Urbanization and industrialization have contributed to the rapid growth of the urban economy and the population and industry have also further agglomerated in cities. But it is undeniable that industrial pollution emission has become one of the primary air pollution sources. The 2016 report on the state of the environment in China shows that 254 of 338 cities do not meet the national ambient air quality standard, accounting for about 75% of the total, and the average number of days not up to the standard accounts for 21.2%. A city is not only an important space for economic agglomeration, but also the main place for human living. The important goal of the 13th Five-Year Plan is giving consideration to both urban environment and economic development and controlling the pollution problem ¹ . With the efforts of the central and local governments, pollution problem in China has been initially controlled and the industrial pollution emission intensity has shown an obvious downward trend as a whole, but there is still a big gap compared with the developed countries. Some policy documents at the national level have clearly pointed out that China’s urbanization in the future is to restrict the increasing agglomeration of the population in large cities and focus on the development of small- and medium-sized cities. Will such an urban development policy with the concept of space organization have a policy dividend on pollution reduction? What kind of spatial policy can achieve regional economic growth and green development consistently? Behind the vigorous process of urbanization in China, it is urgent to explore the rational urban spatial mode for avoiding air pollution caused by excessive emission of industrial pollution due to the agglomeration of economic activities.

In Figure 1, the spatial structure index ² has been distributed according to quartile and there is an obvious positive relationship between the quartile ranking of spatial structure index and pollution emission intensity in China’s 21 city-regions in 2003 and 2016. In Figure 1A, the bar chart of 2003 shows that the average pollution emission intensity of the five city-regions with the smallest spatial structure index (most inclined to polycentric spatial structure) is 147.61 (thousands ton/billion Yuan), and the average pollution emission intensity of the six city-regions with the largest spatial structure index (most inclined to monocentric spatial structure) is 296.33 (thousands ton/billion Yuan). The bar chart of 2016 shows a similar phenomenon: the city-regions with more obvious polycentric spatial structures have lower pollution emission intensity. The bar charts of the quartile ranking of spatial structure index and industrial sulfur dioxide emission intensity have also been plotted in Figure 1B and the results are similar.

Studies at home and abroad have proved that the polycentric spatial structure can promote the local economic efficiency because the cities with geographical proximity and close relations have “borrowed size” (Alonso, 1973). Building on the concept of “borrowed size”, some terms have been conceptualized for external economies at the scale of the regional urban system, such as urban network externalities (Capello, 2000), which can be used as an alternative to the agglomeration externalities in a single city. And under polycentric spatial structure, smaller cities that are part of a city-region have more endogenous ability to control these economic, social, and environmental costs within the acceptable range, while enjoying the agglomeration externality and overcoming the agglomeration diseconomy. Therefore, two questions are worth thinking about: regarding industrial pollution emissions as non-expected outputs, could polycentric spatial structure improve production efficiency while reducing industrial pollution emission intensity? What is the mechanism behind it? The existing research has not yet given theoretical analysis and empirical test. If the polycentric spatial structure has emission reduction effects on industrial pollution, the adoption of polycentric planning strategies will contribute to the realization of economic growth and air quality win-win. Behind the vigorous urbanization process, it is urgent to explore the urban development model and optimize the spatial organization form. It is of great practical and theoretical significance to analyze the urbanization mode of green development with spatial organization.

This study had the following contributions: 1) In previous studies, the impact of spatial structure on the environmental performance of city-regions has mainly played a role through factors such as traffic commuting, infrastructure, and domestic energy consumption; in this study, the relationship between spatial structure and industrial pollution has been investigated from the perspective of polycentricity and monocentricity, which provides the spatial economic activities’ evidence to explain the reduction effect of industrial pollution emission. 2) This study extends the production density model by mathematics, which reveals the non-linear interactive effects of spatial structure and agglomeration on industrial pollution emissions, proposes three mechanisms based on agglomeration theory, and test it by using non-linear threshold regression and mediation model. 3) This study explored the impact of spatial structure on industrial pollution emissions, finding polycentricity is associated with lower industrial pollution emission intensity but the emission reduction effect of polycentricity depends on agglomeration degree.

The study will enrich the research on the relationship between polycentricity and environmental pollution in city-regions. It is expected that this research will help to provide decision-making reference for emerging economies undergoing rapid urbanization to realize high-quality regional development and optimize the spatial layout of urbanization.

