According to new economic growth theory, the regional spillover effect is used to describe the influence of the economic activities in a central region on the economy of other neighboring regions, including the polarization effect and diffusion effect (Huang, 2014). In the field of environmental science, the spatial spillover effect of environmental pollution refers to the impact of local environmental pollution on the ecological environment of adjacent areas, which is essentially a type of mutual conduction relationship of environmental pollution in space (Liu et al., 2015). Spatial spillover is another manifestation of a factor externality, and it is an indirect effect. An externality is an external influence of one economic subject on another. Pigou further developed externalities into the three forms of no externality, negative externality and positive externality on the basis of Marshall’s external economy theory (Demsetz, 2004). When there is a positive externality, because of the existence of a spillover effect, the local income will not be fully owned by the locals, which results in external benefits, such as the improvement of environmental quality; when there is a negative externality, because of damage to the local economy or environment, the adjacent areas must increase the external expenditures to maintain the original development foundation. Industrial development may produce positive or negative externalities to local and adjacent environmental pollution through the effect of scale economies, the structure optimization effect and the technology spillover effect (Ji et al., 2018).

The YRD is one of the most developed urban agglomerations in China. It is located in the lower reaches of the Yangtze River and borders the Yellow Sea and the East China Sea. The terrain is high on the periphery and low in the middle. At the end of 2019, the CPC Central Committee and the State Council issued The Outline of the Yangtze River Delta Regional Integration Development Plan, which designated 41 cities in the YRD, including Shanghai city, provinces of Jiangsu, Zhejiang and Anhui, with a total land area of 358,000 km². As shown in Figure 1, the following 27 cities represent the central areas, with a total land area of 225,000 km²: Shanghai city; Nanjing, Wuxi, Changzhou, Suzhou, Nantong, Yangzhou, Zhenjiang, Yancheng and Taizhou in Jiangsu Province; Hangzhou, Ningbo, Wenzhou, Huzhou, Jiaxing, Shaoxing, Jinhua, Zhoushan and Taizhou in Zhejiang Province; and Hefei, Wuhu, Maanshan, Tongling, Anqing, Chuzhou, Chizhou and Xuancheng in Anhui Province. All of these central cities have rapidly driven the high-quality development of the YRD. As one of the regions with the most active economic development, the highest degree of openness and the strongest innovation ability in China, the YRD plays an important strategic role in China’s overall modernization and opening pattern. The population agglomeration and urbanization process in the YRD has been rapid. By 2017, the total population had reached 223 million, and the population urbanization rate had reached 69.49%. Moreover, the GDP of the entire YRD is 19.53 trillion yuan, with the proportion of primary industry, secondary industry and tertiary industry being 4.2:41.8:54.0, which accounts for approximately 25% of China’s total economy.

With the rapid development of regional integration, environmental losses and ecological damage have become important obstacles for the YRD to achieve the goal of building a world-class urban agglomeration by 2030. In 2017, the total amounts of wastewater discharge, waste gas discharge, industrial solid waste production and domestic waste production in the YRD were 1474.92 million tons, 3.75 million tons, 10.16 million tons and 45.59 million tons, respectively; the land area of the YRD is less than 4% of the entire country, but the proportions of these pollutants are 21.08%, 12.79%, 9.2% and 20.13%, respectively, representing a heavy environmental pollution emission burden per unit area. Due to the spatial proximity and close relationship between cities in the YRD, there is a serious mutual conduction risk of industrial air pollution between cities, which can cause more serious regional air pollution problems. For a long time in the future, water, air and solid waste pollution will be the focus of environmental governance in the YRD.

Against the background of high-quality environmental development, the industrial transformation path of the YRD has gradually shifted from scale reduction and layout optimization to the coexistence of production scale balance, process innovation and technology upgrading, which embodies the changing characteristics of industrial transformation and pollution management in China’s urban agglomerations. Based on the contradiction between industrial development and environmental pollution, the YRD is selected as a case area to analyze the spatiotemporal effect and influencing factors of air pollution from 2008 to 2018 and thus to provide a scientific basis for the realization of regional environmental governance in the integrated development of the YRD.

