Examining the driving factors of industrial CO2 emissions in Chinese cities using geographically weighted regression model

The industrial sector is the sector with the largest CO2 emissions, and to reduce overall CO2 emissions, analysis of the impact factors holds significance. Based on the 2015 industrial CO2 emissions of 282 cities in China combined with economic and social data and a geographically weighted regression model, we analysed the characteristics of the spatial distribution of industrial CO2 emissions and the spatial heterogeneity of influencing factors. The results show that China's urban industrial CO2 emissions present a significant spatial agglomeration state that includes Shandong, Beijing, Tianjin, Shanghai, Zhejiang and Jiangsu, and the core of the coastal areas forms a high–high concentration; a low–low aggregation is formed in less developed areas such as Guizhou, Yunnan, Sichuan and Guangxi. The industrial scale, share of industry in gross domestic product, share of service industry in gross domestic product and level of economic development are factors that promote industrial CO2 emissions. The technological innovation, population density and government intervention are important factors that inhibit industrial CO2 emissions. Among them, the share of service industry in gross domestic product is the most significant factor in the increase of industrial CO2 emissions and the government intervention is the most significant factor in the reduction in industrial CO2 emissions. The effects of FDI, fixed asset investment and financial development are not significant. More importantly, the effects of influencing factors vary across cities, showing obvious spatial heterogeneity.


Introduction
Rising concentrations of greenhouse gases in the atmosphere are the main cause of global warming, and CO 2 is the greatest greenhouse gas emitted. According to data released by the International Energy Agency (IEA), the total global CO 2 emissions in 2018 amounted to approximately 33 billion tons, contributing up to 65% of the greenhouse effect. In particular, with the development of the global economy, the total amount of CO 2 emissions has been maintained at a high level. CO 2 emissions from the industrial sector are the major source of CO 2 emissions. Based on the estimation of IEA, in 2016, industrial CO 2 emissions (ICEs) accounted for 36% of energy CO 2 emissions globally, ranking first among all sectors and being far higher than construction (27%) and transportation (25%) (IEA 2018). At the same time, as China continues to industrialize, the industrial sector consistently ranks first among all sectors in China in terms of energy consumption and CO 2 emissions. According to the National Bureau of Statistics, in 2017, China's industrial energy consumption reached 294,480,400 tons of standard coal equivalent, accounting for 65.7% of the country's total energy consumption, indicating that the industrial sector relies too much on energy consumption. However, China's industrialization is far from complete. Scholars believe that China's industrialization is, overall, in a transitional stage from the middle to the later stage, with approximately 80% of the industrialization level basically having been completed, and there is still nearly 20% room for growth. Based on the experience of the industrialization process and CO 2 emissions of developed countries, as well as the judgement on influencing factors such as China's industrialization and urbanization process, the emissions of the industrial sector will increase gradually and reach a peak around 2040. Even under a low-carbon scenario, China's total industrial sector emissions will peak as soon as around 2030.
At present, the Chinese government is actively formulating and implementing various carbon emission reduction policies to deal with climate change and its impacts. Since 2013, seven regions, including Beijing, Shanghai, Tianjin, Hubei, Guangdong, Shenzhen and Chongqing, have established carbon markets and started substantial trading. Since its inception in 2013, the carbon markets have increasingly grown. By the end of August 2020, the cumulative turnover of the seven pilot carbon market quotas was 406 million tons, with a cumulative turnover of approximately $ 1.34 billion. A total of 2,837 key emitters, 1,082 noncompliant organizations and 11,169 natural persons participated in the pilot carbon markets. In 2016, nearly 200 countries signed the Paris Agreement, and China promised to peak its CO 2 emissions around 2030 and to reduce its CO 2 emissions per unit of GDP by 60%-65% compared to the 2005 level (Feng and Chen 2018). In the face of such a tough emission reduction target, the industrial sector, as the largest carbon emitter, should assume the primary responsibility for CO 2 emission reduction. Only by reducing the CO 2 emissions of the industrial sector can the overall goal of CO 2 emission reduction be achieved. Cities are the core area of relatively complete basic space units and industrial development in China, and they are also the most important source of energy consumption and CO 2 emissions. Excessive energy consumption and serious CO 2 emissions have become important factors hindering urban development (Glaeser and Kahn 2010). Therefore, studying the characteristics of the spatial distribution and influencing factors of ICEs in Chinese cities holds great significance for reducing CO 2 emissions and realizing a lowcarbon way of life.