To answer these above two questions, the purpose of this study is to examine whether polycentricity has more obvious emission reduction effect under the same agglomeration degree, and explain the underlying mechanisms theoretically and empirically. The study proceeds as follows. Section 2 is literature review. Section 3 proposes the theoretical model and mechanism analysis. Section 4 explains the measuring method of spatial structure and introduces empirical methodology of this study. Section 5 discusses the empirical results, and Section 6 concludes the study.

The existing literature has made valuable studies on pollution emissions from the perspectives of environmental regulations (Cole et al., 2005), international trade, foreign investment (Rezza and Alief, 2013), taxation policies (Levinson, 2009), transport infrastructure (Tsakiris et al., 2014), regional economic growth, and agglomeration (Roca et al., 2001; Lalive et al., 2017). Pollution emission is an undesirable output of economic activity, and the spatial density of economic activity is also an important factor affecting pollution emission. Hence, the relative study from the perspective of agglomeration is quite abundant (Zeng and Zhao, 2009; Cheng, 2016), but the empirical study on the relationship between agglomeration and pollution emissions has not yet reached a consistent conclusion. Some studies suggest that the influence of agglomeration on pollution emission is nonlinear and that excessive agglomeration without reasonable space distribution will increase emissions (Newman and Kenworthy, 1989; Kamal-Chaoui and Robert, 2009; Glaser and Kahn, 2010; Qin and Wu, 2014).

Spatial structure first has attracted interest from urban geographers and planners as it is believed to affect economic, environmental, and social performance. Soon, urban economists were keen to discover that the environmental effects of agglomeration may depend on differences in spatial structures and their dynamics (Madlener and Sunak, 2011; Denant-Boemont et al., 2018). Camagni et al. (2002) investigated whether different urban spatial structures could be associated with specific environmental costs and the results found that a “wisely compact” and polycentric urban structure is desirable in the Milan metropolitan areas. Veneri (2014) used a database of 82 Italian metropolitan areas to investigate polycentric spatial structure effects on commuting costs and found that metropolitan areas with a higher degree of polycentricity have lower external costs of mobility which was measured by per capita CO₂ emissions, while other spatial characteristics, such as functional diversification and size could not be proved to have the same effect. Burgalassi and Luzzati (2015) addressed the possible link between urban spatial structures (polycentricity or dispersity) and environmental quality originating from transport and residential heating but the empirical evidence for Italy NUTS three regions' spatial structure on CO₂ and PMs could not be confirmed. Li and Zhou (2019) believed that spatial structure affects air quality through a series of mechanisms, such as commuting distance, energy consumption, and transportation layout, and empirically tested the effect of urban form on urban air quality with air monitoring data of prefecture-level cities in China, indicating that smaller, scattered, and polycentric cities have better air quality.

In addition to the concept of polycentric and monocentric spatial structure, urban compactness and urban sprawl are a set of relative concepts, which are also used to describe the characteristics of urban spatial structure. Urban sprawl refers to the excessive expansion of the urban space beyond the necessary level, which causes the economic activities originally concentrated mainly in the urban center to spread to the periphery of the city. The characteristics of the urban sprawl are dispersed in shape, low population density, polycentric, more dependent on private cars, and so on, which increases the vehicle use, commuting trips, and petrochemical energy consumption (Glaeser and Kahn, 2004). Urban sprawl usually reduces the density of residential and economic activity and increases the demand for buildings (Banzhaf and Lavery, 2010), which leads to increased emissions of dust pollutants during construction. In addition, urban sprawl will erode the green area around the city (Burchfield et al., 2005). Borrego et al. (2006) studied the relationship between urban spatial structure and air quality by simulation and they found that the total amount of air pollutants in compact urban structures was less. Although the urban sprawl is also a way to measure the urban spatial form, it is different from polycentricity which refers to balanced hierarchical relationships among space units, occurring when the vast majority of economic activity is evenly distributed in several space units of comparable size (Burgalassi and Luzzati, 2015). Urban sprawl can only reflect the spatial structure within a single city rather than among cities, so it could not explore the impact of spatial structure on pollution emission on a larger spatial scale beyond the city. In view of the above, this study uses polycentricity to measure spatial structure.

Although many empirical papers deal with air pollution and polycentricity in China (Han et al., 2020; Li et al., 2020; Wang and Zhang, 2020), and have drawn valuable conclusions, some issues still need to be studied. 1) Most of the research in this field is based on urban planning and geography which considers that the reasonable layout of urban function and the appropriate space structure can help to reduce the energy consumption and pollution emission from transport, repeated infrastructure construction, and residential heating for improving the environmental performance. However, few studies have analyzed the influence of urban spatial structure on the industrial pollution emission intensity based on the agglomeration theory. 2) Some studies believed that the agglomeration diseconomies caused by excessive agglomeration are the cause of the non-linear relationship between agglomeration and pollution emissions, but both theoretical and empirical analyses of agglomeration and agglomeration externalities on pollution emission have so far neglected the spatial form of agglomeration. 3) The measurement of polycentricity is multi-dimensional. In terms of economic polycentricity, gross domestic production, labor force, and production outputs are all appropriate indicators to measure the spatial structure of economic activity agglomeration. So, it is necessary to use different spatial structure indexes for empirical testing.

