TW\left(u\mathrm{i},v\mathrm{i}\right)\right]}^{-1}{X}^TW\left(u\mathrm{i},v\mathrm{i}\right)Y, $$

where T represents the matrix transpose operation; Y is the vector of PM2.5 exposure; and W is the spatial weight matrix based on distance.

Results and discussion

Spatial distribution characteristics of PM2.5

The global PM2.5 map was developed in the Geoda software using the administrative boundaries of countries contained in the shape files obtained from the Database of Global Administrative Areas (GADM, version 3.6) (https://gadm.org/data.html). The PM2.5 concentrations in the year 2016 ranged from 5.768 μg/m3 in Brunei Darussalam to 99.182 μg/m3 in Nepal, with the median value of 22.224 μg/m3 observed in Montenegro (Fig. 1). This is in line with previous research that reported that the spatial distribution of atmospheric pollutants varied across countries (Yang et al. 2018). The mean value of PM2.5 concentration worldwide was 27.884 μg/m3, which was markedly high compared to the PM2.5 air quality guideline threshold (10 μg/m3) proposed by the WHO (2005).

Fig. 1
figure 1

Spatial distribution of PM2.5 globally

Full size image

As indicated in Fig. 1, the PM2.5 exposure had a significantly spatial difference. A total of 44 countries (24.859%) suffered from PM2.5 concentrations higher than 35 μg/m3 (the interim target (IT)-1 level of the WHO), which is adverse to the long-term health and mortality risk of humans. The high PM2.5 concentrations were mainly distributed in countries (e.g., Nepal, Saudi Arabia, India, and China) of the Middle East and North Africa and the South Asia regions, which also have frequent dust storms and a tropical desert climate (Kaskaoutis et al. 2018). As Li et al. (2018) found based on Bayesian statistics, countries with severe PM2.5 concentrations from the year 2000 to 2014 were mostly distributed in Asia and Africa. The present study confirmed that this trend was maintained until at least the year 2016. Only 14 countries (7.910%) had PM2.5 concentrations lower than 10 μg/m3 corresponding to relatively excellent air quality. These countries (e.g., Finland, New Zealand, Sweden, and Iceland) were high-income economies and mainly located in Europe and East Asia. It could be concluded that these high-income countries were good at transforming extensive PM2.5 into effective air quality control. To sum up, most countries globally were subjected to substantial PM2.5 threats.

Spatial autocorrelation results

Global spatial autocorrelation result

The global Moran’s I of PM2.5 was 0.390 (Z = 13.491, p < 0.05), suggesting that PM2.5 was not isolated or randomly distributed on the whole. The positive correlation revealed an evident spatial agglomeration of exposure to PM2.5 globally. Consistent with previous literature, this phenomenon clearly demonstrates that countries with close geographical distances have mutually positive effects regarding PM2.5 globally (Fu and Li 2020).

Local spatial autocorrelation result

To further analyze the spatial autocorrelation of PM2.5 locally, the Moran’s I of PM2.5 concentrations for local spatial correlation analysis was drawn, as shown in Fig. 2a, b. Four clusters were derived using the Geoda software.

Fig. 2
figure 2

Local spatial autocorrelation of PM2.5. a The LISA map of PM2.5. b The local spatial autocorrelation significance map of PM2.5

Full size image

For the high PM2.5 concentration cluster of the deep red color, some Asian countries (e.g., India, Nepal, and China) and many African countries (e.g., Central African Republic, Chad, and Cameroon) exhibited significant (p < 0.05) high-high clustering for PM2.5 exposure. The obvious spillover effect of PM2.5 of these countries implicated that PM2.5 diffused from neighboring countries through air mass diffusion. In contrast, in the low PM2.5 concentration cluster of the deep blue color, most countries (e.g., Finland and Sweden) were European countries that presented significant (p < 0.05) low-low clustering for PM2.5 exposure. These countries seem to have remarkably clean air quality. Some countries (e.g., Kazakhstan and Kyrgyzstan) showed low-high clustering for PM2.5 (light blue color), suggesting that these countries might transfer PM2.5 to adjacent countries, which leads to negative air quality nearby. The PM2.5 was spatially uneven throughout the world.