In the existing research, it is believed that the factors influencing CO 2 emissions mainly include the economic growth, urbanization, energy structure, energy consumption, industrial structure, technological innovation, and so on. Some scholars believe that economic growth is positively correlated with CO 2 emissions. For example, Al-Mulali (2012) discussed the influencing factors of CO 2 emissions in 12 Middle Eastern countries, and the research results showed that GDP and the total trade volume were the main reasons for CO 2 emissions. Lin and Benjamin (2019) found that economic growth has a positive effect on CO 2 emissions in the long run. However, some scholars believe that economic growth can help reduce CO 2 emissions. For example, Qin et al. (2019) found that per capita GDP and the CO 2 emissions of northern China and the southeast coastal region present negative correlations, showing that a city's economic development has the potential to reduce carbon dioxide emissions. Some scholars used the environmental Kuznets curve (EKC) to analyse the relationship between economic growth and CO 2 emissions, arguing that CO 2 emissions show an inverted U-shaped relationship with per capita GDP (Ghazali and Ali 2019). For example, Liddle (2015) confirmed the inverted U-shaped relationship between CO 2 emissions and economic growth according to the EKC hypothesis. Other scholars believe that there is no connection between the two. For example, Freitas and Kaneko (2011) studied the relationship between CO 2 emissions and economic growth in the UK using a decoupling index of energy and environmental pressure, and they found that the relationship between the two was completely disconnected.
The impact of urbanization on CO 2 emissions is also a key research area (Wang et al. 2018). Generally, urbanization is considered to be the main driving force for the increase in CO 2 emissions. For example, Zhang et al. (2018) argued that the population and industrial scale brought by urban expansion have expanded the consumption of fossil fuels and that urbanization and industrialization have staged effects on ICEs and the emission intensity. Wang et al. (2019a) argued that different levels of urban development in China would obviously lead to different air quality pollution conditions. Energy structure is considered to be an important factor affecting CO 2 emissions. Increasing investment in clean energy and renewable energy (solar energy) can optimize the energy structure and thus promote carbon emission reduction. For example, Moutinho et al. (2015) found that the optimization of the energy mix would lead to a significant reduction in CO 2 emissions in four regions of Europe. Zhang et al. (2018) found that the coal-based energy structure was the main cause of high ICEs and that adjustment of the energy structure was a powerful measure for reducing environmental pollution and promoting ICEs reduction. The development of clean energy greatly improved the energy structure and reduced the consumption of fossil energy; additionally, adjustment of the energy structure could limit the increase in CO 2 emissions. In other words, adjusting the energy structure has a significant effect on reducing CO 2 emissions, which is of great significance for China's sustainable development strategy (Dong et al. 2018). Some scholars have found that CO 2 emissions are closely related to the industrial structure and have a long-term positive correlation with the proportion of the secondary industry because the secondary industry is considered to be a major user of energy and a major contributor to global CO 2 emissions (Lin and Benjamin 2019). Griffin et al. (2016) found that 19% of the greenhouse gas emissions in the UK came from the steel industry, which had the second largest amount of emissions; the steel industry consumed more energy than the service industry and had a higher carbon emission level. Generally, an increased proportion of the tertiary industry could significantly restrain CO 2 emissions. Most CO 2 emissions come from the secondary industry, especially the electricity, manufacturing and transportation and other industries (Kumbaroglu 2011). Technological innovation can improve energy utilization efficiency and reduce the energy intensity of industry, thus reducing the CO 2 emissions. For example, Dauda et al. (2019) suggested that improving energy technology and related equipment could greatly improve energy efficiency, thereby reducing urban carbon dioxide emissions. Rahman et al. (2017) argued that improving clean energy technologies was crucial for reducing carbon dioxide emissions and that industrial technological innovation should be encouraged, especially to solve the carbon emission problem in carbon-intensive industrial sectors. Therefore, to reduce CO 2 emissions, it is necessary to promote the intensive development of industries with low energy consumption, promote the extensive application of advanced technologies and eliminate energy-intensive industries (Lin and Benjamin 2019).