Environmental pollution and industrial development data were collected from The China Urban Statistical Yearbook (2009-2019), The China Environmental Statistical Yearbook (2009-2019), The City Statistical Yearbook (2009-2019), The Environmental Statistical Bulletin (2009-2019), etc. The industrial patent data of Shanghai, Nanjing, Hangzhou and Ningbo, such as patent applications, patent grants and patents in force, were selected from The Annual Report of Intellectual Property Statistics (2009-2019) provided by the China National Intellectual Property Administration (https://www.cnipa.gov.cn/col/col61/index.html), and the patent authorization data of most other cities were selected from the statistical yearbooks of each city. Then, we combined and sorted the relevant data to obtain the number of valid patents applied for and approved by each city from 2008 to 2018. In addition, to eliminate the impact of price changes and avoid heteroscedasticity and collinearity in the model, we used the GDP deflator index based on 2007 to approximately convert the PGDP and added values of the primary, secondary and tertiary industries of each city over years and then treated all index values logarithmically. Table 1 shows the descriptive statistical results of all experimental data.

The panel data used in this study, with a time series (T) of 11 and a panel series (N) of 41, are short panel data, and an HT unit root test is required. All variables reject the original assumption of non-stationarity after the first-order difference processing, which indicates that there is no unit root variable (Table 2). Then, the Kao, Pedroni and Westerlund methods were used for cointegration tests. Each statistic is significant at the 1% level, which indicates that there is a long-term stable equilibrium relationship between the variables, and they can be used for a regression analysis. Afterwards, by observing the numerical condition of the AIC, BIC and HQIC, the optimal lag order is determined to be the second order. Through the Ganger test (Table 3), it is found that each explanatory variable is the Granger cause of the explained variable ISDE, that is, there is a causal relationship between them.

To assess the spatial clustering and distribution of the different variables more clearly, all variables were visualized on a map (Figure 2). The distribution of indicators in the YRD shows an obvious spatial imbalance, and it forms two heterogeneous regions in the northeast and southwest. Of these areas, the environmental pollution in the cities in the northeast is relatively heavy, and the level of economic and social development, such as industrial development and technological innovation, is higher, while the southwestern cities show the opposite.

Global Moran’s I of the ISDE in the YRD from 2008 to 2018 is calculated according to equation (1), as listed in Table 4. Based on the geographical inverse distance-squared matrix, the ISDE from 2008 to 2018 rejects the original hypothesis of no spatial autocorrelation at the different significance levels, and the correlation coefficient is significantly positive each year, which indicates that there is a significant positive spatial correlation of industrial air pollution in the YRD. Thus, cities with more serious air pollution tend to be adjacent in space, and cities with lower emissions are also adjacent in space. From the interannual variation in the global Moran’s I, although it has a certain volatility, the overall value shows a weakening trend, which suggests that the spatial autocorrelation of ISDE among different cities in the YRD is decreasing, and the spatial difference of regional air pollution is expanding.

Local autocorrelation tests of ISDE in 2008, 2013 and 2018 are carried out according to Equation (2). The results show that the local spatial distribution of industrial air pollution is mainly represented by HH and LL agglomerations. From the perspective of specific city distribution patterns (Figure 3), HH clusters are mainly concentrated in Shanghai, southern Jiangsu and northern Zhejiang, which have a high level of economic development and a large industrial scale. The industrial clusters are dominated by industries with high energy consumption and high pollution, such as equipment manufacturing, pharmaceutical chemistry, papermaking and textiles. LL clusters are located in the southwestern cities of Anhui Province, which focuses on the development of the cultural and tourism industry, and the industrial economy is not prominent. For example, Huangshan city, as an important birthplace of Hui culture, has actively developed its tertiary industry through tourism as the main industry. The contribution rate of manufacturing industry to the city’s economy is only 29.2%, and the pollution emissions caused by industrial production are relatively low.

