Manufacturing agglomeration, urban form, and haze pollution

Manufacturing agglomeration promotes rapid economic development while also causing severe environmental pollution. This paper investigates the impact and mechanism of manufacturing agglomeration on haze pollution from the Chinese city level. Furthermore, we discuss the moderating effect and threshold effect of the three urban forms of urban external shape complexity, urban compactness, and urban fragmentation on the relationship between the two. The result shows the following: (1) The aggregation of the manufacturing industry presents an inverted U-shaped characteristic of promoting first and then inhibiting haze pollution in China’s overall, eastern and central regions. (2) The complexity of the city’s external shape and the city’s fragmentation has a positive moderating effect on the relationship between manufacturing agglomeration and haze pollution. And urban fragmentation shows a negative moderating effect on the relationship between the two when the level of manufacturing agglomeration is on the right side of the inverted U-shaped curve. (3) The urban form shows a significant double threshold characteristic for haze pollution, increasing the complexity of the city’s external shape and the city’s fragmentation. The agglomeration of manufacturing shows the characteristics of first inhibiting and then promoting haze pollution. As urban compactness increases, the inhibitory effect of manufacturing agglomeration on haze pollution increases.


Introduction
Since the Industrial Revolution, the rapid development of economy and urbanization in various world regions has also resulted in severe environmental pollution. Among them, air pollution represented by frequent haze weather is particularly prominent (Wang et al. 2015). Haze pollution will seriously affect the daily life and health of residents but also cause incalculable damage to social and economic development . China is the world's largest developing country, its economic growth relies on traditional manufacturing and fossil energy consumption, and the inefficiency of environmental governance has led to increasingly severe haze pollution (Hao et al. 2022). Therefore, in the face of the threat and trouble of haze pollution, it is necessary to explore the economic root cause of the frequent occurrence of haze pollution and find out the critical factors to control it. By this way, we can take practical measures to improve air quality, thereby safeguarding people's health and promoting sustainable economic development .
Manufacturing plays a crucial role in driving regional economies to grow, increase employment opportunities, and raise living standards (Haraguchi et al. 2017). At the same time, as an organizational form of modern manufacturing, spatial agglomeration brings various positive externalities such as economies of scale and knowledge spillovers. It helps to optimize the allocation efficiency of production factors of enterprises in agglomeration areas (Chen et al. 2020), therefore reducing the production cost of enterprises and enhancing industrial competitiveness. However, industrial agglomeration cannot expand infinitely. When the agglomeration level exceeds the carrying capacity threshold of regional resources and the environment, it may pose a severe threat to the ecological environment (Meng and Xu 2022). China's haze pollution is mainly concentrated in mega-city clusters such as the Pearl River Delta, Yangtze River Delta, and Beijing-Tianjin-Hebei (Song et al. 2017). These regions are also the main agglomeration areas of China's manufacturing industry. So, does this mean that the concentration of manufacturing is the cause of increased haze pollution? Secondly, can industrial agglomeration play a positive role in alleviating haze pollution by various positive externalities such as economies of scale, cost savings, and technology spillovers? Meanwhile, under the influence of regional development differences in China (Dong et al. 2019), is there any heterogeneity in the impact of manufacturing agglomeration on haze pollution in different regions? Therefore, an empirical investigation of the concentration of manufacturing industries on haze pollution in China is conducive to scientifically answering the above questions. Then, we can provide reasonable policy suggestions for the control of haze pollution.
Urban spatial form refers to the spatial pattern of urban elements in the spatial plane and vertical height direction, which is mainly reflected by the spatial heterogeneity caused by the overall shape of the city, land-use diversity, and urban density (Anderson et al. 1996). In fact, existing research has found that fragmented and complex-shaped urban forms produce more air pollution than compact cities with high continuity, low expansion, and low shape complexity (Li and Zhou 2019). In addition, with the urban expansion, the urban form gradually develops toward a high-density and compact model. This further absorbs all kinds of resources into specific areas and provides guarantees of development space and production factors for manufacturing agglomeration (Shao et al. 2019). Thus, it has led to a large number of manufacturing enterprises rapidly gathering in the city, forming manufacturing bases and industrial clusters, which drives China's regional economic development. However, a large amount of energy consumed by resource-intensive manufacturing is also a major cause of environmental pollution . Based on the above analysis, the urban form may have a moderating effect between manufacturing agglomeration and haze pollution. Simultaneously, due to differences in natural conditions, economic development levels, and development history among cities, the changes in urban form are characterized by stages. Therefore, the impact of manufacturing agglomeration on haze pollution will change with the change of urban form, which indicates that there may be a threshold effect on the influence of urban form on the two.
However, previous studies have paid little attention to the impact of urban form on economic activities and environmental pollution. In fact, urban spatial form affects the distribution pattern of population and economic factors in the spatial dimension. Meanwhile, the urban form also reflects the current situation of urban land use, urban expansion, and urban development density. Therefore, the urban form can affect manufacturing agglomeration and haze pollution by affecting land use and urban spatial structure patterns. Moreover, whether the urban form develops in a proper and orderly direction is the key to whether the city can achieve sustainable development. Based on the above analysis, a novelty of this paper is that the urban form is introduced into the analytical framework of manufacturing agglomeration and haze pollution. Furthermore, this study quantifies urban form based on urban external shape complexity, compactness, and fragmentation. Then, we use the moderation effect model and the threshold regression model to examine the possible moderating effects and threshold characteristics of urban form between manufacturing agglomeration and haze pollution. This would help provide a scientific method for the measurement of urban form and identifies the impact mechanism of urban form between manufacturing agglomeration and haze pollution.