Other scholars believe that factors such as the industrial scale, population density, social fixed asset investment and foreign direct investment (FDI) have a certain impact on CO 2 emissions. For example, Dong et al. (2014) argued that industry and fossil fuel combustion were the main sources of carbon emissions. However, whether in the short term or in the long term, there may be a U-shaped relationship between industrial growth and CO 2 emissions (Abokyi et al. 2019). Population density is considered to be positively correlated with CO 2 emissions, and the impact of population in eastern China is significantly greater than that in central and western China. Liu et al. (2021) argued that the total resident population was an important factor to promote CO 2 emissions, and its influence degree had obvious spatial heterogeneity. The capital investment was an important factor to promote CO 2 emissions. Xu and Lin (2021) found that China's CO 2 emissions were mainly driven by domestic investment and economic growth. However, due to the difference of fixed asset investment, the impact of economic growth on CO 2 emissions varied by region. Wang et al. (2019b) found that FDI was significantly positively correlated with CO 2 emissions in Guangdong Province, while the impact of FDI in other regions was not significant. Kahia et al. (2019) also argued that could inhibit CO 2 emissions.
Taking 282 Chinese cities as samples, this paper uses exploratory spatial data analysis (ESDA) to analyse the characteristics of the spatial distribution of ICEs, and it uses a geographically weighted regression (GWR) model to study the spatial heterogeneity of the influencing factors of ICEs. This article makes three major contributions to the existing literature on ICEs. First, existing research basically takes a provincial-or national-level perspective to study ICEs, but data at the national and provincial scales fail to reveal the spatial differences at the city scale. The city is the main body of ICEs, and it is necessary to explore the relationship between ICEs and important measures of social and economic development. Therefore, this article discusses the differences in ICEs at the city level and compensates for the inadequacy of existing research at the ground level and above the city level. Second, the territory of China is vast in size, and cities are interconnected and influenced by each other in geographical space. However, the spatial correlation between cities has been ignored in most existing studies. This paper adopts ESDA to reveal the spatial clustering characteristics of urban ICEs. Third, the existing literature mostly adopts an ordinary least squares (OLS) panel regression model to analyse the influencing factors of ICEs. The model of the estimated coefficient can reflect the influence extent as a whole and cannot reflect the differences in influence between cities. This paper introduces a GWR model to consider the parameters of local spatial heterogeneity and the industrial emissions of social and economic factors based on a more meticulous and comprehensive understanding of spatial scales.

Global spatial autocorrelation
Spatial autocorrelation test can accurately reflect the characteristics of the spatial distribution and agglomeration of ICEs in Chinese cities. Global spatial autocorrelation can reflect the agglomeration characteristics of an economic variable in the whole space. It is represented by Moran's I, and the Z score is calculated to judge the significance of the result. Moran's I is used in this paper to describe the global spatial correlation of ICEs in Chinese cities. The formula is as follows: where n is the number of cities, w ij is the spatial weight matrix, x i and x j are the ICEs of cities i and j , respectively, and x is the average industrial emissions of each city. The value range of Moran's I is [− 1, 1]. When Moran's I > 0, it represents positive spatial correlation. The greater the value is, the more obvious the spatial correlation is. When Moran's I < 0, it represents negative spatial correlation. The smaller the value is, the greater the spatial difference. When Moran's I = 0, it represents the randomness of space.

Local spatial autocorrelation
Global spatial autocorrelation is a global indicator that measures spatial autocorrelation. It reflects only the differences in spatial mean values and may ignore the atypical characteristics of ICEs in some cities in local areas (Elhorst 2012).
Local spatial autocorrelation test can be used to test the local agglomeration characteristics of ICEs. The formula is as follows: where reflects the correlation between a city and its surrounding cities, and the other indicators have the same meaning as those in formula (1). When represents a positive correlation between the ICEs of city i and those of its surrounding cities; that is, the CO 2 emissions present high-high (H-H) or low-low (L-L) agglomeration. When means that the ICEs of city i are negatively correlated with those of its surrounding cities; that is, CO 2 emissions present H-H or L-L agglomeration.