The maximum likelihood is used for the regression analysis in the software platform stata15. According to the results of the Hausman test, the fitting effect of the fixed-effect model is better than that of the random-effect model. The results of the Wald and likelihood ratio tests both significantly reject the original assumption that the SDM can be simplified as an SLM or SEM, i.e., the SDM attains the best fitting effects. The estimation results of these three models are summarized in Table 5. Based on the geographic inverse distance-squared matrix, the regression results of the SDM show that the spatial autoregressive coefficient $\rho$ is significantly positive at the 1% level, and every 1% increase in ISDE locally will lead to a 0.603% average increase in the ISDE in adjacent cities. Combined with the spatial autocorrelation analysis in the previous section, this indicates that there are spatial agglomeration and positive spillover effects of industrial air pollution emissions in the YRD. Cities with close geographical distances interact with one another, and industrial air pollutants form and diffuse among different cities in urban agglomerations.

Furthermore, an increase in the ISCI by 1% in the YRD will lead to a 0.677% average increase in ISDE. Because the local and adjacent cities in the YRD are usually in the same economic division system, with similar industrial structures, production technologies and capital markets, the action mechanism on the ecological environment is also similar. For example, the manufacturing industry is highly concentrated in the central area of the YRD, and it can trigger industrial production activities with high energy consumption through the agglomeration effect, which directly increases regional air pollutant emissions.

In contrast, every 1% increase in the ISTI will reduce ISDE by 0.368%. Industrial structure adjustment drives the improvement of local production efficiency and the technological innovation level. In the YRD, industrial structure adjustment not only means a simple transfer from the secondary industry to the tertiary industry but also extends the low-end equipment manufacturing industry to the high-end industrial chain, such as biomedicine and advanced equipment manufacturing, which greatly reduces production energy consumption and industrial pollution.

For every 1% increase in PGDP, ISDE will increase by 0.045%. This shows that the YRD is still in the stage of a gradual reduction of environmental quality with economic growth, and the economic growth mode at the cost of environmental pollution has not completely changed. Technical and policy variables such as EIT, FDI and ER have negative effects on the ISDE in the YRD, and the correlation coefficients are -0.148, -0.170 and -0.027, respectively. Through the improvement of production technology, the development of the environmental access threshold and the enhancement of the ER intensity, innovative elements such as talent, capital and information have a positive impact on the cleaner production of urban enterprises through the industrial chain.

According to the WX coefficient in Table 5, the explanatory variables all show significant exogenous interaction effects. To explain the spatial autoregressive coefficient of the SDM more specifically, the spatial effects of industrial transformation on air pollution can be decomposed according to equation (8), and the results are summarized in Table 6.

The direct effect represents the influence of the explanatory variables on the explained variables in the local city. The contribution order of different explanatory variables is ISCI> ISTI#>FDI#>ETI#>ER#>PGDP, where $#$ represents a negative effect, and the ISCI, ISTI, ETI and FDI are significant at the different confidence levels. The indirect effect, namely, the spatial spillover effect, represents the average influence of the explanatory variables on the explained variables in adjacent cities. The contribution ranking is ISCI>PGDP#>FDI> ETI>ISTI>ER#, and only the ISTI and ER fail to pass the significance test. Among them, the ISCI, ISTI, ETI and FDI exhibit positive environmental spillover effects, while PGDP and ER have negative environmental spillover effects. To further reflect the spatial interaction of different explanatory variables on the air pollution of local and adjacent cities, the direct effect and spatial spillover effect are taken as the X-axis and Y-axis, respectively, and four quadrants are divided using Origin9.3 software (Figure 4). The four impact patterns are de-fined as follows: ① I-I: a simultaneous increase in the ISDE in local and adjacent cities; ② R-I: a reduction in the ISDE in the local city but an increase in the ISDE in adjacent cities; ③ I-R: an increase in the ISDE in the local city but a reduction in the ISDE in adjacent cities; and ④ R-R: a simultaneous reduction in the ISDE in local and adjacent cities.

