Heterogeneity and Decomposition Analysis of Manufacturing Carbon Dioxide Emissions in China’s Post-Industrial Innovative Megacity Shenzhen

Effectively reducing manufacturing carbon dioxide (CO2) emissions is a vital strategy for China to curb its rapidly rising carbon footprint. Features of such a reduction in manufacturing include an increase in the share of high-tech manufacturing and optimization of the energy consumption structure. This study aims to analyze the case of Shenzhen, a unique post-industrial innovative manufacturing megacity, for its leading experience in China’s manufacturing transition. Disaggregated manufacturing emissions data of Shenzhen, including 27 sub-sectors in four categories, were collected, and driving factors were identified by the logarithmic mean Divisia index (LMDI) method. The results suggest that: (1) CO2 emissions from Shenzhen’s manufacturing show a phased difference between 2008–2012 and 2012–2020. CO2 emissions embodied in electricity consumption have increased by over 30% in the former period and have remained stable at a high level of over 90%. (2) Significant heterogeneity of CO2 emissions in various manufacturing sectors is revealed, with the largest emissions sources being factories that make communication equipment, computers, and other electronic equipment. (3) Lower carbon intensity is the primary factor in reducing CO2 emissions, while the economic activity effect of manufacturing possesses a stimulating impact. (4) The marginal impact of restructuring on CO2 emissions is insignificant since the manufacturing and energy structures of Shenzhen have been upgraded to a low carbon level. Therefore, strengthening the power saving management and improving the energy efficiency of the manufacturing, rather than optimizing the manufacturing and final energy structures, will be a necessary potential solution to the problem of how to reduce CO2 emissions in Shenzhen’s manufacturing.


Introduction
The national carbon peak and carbon neutrality goals ("Dual Carbon Target") are the extensive, far-reaching systemic transition of China's economy and society. With the profound impact of global climate change on the ecological, economic, and social development of human society, a global consensus has formed around the effective management of energy consumption and greenhouse gas (GHG) emissions. Therefore, controlling and reducing CO 2 emissions has become a common goal for academia, industry, government and the international community. After rapid economic growth and a lifestyle evolution in recent decades, China has become the world's largest CO 2 emitter and the leading energy consumer [1]. In 2009, the Chinese government committed to lower CO 2 emissions per unit of GDP (carbon intensity) by 40% to 45% from the 2005 level by 2020. The commitment was achieved in advance. Furthermore, China has also announced a target of reducing its carbon intensity by 60-65% from the 2005 level of reaching the peak of its CO 2 emissions are usually lagging and discontinuous, and there is a lack of input-output tables data in many regions, which limits its wide-scale use. Based on production theory, the PDA method has been gradually applied in the research area of environment and energy after its first presentation by Zhou and Ang [8]. Although some production technology-related components are included in PDA, it cannot give structural effects and its decomposition results can be affected by the number of entities considered. In terms of implementation, the use of PDA to solve the problems associated with linear programming is more complicated than the use of IDA/SDA with algebraic operations. One of the main reasons for choosing the IDA method is that its implementation process is relatively straightforward, the requirements for data in IDA are not high, and its results can be compared with previous studies. With the improvement of IDA, it can overcome the incomplete decomposition defects of the previous decomposition methods and can effectively solve the problem of zero value and negative value of data summary [9]. The biggest advantage of this method is that it can trace the cause of the change of dependent variables and determine the deep factors through the decomposition of subindustries, to provide the basis for the formulation of practical and reliable policy measures. This method has been widely used in the fields of energy economics and environmental economics. However, the IDA method has various sub-methods. Ang has provided a detailed analysis and recommends the use of LMDI-I [10]. Since it has no residuals, this method is easy to use, and the results are easy to interpret. This article uses the method recommended by Ang [10]. LMDI-I has two kinds of decomposition forms, namely additive form and multiplicative form. The only difference between them is the form of representation. To make it easier to understand, this article uses the additive form of LMDI-I.

