Introduction

Environmental issues have become a major bottleneck to China’s sustainable socio-economic development (Liu and Raven, 2010). Since the beginning of the 21st century, China has made significant achievements in economic construction, with growth rates well above the global average and the world’s second-largest economy (Li, 2023; Li et al., 2023). However, this rapid economic development has come at a high environmental cost, resulting in severe pollution as a by-product. Since the implementation of the Air Pollution Prevention and Control Action Plan, Sulfur dioxide (SO2) concentrations in the Beijing–Tianjin-Hebei region and surrounding areas have steadily decreased (He et al., 2023). However, the reduction of secondary components of particulate matter nationwide has been significantly slower than that of primary pollutants, often associated with the rapid growth of secondary components such as sulfates. SO2 is a major air pollutant produced by human activities that burn fossil fuels and is a primary precursor of particulate matter, so continued improvement in SO2 concentrations remains critical (Aldekheel et al., 2024; Huang et al., 2019). However, existing emission reduction policies are relatively stand-alone, focusing only on direct emissions from specific regions or sectors, while neglecting the complexity of socio-economic systems. This complexity arises from the close economic links between regions and industries, resulting in the transfer of pollutant emissions through cross-regional and cross-industry trade activities. In addition, ongoing trans-regional power dispatch and large-scale projects such as coal transport have further facilitated extensive pollution transfer across regions (Liu et al., 2017; Zhang and Peng, 2017). The latest “Action Plan for Continuous Improvement of Air Quality” issued by the Chinese government explicitly states the need to strengthen regional cooperation mechanisms for air pollution prevention and control. Therefore, air pollutant emissions transferred through inter-regional trade should be taken into account when formulating pollution control policies in order to enhance regional cooperation.

To map the “transfer” pathways of inter-regional pollutant emissions embedded in product and service trade networks, some studies have adopted the gravity model approach (Dong and Li, 2022, Huo et al., 2022; Wang et al., 2018). In contrast, the multi-regional input–output (MRIO) model, which is commonly used to assess the environmental impacts of trade, can more accurately quantify inter-regional pollution transfers. The MRIO model uses inter-regional trade matrices to track value chains and their feedback effects among multiple trading partners, thereby identifying the final consumers of products and the associated resource and environmental costs (Wang et al., 2024c; Dorninger et al., 2021). Most existing research focuses on the pollutant emissions embodied in trade (Wang et al., 2017; Zhao et al., 2015), their contributions to air quality or public health, and, to some extent, the unequal relationships between developed and underdeveloped regions in China (Qian et al., 2019, Xu, 2023; Zhao et al., 2023). For example, Zhang et al. (2018) used the MRIO model and provincial-level industrial air pollutant emission inventories to explore environmental inequality and found that 62–76% of consumption-based air pollutant emissions in wealthier regions (Beijing–Tianjin, East Coast, and South Coast) are outsourced to other regions, while about 70% of value added is retained. Zhang et al. (2021) used an inter-sectoral input–output framework to analyze SO2 and NOx emissions and drivers from an industry linkage perspective, but further inter-regional analysis was lacking. Overall, previous studies have contributed significantly to understanding the origins of GHG and pollutant emissions challenges from a consumption-based perspective. However, few have examined the characteristics and drivers of regional positions within the implicit SO2 emissions network. To gain a more comprehensive understanding of the intertwined SO2 emission flows, rather than focusing only on the two ends of the supply chain (production or consumption-based perspectives), complex network analysis can be used to further explore the characteristics of implicit emissions obtained by the MRIO model (Liu et al., 2023). Quantifying the regional positions, linkages, and changes in driving factors for pollution emission transfers not only provides a basis for determining the long-term pollution control responsibilities of each province in China and making future projections, but also provides scientific guidance for multi-regional cooperation in emission reduction.

Ecological Network Analysis (ENA) stands out as a powerful method for analyzing and understanding the components and overall structure of energy consumption or emission networks (Ren et al., 2022). Rooted in input–output analysis, ENA adopts a systemic approach that unveils the control or interdependency linkages among different regions from a holistic viewpoint (Ulanowicz, 2004; Yang et al., 2012). Presently, ENA has found successful application in various environmental impact domains, including the urban land-carbon relationship (Wei et al., 2022; Xia and Chen, 2020), carbon metabolic flow analysis (Zhang, Li, et al., 2018), and agricultural freshwater impacts (Gao et al., 2018). Primarily, ENA serves to quantify the absolute relationships of nodes (provinces) within the entire emission network and among other nodes, facilitating the formulation of pertinent policy recommendations and furnishing a theoretical and practical groundwork for further inquiry. On the other hand, Social Network Analysis (SNA) is a method used to map out the network coherence and hierarchy forged by interconnections between nodes, elucidating spatial connectedness and relationships (Bai et al., 2020; Sun et al., 2020). This approach creates a spatial relationship matrix by analyzing the features of the research topic and object. It evaluates the overall network characteristics, individual network attributes, and spatial clustering aspects of the research subject (Dong et al., 2023). From its definition and evaluation outcomes, SNA excels in evaluating the overall state of the emission network and the relative positioning among nodes. Thus, while this study primarily relies on ENA methodology, it also integrates the conceptual framework of SNA to amalgamate analytical methods, aiming to garner a more comprehensive understanding of the transfers in the embodied SO2 emission network over the past decade.