Literature review and theoretical hypotheses
The theory of externalities suggests that industrial agglomeration has positive and negative externalities. And its combined results determine the environmental effects of industrial agglomeration (Porter 1990).
The first type of research believes that manufacturing agglomeration will aggravate ecological pollution, and its mechanism is manifested in three aspects: (1) crowding effect. When a large number of enterprises or labor are concentrated in a certain area, it will cause excessive consumption of resources and cause traffic congestion, which will significantly increase environmental pollution (Henderson 2003a, b). At the same time, excessive industrial agglomeration may hinder the entry of new knowledge and new technologies, which is not helpful in reducing haze pollution (Martin and Sunley 2006). (2) Industrial structure effect. In the preliminary phase of manufacturing agglomeration, the industrial structure is usually transformed into a "pollutionintensive" direction, resulting in increased energy consumption and pollutant emissions (Chen et al. 2019). (3) Downward competition effect. In order to attract enterprises to invest and drive local economic development, local governments compete to lower the environmental access threshold. The companies that entered at this time brought relatively low-level industries and technologies. As a result, the region falls into a vicious circle of "high energy consumption, high pollution" industrial agglomeration, resulting in increased emissions (List and Co 2000).
The second type of study argues that manufacturing agglomeration can improve environmental pollution and its mechanism of action is manifested in three main ways: (1) economies of scale effect. Agglomeration of manufacturing industries motivates the accumulation of capital, information, labor, and other factors of production in the aggregation area, which improves the effectiveness of resource distribution and production. The resulting economies of scale are conducive to forming a synergy of centralized production and pollution control, thus reducing the cost of pollution control for enterprises (Meng and Xu 2022). (2) Sharing effect. On the one hand, manufacturing enterprises concentrate their production activities in one region so that enterprises can share labor in this area. This helps companies select labor that matches high-level pollution technologies flexibly and improve their pollution treatment technology (Fang et al. 2020). On the other hand, regions with a high concentration of manufacturing industries also have more complete facilities and equipment. Enterprises can improve the pollution control level of the whole region by sharing energy-saving and emission reduction facilities (Helsley and Strange 1990). (3) Learning effect. Manufacturing agglomeration can promote information exchange between enterprises. In this way, it is helpful for the sharing and spillover of pollution control experience and pollution treatment technology (Glaeser et al. 1992), hence alleviating the environmental pollution.
Meanwhile, manufacturing agglomeration has a life cycle, which means that manufacturing also goes through stages of emergence, development, and decline (Ingstrup and Damgaard 2013). And different manufacturing agglomeration stages have different characteristics in terms of resource allocation efficiency, R&D efficiency, competition level, and public facility construction, which directly affects the relationship between manufacturing agglomeration and haze pollution. Based on the above analysis, we think that the agglomeration of manufacturing industries on haze pollution may show an inverted U-shaped of "promoting first and then inhibiting." At the same time, there are vast variations in economic development, government policies, and levels of marketization across China's regions (Li et al. 2021). As a result, there may also be heterogeneity in manufacturing agglomeration and the stage of manufacturing development in different areas. Therefore, we propose the following two hypotheses.
Hypothesis 1. The influence of manufacturing agglomeration on haze pollution has a nonlinear inverted "U" curve relationship. Hypothesis 2. The impact of manufacturing agglomeration on haze pollution is heterogeneous in different regions of China.
This study believes that urban spatial form should comprehensively consider cities' compactness, density, and external shape characteristics in development (Fortuna et al. 2006).
(1) Urban exterior shape complexity. The external shape complexity of the city represents the degree of connectivity of each patch in the city. And the high complexity refers to severe segmentation among urban patches with low connectivity and accessibility (McCarty and Kaza 2015). On the one hand, complex urban boundaries may undermine the mobility of factors in production between regions, such as labor and capital. This leads to the less efficient allocation of resources and higher communication costs between enterprises (Ma et al. 2015). On the other hand, in the expansion process, the city may present a complicated external shape due to the influence of natural conditions such as mountains and hills. This can result in poor urban ventilation, which is detrimental to the dispersion of pollutants and enhances urban environmental pollution (Sha et al. 2018). Accordingly, the expansion of output scale resulting from manufacturing agglomeration has intensified the consumption of resources and energy. This process will bring about an increase in pollution emissions. In addition, complex urban boundaries are not conducive to the diffusion of pollutants, thus intensifying the negative externalities of manufacturing agglomeration on the environment. (2) Urban compactness. Urban compactness refers to highdensity agglomeration structures that maximize urban economic benefits. It emphasizes the "mixing of land use functions" under the high-intensity land use development model (Neuman 2005). One view holds that the increase in urban compactness helps to exert the inhibitory influence of the agglomeration of manufacturing on haze pollution. And the compactness of cities strengthens the intensity of urban land use. It is beneficial to the agglomeration of construction land in the city, forming economies of scale (Brülhart and Sbergami 2009), thereby attracting manufacturing enterprises to agglomerate. At the same time, high-density urban development can prevent the blind expansion of urban construction land. It is beneficial to protecting the grasslands, rivers, and other lands outside the city. It plays the role of greening the environment and absorbing pollutants (Banzhaf and Walsh 2008), which can effectively alleviate haze pollution. Another view is that the development of urban compactness exacerbates the negative influence of the agglomeration of manufacturing on environmental pollution. Within a fixed urban area, increasing compactness may increase urban land prices and labor costs (Gaigné et al. 2012). Manufacturing companies have scattered to the suburbs in pursuit of lower land and wage costs. And industrial suburbanization will weaken the economic "agglomeration effect," resulting in higher commuting costs and transportation costs, which will lead to increased pollutant emissions (Henderson 2003a, b). At the same time, it is difficult for manufacturing enterprises to exert the positive effects of agglomeration, such as technology spillover, centralized treatment of pollutants, and industrial structure upgrading, thus further aggravating environmental pollution (Zhao et al. 2022).
(3) Urban fragmentation. In the process of urban development, the urban form shows the characteristics of fragmentation in order to protect natural spaces such as rivers, grasslands, and forests (Wang et al. 2020). One view is that urban fragmentation exacerbates the negative effects of the concentration of manufacturing on environmental pollution. Urbanization is the primary driver of land fragmentation in China ). In the urbanization process, urban spaces that are not suitable for development and living may appear, such as parks, nature reserves, or industrial parks developed across regions (Wang et al. 2020). As a result, it is tough for manufacturing enterprises to form agglomeration economic advantages. The optimal allocation of production factors cannot be achieved, hence inhibiting the level of manufacturing agglomeration (Glaeser and Kahn 2010). Another view is that urban fragmentation helps to exert the inhibitory effect of manufacturing agglomeration on air pollution. The increase in urban fragmentation of cities will promote the transformation of cities from single-center to multi-center. The formation of polycentric cities is conducive to guiding manufacturing enterprises to transfer to sub-central cities. This can effectively alleviate the congestion effect and environmental pollution problems in central cities (Kloosterman and Lambregts 2001). In addition, the fragmentation caused by the increase of natural space can improve the ecological quality of the city. It is beneficial to enhancing the city's attractiveness, promoting the city's agglomeration effect and economies of scale, which in turn enhances the level of manufacturing agglomeration (Banzhaf and Walsh 2008). Therefore, the development of urban fragmentation can reduce the crowding effect of central cities and promote the improvement of air quality. This can play the role of various positive externalities of manufacturing agglomeration to reduce haze pollution, thus forming a virtuous circle of improving the local ecological environment.
In summary, in the context of the rapid development of manufacturing, one view holds that the compact, fragmented, and complex development of urban forms will lead to a decline in the level of manufacturing agglomeration. Under such circumstances, it is difficult to take advantage of the positive externalities such as economies of scale, technology spillovers, circular economy, and labor sharing of manufacturing agglomeration, thus aggravating haze pollution. Another view is that changes in urban form can directly improve the level of manufacturing agglomeration. On the other hand, it can also form a positive feedback effect on manufacturing agglomeration through environmental improvement. This can bring about the mitigation effect of manufacturing agglomeration on haze pollution. Based on the above analysis, we propose hypothesis 3: Hypothesis 3. The urban form has a moderating effect on the relationship between manufacturing agglomeration and haze pollution.
Urban form has the characteristics of dynamic change. When the complexity of the external shape of the city and the degree of urban fragmentation is low, its impact on manufacturing agglomeration is relatively tiny. At this time, manufacturing agglomeration can play various positive externalities to reduce the level of haze pollution (He 2006). But with the increase in the complexity of the external shape and the fragmentation of the city, manufacturing agglomeration has been suppressed. And the crowding effect of manufacturing agglomeration is greater than the effect of economies of scale, thus showing a promoting effect on haze pollution (Yuan et al. 2020). When urban compactness is low, disordered, lowdensity urban sprawl can negatively impact economic development. It will not only lead to inefficient allocation of production factors but also weaken the ability of cities to divide labor and cooperate, resulting in increased production costs and transaction costs (Fallah et al. 2011). The increase in cost will inhibit the agglomeration effect and scale economy effect of manufacturing enterprises, thereby worsening the haze pollution. However, when the level of urban compactness reaches a certain level, various positive effects of manufacturing agglomeration gradually become dominant, significantly reducing haze pollution (Stone 2008). Therefore, due to the different development stages of different urban forms in terms of scale, structure, and shape, the effect of aggregation of manufacturing industries on haze pollution may also be quite different in different urban forms (Wang et al. 2020). This implies that there may be a threshold effect between manufacturing agglomeration and haze pollution under the effect of urban form. Thus, we propose: Hypothesis 4. Under the influence of urban form, the impact of manufacturing agglomeration on haze pollution presents threshold characteristics.
To summarize, we use a flowchart to illustrate the research ideas of this paper in detail (Fig. 1).