Driving Factors of CO 2 Emission Change
Representative studies about CO 2 emission decomposition of the LMDI method are summarized in Table 1. These four factors, namely total production activity, industrial structure, energy structure, and energy/carbon intensity, have been the focus of existing studies. The majority of the literature about manufacturing CO 2 emissions focuses on developed European countries. For example, Akbostanci et al. have revealed that coal is the decisive factor and that the iron and steel industry is the most polluting portion of Turkey's manufacturing sector [11]. Hammond and Norman have found that the nonenergy-intensive subsector of the UK's manufacturing sector has a greater reduction than the energy-intensive subsector because of the greater improvement in energy intensity [12]. Jeong and Kim decomposed the GHG emissions of the UK's manufacturing sector and found that the structure and intensity effects play roles in it [13]. In the developing world, Román et al. have shown that the key drivers of emissions in Colombia during 1990-2012 were the effects of wealth and population increase, and that the improvement in energy efficiency failed to offset the increasing effect of other factors [14]. Mousavi et al. have indicated that the major driver of Iran's CO 2 emissions is its increased consumption, and that the effect of technology-related improvements is very weak [15]. At present, China's CO 2 emissions have attracted the attention of many researchers, but most of them have focused on the national or provincial level. Liu et al. have shown that, in China, the ferrous metal smelting and rolling industry is the largest source of CO 2 emissions, followed by chemical raw materials and products and then non-metallic minerals [5]. Wang and Feng have found that the secondary industry accounts for approximately 80% of total CO 2 emissions with provincial panel data of three sectors in China [16]; however, the industrial structure effect and energy structure effect varied considerably over the years without showing clear trends [16]. Wu et al. found that, in Inner Mongolia, China, energy-intensive industries, such as energy production and processing, are the main sources of industrial emissions, and coal is the most important source of energy in the energy mix [17]. Wang et al. have indicated that the energy intensity effect plays a dominant role in promoting the decoupling status between CO 2 emissions and industrial growth in Taiwan of China [18]. Feng et al. have shown that economic and demographic growth has positive effects on the CO 2 emissions of Guangdong province [19]. The existing studies about manufacturing emissions mostly conducted index decomposition analysis at the national level or the provincial level. Only a few studies have investigated the emissions at the city level [20][21][22] and have focused on Beijing, Shanghai, Tianjin, Chongqing, and Zhuhai in China [23,24]. Gu et al. studied the determinants of Shanghai's CO 2 emission change during the period 1995-2015 and identified that energy intensity is the main factor for carbon mitigation, followed by economic structure, residential energy intensity, and emission coefficients [25]. Liu Note: "×" indicates that the document does not belong to this option; " √ " means the item is in the document; "+" and "−" represent the positive and negative effects, respectively.
In summary, we find that a large number of existing studies have the following deficiencies. First of all, in terms of the research area, most of the current related literature is focused on the national or inter-provincial level of China, or some heavy industrial cities, but rarely study innovative megacities such as Shenzhen. Shenzhen uses less fossil energy but more electricity to generate and create more economic output, and has not received enough research attention. Second, in terms of the research industry, most studies on the energy structure and emission intensity are concentrated on the thermal power generation industry, the transportation industry, or the whole economic sector. Furthermore, those city-level studies only considered the aggregate-level emissions when performing the decomposition analysis and none of them studied the inner structure of manufacturing in an innovative megacity. It is necessary and important to study the decomposing analysis with disaggregated manufacturing data. Third, many empirical studies on the drivers of industrial CO 2 emissions only consider direct CO 2 emissions from the consumption of fixed energy sources (e.g., coal, oil, fuel oil, diesel, gasoline, natural gas, etc.), without taking into account indirect emissions from electricity use [29]. This means that the actual CO 2 emissions may have been greatly underestimated. Fourth, many previous studies have neglected the decomposition of different stages. Consistent results about the emission drivers have also not been achieved yet. It is necessary and important to decompose the drivers of CO 2 emissions by different stages according to the emission characteristics of manufacturing in the megacity.