With the increase in energy consumption, the proliferation of motor vehicles, and the expansion of industrial scale, China’s atmospheric environmental challenges have become increasingly pronounced. China has emerged as an indispensable partner in shaping global policy frameworks aimed at curbing pollutant gas emissions (Wang et al., 2024b; Xing et al., 2020). However, SO2 emissions exhibit wide disparities across China’s provinces and industrial sectors, with complex emission sources. Hence, a precise and comprehensive quantification of regional and sectoral SO2 emissions, coupled with a thorough exploration of underlying emission sources and network correlations, is imperative for China to devise national SO2 emission reduction strategies and inter-regional collaborative emission reduction strategies. The novelty and significance of this study manifest in three key aspects. Firstly, leveraging the Multi-Regional Input–Output (MRIO) table and provincial sectoral emission inventories, this study constructs a more nuanced region-to-region emission transfer network. This rectifies the inadequacies of previous imprecise provincial sectoral emission estimations (Qian et al., 2020; Shao et al., 2019), which resulted in the loss of region-to-region emission transfer data (Qian et al., 2022). The elucidation of pollutant transfer relationships among regions and sectors within the economic system identifies pivotal regions and sectors driving pollutant transfers, facilitating effective emission reduction strategies. Secondly, this study innovatively uses multi-year data and comprehensive network analysis methods to examine the characteristics and interrelationships of provincial nodes in the SO2 emission network, which provides a basis for determining the long-term pollution control responsibilities of nodes and predicting structural changes in the network. Finally, the Quadratic Assignment Procedure (QAP) method is used to investigate the factors influencing SO2 emission transfers and their variations, with the aim of providing insights for future regional cooperative governance.

Materials and methods

Emission inventories and inter-regional SO2 emission transfer networks

The China Environmental Statistics Database (CES) stands as the foremost authority on pollution survey data in China (Cai et al., 2019). For instance, the database’s industrial pollution source survey encompassed 40 major industries categorized under the “industrial classification of national economic activities,” along with the water production and supply sector. However, a notable limitation of CES lies in its lack of breakdown statistics for different sectors within regions, potentially undermining the region-sector relationship characteristics essential for emission network analysis (Qian and Wu, 2020). This study aligns the input–output tables with the SO2 emission inventories containing sectoral information (see Fig. S1). Emissions from sectors other than industry were nullified in the MRIO tables.

Input–output modeling is a quantitative economic technique employed to analyze the exchange of commodities and services among economic sectors in a specific region or country, utilizing input–output tables (Liang et al., 2023; Wiedmann et al., 2011). The environmentally extended Multi-Regional Input–Output (MRIO) approach discerns both direct and indirect emissions (via supply chains) stemming from the final consumption of regional goods and services (Duan et al., 2018). Consequently, an inter-regional SO2 network model was formulated, leveraging the direction and magnitude of SO2 flows derived from MRIO data. The fundamental linear equations of MRIO are delineated as follows:

$$x={Ax}+Y$$
(1)
$$x={\left(I-A\right)}^{-1}Y$$
(2)

where x, A, and Y denote the matrices of inputs, direct input coefficients, and final demand, respectively; Eq. (2) is a transformed expression of Eq. (1), and \({\left(I-A\right)}^{-1}\) represents the Leontief inverse matrix, which signifies the direct and indirect impacts of final consumption of goods and services within the region—essentially, the infinite-round production effects triggered by final demand. I denotes the unit matrix. To calculate the emissions associated with the consumption of commodities and services, Eq. (2) can be enhanced by using the SO2 emission intensity matrix, resulting in the inter-regional embodied SO2 transfer network. The comprehensive equation is articulated as follows:

$$\left[\begin{array}{cc}\begin{array}{cc}{{EC}}^{11} & {{EC}}^{12}\\ {{EC}}^{21} & {{EC}}^{22}\end{array} & \begin{array}{cc}\cdots & {{EC}}^{1n}\\ \cdots & {{EC}}^{2n}\end{array}\\ \begin{array}{cc}\vdots & \vdots \\ {{EC}}^{n1} & {{EC}}^{n2}\end{array} & \begin{array}{cc}\ddots & {{EC}}^{{rn}}\\ \cdots & {{EC}}^{{nn}}\end{array}\end{array}\right]=\left[\begin{array}{ccc}{f}^{1} & 0 & \begin{array}{cc}\cdots & 0\end{array}\\ 0 & {f}^{2} & \begin{array}{cc}\cdots & 0\end{array}\\ \begin{array}{c}\vdots \\ 0\end{array} & \begin{array}{c}\vdots \\ 0\end{array} & \begin{array}{cc}\begin{array}{c}\ddots \\ \cdots \end{array} & \begin{array}{c}\vdots \\ {f}^{n}\end{array}\end{array}\end{array}\right]\left[\begin{array}{cc}\begin{array}{cc}{B}^{11} & {B}^{12}\\ {B}^{21} & {B}^{22}\end{array} & \begin{array}{cc}\cdots & {B}^{1n}\\ \cdots & {B}^{2n}\end{array}\\ \begin{array}{cc}\vdots & \vdots \\ {B}^{n1} & {B}^{n2}\end{array} & \begin{array}{cc}\ddots & {B}^{{rn}}\\ \cdots & {B}^{{nn}}\end{array}\end{array}\right]\left[\begin{array}{cc}\begin{array}{cc}{Y}^{11} & {Y}^{12}\\ {Y}^{21} & {Y}^{22}\end{array} & \begin{array}{cc}\cdots & {Y}^{1n}\\ \cdots & {Y}^{2n}\end{array}\\ \begin{array}{cc}\vdots & \vdots \\ {Y}^{n1} & {Y}^{n2}\end{array} & \begin{array}{cc}\ddots & {Y}^{{rn}}\\ \cdots & {Y}^{{nn}}\end{array}\end{array}\right]$$
(3)

where n is the number of economic sectors; the diagonal of the f-matrix represents the sectoral emissions per unit of currency for each region; the B-matrix is the Leontief inverse matrix; the Y-matrix is the final demand matrix for all sectors in the various regions; and \({{EC}}^{{rn}}\) represents the transfer of SO2 from the r-region to the n-region. The diagonal elements of the EC matrix represent emissions originating from both production and consuming activities within a certain region. Due to the large amount of data in the EC matrix (if the MRIO table has 31 provinces and 40 sectors, it is a 1240*1240 matrix). This work focuses on calculating the net SO2 transfers across regions and sectors in the EC matrix, as well as the total embodied emissions resulting from final consumption in a region by summing the column vectors of the EC matrix. The consumption-based SO2 network reflects the nationwide economic activity that is triggered to meet regional demand for goods and services, resulting in the distribution of remote emissions.

Comprehensive network analysis

This study primarily employs ecological network analysis methods, while drawing on concepts from social network analysis, to comprehensively understand the changes in the implicit SO2 emission network over the past decade. The integrated network analysis includes through-flow analysis (TA), network centrality analysis (NCE), and network control analysis (NCA).

The formulas for each network analysis indicator are shown in Table 1. The TA analysis focuses on the percentage of implicit pollutant flow from each region in China’s trade relative to the total implicit emissions (P), which indicates the influence of nodes on the overall emissions network. High-value nodes are prioritized for SO2 abatement. The network centrality analysis uses three social network analysis indicators to measure the relative position of provinces. Degree centrality counts the number of times each province receives emissions; a higher number indicates that the node is transferring SO2 emission responsibilities to more provinces, resulting in a more complex reduction cooperation network. Closeness centrality measures a node’s independence from others in the network; a higher value indicates less control by other nodes and more direct emission transfers. Betweenness centrality assesses whether a node acts as a “bridge” in the network; a higher value indicates that the province plays a critical role in pollution transfer, making reduction efforts more difficult. NCA, derived from ecological network analysis, assesses the control and dependency capacities of network components based on paired total flows, indicating the degree of control or influence between regions in the network.

Table 1 The equation used to calculate the network-related indices.
Full size table

Multiple regression-quadratic assignment process (MRQAP)

After scrutinizing the characteristics of the SO2 network, this paper employs Multiple Regression Quadratic Assignment Procedure (MRQAP) to delve into the factors driving SO2 emissions within the network. MRQAP remains a widely utilized network regression technique in regional economic and innovation studies (Briseño-García et al., 2022; Park and Koo, 2021; Zagenczyk and Powell, 2023). It employs a ranking approach to evaluate the statistical significance and interdependence of relationship variables. Model fit coefficients and regression coefficients for observed data are juxtaposed with coefficients derived from an extensive ranking of rows and columns. For instance, original coefficient estimates are deemed significant at the 0.05 level if initial estimated coefficients still surpass 95% of estimates obtained through alignment (Borgatti et al., 2024). Given the presence of multicollinearity among variables affecting the SO2 network, such a condition may influence regression model estimation results. In contrast, the MRQAP model does not necessitate assumptions of independence and normal distribution, hence offering improved handling of relational data covariance issues. Additionally, Logistic Regression Quadratic Assignment Procedure (LRQAP) is employed to further scrutinize MRQAP model outcomes, enhancing result robustness (Golra et al., 2023).