Spatial econometric model results

For spatial econometric model adaptability, we identified whether spatial effects exist in our model. Following Anselin et al. (2006) and López-Hernández (2013), a diagnostic for spatial dependence with the Moran’s I test in the error term was applied to examine whether the non-spatial regression model was enough or a spatial econometric model was required. The value of Moran’s I was 3.643 (p < 0.001), indicating a strong positive spatial autocorrelation in the errors. Therefore, the spatial econometric model was appropriate for further regression estimation. In terms of model selection, the Lagrange multiplier (LM) test for the lag and error is typically used to determine the optimal model (López-Hernández 2013). According to the LM test, the LM error value 4.094 (p < 0.05) was higher than LM lag value 3.437 (p > 0.05), indicating that the SEM was more appropriate than SLM. Excluding missing values of the variables selected above, a total of 132 countries were used for model analysis. In order to eliminate differences in numerical magnitudes, we standardized all the variables. The SEM results are shown in Table 3. The model collinearity was 4.780, which was significantly lower than the diagnostic criterion of 30 in Geoda, indicating that the variables did not exit notable collinearity. The R2 of the SEM was 0.587, and the spatial error coefficient lambda was 0.427 (p < 0.05), implying that the spatial econometric model was valid and those social vulnerability variables played vital roles in the generation of PM2.5 pollution.

Table 3 The results of the SEM for PM2.5
Full size table

As illustrated in Table 3, population, education inequality, biocapacity, and social governance displayed significant influence on country level PM2.5, whereas other variables did not present significant relationships. In the dimension of sensitivity, total population and inequality in education were two important indicators that significantly increased the PM2.5 (p < 0.001). Population exhibited a positive effect (0.270) on PM2.5, implying the production of air pollution is caused by humans, consist with previous research (Branis and Linhartova 2012). It could be inferred that the large size of the population has brought about much air pollution due to the excessive utilization and consumption of environmental resources. The density of population, however, did not significantly correlate to PM2.5, implying that PM2.5 was induced by complex productive and consumptive activities determined by the deep structure of society. Different from previous studies (Fu and Li 2020; Han et al. 2018), urbanization was not found to be significantly associated with PM2.5 worldwide, which suggested that population expansion of urban cities did not certainly lead to high level of PM2.5 pollution. Because of the complexity of the urbanization process, the impact of urbanization on environmental pollution production is not completely known. As Han et al. (2016) reported, the relationship between urbanization and air pollution substantially varied among countries, and an optimum urban size might be suitable for environmental sustainability.

It should be noted that inequality in education had positive impact (0.350) on the generation of PM2.5 pollution. This novel discovery indicated that inequality in education was one of the major social vulnerability factors for PM2.5 generation. As the social inequality proposition argued in disaster vulnerability, social inequality may lead to the uneven distribution of environmental disasters (Bolin 2007). To some extent, the current study demonstrated that the social inequality proposition of education inequality in PM2.5 pollution appeared worldwide in the year 2016. As mentioned in the theory construction section, inequality in education was closely linked to occupational resource competition. The high degree of education inequality stimulated the unequal allocation of opportunities for jobs. Besides, the majority of low-educated individuals tended to engage in fundamental and pollution-oriented job environments, leading to more environmental pollution troubles for them. Therefore, inequality in education could be a major factor in the increase of PM2.5 pollution. Compared with education inequality, no significant influence of income inequality on PM2.5 pollution was found, indicating the income gap between the rich and the poor did not link with the generation of PM2.5. Although some previous studies had found that income inequality increased air pollution (Coondoo and Dinda 2008; Golley and Meng 2012), some other research reported with the quite opposite conclusion that income inequality decreased air pollution (Demir et al. 2019). This controversial consequence somewhat indicated the complexity of income inequality influence mechanism on air pollution. As Grunewald et al. (2017) verified based on substantially large data of regional and temporal coverage, the ambiguous relationship between income inequality and pollution might vary with the income level of countries. The marginal propensity to emit air pollution between rich and poor was uncertain, varying from country to country. In addition, it had also been confirmed that the production-based pollution model was different among countries (Wang and Zhou 2020), indicating that pollution production was closely related to the economic development level of counties. The impact of social inequality on PM2.5 pollution was summarized as follows: the inconsistent results between education inequality and income inequality on the PM2.5 were due to their different influence mechanisms on the generation PM2.5 pollution. From the point view of impact mechanism, education inequality was related to occupational differentiation, while income inequality was related to consumption differentiation. Further, the industrial air pollution production derived from occupational differentiation leaded to much PM2.5 pollution. But for the consumption competition approach of income inequality on PM2.5 generation, the rich and poor did not exhibit different marginal propensity to emit air pollution in the process of transportation and resource consumption.