Quantification of Carbon Dioxide Emissions
According to Shenzhen's statistical yearbook, raw coal, crude, gasoline, kerosene, diesel oil, fuel oil, liquid petrol gas (LPG), natural gas and electricity are the nine major energies used in the manufacturing industry in Shenzhen. Energy activities account for more than 95% of Shenzhen's GHG emissions [30]. Therefore, almost all sources of CO 2 emissions from manufacturing are covered in this study. The applied method of CO 2 emission quantification followed the guidelines of the Intergovernmental Panel on Climate Change (IPCC) [31] and the Guidelines for Provincial Inventory of Greenhouse Gas Emissions. The CO 2 emissions arising from fossil energy combustion were calculated by the parameters of their energy calorific value and carbon emission factor [5,29,32]. Referring to other researchers [5,28], we calculate the CO 2 emissions coefficient of each form of fossil energy, which is a product of net calorific value, carbon content, oxidation rate, and 44/12. It is then directly multiplied by the amount of physical energy consumption to get the direct CO 2 emissions. In addition to the CO 2 emissions from the combustion of fossil energy, this paper also considers the CO 2 emissions embodied in the electricity consumption of manufacturing. We calculate the CO 2 emissions of each kind of energy in each sector and then add them together to obtain the total emissions of each sector in manufacturing. The formula we use to calculate CO 2 emissions of each manufacturing sector is shown in Formula (1). (1) where: i: the type of energy from 1 to 9 (See Table 2) j: the sub-sector of manufacturing from 1 to 27 C j : CO 2 emissions of sector j in the manufacturing sector AD i j : the physical amount of consumption of the i type of energy in sector j of manufacturing C ij : the CO 2 emissions of the i type of energy in sector j of manufacturing LCV i : the lower calorific value of the i type of energy CF i : the carbon content of the i type of energy O i : the oxidation rate of the i type of energy f i : the CO 2 emission coefficient of the i type of energy 44/12: the relative atomic mass of CO 2 and C Generally, the CO 2 emission coefficient of fossil energy combustion will not change significantly in a short period of time. The calculated CO 2 emission coefficient for various energy sources in this study are shown in Table 2. LCV i , CF i , and O i are from the Guidelines for Provincial Greenhouse Gas Inventory (provided by the National Development and Reform Commission of China) and the appendix of the "China Energy Statistical Yearbook 2018". The CO 2 emission coefficients of main fossil energy from IPCC [31] are also presented in Table 2. Since the electricity of Shenzhen is mainly provided by China Southern Power Grid Corporation, the emission coefficient of the electricity of China Southern Power Grid is based on the report of the regional grid baseline emission factor and the ratio of non-fossil energy. The thermal power carbon emission coefficient reported in 2012 is about 0.9344 tons CO 2 /MWh [33], and the non-fossil energy of the Southern Power Grid accounts for 44% of the total generated electricity [34]. Then, the final calculated emission coefficient of electricity use is about 0.5233 tons CO 2 /MWh. This paper assumes that the CO 2 emission coefficient for electricity remains the same for different years.