This study synthesizes existing literature and empirical characteristics of SO2 emissions in industrial production (Costantini et al., 2013; Wang et al., 2024a; Zhou et al., 2021) to identify key factors driving SO2 emissions and their inter-regional transfers. Based on this, it constructs the following model:

$${{ST}}={{F}}({{SP}},{{PGDP}},{{IL}},{{EG}},{{ECS}},{{PS}},{{IS}},{{EP}},{{RE}}),$$

where ST denotes the SO2 transfer network. SP represents spatial proximity, captured by an interprovincial weight matrix in which adjacent provinces are assigned a value of 1 and all others 0. PGDP indicates the level of economic development, measured by GDP per capita. IL refers to innovation level, proxied by the intensity of research and development (R&D) investment. EG captures environmental governance, expressed as the ratio of investment in pollution control to GDP. ECS reflects the energy consumption structure, represented by the proportion of coal in total energy use. PS denotes public satisfaction, based on survey evaluations of environmental quality. IS stands for industrial structure, measured by the share of the tertiary industry in GDP. EP corresponds to environmental protection investment, assessed by the amount of funding allocated annually. Lastly, RE reflects regional resource endowment, defined by the basic reserves of key non-ferrous and non-metallic minerals. These factors were standardized using the extreme difference method, and an inter-regional difference matrix was constructed. In this study, MRQAP routines were executed with 10,000 permutations.

Data sources

The SO2 emission inventories utilized in this paper were sourced from the China Industrial Environmental Database (CIED) (Qian et al., 2022) and the China Environmental Statistics Yearbook. The CIED uses cross-entropy to effectively balance top-down statistical data with bottom-up micro-environmental data from enterprises, yielding SO2 emission information for different sub-sectors in different provinces, and demonstrating a good fit with publicly available sector-level emission observation data. The MRIO tables are obtained from the China Emission Accounts & Datasets (http://www.ceads.net/), which publishes inter-regional input–output tables for 31 provinces in mainland China (Li et al., 2016; Zheng et al., 2021). Since the MRIO tables are published every five years, with the most recent year being 2017, this study only covers the period from 2007 to 2017. Socio-economic data are mainly obtained from the China Statistical Yearbook and the China Energy Statistical Yearbook. Public satisfaction data are taken from the annual report on the evaluation results of ecological civilization published by the National Bureau of Statistics.

Results and discussion

China’s interprovincial SO2 transfer network

Figure 1a–c illustrates the embodied SO2 flows among regions in China driven by final demand. Node information within the network denotes the total net SO2 transfer of each province, while lines represent differences in emission transfers between provinces. As depicted, total SO2 emissions from China’s industrial sector steadily declined from 2007 to 2017, with the highest magnitude of interprovincial transfers decreasing from 56,431 tons to 11,006 tons. Over the decade, provinces bearing net consumer responsibility for emissions (CBE > 0) initially dwindled in number before rebounding to their initial levels. In terms of distribution, such provinces were predominantly situated in eastern coastal regions in 2007 and 2012, later spreading to western provinces like Yunnan, Sichuan, and Shaanxi, likely correlated with the economic upsurge of certain western provinces. By 2017, provinces responsible for production emissions (CBE < 0) became concentrated in northern and central provinces, primarily serving as crucial energy producers, notably in electric heat production. In contrast to previous findings on implied gas transfers (Wang and Hu, 2020; Guo and Rong, 2021), which suggest that regions such as Beijing–Tianjin, the eastern coast, and the southern coast consistently experienced net embodied carbon inflows through trade, this study identifies a shift in patterns. Several coastal provinces, or those adjacent to them, have gradually transitioned from being net consumers to net producers of emissions. Notable examples include Fujian, Jiangxi, Anhui, and Shandong. Additionally, Tianjin, Jiangsu, and Zhejiang provinces witnessed a decline in consumer emissions. Figure 1d also displays cumulative net SO2 transfers by province over the decade, highlighting Guangdong in South China, Jiangsu, Zhejiang, and Shanghai in East China, Beijing and Tianjin in North China, and Jilin in Northeast China as top-tier provinces bearing consumer emission responsibility. Meanwhile, provinces like Henan, Shanxi, Hebei, Inner Mongolia, Liaoning, and Guizhou belong to the first tier of producer responsibility. Notably, these two types of provinces are closely located, warranting further discussion on the relationship between distance and emissions in subsequent sections.

Fig. 1: Interprovincial embodied SO2 net transfer networks in China.
figure 1

a–c Embodied net transfer of SO2 between provinces in China in 2007, 2012, and 2017, respectively. CBE stands for consumption-based emissions. A CBE greater than 0 indicates a net inflow, while a CBE less than 0 indicates a net outflow. d Cumulative net transfer of SO2 between provinces over the decade.

Full size image

This paper employs the natural breakpoint method to categorize interprovincial SO2 emission transfers. Notably, transfer relationships characterized by high levels in 2007 predominantly occurred in pairs, such as Guizhou-Guangdong, Henan-Zhejiang, Shanxi-Shandong, and Inner Mongolia-Jilin. However, starting from 2012, transfer relationships have evolved into triangular loops, primarily manifested as multi-party connections between coastal provinces and their hinterland counterparts (Song et al., 2022). Nevertheless, it is imperative to conduct further analysis of the SO2 network to discern relationships between key emission reduction regions and pivotal sectors influencing pollutant gas emissions nationwide.