Geographically weighted regression model
In traditional regression analysis, the OLS method is commonly used, but the regression coefficient obtained by an OLS model is the average value of the whole study area, and local regional characteristics cannot be obtained. A GWR model is a local form of a linear regression model that is used to analyse spatial variation relationships. When analysing influencing factors, a GWR model sufficiently considers the spatial effects and can reflect the heterogeneity of the regression coefficients in different spaces. Due to the vast size of China, there are huge differences in industrial development and ICEs between cities. Therefore, a GWR model is adopted in this paper to analyse the influencing factors of ICEs. The formula is as follows: where Y i represents the ICEs value of city i , u i , v i ) represents the latitude and longitude coordinates of are coordinate functions, and i represents the residual term. The GWR tool of Arcgis10.2 is used in this paper. Gaussian function is used to construct the spatial weight matrix, and the Akaike Information Criterion (AIC) method is used to select the optimal bandwidth, and then regression calculation is carried out. (3)

Explained variable: industrial CO 2 emissions (ICEs)
In this paper, data from the China High Resolution Emission Gridded Database (CHRED), composed of 76 units and 137 researchers, are used. Covering the enterprise, industry and urban levels, these data on the levels of city greenhouse gas emissions from activities are collected, sorted and cleaned. At the same time, they are used to carry out a large number of cross-validation and data analyses. A data set on the 2015 greenhouse gas emissions of Chinese cities is ultimately established. The CHRED references the international mainstream bottom-up spatial analysis method. It combines the actual situation in China and the characteristics of the data. The spatial analysis method is based on point emission sources from the bottom-up (industrial enterprises, sewage treatment plants, landfills, livestock and poultry farms, coal mining, water transport ships, etc.) and other line source (traffic source) and nonpoint source (such as agriculture, life source) data, as well as other greenhouse emissions grid data at a 1 km resolution and methods of spatial data accuracy and uncertainty analysis. The spatial position accuracy of point source data is controlled by two methods: the longitude and latitude data on emission sources and spatial coordinates, and address matching verification is carried out based on application programming interface (API) geocoding technology. CHRED data highlight the spatialization and spatial distribution pattern of emissions and emphasize the spatial accuracy of emissions data.

Explanatory variables
On the basis of previous studies, this study takes ten indicators, i.e. the industrial scale (IS), the share of industry in GDP (IR), the share of the service industry in GDP (RGDP) (TR), the economic development level, technology innovation (TI), population density (PD), foreign direct investment (FDI), social investment in fixed assets (FA), government intervention (FE) and financial development (FD), as influencing factors of China's urban ICEs. These indicators and their impact on ICEs can be simply described as follows: Industrial scale is expressed by the total industrial output value of each city. In general, the larger the industrial scale is, the higher the level of industrial output will be, and the higher the ICEs will naturally be. The share of industry in GDP is expressed by the proportion of industrial added value in GDP, reflecting the impact of the industrial structure on ICEs. The higher the proportion of the secondary industry is, especially the higher the proportion of heavy industry, the higher the ICEs will be. The share of the service industry in GDP uses the proportion of the added value of the service industry to GDP, reflecting the effects of ICEs from the upgrading of the industrial structure of the industry, especially the modern service industry relative to traditional industry, The service sector is characterized by low energy consumption, low pollution and the accelerated development of finance, tourism, education, culture and other industries that can reduce ICEs. Economic development (RGDP) is measured by per capita GDP. According to the EKC hypothesis, due to different levels of economic growth, economic growth may increase or reduce ICEs, and there may be an inverted U-shaped relationship between the two. Technology innovation (TI), represented by the number of patents granted per 10,000 people, improves energy efficiency and promotes the development of high-tech industries, thereby effectively reducing ICEs. Population density is represented by the number of people per square kilometre. Human activities will result in an increase in ICEs. In general, population density is positively correlated with ICEs, but population agglomeration can also bring energy-intensive use, thus reducing ICEs. FDI, expressed in terms of the actual use of foreign capital as a share of GDP, brings advanced production technologies and managerial expertise, promotes technological progress in host-country industries, and thus increases and reduces ICEs. Social fixed asset investment is expressed by the per capita fixed asset investment of the whole society. When investment is inclined towards industrial upgrading, it can bring advanced production equipment and technology to enterprises, thus reducing ICEs. Government intervention is expressed by the share of fiscal expenditure in GDP, and it reflects the government's intervention in economic activities. The government uses fiscal allocations to encourage enterprises to increase their research and development (R&D) and innovation, improve their energy efficiency and reduce their ICEs. In addition, China's fiscal decentralization system enables local officials to control the power of fiscal expenditure and resource allocation. The existing performance appraisal mechanism, with GDP as the core, also enables local governments to relax environmental regulations on ICEs and to even support high-pollution industries, leading to an increase in ICEs. Financial Development is expressed by the proportion of the total amount of deposits and loans in GDP. Efficient financial intermediaries are usually more conducive to the loan activities of consumers, making it easier for them to buy automobiles, refrigerators, air conditioners and other big-box goods and increasing the amount of ICEs. At the same time, the development of the financial market helps enterprises reduce their financing costs, increase their financing channels and optimize the structure of their assets and liabilities to introduce new equipment, increase technological innovation and thus reduce ICEs.