LMDI Method
Ehrlich and Holdren put forward the famous IPAT identity, a formula that states that environmental pressure comes from population size, per capita wealth, and the level of technology related to the environment [35]. Waggoner and Ausubel further decomposed the technology level T as the product of technology per unit of GDP and the environmental impact per unit of technology [36]. Kaya proposed the Kaya identity, a formula which links the factors of economy, policy, and population to CO 2 emissions from human activities [37]. Grossman and Krueger proposed that economic growth affects environmental quality through scale, technology, and composition effects [38]. Manufacturing activities also affect industrial CO 2 emissions mainly through scale, structure, and technology effects. First, the expansion of manufacturing activity increases the use of energy and then increases the amount of CO 2 emissions if the nature of manufacturing activity remains unchanged. Second, different energies produce the same power with different CO 2 emissions, and different industries produce the same economic output with different CO 2 emissions. Therefore, the adjustment of the energy structure and the optimization of the industrial structure tend to reduce the proportion of carbon-intensive energy and polluting industries, while the proportion of clean industries and energy increases, resulting in the reduction of emissions. Finally, modern manufacturing technologies are typically cleaner than older technologies because of the growing awareness of CO 2 emission reduction and energy conservation in production.
Referring to Ren et al. [29], Gu et al. [25], and Tu et al. [20], we use the LMDI method to identify the driving factors of the total CO 2 emissions from Shenzhen's manufacturing sector as follows: where: C mfr .: the total CO 2 emissions of manufacturing Y j : the added value of the industry j of manufacturing Y: the gross domestic product of manufacturing, with 2010 as the base year other variables are the same as the Formula (1). Among Equation (2), the corresponding multipliers represent different influencing factors. The meanings and abbreviations of each multiplier are shown in Table 3 as follows. Based on Equation (2), we divide the factors affecting CO 2 emissions of Shenzhen's manufacturing into the following four categories: the emission structure of different energies (CS), CO 2 emission intensity (CI), the added value structure of manufacturing (MS) and economic growth of manufacturing (Y). The CO 2 intensity uses the ratio of the CO 2 emissions to the added value of the manufacturing. It represents the level of CO 2 emission efficiency by different sectors. The lower the CO 2 emission intensity, the better the carbon-reducing technology in manufacturing industries and vice versa. The manufacturing structure adopts the ratio of the added value of each manufacturing industry to the total added value of manufacturing. The added value of manufacturing is the level of economic activity of manufacturing. The higher the economic activity, the higher the CO 2 emissions. All zeros in the data set may be replaced by a small positive constant [9]. Table 3. Definitions of the variables.

Multipliers
Abbreviation Meaning Units Therefore, according to the above definition, we rewrite Equation (2) as follows: Based on Equation (3), we use the additive form of the LMDI method to decompose the change in CO 2 emissions of manufacturing into four corresponding factors between the base year and the T year as follows: The emission energy structure effect (∆C CS ) means the impact of changes in the structure of different energies on the total CO 2 emissions. The CO 2 emission intensity effect (∆C CI ) can reflect changes in CO 2 emissions due to increased emission efficiency, such as low-carbon technologies. The manufacturing structure effect (∆C MS ) is the impact of changes in the added value proportion of each sector on CO 2 emissions, usually indicating a rise in high-tech manufacturing. The economic activity effect of manufacturing (∆C Y ) indicates that changes of the manufacturing economy affect the changes in CO 2 emissions. The total of each effect is then equal to the change in total CO 2 emissions from manufacturing. The corresponding effects are calculated as follows: represents the average logarithmic weight.

Data Sources
Because of the data limit, our research period is from 2008 to 2020. The energy activity data of different manufacturing sectors are from the Shenzhen Statistical Yearbook 2009-2021 published by the Shenzhen Statistical Bureau [39]. The total added value data of the industry is also from the Shenzhen Statistical Yearbook 2021. All value-added data are at the 2010 price based on the industrial value-added index. In this study, we used the structure of the added value of the above-sized manufacturing. The added value data of above-sized manufacturing by different sectors during 2008-2020 in Shenzhen come from the Guangdong Statistical Yearbook 2009-2021. The added value of industry above the designated size accounts for about 90% of the total industrial added value. There are very small changes in the definition of manufacturing classification after 2013. For the format consistency of manufacturing data in different years, we divide Shenzhen's whole manufacturing economy into 27 sectors and 4 categories, as shown in Table 4.