Spatial econometric models
In this research, the existence of spatial dispersion of pollutants is considered. The haze pollution in a region mainly comprises the actual generation of haze pollution in the area and the diffusion amount from other regions (Sun et al. 2014). And part of the real production of smog pollution in this region will spread to other areas. This part does not affect local haze pollution, so we subtract it (Shao et al. 2016). And the diffusion of haze pollution from other areas and the dispersal of the region to other areas reflect the spatial dependence of haze pollution in each region (Yang et al. 2021). Based on this, this study will establish spatial econometric models. And we introduce the squared term of manufacturing agglomeration (sMan) to investigate the nonlinear relationship between manufacturing agglomeration and haze pollution. There are three forms of spatial econometric models: the spatial autoregressive model (SAR) is mainly applied to investigate the spatial dependency characteristics between explained variables. The spatial error model (SEM) is used to research the spatial association of omitted variables or unobservable random shocks not included in the explanatory variables. And the spatial durbin model (SDM) takes into account the situation where both spatial lag and spatial error may exist simultaneously. Since SAR and SEM are more specific than general-purpose SDM, this paper will focus on examining the applicability of SAR and SEM to explore the spatial effect of the agglomeration of manufacturing on haze pollution and its transmission mechanism. The two models mentioned above are represented as follows: (1) In this paper, PM is used to denote the dependent variable, haze pollution; Man is the independent variable of this paper, manufacturing agglomeration; sMan is the squared term of manufacturing agglomeration; X represents the vector composed of control variables, and α and β are their corresponding coefficient vectors; υ i and γ t denote spatial individual effect and spatial time effect respectively. In addition, u it conforms to the disturbance term of the normal distribution; The current period spatial lag coefficient is denoted by ρ, and the current spatial error term coefficient is denoted by λ.