With regard to the capacity dimension, biocapacity had a negative effect (−0.138) on the generation PM2.5, indicating that biocapacity was a useful ecological resource to decrease PM2.5 pollution. As described in “The theoretical construction of social vulnerability to PM2.5,” biocapacity is the essential social-ecological wealth that functions to degrade garbage and absorb pollution (Marti and Puertas 2020; Niccolucci et al. 2012). Scholars have also found that areas with less or no vegetation cover were more vulnerable than densely vegetated regions (Nguyen and Liou 2019). Similarly, Liu et al. (2017) found that green land had both direct and indirect positive effects on air quality improvement. PM2.5, especially, as a volatile organic pollutant, is influenced by the regional environmental carrying capacity relating to atmospheric removal, dilution, and deposition (e.g., thermodynamic stability, cloud scavenging, and precipitation) (Chen et al. 2018). Effective maintenance of country-level biocapacity is an important ecological guarantee against PM2.5. The results strongly indicated that social governance significantly decreased (−0.297) PM2.5. It could be inferred that good social governance was an effective instrument for PM2.5 mitigation. In latest sustainability assessment methodologies, good governance has been viewed as an indicator for a well-balanced society and a reliable tool to adapt to various challenges, including environmental pollution (Singh et al. 2009). Social governance emphasizes the process by which governments are functioned and monitored and the ability of the government to propose and effectively implement social policies (Ward and Dorussen 2015). It is also an important topic in social policy formulation and in devising approaches to control environmental pollution (Bos and Gupta 2019). In our study, we may infer that improvement of social governance helped to reduce PM2.5 throughout the world. For GNI per capita, there was no strong evidence that it significantly related to PM2.5 globally. Although some previous studies have reported that environmental pollution is correlated with high economic development (Iwata et al. 2011), the relationship between economic development and environmental pollution is really complex and may be disturbed by other social factors (Altıntaş and Kassouri 2020). Rather than a simple linear relationship, more complex relationships such as an inverted U-shaped or N-shaped curve might occur between economic development and environmental pollution (Apergis and Ozturk 2015; Friedl and Getzner 2003). The relationship between economic development and PM2.5 is closely dependent on the social development stage and economic development level of each country (Fu and Li 2020). As indicated by Speak et al. (2012), high-income countries had effectively controlled air particulate pollution by energy efficiency improvements and green-city development strategies. Unfortunately, low-income countries tend to produce huge volumes of air pollutants that exceed the carrying capacity of their natural ecosystems (Han et al. 2018). In the following section, we present a GWR model that considers spatial variability.

Geographically weighted regression model results

Although the global impacts of social vulnerability on PM2.5 have been examined by the spatial econometric model, the local effects of the social vulnerability indicators remain unknown. Therefore, the geographically weighted regression model was analyzed to further clarify the regional heterogeneity between the social vulnerability variables and PM2.5. The GWR model (R2 = 0.610, AIC = 271.686) was better fitted than the OLS model (R2 = 0.558, AIC = 276.004), indicating that the GWR model had a better goodness-of-fit than the classical OLS model. The standardized geographically weighted results obtained through the MGWR software are presented in Table 4.

Table 4 The results of the GWR on PM2.5
Full size table

As shown in Fig. 3, the local R2 ranged from 0.529 to 0.673, exhibiting different degrees of fit for the 132 counties. Most countries (e.g., Norway, Sweden, Belarus, and Netherlands) with high R2 were located in Europe and Central Asia, and countries with low R2 were distributed in Latin America and the Caribbean (e.g., Chile, Argentina, and Uruguay) and Sub-Saharan Africa (e.g., South Africa, Lesotho, and Botswana). In general, the relationship of social vulnerability and PM2.5 was better captured in Europe and Central Asia regions by the regression model.