Characteristics of CO 2 Emissions from Shenzhen's Manufacturing
(1) Overview of total CO 2 emissions from manufacturing. Figure 1 illustrates the total CO 2 emissions from Shenzhen's manufacturing from 2008 to 2020. CO 2 emissions from Shenzhen's manufacturing in 2020 are only 58% of those in 2008. This change is equivalent to an average annual reduction rate of 4.42%. However, there is a significant difference between the emission rate from 2008 to 2012 and that from 2012 to 2020. The rapid decline of CO 2 emissions during 2008-2012 is because, since the global financial crisis in 2008, Shenzhen has been actively pushing for industrial transition and upgrading, especially the manufacturing. Another reason is Shenzhen's early attention to the air pollution problem of megacities in China [30] and the rapid electrification of its manufacturing. Since CO 2 emission control and air pollution control are highly rooted in the same origin in China, the reduction of CO 2 emissions is widely regarded as one of the synergies of air pollution control [22,30]. As a consequence, low-end industries with high fossil energy consumption and pollution are being quickly phased out. Since 2012, Shenzhen has been continuously introducing policies to support strategic emerging industries, forming several major industries such as the internet, cultural creative, new energy, and new materials. CO 2 emissions have since reached 15.40 million tons in 2020 with a small, negligible increase each year. The reason for the subtle increase during 2012-2020 is mainly due to the rapid development of high-tech manufacturing industries such as electronic information and precision manufacturing.
being quickly phased out. Since 2012, Shenzhen has been continuously introducing policies to support strategic emerging industries, forming several major industries such as the internet, cultural creative, new energy, and new materials. CO2 emissions have since reached 15.40 million tons in 2020 with a small, negligible increase each year. The reason for the subtle increase during 2012-2020 is mainly due to the rapid development of high-tech manufacturing industries such as electronic information and precision manufacturing.  (2) CO 2 emissions under the energy structure. Figure 2 shows the energy structure of CO 2 emissions from Shenzhen's manufacturing during the period 2008-2020. Electricity is obviously the primary source of emissions in Shenzhen's manufacturing, showing a completely opposite development trend to oil and gasoline. Specifically, the embodied CO 2 emissions of electricity accounted for nearly 60% of total emissions in 2008 and increased to 95% in 2020. After 2011, its proportion exceeded 90% of total manufacturing CO 2 emissions. The sum of the CO 2 emissions proportion of diesel and fuel oil was 37.6%, 34.8%, and 24.0% in 2008, 2009, and 2010, respectively. After 2011, the CO 2 emission proportion of the two energies fell to a range of between 2% and 4%. The proportion of electricity in Shenzhen's manufacturing stays at a relatively high level. Electrification reduces direct CO 2 emissions and air pollutants and improves the living environment. The improvement in air quality will have significant impacts on health benefits [40]. This may be one of the reasons why Shenzhen has attracted a large number of high-tech talents, and the inflow of talents has helped its rapid development in high-tech industries. (2) CO2 emissions under the energy structure. Figure 2 shows the energy structure of CO2 emissions from Shenzhen's manufacturing during the period 2008-2020. Electricity is obviously the primary source of emissions in Shenzhen's manufacturing, showing a completely opposite development trend to oil and gasoline. Specifically, the embodied CO2 emissions of electricity accounted for nearly 60% of total emissions in 2008 and increased to 95% in 2020. After 2011, its proportion exceeded 90% of total manufacturing CO2 emissions. The sum of the CO2 emissions proportion of diesel and fuel oil was 37.6%, 34.8%, and 24.0% in 2008, 2009, and 2010, respectively. After 2011, the CO2 emission proportion of the two energies fell to a range of between 2% and 4%. The proportion of electricity in Shenzhen's manufacturing stays at a relatively high level. Electrification reduces direct CO2 emissions and air pollutants and improves the living environment. The improvement in air quality will have significant impacts on health benefits [40]. This may be one of the reasons why Shenzhen has attracted a large number of high-tech talents, and the inflow of talents has helped its rapid development in high-tech industries.