Under an agglomeration economy, industry and labor are concentrated constantly, and finally, the spatial organization structure of economic activities is formed. In this study, the relatively balanced spatial organization relationship is called polycentric spatial structure. Based on agglomeration theory, the relationship between polycentric spatial structure and regional pollution emission is discussed. Agglomeration can bring a variety of spillover effects through sharing, matching, and learning, so as to produce a certain positive externality. Ciccone and Hall (1996) constructed a production density model allowing increasing returns to scale which provided a simple and reasonable theoretical framework for agglomeration theory. Ke (2010) simplified and improved this model, and this study uses his method for reference to introduce the urban spatial structure into the model. q i = Q i / A i = π i [ l i β k i 1 − β ] α [ S i ⋅ ( Q i / A i ) ] ( λ − 1 ) / λ (1) where q _i is the output per unit area in the ith city, Q _i and A _i are the total output and total urban built-up area of the ith city, and Q _i /A _i obviously is the spatial density of output. π _i is the total factor productivity, l _i is the employment density, k _i is the physical capital input per unit area, α is the return on capital and labor per unit area, 0 < α ≤ 1, indicating diminishing marginal productivity due to congestion, and β is the contribution rate of labor input to output per unit area, 0 < β ≤ 1. S _i is the variable for urban spatial structure. The larger the S _i is, the more monocentric the city is, and conversely, the smaller the S _i is, the more polycentric the city is. Therefore, spatial density of output Q _i /A _i for reflecting the agglomeration degree of economic activities, and urban spatial structure S _i for indicating spatial organization of agglomeration will jointly determine the externality of city-region by this model setting. λ is the parameter of output density. If λ > 1, agglomeration bring about positive externalities and increase more output per unit area in cities. Considering that the general production process is usually accompanied by pollution emissions, this study assumes that the Q unit output can be divided into two parts: the Q ₀ unit expected output and the w unit non-expected output (pollution emission).

Therefore, Eq. 1 can be written as q i = ( Q 0 + w ) / A i = ( Q 0 / A i ) ( 1 + w / Q 0 ) = π i [ l i β k i 1 − β ] α [ S i ⋅ ( Q 0 / A i ) ] ( λ − 1 ) / λ ( 1 + w / Q 0 ) ( λ − 1 ) / λ (2)

Given the lack of the capital data for cities, it can be assumed that the expression of capital requirements is as follows: k i = α ( 1 − β ) ( Q 0 + w ) / γ i (3)

where γ is the capital price of the ith city. From Eq. (3) into Eq. (2) the following equation is obtained: 1 + w / Q 0 = π λ 1 − α ( 1 − β ) λ ( Q 0 L i ) − α β λ 1 − α ( 1 − β ) λ ( S i ⋅ Q 0 A i ) α λ − 1 1 − α ( 1 − β ) λ [ α ( 1 − β ) A i γ i ] α ( 1 − β ) λ 1 − α ( 1 − β ) λ (4) where w/Q ₀ , the left-hand of Eq. 4, is pollution emission per unit output and the right-hand side of the equation includes labor productivity (Q ₀ /L _i ), urban spatial structure (S _i ), and spatial density of expected output (Q ₀ /A _i ). Eq. 4 shows that the pollution intensity of the city is closely related to the labor productivity, spatial structure, and the agglomeration degree measured by output density. Eq. 4 can then be rewritten as: ln ( 1 + w / Q 0 ) = [ λ 1 − α ( 1 − β ) λ ] ln ⁡ π i + − α β λ 1 − α ( 1 − β ) λ ln ( Q 0 L i ) + α λ − 1 1 − α ( 1 − β ) λ ln ( S i ⋅ Q 0 A i ) + α ( 1 − β ) λ 1 − α ( 1 − β ) λ [ ln ⁡ α ( 1 − β ) A i − ln ⁡ γ i ] ( 5 ) And then, the equation yields the expression w / Q 0 ≈ [ λ 1 − α ( 1 − β ) λ ] ln ⁡ π i + − α β λ 1 − α ( 1 − β ) λ ln ( Q 0 L i ) + α λ − 1 1 − α ( 1 − β ) λ ln ( S i ⋅ Q 0 A i ) + α ( 1 − β ) λ 1 − α ( 1 − β ) λ [ ln ⁡ α ( 1 − β ) A i − ln ⁡ γ i ] (6)