The ISCI is located in the I-I quadrant. Combined with the decomposition coefficient value in Table 6, the direct and indirect effects of the ISCI are 0.745 and 1.531, respectively, which are significant at the 1% and 10% levels. This result means that each 1% increase in the local ISCI will lead to a 0.745% and 1.531% increase in the ISDE in local and adjacent cities, and the spatial spillover effect of the ISCI is obvious. The industrial structure of the YRD is still dominated by secondary industry, and the proportion of polluting industries, such as chemical materials, nonmetallic minerals and metal smelting, remains high. The expansion of the production scale of these high-polluting industries will inevitably lead to the aggravation of local air pollution. Moreover, the distribution of production factors in the YRD often has the characteristics of spatial agglomeration. In Suzhou, Wuxi and Changzhou, a large number of industrial parks have biomedicine and equipment manufacturing as the main industries, and the industrial chain runs through these three cities. The cluster development mode between cities makes the scale effect of industrial agglomeration more prominent, which leads to the spatial spillover effect of promoting emissions of air pollutants from adjacent cities.

The ISTI is located in the R-I quadrant, which indicates that every 1% increase in the ISTI reduces the local ISDE by 0.367% but increases the ISDE of adjacent cities by 0.113%. In the YRD, especially in core cities such as Shanghai, Suzhou and Hangzhou, the purpose of industrial structure adjustment is more to realize the balance between the manufacturing and service industries and complete industrial upgrading within the manufacturing industry. Under such targeted control, local cities introduce more industries at the high end of the industrial chain and control pollutant emissions from the source through technical effects. Moreover, some high-energy consumption and high-pollution industries in core cities will be transferred to adjacent cities, and the scale of these eliminated industries will be forced to grow continuously, which will directly aggravate pollutant emissions.

In terms of the spatial spillover effect of the other control variables, PGDP is in the I-R quadrant, and for every 1% increase in GPDP, the ISDE in adjacent cities will decrease by 1.306%, which implies that urban economic agglomeration has a positive externality and drives improvements in the green production efficiency of surrounding cities. Both ETI and FDI are located in the R-I quadrant, and the direct effect is significantly negative, while the spatial spillover effect is significantly positive. Thus, ETI and FDI have a reduction effect on the ISDE of the local city but aggravate the ISDE of adjacent cities to varying degrees. The distribution of innovation factors in the YRD is not balanced. After the main body of innovation flows into the central city, it will inevitably lead to the reduction of innovation factors in adjacent cities, which will result in differing air pollution between the surrounding cities. The spatial spillover effect of FDI is related to the quantity and quality of foreign investment. At present, the negative externality of scale spillover is still stronger than the positive externality of technology spillover in the YRD. Therefore, local cities have changed their mode of production and industrial structure by attracting foreign capital, but traditional industries are transferred, which aggravates the air pollutant emissions of adjacent cities. ER is located in the R-R quadrant, and the coefficients of the direct and indirect effects are -0.022 and -0.019, respectively. Under the policy constraints of high-quality environmental development, the high standard ER in the YRD will stimulate enterprises to promote technological innovation to offset the cost of environmental emission reduction to achieve a win-win situation between environmental protection and economic development. Moreover, ER has a certain radiation range (Liu et al., 2021). There is ER competition in adjacent cities. Adjacent cities will follow suit and strengthen their regulation intensity, resulting in the simultaneous reduction of industrial emissions.

In this study, 41 prefecture-level cities in the YRD in China were adopted as spatial units, and ESDA and SDM were used to test the spatial spillover effects of industrial transformation on industrial air pollution in urban agglomerations. The findings are as follows.