Characteristics of the spatial distribution
The spatial distribution of ICEs in Chinese cities is shown in Fig. 1

Global and local correlation
Based on formulas (1) and (2)

OLS regression results
First, an OLS model was used for parameter estimation, and the results are shown in Table 2. The results of the ANOVA F test show that the model is strongly significant, and the variance inflation factor (VIF) values are less than 10, indicating that there is no serious collinearity problem in the model. The regression results show that the industrial scale, the share of industry in GDP, the share of service industry in GDP and the level of economic development can significantly promote urban ICEs, and for each 1% increase in these four factors, the ICEs increase by 0.66%, 1.45%, 1.86% and 1.13%, respectively. Thus, the share of service industry in GDP is the most significant factor in the increase of ICEs. At present, China's industrial output is as high as 39.0% of GDP, and industry is the sector with the largest CO 2 emissions. With the expansion of the industrial scale and the increase of the share of industry in GDP, the input of raw materials will increase, and in particular, energy consumption will greatly increase, which will naturally result in an increase in ICEs. The coefficient of the service industry is not consistent with expectations, which may be due to the high proportion of China's transportation industry in the service  industry. In particular, in recent years, the rapid development of China's e-commerce and logistics industry has promoted the development of the transportation industry, which is the industry with the second largest ICEs after industry. With improved economic development, people will pursue better living standards, and the demand for household appliances, computers, automobiles and other industrial supplies will greatly increase, which will indirectly promote ICEs. At the same time, the figure shows that China is still on the left side of the EKC and has not reached the critical point of economic development to improve the environment. Technological innovation, population density and government intervention are all important reasons for restraining ICEs; for every 1% rise in the four factors, accordingly the ICEs are reduced by 0.10%, 0.25% and 0.38%, respectively. Thus, the government intervention is the most significant factor in the reduction in ICEs. Technological innovation, especially in energy conservation, environmental protection and low-carbon technologies, can improve energy efficiency and thus reduce carbon dioxide emissions. The higher the population density is, the more intensive the energy use, potentially reducing ICEs. An increase in government spending could encourage industrial enterprises to carry out technological innovations or to introduce advanced production lines to improve their energy efficiency and reduce their ICEs. The effects of FDI, fixed asset investment and financial development on ICEs are not significant.

GWR regression results
In this paper, a GWR model is used to analyse the local heterogeneity of factors affecting ICEs. The R 2 values are shown in Fig. 3. These values vary between 0.3731 and 0.8186, and there are significant differences in the fitting degree of different cities. The R 2 basically decreases from the marginal areas of Northeast China, Northwest China and the coast of Eastern China to Central China. The GWR model had the best fitting effect in Xinjiang, Gansu, Heilongjiang and Jilin, while the vast central region had the worst fitting effect. This result indicates that the relationship between the driving factors and ICEs is better reflected by the regression model in Xinjiang, Gansu, Heilongjiang and Jilin provinces. The GWR model can calculate the coefficient of the influence of the explanatory variables on the explained variables in each city, making it possible to more clearly analyse the local characteristics of the coefficient of influence. The descriptive statistics of the coefficient of the influence for each variable on ICEs is shown in Table 3. The standard deviation of the estimated coefficient varies over a wide range, with the minimum (FD) value being 0.026 and the maximum (TI) value being 0.147. The greater the variation range of the coefficient is, the greater the spatial difference in the influence extent of various factors on ICEs, which better reflects the superiority of the GWR model. Table 3 also calculates the proportion of the number of cities with The article also describes the spatial characteristics of the estimated coefficients of each influencing factor, as shown in Fig. 4. Cities whose estimated coefficients are significant at the 10% level are also marked with slashes. In all cities, the industrial scale, the share of industry in GDP and the share of service industry in GDP all promote ICEs, while technological innovation, population density and social fixed asset investment all restrain ICEs. In contrast, the level of economic development, FDI, financial development and government intervention have a two-way impact on ICEs, and there are significant differences in different cities. which is inconsistent with the results estimated by the OLS regression model. These results are different from those of OLS regression model. The specific analysis is as follows.