Fig. 3
figure 3

The local R-squared effects of social vulnerability variables on PM2.5

Full size image

Figure 4 shows the local coefficients of the social vulnerability variables that significantly correlated with PM2.5. As shown in Fig. 4a, the total population in Angola, Cameroon, and many other countries in Africa exerted the most influence on PM2.5 because these countries had a large population and a relatively low economic development level. As indicated by Speak et al. (2012), high-income countries were more likely to have effective air pollution control strategies, whereas low-income countries were poor at environmental governance. For education inequality shown in Fig. 4b, Thailand, Pakistan, India, and other countries concentrated in south and east Asia exhibited a significant effect on PM2.5. Education inequality in these countries contributed greatly to the creation of PM2.5. It can be inferred that education resources in such developing countries are limited, and people with less education might work in energy-driven and labor-intensive industries that cause mass air pollution. The impact of biocapacity on PM2.5 in India, Vietnam, Philippines, and many other lower-middle-income countries displayed effective capacity for PM2.5 reduction (Fig. 4c). We supposed that PM2.5 in these less-developed countries was mainly absorbed by their ecosystem naturally. Furthermore, the impact of social governance in lower-middle-income countries such as India and Cambodia strongly contributed to eliminating PM2.5. Similarly as biocapacity, social governance effectively alleviated PM2.5 concentration, especially in less-developed and developing countries (Fig. 4d). Some of these developing countries, such as India and Nepal, were regions with the most severe PM2.5 exposure. As argued by scholars, PM2.5 in India has been increasing notably in the past years and India has already been recognized as an air pollution hotspot country (Pant et al. 2017). According to a previous study, the lack of powerful environmental laws and sound protection policies in undeveloped countries might inevitably lead to mass air contamination (Han et al. 2018). Therefore, our result signified that PM2.5 in developing countries should be emphasized and controlled by social governance. The capacity of social governance exhibited a prominent role in most lower-middle-income countries, implying that PM2.5 reduction in lower-middle-income countries could be obviously improved by social governance. The likely reason is that most lower-middle-income countries were developing countries that sustained development by relying on secondary industries without adequate economic investments and high-technology achievements on air pollution control. Therefore, policymakers in these lower-middle-income economies need to recognize that maintaining the balance between economic development and environmental sustainability as well as encouraging technological progress and innovation can reduce PM2.5.

Fig. 4
figure 4

The local social vulnerability effects on PM2.5. a The local population effects on PM2.5; b the local education inequality effects on PM2.5; c the local biocapacity effects on PM2.5; and d the local social governance effects on PM2.5

Full size image

Conclusions and policy implications

The present study constructed and examined social vulnerability to PM2.5 on a global level. Based on the social vulnerability concept and the relationship between social vulnerability and PM2.5, sensitivity and capacity dimensions with a range of appropriate indicators were selected for social vulnerability assessment using a theoretical construction approach. From spatial autocorrelation analysis, a positive global spatial autocorrelation and different local spatial clusters of PM2.5 were found, indicating a remarkably spatial heterogeneity worldwide. Then, a spatial econometric model and a geographically weighted regression model were used to explore the influence of social vulnerability indicators on PM2.5 with the consideration of spatial autocorrelation and spatial heterogeneity. The major findings of the current research are presented as follows: Population and education inequality in the sensitivity dimension exhibited a significant positive impact on PM2.5, and biocapacity and social governance in the capacity dimension strongly contributed to the decrease of PM2.5 on the whole, as found through the SEM analysis. Furthermore, the geographically weighted regression model revealed spatial heterogeneity in the effects of social vulnerability variables on PM2.5.

Based on the current study, several PM2.5 pollution governance policy implications can be proposed, as follows. First of all, social vulnerability evaluation is a powerful analytic approach for effective PM2.5 pollution mitigation and adaption. This conclusion could provide a useful reference for governance of other similar environmental pollutions. With respect to social structure, focusing on the elimination of social inequality, especially education inequality, is essential for PM2.5 decrease. Meanwhile, the elimination of education inequality at the country level is valuable for the PM2.5 pollution reduction by optimized occupational resource allocation approaches. As reducing and eliminating social inequalities among countries is the primary intention of the Sustainable Development Goals (UNDP 2016), narrowing the social-structural gap might reduce environmental pollution and improve the regional ecological quality.

Secondly, it is imperative to improve social governance for PM2.5 reduction. A society with good governance would tackle PM2.5 pollution competently before it translated into a serious environmental crisis or event. Therefore, it is vital to take proactive social governance strategies rather than reactive actions, especially in developing countries. Developing countries have also been indicated to take more air pollution reduction responsibility. It is therefore vital for developing countries to strengthen the general social governance capacity in order to provide better social environment for powerful PM2.5 pollution control.

Finally, the positive spatial correlation of PM2.5 globally signified that international cooperation could help to cope with PM2.5 pollution among countries. Joint regional PM2.5 prevention and control measures should be established for effective PM2.5 governance across countries to deal with this global pollution. The spatial spillover effect of PM2.5 promoted the collaborative governance among countries and became an attractive way to achieve win-win cooperation for better air quality. Besides, in consideration of the heterogeneity of regional economic development levels and social structure conditions, targeted PM2.5 pollution governance policies should not be ignored within each country.