Spatial econometric model based on moderating effect
To examine whether manufacturing agglomeration affects haze pollution through urban form, we introduce the urban form (Cform) into the spatial measurement model and establish the following moderating effect model, as shown in formulas (3)-(4): (2) Fig. 1 The mechanism of manufacturing agglomeration on haze pollution

Establishment of threshold regression model
Among the variables that reflect causality, there may be some threshold variables. We divide into different sample groups according to the threshold value, which may have significantly different effects on the correlation between independent and dependent variables (Hansen 1999). To test the threshold effect of urban form (Cform) between haze pollution and manufacturing agglomeration, we first set a single threshold regression model, which can be expressed as: In formula (5), PM is chosen to represent the dependent variable, and X is composed of control variables, Cfrom represents the threshold variable urban form, represents a fixed threshold value, α is the coefficient of the effect of control variables on the dependent variable, β1 and β2 represent the influence coefficients of manufacturing agglomeration on haze pollution under different threshold levels, the constant term is C, it ∼ (0, 2 ) means a random disturbance term, and I(·) is an indicator function. Similarly, the formulas of the double threshold test and the triple threshold test that we set up are as follows. Accordingly, the meanings of β3 and β4 are analogous to that of β1.

Data description
Dependent variable: PM2.5 (PM) PM2.5 is considered to be the main component of haze pollution in China. It has a significant adverse influence on air quality, residents' health, and economic sustainability (Zhu et al. 2019). Based on this, we use the annual average concentration of PM2.5 to characterize the urban haze pollution level. The data in this article are from raster data published by the Center for Socioeconomic Data and Applications at Columbia University (Van Donkelaar et al. 2015). Then, we use ArcGIS software further to parse the annual average PM2.5 concentration data.
At the same time, this paper uses ArcGIS10.2 to visualize the spatial evolution characteristics of haze pollution in China from 2006 to 2016. As shown in Fig. 2, China's urban haze pollution showed the characteristics of spatial agglomeration. The areas with more serious haze pollution were mainly concentrated in the Harbin-Changchun urban agglomeration, the Beijing-Tianjin-Hebei urban agglomeration, the Central Plains urban agglomeration, and the Yangtze River Delta urban agglomeration, which were also manufacturing clusters. From 2006 to 2016, the haze pollution levels in most cities south of the Yangtze River had improved, but China as a whole still had haze pollution problems.

Independent variable: manufacturing agglomeration
In this paper, we refer to the method of O'Donoghue and Gleave (2004) and measure the agglomeration level with the location entropy. This indicator is simple and easy to implement. And it can accurately reflect the comparative advantages of the national urban manufacturing agglomeration, which can better eliminate the endogenous effects of different regional scales. The formula for calculating the location entropy of industry i in region j is as follows: In the formula, x ij is the number of employed persons of industry i in area j, x j is the number of employed persons of all industries in area j, x i is the employment in the industry i at the national level, and x is the employment in all industries at the national level.

Moderating variables and threshold variables: urban form
Urban exterior shape complexity The urban spatial form has irregularity and self-similarity. Fractal dimension (FD) quantifies the irregularity and self-similarity of the urban spatial form (Frankhauser 2015;Terzi and Kaya 2011). FD will increase if cities become irregular and have uneven boundaries as they expand. Therefore, we use FD to characterize the urban land shape complexity.
Urban compactness Compact cities are becoming the direction of urbanization. It represents a model of urban construction and spatial development layout with high density and efficient mixed land use (Xu et al. 2017). In this paper, the largest patch index (LPI) is introduced into the model to characterize the level of urban compactness. The calculation method is the proportion of the largest urban land patch in the total urban land area.
Urban fragmentation Fragmentation can be quantitatively analyzed by patch density, which can better characterize the degree of urban leapfrog development (Bechle et al. 2011). We use the number of urban patches and the ratio of the total urban land area for the calculation.