The coefficient of the IS is all positive and significant in 100% of the cities under study, indicating that industrialscale expansion significantly promotes ICEs, which is basically consistent with the OLS model estimation results. Among them, in cities such as Gansu and Xinjiang, the influence of the industrial scale on ICEs is the highest, while the influence on the ICEs of coastal areas such as Guangxi, Guangdong and Fujian is relatively low. Overall, the coefficient of influence for northern cities is obviously higher than that for southern cities. One possible reason for this result is that in recent years, with upgrading of industrial structure, energy-intensive enterprises in the eastern coastal areas have gradually moved to western regions. Gansu and Xinjiang are the two major energy and chemical bases in China, and the scale of energy-intensive industries continues to expand; thus, the impact on ICEs is higher than that in other regions.
The coefficient of the IR is positive and significant in 86.01% of the cities under study, and the influence extent gradually decreases from west to east. Among them, the share of industry in GDP has the highest influence on ICEs in Inner Mongolia, Gansu, Ningxia, Shanxi, Shaanxi and other western regions. This result may be because Western China is rich in energy resources, and the energy-intensive industries related to these resources, such as the energy and chemical industry, the equipment manufacturing industry and the nonferrous metal smelting and processing industry, are relatively developed. Therefore, an increase in the share of industry in GDP will promote ICEs. In addition, to catch up and go beyond, western cities tend to weaken their environmental regulations when undertaking industrial transfer from eastern regions, and they introduce some high-energyconsuming enterprises into local areas. Therefore, with the development of industry, especially heavy industry, ICEs will inevitably increase.
The coefficient of the TR is positive and significant in 94.76% of cities, indicating that an increase in the share of the service industry in GDP will promote ICEs, and the influence extent successively decreases from west to east. In western cities such as those in Inner Mongolia, Gansu and Shaanxi, the share of the service industry in GDP has the highest impact on ICEs, while in eastern cities, the impact is much smaller. One possible reason is related to the internal structure of the service industry. In Western China, lowend service industries such as transportation, storage and logistics, accommodation and catering account for a higher proportion. In Eastern China, financial, legal consulting, real estate and other services are more developed. Therefore, an increase in the share of the service industry in GDP in Western China will cause more ICEs compared to Eastern China.
The level of economic development has a two-way effect on ICEs. In Jiangsu, Zhejiang, Fujian, Guangdong, Hunan and other eastern coastal cities, the coefficient is significantly negative, showing that economic development can restrain ICEs, which conforms to the law described in the right part of the EKC. The main reason may be that the economic level of the eastern coastal areas has reached a certain degree and exceeded a certain critical point. Therefore, people will pursue better living standards and have a higher demand for green and low-carbon products, the government will have strong environmental regulations, and enterprises will tend to adopt green and low-carbon technologies; thus, ICEs can be restrained. The impact coefficient of RGDP in other cities is not significant, indicating that economic development in these cities has difficulty curbing ICEs. The coefficient of TI is negative and significant in 93.36% of the cities under study, and the influence extent increases successively from east to west. In western regions such as Inner Mongolia, Gansu, Ningxia and Qinghai, technology innovation has the highest impact on ICEs, while its impact is relatively low in Heilongjiang, Hainan and other places. This result may be because of the rich resources in Western China; the high proportion of energy-intensive industries, such as flue gas enterprises; and the comprehensive utilization of solid waste, the energy savings and the reduced consumption in key areas, such as the independent development of new technologies, new processes and new equipment, which promote the development of low-carbon technology, improve the efficiency of energy utilization and curb ICEs.
The impact coefficient of PD is negative and significant in only 9.79% of cities, indicating that PD can suppress ICEs in very few cities, which are mainly concentrated in western regions such as Ningxia, Gansu, Shaanxi and Inner Mongolia. One possible reason is that Western China is sparsely populated, with a low population density, and CO 2 emissions mainly come from the bioenergy in rural. With the advancement of urbanization and the increase in population density, energy is utilized intensively, and the utilization efficiency is greatly improved; thus, ICEs will also be reduced accordingly.