Evolution characteristics of SO2 transfer network

Through-flow analysis

Figure 2 illustrates the strength of individual provinces’ control over the overall network flow. Overall, the provincial control curves exhibit relatively consistent patterns over the decade, reflecting a persistent path dependence in regional energy consumption and industrial organization. Each macro-region maintained one or more dominant provinces in terms of control intensity, such as Hebei in North China, Liaoning in Northeast China, Jiangsu and Shandong in East China, Guangdong in South China, Henan in Central China, Sichuan in Southwest China, and Shaanxi in Northwest China.

Fig. 2
figure 2

Evolution of SO2 network control intensity by province in China.

Full size image

In 2007, 17 provinces had a network control intensity above 3%, with 7 exceeding 4%, including Shandong (7.95%), Guangdong (6.53%), Zhejiang (5.62%), Jiangsu (5.42%), Sichuan (5.26%), Henan (5.07%), and Hebei (4.98%). By 2012, these numbers slightly declined to 16 and 6, respectively, with Guangdong (7.30%), Shandong (6.90%), Jiangsu (5.17%), Henan (5.10%), Hebei (4.51%), and Liaoning (4.32%) remaining above the 4% threshold. In 2017, the number of provinces above 3% remained at 16, but those above 4% increased to 10, including new entrants such as Yunnan (5.90%), Hunan (4.61%), and Xinjiang (4.13%), alongside Guangdong (6.49%), Jiangsu (6.08%), Shandong (5.88%), Henan (5.69%), Hebei (5.41%), Zhejiang (5.04%), and Sichuan (4.72%).

This shift suggests a gradual diffusion of control intensity from traditional industrial hubs toward inland regions, driven by the spatial reallocation of high-emission industries and evolving interprovincial trade relationships. While Beijing, Shanghai, and Jilin remain prominent in terms of net transfer volumes, their overall network control has weakened over time. In contrast, provinces in the southwest and northwest—such as Yunnan, Guizhou, and Xinjiang—have steadily enhanced their control capacity within the emissions network, likely reflecting broader trends in industrial restructuring and gradient-driven relocation of polluting sectors.

Under the framework of the Western Development Strategy, the role of Western provinces in the national emissions landscape has changed significantly. In Yunnan, the enhanced network control is primarily attributed to rising final demand associated with rapid urbanization and infrastructure expansion, while the bulk of high-emission production remains concentrated in other provinces. Conversely, Guizhou and Xinjiang have emerged as net exporters of SO2 emissions. As major national suppliers of energy and raw materials, these provinces host emission-intensive industries such as coal-fired power generation and non-ferrous metal smelting, with a substantial portion of emissions resulting from production activities that serve external demand.

Network centrality analysis

A comparative analysis of the control intensity curves depicted in Fig. 2 and the centrality curves shown in Fig. 3 reveals a misalignment between the relative and absolute positions of each province within the network. While control intensity reflects a province’s absolute influence on the network, centrality indicators capture structural characteristics such as interconnectivity and intermediary roles. A higher relative position in centrality terms suggests more complex linkages within the network, potentially posing greater challenges for emission governance due to increased exposure to interprovincial flows.

Fig. 3
figure 3

SO2 net transfer network centrality curves.

Full size image

Among provinces with control intensity exceeding 4%, Zhejiang, Guangdong, and Jiangsu consistently exhibit high degree centrality, reflecting both substantial emission volumes and dense connectivity within the network. In contrast, Hebei shows a gradual decline, suggesting a weakening in its transfer linkages, likely due to the ongoing compression of steel production capacity under national environmental and industrial restructuring policies. Henan, on the other hand, shows a steady upward trend, indicating its growing involvement in regional emission exchanges.

Shanxi and Inner Mongolia consistently maintain high proximity centrality values, underscoring their strong direct emission linkages with other provinces. This pattern reflects their status as major energy-producing regions with significant upstream emissions. Inner Mongolia, in particular, plays a strategic role as a national coal base, thermal power hub, and a key region in west-to-east energy transmission, including coal, electricity, and natural gas.

The sharp increase in closeness centrality observed around 2012 coincides with a broader period of economic adjustment in China (Wang et al., 2019). During this time, energy-intensive sectors such as steel, building materials, photovoltaic solar energy, and wind power experienced contraction and restructuring. As a result, many provinces reduced the number of trading partners and shifted toward more stable and centralized supplier–consumer relationships. This transition is reflected in the network structure as a drop in degree centrality and a simultaneous rise in closeness centrality, indicating a tightening of transfer pathways and greater reliance on fewer nodes.

The betweenness centrality curves further illustrate how provinces’ intermediary roles in SO₂ transfer have evolved. In particular, consumer provinces such as Guangdong, Jiangsu, Zhejiang, Shanghai, Beijing, and Jilin exhibit a declining trend in betweenness centrality. This suggests a reduction in their role as critical bridges within the emission transfer network. For these provinces, the simplification of interprovincial emission relationships may lower the complexity of policy coordination and reduce their burden in facilitating cross-regional emission reductions.