Control variables
On the basis of referring to the existing research, we introduce the following control variables to reduce the influence of missing variables on the regression results.
(1) Transportation (Tra). The number of road kilometers (km) is used to measure the city's transportation intensity. The long-term growth rate of the road is lower than the growth rate of the vehicle that will increase the traffic flow on the road, which will easy cause slow traffic flow and even traffic congestion. And this leads to inefficient combustion of vehicle fuels, which largely contributes to air pollution in cities ( Wrobel et al. 2000).
(2) Open level (Open). International knowledge transfer and emission reduction technologies brought by foreign investment can effectively enhance the confidence and ability of host countries to control haze (Gao et al. 2017). Thus, we use foreign direct investment (10 3 yuan) to characterize the city's openness level.
(3) Population density (Pop). Overpopulation can have a crowding effect. It will not only increase the consumption of resources and energy but also cause a series of urban problems such as traffic congestion, which will cause the urban environment to deteriorate (Li et al. 2021). In this paper, the number of people per unit of the urban area is used to measure the effect of urban population concentration on haze pollution. (4) Energy consumption structure (Es). Pollutants generated by high-intensity energy consumption are the main cause of haze pollution (Kouchaki-Penchah et al. 2016). To control the impact of energy consumption on smog pollution, this paper measures energy consumption using the share of coal consumption in total energy consumption. (5) Technical level (Tec). Technological innovation can reduce energy consumption and emission reduction costs by improving energy efficiency in the production process (Sohag et al. 2015). In the long run, technological progress has a suppressing effect on haze pollution. Therefore, we use the number of three patent applications (pieces) to reflect the technical level of the city. (6) Environmental regulation (Er). Environmental regulation will inhibit the pollutant discharge behavior of enterprises. At the same time, excessive ecological regulations will force companies to relocate, hence reducing environmental pollution associated with economic growth (Shapiro and Walker 2018). Thus, we introduce the proportion of environmental protection expenditure in GDP into the study to examine the effect of environmental regulation intensity on haze pollution.

Spatial weight matrix
Geographical closeness and economic relationship are essential factors affecting the spatial layout of manufacturing and haze pollution. Therefore, the setting of the weight matrix needs to consider the effects of geographic distance and economic connection. Based on this, this paper constructs a citybased geographic distance spatial matrix (W1). And element w ij denotes the number of nearest road miles between city i and j. In addition, to verify the results of the robustness of the spatial distance weight matrix selection, we also construct an economic distance weight matrix (W2) for robustness testing.

Data sources
This paper takes into account data availability and references to existing literature. And we have made the following choices for the variables mentioned in the paper (Table 1).  Table 3 shows the regression results of manufacturing agglomeration on haze pollution by OLS, OLS-FE, SAR, and SEM. The estimated coefficients of manufacturing agglomeration in M1 and M2 without considering the spatial correlation are not significant. Therefore, not considering spatial correlation may get biased estimation results. The estimation results of M3 and M4, considering the spatial correlation, show better statistical characteristics. Further from the comparison between M3 and M4, since R 2 in M3 is slightly higher than in M4, it proves that the SAR fits the sample data better than the SEM.   198 According to the regression results, the spatial lag term coefficient in model 3 is significant at the 5% level. It shows an apparent spatial correlation feature of urban haze pollution in China. On the one hand, under the dual effects of natural reasons such as atmospheric circulation and water flow and economic activities such as regional economic integration and regional industrial transfer, geographical proximity is strongly correlated with the level of haze pollution. This suggests the necessity for strengthening regional cooperation in the fight against haze pollution. Otherwise, it will significantly undermine the level of regional haze control. On the other hand, there are strategic interactions between local governments in environmental regulation. And the haze control effect of environmental regulations in neighboring areas can affect the haze pollution concentration in the region through haze pollution spillover. In addition, the regional haze pollution control model and successful experience will also form a strong demonstration effect on neighboring regions through information exchange and technology spillover.

Spatial econometric regression model results
The primary term for manufacturing agglomeration is significantly positive at 1%. And the squared term of manufacturing agglomeration is highly negative at 1%. This implies that manufacturing agglomeration has a significant inverted "U" curve relationship with haze pollution, which confirms hypothesis 1. When manufacturing agglomeration is in its infancy, it has a significantly aggravating influence on the pollution of haze. But with the improvement of the agglomeration level of the manufacturing industry, the positive externality of manufacturing agglomeration has gradually become prominent, thus showing a suppression effect on haze pollution. The initial stage of the agglomeration of the manufacturing industry usually indicates the characteristics of rigid energy demand and rapid growth of energy consumption. At this time, the pollution emission effect brought by the concentration of manufacturing industries is greater than the economy of scale effect and knowledge spillover effect. Therefore, the rapidly advancing manufacturing agglomeration is usually accompanied by an extensive economic growth model of higher energy consumption and emissions, which leads to frequent haze pollution problems. When the level of manufacturing agglomeration reaches a certain threshold, the positive externalities such as resourcesaving and sharing effect of manufacturing agglomeration are gradually becoming more prominent. At this time, the aggregation of manufacturing industries began to show a suppressing effect on haze pollution. The reason is that the concentration of manufacturing industries is conducive to enterprises sharing energy-saving and emission-reduction treatment facilities or technologies. This can reduce the level of haze pollution. In addition, the unit cost of government regulation of pollutant emissions will be lower due to the spatial concentration of manufacturing.
The control variable results in Table 3 show that for every 1% increase in transportation volume, the haze pollution intensity will increase by 0.0855%. This indicates that the expansion of China's transportation scale in recent years has promoted haze pollution. The opening level shows an increasing effect on haze pollution at the 5% significance level, which supports the "pollution paradise" hypothesis of China's haze pollution established. In addition, the influence of population density on haze pollution is positive and significant. It means that the increase in population density will generate a large amount of energy and transport demand, thus increasing haze pollution. And the energy consumption structure exhibits a significant effect on decreasing haze pollution, which indicates that the "green" upgrade of the energy consumption structure being promoted in China helps mitigate haze pollution. Meanwhile, the level of technology has a significantly negative impact on haze pollution. It shows that technological innovation can reduce pollution emissions, thereby inhibiting haze pollution. Environmental regulation shows a significant promoting influence on haze pollution. The reason is that local governments have adopted relatively lax environmental regulation standards to promote economic development. And this method will attract a lot of polluting firms to invest, making the place a "pollution sanctuary," thus aggravating smog pollution.