FDI has a two-way impact on ICEs. The impact is significantly positive in Hainan, Guangdong, Fujian and other south-eastern regions, indicating that FDI can promote ICEs. However, it is significantly negative in Heilongjiang, Jilin, Liaoning and other north-eastern regions, indicating that FDI can curb ICEs. This means that FDI has an impact on ICEs only in specific cities and that OLS models cannot reveal this feature. This may be because in the eastern coastal areas with developed industries, FDI inflows are mainly to industrial enterprises; thus, an increase in FDI will promote ICEs. In Northeast China, traditional industries, with high energy consumption and backward technology, are dominant. FDI inflow can bring advanced technology and management experience, promote the technological progress of local industrial enterprises and thus reduce ICEs.
Social fixed asset investment has a two-way impact on ICEs. The coefficient of influence for Inner Mongolia, Ningxia, Gansu and other northwest regions is significantly negative, while in most other cities, it is not significant. This result shows that social fixed asset investment is an important factor in restraining ICEs in Northwest China, possibly because the industrial enterprises in Northwest China are at a critical stage of eliminating their backward production capacity and transforming and upgrading. A large amount of social fixed asset investment can bring advanced production equipment and technology to enterprises, which may increase their economic income and give them more funds for environmental governance, thereby reducing their ICEs.
The coefficient of FE is negative and significant only in 22.73% of the cities under study, which are mainly located in eastern regions such as Shanghai, Shandong, Zhejiang, Anhui and Jiangsu. The impact of government intervention is not significant in most other cities. This result suggests that government intervention is currently able to curb ICEs in Eastern China. One possible reason is that the eastern region has a relatively developed economy, and local governments have relatively high financial revenues. They typically use tax incentives and subsidy policies to encourage enterprises to innovate. Under the encouragement of preferential policies from the government, enterprises will invest more money in scientific innovation and technological innovation and introduce green and low-carbon technologies to reduce their CO 2 emissions.
The influence of FD on ICEs is negative and significant only in 27.97% of the cities under study. This result shows that financial development could inhibit ICEs. The cities in which financial development has an impact are mainly in Chongqing, Hunan, Guangdong, Guangxi, Yunnan, southern Sichuan and Guizhou; the influence of financial development is not significant in most other cities. One possible reason is that the degree of regional financial development is higher, and therefore, financial institutions provide financial support for green credit business, technological transformation and innovation, especially in terms of green technology R&D and promotion. Financial institutions will give preferential access to credit, low interest rates and other support, improving the hatching rate and survival rate and improving technology innovation to reduce ICEs.

Conclusion
In this paper, the spatial and temporal distribution characteristics of ICEs are discussed by ESDA method, and the spatial heterogeneity of influencing factors of ICEs is studied by GWR model. The main conclusions are as follows.
First, China's urban ICEs show significant spatial clustering characteristics. Areas with high ICEs are mainly concentrated in Chongqing, Tianjin, Shandong, Shanghai, Jiangsu, Zhejiang, Hebei, Shanxi, Yinchuan, Inner Mongolia, Shaanxi and other provinces, while low-emission areas are mainly distributed in less developed cities in Central and Western China, such as Sichuan, Gansu, Guizhou and Yunnan. Therefore, coordinated emission reduction policies should be formulated for areas with high ICEs.
Second, regarding ICEs, based on the influencing factors of the overall situation, industry scale, the share of industry in GDP, the share of service industry in GDP and the level of economic development play a major role in promoting ICEs, while the technology innovation, population density, government intervention play a major role in inhibiting ICEs. Among them, the share of service industry in GDP is the most significant factor in the increase in ICEs and the government intervention is the most significant factor in the reduction in ICEs. The effects of FDI, fixed asset investment and financial development on ICEs are not significant.