The coefficient we are interested in is (αλ-1)/[1-α (1-β) λ], the coefficient of the third term on the right of the equation, which indicates the interactive effect of spatial density and spatial structure on pollution emission. (αλ-1)/[1-α (1-β) λ] may be positive or negative, depending on the parameters α, β, and λ which are closely related to the stage of regional economic development and the agglomeration degree. It is clear that a linear interactive effect is not enough to describe the complex relation, and the effect of the urban spatial structure on the pollution emission may also have a non-linear threshold effect due to the different agglomeration degrees.

The environmental effect of agglomeration is the result of increasing effect and emission reduction effect. On the one hand, agglomeration is accompanied by the expansion of total output and pollution emission (Yang, Y., 2021); on the other hand, agglomeration promotes the decline in pollution emission intensity through the professional division, pollution control cost sharing, and pollution control technology learning (Glaser and Kahn, 2010). Therefore, whether the emission reduction effect of industrial agglomeration dominates depends on a variety of influencing factors and presents spatial heterogeneity and non-linear characteristics (Yang et al., 2021). When the regional agglomeration degree is low, the inconveniences of polycentricity could be even more prominent. Because, under these circumstances, the polycentric structure is a dispersed spatial organization form, it is difficult to produce the scale effect of pollution control and the spillover effect of technology. The polycentric structure not only loses the advantage of emission reduction but also increases the expenditure on infrastructure construction and social governance. Thus, the overall spatial efficiency of the city-regions is reduced. But, when the regional agglomeration degree is low, the monocentric spatial structure can centralize the various factors of production and supporting facilities, which will help to reduce the enterprises' physical distribution costs and promote the production efficiency by matching labor force and the spillover of knowledge and technology (Yuan et al.2021). At this time, the positive agglomeration externality can be brought into full play. In addition, the unit pollution abatement cost can be reduced by sharing the pollution treatment equipment. Therefore, in the areas with low level of agglomeration, the monocentric spatial structure is conducive to promoting the sharing of facilities, cost saving, and efficiency improvement, which shows the emission reduction effect.

When the economy is in rapid development stage, the agglomeration of economic activities leads to the expansion of the capacity scale, resulting in a sharp increase in energy demand, which in turn leads to an increase in the intensity of pollution emissions. In this case, the polycentric spatial structure can play a role in pollution emission reduction through three mechanisms. 1) environmental regulation integration effect. Polycentricity is not a hierarchical system, but a city network system connected by a traffic trunk, which is not only conducive to deepening the industrial division, building closer functional links, and facilitating the flow of people, goods, and information (Glaeser et al., 2016) but also improving the production efficiency (Zhang et al., 2017). In addition, it is also possible to promote equitable distribution of educational and medical resources and infrastructure (Eraydin, 2008), which also plays a positive role in establishing a more equal cooperative relationship among regions. Therefore, polycentricity is not only a term describing spatial structure but also a policy option to promote coordinated regional development and enhance regional integration. When the level of regional agglomeration is high, the production efficiency of each spatial unit in polycentric areas has been generally improved; the flows of population, capital, and information across the spatial units have been more and more rapid; and the income gap has narrowed. Such an economic integration will also contribute to the unification of environmental regulation standards and the consensus of non-governmental environmental protection in the region, so as to avoid the decline in the overall regional environmental quality caused by pollution haven. Therefore, polycentric structure has a significant emission reduction effect. The monocentric structure has an obvious “center-periphery” distribution pattern. When the polarization effect is greater than the diffusion effect, the agglomeration center will restrain the economic growth of the marginal area, called “agglomeration shadow” or “dark under lamp,” which produces obvious boundary effect and hinders the process of regional integration. Environmental awareness in the periphery is weak and opportunities for economic growth are urgently needed, so these peripheral areas, which trade for investment at the expense of the environment, often become “havens” for polluting enterprises. Even if the center area has the emission reduction effect of agglomeration, the increase in unit pollution emissions in the peripheral areas (the pollution control facilities are generally poor, and the emission reduction efficiency is low) will lead to a decline in the overall environmental quality. 2) Energy utilization efficiency effect. When the city networks are formed by geographically adjacent and strongly functionally linked small cities, they can enjoy the same urban functions as a single megacity (Meijers et al., 2016). That is to say, based on industrial division, knowledge spillover, labor pool, and other forms of “borrowed size,” these spatial units can also produce increasing economic benefits in place of the “scale effect”. “Borrowed size” is often used as a mechanism for explaining polycentric structure to improve regional productivity. In fact, the “borrowed size” mechanism also has the effect of reducing emissions. Polycentric structure helps the spillover of industrial energy-saving technology through “borrowed size”, which is conducive to improving the energy utilization efficiency and reducing the industrial energy consumption intensity so as to realize the pollution emission reduction in the region. 3) Environmental treatment capacity effect. The unreasonable spatial structure in megacities leads to the excessive agglomeration of production factors, resulting in various symptoms of “urban problem”, which should be solved by optimizing the spatial structure and easing city functions. The space units that form polycentric structures are relatively small in size, so they have greater endogenous capacity to keep these economic, social, and environmental costs within affordable limits (Capello and Camagni, 2000). Therefore, under the same density of city-region, the polycentric structure can effectively alleviate the deterioration of urban problems and industrial problems caused by excessive agglomeration under monocentric structure. The theoretical framework for the effect of polycentric spatial structure on pollution emission was briefly constructed by the above analysis, as shown in Figure 2.