Network control analysis (regional-regional, regional-sectoral)

Figure 4 presents the average control allocation (CA) matrix by region from 2007 to 2017. Higher values indicate a greater intensity of control between regions, where consumption in the controlling region results in a heavier burden of production emissions in the controlled region. Overall, most regions are subject to control by other regions in multiple ways, except for Beijing and Tibet. In contrast, provinces like Shanxi and Inner Mongolia in northern China, Guizhou in southwestern China, and Gansu, Qinghai, Ningxia, and Xinjiang in northwestern China exhibit little or no control over other regions. These provinces are generally of the producer responsibility for emissions type. Developed provinces like Beijing, Shanghai, Zhejiang, and Guangdong exert significant control over individual regions. Notably, these provinces not only control emissions from developing provinces but also from some developed provinces. Specifically, Beijing controls 100% of Tibet’s SO2 emissions, 92% of Shanghai’s, 40% of Guangdong’s, 38% of Tianjin’s, 28% of Hainan’s, 27% of Zhejiang’s, and 24% of Fujian’s. Shanghai controls 52% of Tianjin’s, 22% of Guangdong’s, 21% of Fujian’s, and 21% of Zhejiang’s SO2 emissions. Zhejiang controls 26% of Guangdong’s, 23% of Jiangsu’s, 22% of Yunnan’s, 20% of Tianjin’s, and 20% of Shanxi’s SO2 emissions. Conversely, Guangdong controls 48% of Hainan’s, 36% of Yunnan’s, 33% of Guangxi’s, 30% of Chongqing’s, 28% of Hunan’s, 25% of Fujian’s, 22% of Sichuan’s, and 20% of Shaanxi’s SO2 emissions.

Fig. 4
figure 4

Region-region control average matrix.

Full size image

As depicted in Fig. 5A, different regions exert control over SO2 emissions from various industrial sectors at the national level. Overall, each industrial sector is primarily controlled by several regions and exhibits certain regional characteristics. Hainan and Tibet show lower control intensity in the entire network and have minimal control over any industrial sectors. Figure 5B illustrates that the industrial sectors responsible for SO2 emissions largely overlap across provinces, mainly due to the similar product consumption structures in each province. Over time, implied gas emissions have increasingly concentrated in a few key industrial sectors (Ma et al., 2019; Wang et al., 2021). Electricity and heat production (sector 22), metal processing (sector 13), non-metallic mineral products (sector 13), chemical products (sector 11), and processed petroleum products (sector 10) are sectors with high SO2 emissions. For instance, Shandong, Guangdong, Jiangsu, Henan, Liaoning, Hebei, and Inner Mongolia control 8.23%, 7.04%, 5.34%, 5.24%, 5.17%, 4.49%, and 4.11% of the SO2 emissions from the national electric heat generation sector, respectively. Similarly, Guangdong, Henan, Hebei, Jiangsu, Sichuan, Zhejiang, and Shanxi control 6.47%, 5.97%, 5.90%, 5.65%, 5.18%, 4.98%, and 4.90% of the national SO2 emissions from the metalworking sector, respectively. This pattern repeats across various sectors, with representative provinces in each major region dominating control. Provinces with stronger overall network control also exhibit higher levels of control over industrial sectors. Guangdong, Shandong, Yunnan, Jiangxi, Hebei, Sichuan, and Chongqing control 7.29%, 5.87%, 5.37%, 5.12%, 4.98%, and 4.60% of the country’s SO2 emissions from the non-metallic mineral sector, respectively. In the chemical products sector, Guangdong, Henan, Sichuan, Shandong, Yunnan, Jiangsu, and Hubei control 7.14%, 5.71%, 5.44%, 5.38%, 5.31%, 5.05%, and 4.90% of the national SO2 emissions, respectively. Moreover, in the petroleum processing sector, Guangdong, Shanghai, Shandong, Henan, Yunnan, Hunan, and Jiangsu control 8.00%, 6.76%, 5.39%, 5.15%, 5.13%, 4.91%, and 4.83% of the national SO2 emissions, respectively. Just like the overall control of the network, specific provinces in each major region also have significant influence over these sectors. Provinces with stronger overall grid control also exert a higher amount and intensity of control over the industrial sector.

Fig. 5: Regional-sectoral control and dependency average matrix.
figure 5

A Average control matrix for regions and sectors. The darker the color in each row, the higher the degree to which emissions from that sector are controlled by that region. B Average dependency matrix for regions and sectors. The darker the color in each column, the higher the degree of influence of emissions from that sector on that region.

Full size image

Driving forces of SO2 transfer network

MRQAP regression analysis was conducted in accordance with the established protocol, the results of which are displayed in Table 2. Throughout the period 2007–2017, the significance level of the model was set at 1%. The selected factors explained more than 50% of the variation in the embodied SO2 transfer network. Comparing the explanatory power of the selected variables in numerous studies for the spatial correlation network, it is evident that the fitting effect is superior (Golra, et al., 2023; Ma et al., 2023; Zheng et al., 2020).