First, there is a significant spatial autocorrelation of air pollution in the YRD, and the agglomeration characteristics of the cities in the eastern YRD are obvious. According to the spatial regression coefficient, a 1% increase in ISDE locally will lead to a synchronous increase of 0.603% in the ISDE of adjacent cities, which suggests that ISDE has a significant positive spatial effect at the scale of urban agglomeration. This conclusion is similar to the research results of Liu et al. (2018) in the Bohai Rim region, where it was shown that PM_2.5 emissions have significant spatial agglomeration and diffusion, and there is a positive interaction between cities. Our study confirms that regardless of the industrial development stage or mode of urban agglomeration, pollutant diffusion from local to adjacent areas is objective. As air mobility is large, pollutant emissions easily spread between regions, which leads to cross-regional air pollution problems. Moreover, the distribution of economic and social production factors is spatially related. For example, companies, capital investment, etc. tend to have spatial convergence. The agglomeration of high energy-consuming and high-polluting companies is likely to cause high-emission air pollution problems, while the agglomeration of green production factors will reduce regional air pollutant emissions, both of which cause air pollutant emissions to have a significant spatial spillover effect.

Second, for every 1% increase in the ISCI and ISTI, the ISDE of the entire YRD will increase by 0.677% and decrease by 0.368%, respectively, with a significant spatial effect. Previous studies have shown that economic agglomeration has a different degree of spatial correlation with pollutant emissions, and it leads to environmental degradation at the watershed and provincial scales (Liu et al., 2017; Liu et al., 2014). High-quality industrial transformation is conducive to improving labor productivity, promoting stricter environmental regulation and reducing pollution (Hosoe et al., 2006; Wang et al., 2018). Our study shows that in urban agglomerations with convenient transportation and capital circulation, such as the YRD, the process of industrial economic agglomeration will also aggravate air pollution. The scale of the traditional manufacturing industry has not been significantly reduced in the current transformation stage; therefore, the problem of high pollution emissions has not been completely solved. However, under the guidance of high-quality environmental development policies, the industrial structure of the YRD is gradually changing to green and digital, and the low-energy production mode has restrained pollutant emissions to a certain extent.

Third, the air pollution effect of industrial transformation can be decomposed into the local direct effect and the spatial spillover effect. The contribution order of the direct effect is ISCI>ISTI>FDI>ETI>ER>PGDP, and the contribution order of the spatial spillover effect is ISCI>PGDP>FDI>ETI>ISTI>ER. All explanatory variables can be divided into four impact patterns, namely, I-I, R-I, I-R and R-R, where I means increasing emissions, and R means reducing emissions. The current study further decomposes industrial transformation into industrial agglomeration and structure adjustment. Through an SDM analysis, it is found that industrial transformation does not blindly promote pollutant emissions, and not all factors produce significant spatial spillover effects. We should focus on the core factors with strong significance, such as industrial agglomeration, to alleviate the process of environmental stress. The EKC hypothesis proposes that economic growth affects environmental quality in three ways, specifically, through scale, technology and structure effects (Copeland and Taylor, 2004). The traditional EKC takes the level of economic development as the independent variable to fit the curve (Zhou et al., 2019), but an increasing number of studies take the level of industrial development as the main variable in theoretical discussions (He et al., 2016). In this study, to explain the rationality of the EKC hypothesis from different aspects, the ISCI and PGDP are selected to measure the scale effect, and the ISTI is selected to measure the structure effect. In addition, ETI and FDI are selected to measure the technology effect. Through an empirical analysis of the YRD, it is verified that the scale effect worsens the environment, while the structure and technology effects improve the environment.