Robustness tests
In this paper, we mainly select four methods: replace space weight matrix, replace model, use dynamic space measurement model, and replace variable calculation method to test the robustness of the regression results of the spatial measurement model. The specific process is as follows: First, we choose the economic distance matrix (W2) to substitute the geographic distance space weight matrix (W1) previously used in the regressions for robustness testing. Secondly, based on the GS2SLS model, this study uses the spatial lag term as an instrumental variable. The model is re-estimated based on W1, and the endogeneity of the model is tested. Third, re-estimate based on dynamic SAR, replacing the previously used first-order spatial lag term in the regression as the explanatory variable. Finally, the calculation method of manufacturing agglomeration is changed. And the spatial Gini coefficient is used to calculate the level of urban manufacturing agglomeration. Table 4 is the result of the robustness test regression. The spatial lag term of haze pollution remains significant. And the global Moran index reported by GS2SLS is also still substantial. In addition, the impact of manufacturing agglomeration on haze pollution shows an inverted U-shaped relationship. The above results show that the previous benchmark regression has strong robustness and once again verifies that hypothesis 1 is established.

Different regional tests
According to the regression results in Table 5, the agglomeration of manufacturing industries in the eastern and central regions has a significant inverted U-shaped effect on haze pollution. In contrast, manufacturing agglomeration in the west does not have a significantly impact on haze pollution. Thus, it shows that there are regional differences in the impact of the concentration of manufacturing industries on haze pollution in China, which confirms hypothesis 2. First of all, the eastern and central regions not only have abundant human capital but also have better conditions in terms of openness and infrastructure, which attract a large concentration of manufacturing enterprises. In the stage of rapid development of the manufacturing industry, energy consumption usually increases rapidly, and the manufacturing industry at this time has the characteristics of high pollution and high energy consumption. And the government adopts relatively lax environmental regulation standards to promote rapid economic development during this period. This causes manufacturing enterprises to ignore the investment in green technology innovation in the production process and attracts a large number of highly polluting enterprises to move in. As a result, the agglomeration of manufacturing industries exacerbates haze pollution. However, with the increasingly prominent negative externalities  of manufacturing agglomeration, the government gradually realizes the importance of the ecological environment. And at this stage, they began to implement strict environmental regulation policies, forcing enterprises to invest more in green technology innovation. In addition, the positive impact of the industrial structure upgrading effect and technology spillover effect of manufacturing agglomeration during this period also began to dominate, thus significantly reducing the problem of haze pollution. Secondly, the primary coefficient of manufacturing agglomeration in the eastern region is lower than in the central region. This may be due to the fact that in the economic development of the central region, it has undertaken many pollution-intensive industries from the east. At the same time, the development level of the manufacturing industry in the western region may not reach the carrying capacity of the environment, so the influence coefficient is not significant. Table 6 reports the moderating effect of urban form on the relationship between manufacturing agglomeration and haze pollution. In model (1), the interaction coefficient between the complexity of urban exterior shape and manufacturing agglomeration is positively significant at 5%. This means that as the external shape of the city develops in a complex way, it positively moderates the relationship between manufacturing agglomeration and haze pollution. And the concentration of manufacturing industries will further aggravate haze pollution. The product term coefficient of the external shape complexity of the city and the squared term of manufacturing agglomeration is negative and insignificant. This shows that when the manufacturing agglomeration level is on the right side of the inverted U-shaped curve, the urban exterior shape complexity has no moderating effect on the relationship between the agglomeration of manufacturing and haze pollution. The reason is that, on the one hand, the complex development of urban boundaries may lead to traffic congestion and poor flow of production factors. This will increase the production cost of manufacturing enterprises, which is not conducive to improving the level of manufacturing agglomeration. At this time, it is difficult for manufacturing companies to exert the agglomeration effect on the mitigation of haze pollution. On the other hand, complicating urban boundaries can adversely affect the permeability of cities. This has led to the accumulation of many pollutants emitted by the manufacturing industry in the city, thus increasing haze pollution.   The interaction term between urban compactness and manufacturing agglomeration shows insignificant positive characteristics. And the interaction term of the urban compactness and the squared term of manufacturing agglomeration is not significant in a negative way. This demonstrates that compact urban form does not have a moderating effect on the concentration of manufacturing and haze pollution. The reason is that the increased compactness of the city is conducive to a higher level of manufacturing agglomeration. On the one hand, the participation of many industrial enterprises will have a siphon effect on energy consumption, thus aggravating the local haze pollution level. However, increased urban compactness can also drive business and population agglomeration through compact land use and transportation. The agglomeration of production factors can promote the spillover effect of technology through sharing, matching, and learning mechanisms, thereby reducing haze pollution. Therefore, the compact development of urban form has simultaneously negative and positive externalities, and the two form a balanced development trend in the game process. This indicates that the moderating effect of urban form compactness on the agglomeration of manufacturing and haze pollution is not significant.