Table 2 MRQAP regression results.
Full size table

The regression results reveal that the coefficient of spatial proximity is significantly positive in 2007 and 2017, suggesting that the transfer of embodied SO2 is more probable between neighboring areas. However, the influence of spatial proximity on SO2 transfer diminished over time. While differences in the level of economic development were insignificant in 2007 and 2012, they turned significantly negative in 2017. It is noteworthy that 2017 marked a period of deepening reform and high-quality development since 2013, during which the market mechanism gradually took effect. When economic development levels are similar, regions tend to have comparable land use patterns and similar needs in industrial production technology and labor (Roost and Jeckel, 2021). Consequently, factors influencing the transfer of SO2 are more likely to flow between geographic regions with similar levels of economic development.

From 2007 to 2017, regional differences in environmental governance exhibited a significantly negative correlation with embodied SO2 transfer. This suggests that greater disparities in environmental governance between regions correspond to less favorable conditions for embodied SO2 transfer. Conventionally, it’s believed that when there are substantial differences in environmental regulations, industrial enterprises in regions with less stringent regulations tend to produce more products for regions with stricter regulations, leading to embodied SO2 transfer. However, it’s worth noting that some provinces with less stringent environmental regulations also account for a higher volume of emissions in terms of consumption. These provinces are less inclined to manufacture highly polluting products and thus may have fewer environmental regulations in place. Such mismatches in identity can result in outcomes contrary to the intended transfer of SO2, influenced by cognitive mechanisms.

From 2007 to 2017, regional differences in innovation levels exhibited increasing significance and coefficients, with significant positive performance in 2012 and 2017. The bigger the difference in technical innovation levels between locations, the more favorable the transmission of SO2, and this influence is increasing annually. Provinces with greater responsibility for measuring consumption-related emissions tend to possess stronger scientific and technological innovation capabilities. Moreover, a notable inverse relationship was observed between the difference in energy consumption structure over the ten-year period and its degree of influence, which gradually increased. This suggests that regions with a smaller difference in energy consumption structure have a more favorable embodied SO2 transfer. This article analyzes the energy consumption pattern by assessing the proportion of coal usage in relation to total energy consumption. When the difference in energy consumption structure is smaller, the energy factor demand and industrial chain of the two regions are closer, thereby increasing the likelihood of promoting the embodied transfer of SO2.

In 2017, the resource endowment differential had a notably unfavorable influence on the embodied transfer of SO2, which may be comparable to the impact of the difference in energy consumption structure. The discrepancies in resource endowment between 2007, 2012, and 2017 are not consistent due to the lag in statistics and release of major non-ferrous and non-metallic minerals. The model does not show a substantial difference in industrial structure due to the lack of obvious distinction in industrial structure values among provinces with varying emission obligations. The results suggest that bottom-up feedback is not effective in influencing the embodied transfer of SO2 in trade, as there are no significant differences in public satisfaction and ecological protection.

Regarding impact pathways, the most significant drivers of SO2 transfer networks are the level of innovation, the structure of energy consumption, and regional differences in environmental governance. The level of innovation reflects the actual level of local consumption and production structures. The production of high-pollutant gas-emitting products in low-innovation regions serves high-innovation regions, resulting in the transfer of embodied SO2. The regression results indicate that the difference in industrial structure is not an effective differentiator between production-measured and consumption-measured emissions in the two types of provinces. In the process of trading goods between regions, the structure of energy consumption and environmental regulations can impact the transfer of industrial emissions. Provinces with similar energy consumption structures and environmental regulations are more likely to cooperate in trade and contribute to the transfer of SO2 due to shared energy demands and industrial chains.

Uncertainty and limitations

Although this study systematically investigates the interprovincial transfer of embodied SO₂ emissions and its driving factors using the MRIO model and network analysis, several limitations remain. First, the model relies on existing input–output tables and environmental datasets, which are updated periodically and may not reflect real-time industrial or environmental changes. Future research could incorporate higher-frequency data sources, such as satellite remote sensing or firm-level emissions data, to improve the timeliness and accuracy of the analysis. Second, this study adopts a static network analysis approach, which limits the ability to capture temporal dynamics in the evolution of the transfer network. Future extensions may consider using dynamic network models or methods from complex network evolution theory to address this issue.

To assess the robustness of the results, LRQAP analysis was conducted following the approach of Pinheiro et al. (2016) and Borgatti et al. (2024). All models were re-estimated, and the LRQAP outcomes were compared to those of the MRQAP by examining the consistency in the direction and significance of the regression coefficients. The robustness of the findings is supported by the stable significance of innovation level, energy consumption structure, and environmental regulation across both models. However, LRQAP results (see Supplementary Tables 1–3) do not confirm the significance of geospatial proximity, which, together with the network pattern illustrated in Fig. 1, suggests that spatial proximity is not a major determinant of embodied SO₂ transfers. Additionally, although the coefficient for resource endowment shows a consistent sign, it does not reach the 90% confidence level. These findings indicate that while core drivers are robustly identified, certain explanatory variables may be sensitive to model specification, and future studies may benefit from incorporating alternative spatial or sectoral perspectives.