Third, the influencing factors of ICEs have spatial heterogeneity. The influence of the industrial scale on ICEs is always positive and significant in all cities evaluated, and the influence extent gradually decreases from north to south. The influence of the share of industry in GDP on ICEs is always positive and significant in 86.01% of the cities under study, and the influence extent gradually decreases from west to east. The coefficient of the share of service industry in GDP on ICEs is positive and significant in 94.76% of the cities, and the influence extent successively decreases from west to east. The economic development level has a two-way impact on ICEs. This impact is significant in 53.8% of the cities, and the inhibitory effect of the economic development level is the strongest in eastern coastal cities. The impact of technology innovation on ICEs is negative and significant in 93.36% of the cities, and the influence extent increases from east to west. The impact of population density on ICEs is negative and significant in only 9.79% of the cities, which were mainly concentrated in Ningxia, Gansu, Shaanxi, Shanxi, Hebei and other regions. FDI has a two-way impact on ICEs, and this impact is significantly positive in Hainan, Guangdong, Fujian and other south-eastern cities and significantly negative in Heilongjiang, Jilin, Liaoning, Hebei, Inner Mongolia and other north-eastern regions. Social fixed asset investment has a two-way impact on ICEs. This impact is significant in only 5.24% of the cities, which are mainly located in Inner Mongolia, Shaanxi, Ningxia and other north-western regions, while in most other cities, the impact of social fixed asset investment is not significant. Government intervention has a two-way impact on ICEs that is significant in only 22.73% of the cities, which are located in eastern regions such as Shanghai, Shandong, Zhejiang, Anhui, Jiangsu and Fujian. The impact of government intervention is not significant in most other cities. Financial development has a negative impact on ICEs, and this impact is in 27.97% of the cities, which are located in southern regions such as Chongqing, Hunan, Guangdong, Guangxi, Yunnan, Sichuan and Guizhou.

Policy implications
According to the above conclusions, the influences of different factors on ICEs are spatially heterogeneous. Therefore, policies should be made according to the development characteristics of cities and according to local conditions. Therefore, the following suggestions are proposed: the industrial production scale and excessive proportion of industry are still one of the main sources of ICEs, and local governments should strictly control the industrial structure, especially the heavy industry output value, and maintain it within a reasonable scope. The choice of low-pollution, low-emission and green industries should encourage enterprises in Central China to optimize and upgrade the industrial structure, reduce the proportion of the secondary industry in GDP and actively develop the low-carbon economy. On the premise of encouraging industrial structure optimization, local governments should focus on developing high-end low-carbon service industries, such as finance, medical care, culture and entertainment, and information technology, while vigorously developing the service industry to avoid a shift in ICEs from industry to the service industry. As economic development has a significant restraining effect on ICEs, Central and Western China should vigorously develop their economy, improve people's living standards, alleviate the contradiction between people's growing demand for a better life and economic development and promote people to cultivate green and low-carbon living habits. Technological innovation should be accelerated, with emphasis on the application of new energy technologies and low-carbon technologies, flue gas treatment and the comprehensive utilization of solid waste. Market mechanisms should be introduced to ensure the application of low-carbon technologies and to improve energy utilization efficiency. The government should encourage an appropriate increase in the population density of western cities and guide the population to flow into Western China through industrial transfer, talent introduction and other measures because this change in the population structure can not only relieve the environmental pressure brought by the population in Eastern China but also promote the balanced development of urbanization in Central and Western China. For the eastern coastal areas, foreign capital should be encouraged. When introducing foreign capital, attention should be paid to the investment structure of foreign investors, and they should be guided to flow to high-tech industries and service industries. For less developed regions such as Northeast China and Northwest China, the business environment should be further optimized to attract high-tech foreign-funded enterprises, and local enterprises should be encouraged to actively introduce advanced foreign technologies and conduct cooperations and collaborative innovation with foreign high-tech enterprises. Western China should increase the scale of social fixed asset investment, pay attention to optimizing the investment structure, accelerate the transformation and upgrading of traditional industries in industrial investment, and encourage enterprises to introduce advanced production equipment and technological transformation. We will quickly promote investment in medium-and high-end manufacturing, reduce overcapacity and high energy consumption and improve the quality of industrial investment. In Central and Western China, we should increase government financial expenditure, especially on science and technology and environmental governance. We should also set up innovation funds, use tax and fiscal subsidy policies, guide and encourage enterprises to adopt green and low-carbon technologies, and increase R&D. We should strengthen the financial development in the northern cities of China, give full play to the role of financial capital in allocating resources by means of the market, encourage financial institutions to provide financial support for innovative and green enterprises and even offer certain preferential interest rates to alleviate the financing pressure of enterprises. For zombie enterprises with an excessive production capacity and severe pollution, loans should be prohibited.
This study still has some limitations. First, limited by the integrity and difficulty of urban data acquisition, other factors such as energy efficiency, energy consumption structure, environmental regulation and carbon trading discussed in other literatures are not included in the model, which can be used as the choice for subsequent research. Second, emission reduction policies based on many driving factors need to be further refined in combination with the actual situation and economic development level of each city.