Moderating effect regression results
The interaction coefficient between urban fragmentation and manufacturing agglomeration is 2.9504, which is statistically significant and positive at the 1% level. This shows that increased urban fragmentation has a positive moderating effect between manufacturing agglomeration and haze pollution. Thus, manufacturing agglomeration will further intensify haze pollution. And the coefficient of the interaction term between urban fragmentation and the squared term of manufacturing agglomeration is − 0.7113, which is quite negative. When the level of manufacturing agglomeration is on the right side of the inverted U-shaped curve, urban fragmentation has a negative moderating effect on the relationship between the agglomeration of manufacturing and haze pollution. And it is due to the fact that the development of land fragmentation will promote the formation of multiple central cities. The increase in urban polycentricity will make the distribution of regional resources more decentralized. It is difficult to form an urban agglomeration economy, which makes it difficult for manufacturing enterprises to create economies of scale, thus promoting haze pollution. In addition, many industrial zones developed across regions will cause the emission of pollutants, mainly including smoke and dust, which induce haze pollution. At the same time, China is currently in rapid industrialization and urbanization. The environmental improvement effect of natural spaces such as rivers and grasslands is insufficient to reduce pollution emissions from manufacturing agglomeration significantly.
In summary, urban form has a moderating effect on the relationship between manufacturing agglomeration and haze pollution, which indicates that hypothesis 3 is established.

Threshold effect regression results
This paper uses Stata15.1 software to perform regression analysis on panel data. Under the Bootstrap method of repeated sampling 300 times, we get the test and calculation results through formula (5) and stata15.1 software.
According to Table 7, three urban forms characterized by urban exterior shape complexity (FD), urban compactness (LPI), and urban fragmentation (PD) are threshold variables. This study concludes that FD, LPI, and PD are significant at the 1% level, whether a single threshold test or a double threshold test (Table 7). In addition, the double threshold F statistic value of the three urban forms is the largest. Therefore, this study adopts a double threshold for analysis (Table 8).
According to Table 9, the impact of urban exterior shape complexity (FD) on haze pollution has a significant double threshold feature. When FD < 1.619, its influence coefficient on haze pollution is − 0.3511; when 1.619 < FD < 1.703, the regression coefficient is − 0.0459; when FD > 1.703, its corresponding regression coefficient becomes 0.0993. And the above three regression coefficients are statistically significant at the 1% level. It follows that FD's negative effect on haze pollution increases gradually with the complexity of the external shape of the city. When the threshold of 1.703 is crossed, FD shows a significant contribution to haze pollution. The reason is that the high complexity of construction land around the city will cause traffic congestion. And this will lead to slower vehicle speed and long driving time, which will accelerate exhaust emissions, resulting in severe haze pollution. In addition, the increase in FD will also cause poor communication between regions. It hinders companies from taking advantage of economies of scale through technology spillovers and sharing infrastructure, which is not beneficial in mitigating haze pollution. To sum up, the increase in the complexity of the external shape of the city will play a negative externality effect on aggravating the haze pollution.
The impact of urban compactness (LPI) on haze pollution has a significant double threshold feature. When LPI < 0.100, the impact coefficient of LPI on haze pollution is 0.2653 and passes the test of significance at the 1%; The impact coefficient of LPI on haze pollution is 0.1015 when 0.100 < LPI < 0.414, and its effect is significantly promoted; When LPI > 0.414, the regression coefficient of LPI on haze pollution becomes − 0.0490 and is significant at the 1% level. In conclusion, the contribution of LPI to haze pollution shows a decreasing trend. When LPI crosses the second threshold, LPI positively impacts suppressing haze pollution. And the higher the LPI value, the more compact the urban spatial form is. On the one hand, the improvement of urban compactness is conducive to giving full play to the positive externalities such as the economies of scale of manufacturing agglomeration and the centralized treatment of pollution. This can effectively alleviate the problem of haze pollution. On the other hand, the diversified development models and convenient urban transportation of compact cities will strengthen innovative talents and enterprises' attractiveness. And diversified agglomerations arising from the diversification of land use will also contribute to the spillover effect of technology. Thus, increased urban compactness will reduce air pollutant emissions by improving urban technology levels.
Urban fragmentation (PD) has a significant double threshold characteristic for haze pollution. When PD < 0.050, the influence coefficient of urban fragmentation on haze pollution is − 0.3402; When 0.050 < PD < 0.077, the influence coefficient of PD on haze pollution is − 0.1574; when PD > 0.077, the influence of PD on haze pollution changes from negative to positive, and the corresponding regression coefficient is 0.0571. The above three regression coefficients all pass the 1% significance test. To sum up, when the degree of urban fragmentation is in different ranges, its impact on haze pollution is quite different. When the urban fragmentation is low, PD positively affects haze pollution. However, as urban fragmentation develops, this positive effect has diminished. When the threshold value of 0.077 is crossed, the increase in PD will significantly promote haze pollution. When the urban fragmentation is low, manufacturing enterprises can play a positive externality role in the agglomeration effect. Specifically, improving the manufacturing agglomeration level is conducive to improving resource utilization efficiency, saving costs, and sharing pollution treatment facilities, hence effectively reducing haze pollution. However, the fragmented development of urban form may inhibit the increase in manufacturing agglomeration and reduce its various positive effects. In addition, the growth of urban fragmentation will promote the accumulation of population and economy in multiple core cities, resulting in an increase in pollution sources and pollution emissions. Therefore, urban form has a threshold effect on the impact of manufacturing agglomeration on haze pollution, and hypothesis 4 has been verified.