Finally, the spatial spillover effect of industrial transformation on air pollution in the YRD is also affected by foreign investment, innovation, ER and other factors, and these effects should not be completely ignored. Some scholars have studied the PHH in ASEAN, Europe and Latin America and have verified the validity of the PHH by analyzing the distribution law and externality of FDI (Mulatu et al., 2010; Baek, 2016; Sapkota et al., 2017). To explore whether the PHH is true in the YRD, this study adds the FDI index to the control variable. The results show that the spatial coefficient of FDI is -0.170, which is negative and significant at the 1% level; thus, FDI results in emission reductions, and the PHH does not hold in the YRD. The reason is that the YRD has a high intensity of ER, which consists of a strict environmental protection access system for foreign-invested enterprises and open information evaluation standards concerning the environmental behavior of all enterprises. Most of the foreign-invested enterprises choose to invest in industries at the back-end of the industrial chain with low pollution. This type of investment brings more technological progress and innovation upgrading, which to a certain extent inhibits the negative impact of FDI on the regional environment. In addition, by introducing the ER index, it can be found that Porter’s hypothesis is more suitable for the YRD, which is an area with high requirements of ER and a relatively successful industrial transformation. The innovation effect of ER has brought a win-win situation between economic growth and environmental protection.

Under the guidance of a high-quality environmental development strategy, the YRD should adhere to the green and sustainable economic growth mode, strengthen the positive roles of the technology and structure effects, and weaken the negative scale effect. The YRD should give attention to the importance of industrial agglomeration, industrial transfer and industrial transformation in the process of economic development and the spatial interaction between them and air pollution emissions to address the environmental problems caused by the traditional industrial transformation mode.

First, the YRD needs to optimize the industrial transformation path with a focus on structural emission reduction. The YRD should give attention to the flow direction of foreign investment and increase the investment proportion of high-end technology industries and other environmentally friendly industries by setting higher market access thresholds and other ER means. In addition, the YRD needs to eliminate traditional enterprises with high pollution and backward technology, speed up the development of modern service and strategic emerging industries, and actively play the positive externality of the scale and structure effects through industrial renewal.

Second, the YRD should achieve ultralow emission transformation with a focus on technology upgrading. The measures include improving the treatment of industrial waste gas from pollution-intensive industries, such as chemical materials, metal smelting byproducts and nonmetallic mineral products, and gradually implementing specific emission limits for air pollutants in the national emission standards for these industries. At the same time, the YRD should carry out advanced technologies and cleaner production concepts in key industries, such as steel and cement, increase environmental protection investment, and improve the positive externality of the technology effect.

Third, a regional joint prevention and control mechanism focused on spillover governance should be formed. With the YRD as the carrier, the negative spatial spillover effect in pollution control should be eliminated. Local and adjacent cities should reach a consensus on pollution emission standards and total emission reductions, actively promote policy coordination and resource sharing among cities, and establish deep-level emission reduction modes for the entire YRD, such as environmental access, pollution payments, and cross-border early warnings.

By introducing the factors of industrial agglomeration and structure adjustment, our study systematically discusses the relationship and impact mechanism between industrial transformation and environmental pollution in urban agglomerations from the perspective of spatial spillover. It is concluded that the EKC hypothesis can still reasonably explain the spatial interaction between industrial transformation and air pollution emissions under different transformation paths through scale, structure and technology effects. This empirical study of typical regions in the YRD also provides case evidence for controlling the spatial effect of environmental pollution by using the industrial development in developing countries. However, economic development and environmental quality vary greatly among different regions of China. The YRD is representative of the developed regions in China. The conclusions are related to the characteristics of this region. In the future, the SDM can be applied to other regions to verify the applicability of the model and better reveal the regional differences in the environmental spatial effects. In addition, environmental pollution mainly focuses on air pollution, which has strong mobility. Other common environmental pollutants, such as wastewater and solid waste, have not been studied in depth here and can be further discussed in the future. Furthermore, this study only divides the industry into primary, secondary and tertiary industries and does not subdivide the next level of industry sectors. The environmental spatial effects brought by the transformation of these various subindustries are also different, which is worthy of in-depth research.