In this study, two pollutants, industrial sulfur dioxide and industrial smoke dust, are used for empirical regression to provide convincing and robust results. In city-regions samples, the number of individuals is 21, and the total observations are 294. This study uses pooled regression with cluster-robust standard error and uses the centralization method for both high- and low-order terms of interaction terms to solve the possible multiple collinearity problems.

Based on the panel data of city-regions in China from 2003 to 2016, the study measures the spatial structure of city regions in terms of economic scale and labor force size. Based on this, the study explores the impact of spatial structure of urban regions on industrial air pollution by GMM and threshold regression methods. The main conclusions are as follows: (1) The interactive emission reduction effect of polycentric spatial structure is more and more obvious with the increase in agglomeration degree. (2) The emission reduction effect of polycentric spatial structure depends on regimes of the regional agglomeration. When the degree of agglomeration is low, spatial structure has no significant effect on pollution emission intensity, while the areas with a higher agglomeration degree, polycentric spatial structure would have significant emission reduction effect, and this effect is increasing. (3) On the scales of city-regions, the emission reduction effect of polycentric spatial structure will decrease with the lengthening of traffic distance and time across city-regions. After controlling the spatial organization form and geographical compactness of the city-regions, the intra-provincial city-regions have more emission reduction effects than the inter-provincial city-regions. (4) The emission reduction effect of polycentric spatial structure can be explained by three mechanisms: regional environmental regulation integration, energy utilization efficiency, and environmental treatment capacity.

The Chinese government has always played a dominant role in the evolution of the city-regions’ spatial structure. Even if a consensus on this issue could be formed that the rational spatial structure has a dual policy dividend of coordinating regional development and improvement of air quality, some real obstacles still need to be solved by the government, so as to achieve the expected emission reduction effect. The great practical significance of this study is as follows: (1) It is worth noting that the emission reduction effect of polycentricity depends on agglomeration degree. When the agglomeration is low, the monocentric spatial structure will play a better emission reduction effect; even if a polycentric spatial structure is achieved through administrative forces, the effect of this action on emissions reduction may be negligible. (2) The government should promote the transformation of the spatial structure from a monocentricity to polycentricity according to the actual situation of the region. Positive measures that the government can take includes rationally planning infrastructure, strengthening cooperation mechanism, breaking down the administrative barrier, and eliminating technology gap, to ensure that the expected emission reduction effect can be achieved. (3) With the increasing traffic distance and time in the city-regions, the emission reduction effect of polycentricity decreases. The city-regions should focus on the construction of efficient and fast transportation networks to promote the optimal allocation of production factors. At the same time, the joint construction and infrastructure sharing in city-regions will reduce low-level repetitive construction by central government planning guidance. (4) In addition to the geographical distance, the administrative barriers will weaken the regional connection and hinder the efficiency of polycentric spatial structure. Therefore, city-regions should actively establish a cooperation mechanism to fairly divide emission reduction responsibilities considering the industrial division, narrow the green technology gap, form a unified consensus on emission reduction, and set up the compensation mechanism for technology transfer and funds support.

Although this study examines in detail how the polycentricity in city-regions affects industrial pollution emissions and analyses the mechanisms, it considers the polycentricity mainly in terms of the level of economic development and the labor force. Richer conclusions could be drawn if functional polycentricity or the polycentricity of transport networks was used to measure the polycentric spatial structure.