Conclusion and policy implications

Conclusion

This study investigates the transfer of embodied SO2 among Chinese provinces using an MRIO model and comprehensive network analysis. Constructing an MRQAP model based on the embodied SO2 network, the study uncovers the drivers of SO2 transfer, offering fresh insights for regional cooperation in emission reduction and equity. The primary findings are as follows:

  1. (1)

    Over the past decade, the total SO2 emissions from China’s industrial sector have gradually decreased. The provinces with production emissions are primarily concentrated in certain northern and central provinces, which are significant energy production areas, especially for electricity and heat. Some coastal provinces or regions close to the coast, such as Fujian, Jiangxi, Anhui, and Shandong, have also gradually transitioned from consumer to production emission provinces.

  2. (2)

    There is a certain degree of path dependence in the transfer of implicit emissions over the past decade, with representative provinces in each major region holding dominant control positions in the emission network, such as Hebei in North China, Liaoning in Northeast China, Jiangsu and Shandong in East China, Guangdong in South China, Henan in Central China, Sichuan in Southwest China, and Shaanxi in Northwest China.

  3. (3)

    Embodied SOâ‚‚ emissions have become increasingly concentrated in several key industrial sectors, including power and heat production, metal processing, non-metallic minerals, chemical products, and petroleum refining. Similar to the overall network control structure, each major region features representative provinces that dominate the control of these sectors.

  4. (4)

    Regional differences in innovation levels, energy consumption structures, and environmental regulations are the most influential factors in explaining SO2 transfers, although their pathways of influence vary.

Policy implications

This study provides a quantitative assessment of the embodied SO2 transfer between regions and elucidates the characteristics of SO2 transfer networks. By examining the status (both absolute and relative) of provincial nodes and network connections, it enables the identification of sectors and regions with the greatest potential for emission reductions, as well as the challenges or broader implications associated with achieving them. Policymakers can leverage this tool to pinpoint the most pertinent provinces and devise effective strategies for optimizing industrial layout and regional trade.

The research indicates that Guangdong in South China, Jiangsu, Zhejiang, and Shanghai in East China, along with Beijing and Tianjin in North China, and Jilin in Northeast China, are in the leading tier for consumer emission responsibilities. Meanwhile, provinces like Henan, Shanxi, Hebei, Inner Mongolia, Liaoning, and Guizhou rank in the top tier for producer emission responsibilities. The results of the centrality assessments show a downward trend in degree centrality and betweenness centrality, while those with higher closeness centrality emerge as effective intervention points for achieving system-wide emission reductions and should be prioritized in policy implementation.

Shanxi and Inner Mongolia continue to be significant sources of SO2 emissions in China, particularly in non-power sectors such as coal-based chemicals, metallurgy, and non-ferrous metal processing. While both provinces have largely completed ultra-low emission retrofits in the power sector, the next stage should focus on deep governance of non-electricity industries by strengthening process-level pollution control, optimizing industrial structure, and enhancing regional coordination mechanisms. In particular, steel, cement, and non-ferrous metal industries should integrate desulfurization with broader process optimization, adopt cleaner production technologies, and advance coordinated control of multiple pollutants along the entire supply chain. As China’s air pollution governance enters a new phase that emphasizes the coordinated control of PM2.5 and O3, strengthening regional cooperation on SO₂ mitigation can advance both single- and multipollutant control objectives, thereby contributing to broader improvements in air quality.

When designing joint regional air pollution control strategies, special attention should be given to provinces where SO₂ production responsibility overlaps with severe PM2.5 and O3 pollution. Provinces such as Shanxi, Hebei, and Henan not only emit large quantities of SO2 but are also located in the Fenwei Plain and the Beijing–Tianjin–Hebei region—areas frequently affected by complex air pollution. These provinces should be prioritized in multipollutant governance efforts, with strengthened integration of desulfurization, NOX and VOCs control, and clean energy substitution, alongside the establishment of more systematic regional coordination mechanisms to achieve greater marginal benefits in air quality improvement.

From the perspectives of supply chain and final consumption, it is possible to identify emissions from specific regions and sectors that are controlled by other areas, thereby guiding regional cooperative reduction policies. Provinces that dominate emissions in their regions should strengthen collaborative monitoring, prevention, and response mechanisms for regional air pollutants and improve the atmospheric environmental management system. Economically developed provinces like Beijing, Shanghai, Zhejiang, and Guangdong contribute significantly to production emission burdens across multiple provinces, necessitating further promotion of clean energy use and advocacy for green lifestyles. Ecological compensation initiatives should incorporate implicit environmental impact assessments for pollution cost transfers (such as emission equivalents, costs, and taxes) alongside comparative analyses of trade value transfers.

At the sectoral level, key control provinces in coal-fired power, metal processing, non-metallic minerals, chemical production, and petroleum refining should jointly develop full-chain SO2 reduction partnerships. These efforts should focus on technological integration, process-level pollution control, and coordination along green supply chains. Based on the identified drivers of SO2 transfer, local governments should further strengthen environmental regulations and green innovation capacity, increase the share of clean energy in electricity, transportation, and residential sectors, and improve their position in the national emission transfer network, thereby contributing to a fair, efficient, and collaborative regional emission reduction framework.