Conclusions and policy recommendations
This paper takes 283 prefecture-level cities from 2006 to 2016 in China as the research sample. We use the spatial econometric model, panel threshold model, and moderating effect model to study the relationship and mechanism among manufacturing agglomeration, urban form, and haze pollution. The main conclusions and policy recommendations of this study are as follows: (1) The impact of manufacturing agglomeration on haze pollution shows an inverted U-shaped relationship that promotes first and then inhibits. Therefore, we need to adopt targeted agglomeration policies according to different levels of manufacturing agglomeration. When manufacturing agglomeration is low, the government needs to introduce strict environmental regulation policies to force manufacturing enterprises to improve their technological innovation capabilities. In addition, we should enhance the spatial concentration of the manufacturing industry so that the level of manufacturing agglomeration can reach an ideal stage where it can play a significant emission reduction effect. When the degree of manufacturing agglomeration exceeds the critical value, the government must be vigilant about the crowding effect that may be brought about by excessive agglomeration of manufacturing industries. And the negative impact of excessive agglomeration on the economy and the environment can be mitigated through industrial transfer policies.
(2) The regional empirical results demonstrate that the impact of manufacturing agglomeration on haze pollution in eastern and central China shows an inverted U-shaped curve relationship of first promoting and then inhibiting. However, there is no inverted U-shaped relationship between the impacts of manufacturing agglomeration on haze pollution in the western region. On the one hand, the eastern and central regions should give full play to the advantages of agglomeration of innovative elements, accelerate the creation of world-class advanced industrial clusters, and promote the development of emerging industries. On the other hand, the western region should attract enterprises, capital, and technology based on its advantages. In this way, the western region can improve the level of manufacturing agglomeration and accelerate the process of new industrialization. At the same time, strict environmental protection reviews should be set up for newly entered enterprises and technologies to avoid the influx of polluting enterprises. (3) When the level of manufacturing agglomeration is on the left side of the inverted U-shaped curve, the complexity of urban exterior shape and urban fragmentation have a significant strengthening effect on the relationship between manufacturing agglomeration and haze pollution. However, the positive moderating effect of urban compactness on the two is not significant. When the level of manufacturing agglomeration crosses the inflection point, only urban fragmentation has a significant negative moderating effect on the relationship between manufacturing agglomeration and haze pollution. Based on this, we must first strengthen infrastructure construction, especially to expand to the city boundary. This can avoid reducing manufacturing agglomeration levels due to the crowding effect on the city boundary. Secondly, the government should establish a compact urban development concept in urban planning to avoid disorderly urban sprawl. Finally, local governments should speed up the flow of factors between cities and strengthen the division of labor and cooperation in the fields of industry and knowledge innovation, which can weaken the negative impact of urban fragmentation. (4) The urban form has a significant double threshold effect on haze pollution. With the increase in the complexity of the external shape of the city, it can effectively suppress the haze pollution, but the influence coefficient gradually weakens. When it crosses the double threshold, the urban exterior shape complexity shows a promoting influence on haze pollution. And the increase in urban compactness is conducive to reducing haze pollution. Moreover, the effect of increasing urban fragmentation on haze pollution shows the characteristics of first inhibition and then promotion. On the one hand, enhancing the ventilation capacity of cities can effectively alleviate haze pollution. Thus, the government can try to build urban ventilation corridors and adjust the pollution sources to locations with better air circulation. On the other hand, the government should take into account land expansion, infrastructure, and greening construction in urban construction, and pay attention to the coordinated promotion of economic and ecological benefits. Finally, promote the optimal allocation of factors among different cities, and guide the rational accumulation and distribution of industries.
Author contribution All the authors contributed extensively to the work presented in this paper. Jishi Wei: conceptualization, software, writing-original draft. Yunling Ye: literature review, grammar. Haichao Yu: conceptualization, methodology.
Funding This work was supported by the Postdoctoral Innovation Program of Chinese Academy of Social Sciences.

Declarations
Ethics approval Not applicable.

Consent to participate Not applicable.
Consent for publication All authors have read and agreed to the published version of the manuscript.

Competing interests
The authors declare no competing interests.