Heterogeneity of CO2 Emissions from Shenzhen's Manufacturing
(1) Industry heterogeneity. Figure 3 shows the industry heterogeneity of CO2 emissions from manufacturing in Shenzhen during 2008-2020. According to industry charac-

Heterogeneity of CO 2 Emissions from Shenzhen's Manufacturing
(1) Industry heterogeneity. Figure 3 shows the industry heterogeneity of CO 2 emissions from manufacturing in Shenzhen during 2008-2020. According to industry characteristics, 27 sub-sectors are divided into four categories, revealing the sector structure of CO 2 emissions from Shenzhen's manufacturing. Specifically, electronic equipment manufacturing is the primary sector responsible for energy emissions, accounting for nearly 34.93% of total manufacturing emissions in 2008 and increasing to 47.43% in 2020. Before 2012, the proportion of its emissions was slightly reduced to about 30% but has gradually increased since. The emission proportion of light industry to the total proportion of manufacturing has dropped from 16.86% in 2008 to 8.26% in 2020. During the period under study, the emission share of equipment manufacturing and others in manufacturing decreased from 32.1% to 27.73%. The emission proportion of the raw material processing industry increased slightly and then decreased slightly, and the proportion in 2020 (16.57%) and 2008 (16.11%) is basically the same.  (2) Electricity heterogeneity. Table 5 shows the proportion of CO2 emissions embodied in electricity to the total CO2 emissions in individual subsectors during 2008-2020. After 2011, their values increased significantly. This is the same conclusion obtained in Figure 2. In 2020, the proportion of CO2 emissions embodied in electricity from the smelting and pressing of ferrous metals (Sector ID_18), processing of petroleum, coking, and nuclear fuel (Sector ID_12), and manufacturing of wine, beverages, and refined tea (Sector ID_3) is around 65%. The proportion of CO2 emissions contained in electricity in other individual industries is above 70%. There are more than 50% of subsectors whose electricity-triggered CO2 emissions occupied over 90% of their total emissions. Among these, three subsectors (sector ID_27, ID_24, and ID_16) reach as high as 95%. Therefore, in Shenzhen, a city in the post-industrial era, indirect CO2 emissions triggered by electricity usage are the most important emission source in its manufacturing industries.
An important feature of the emission evolution in Shenzhen's manufacturing is the shift from direct emissions from high-emitting heavy industries to indirect emissions from high-tech manufacturing industries. At present, high-tech manufacturing is widely regarded as a low-carbon industry, which can reduce emissions by replacing heavy industry [29,41]. However, attention should be paid to indirect emissions in the supply chain of high-tech manufacturing [26,42]. This paper shows more specifically that high-tech manufacturing industries have low direct emissions from fossil energy combustion, but they may have more indirect emissions through the massive consumption of electricity. Table 5. The proportion of CO2 emissions embodied in electricity to total emissions in 27 manufac- (2) Electricity heterogeneity. Table 5 shows the proportion of CO 2 emissions embodied in electricity to the total CO 2 emissions in individual subsectors during 2008-2020. After 2011, their values increased significantly. This is the same conclusion obtained in Figure 2. In 2020, the proportion of CO 2 emissions embodied in electricity from the smelting and pressing of ferrous metals (Sector ID_18), processing of petroleum, coking, and nuclear fuel (Sector ID_12), and manufacturing of wine, beverages, and refined tea (Sector ID_3) is around 65%. The proportion of CO 2 emissions contained in electricity in other individual industries is above 70%. There are more than 50% of subsectors whose electricity-triggered CO 2 emissions occupied over 90% of their total emissions. Among these, three subsectors (sector ID_27, ID_24, and ID_16) reach as high as 95%. Therefore, in Shenzhen, a city in the post-industrial era, indirect CO 2 emissions triggered by electricity usage are the most important emission source in its manufacturing industries.

Decomposition Analysis of CO 2 Emissions in Shenzhen's Manufacturing
(1) Decomposition of different driving factors. The decomposition of CO 2 emissions changes and the contributions of various driving factors from 2008 to 2020 are shown in Table 6. The contributions of various driving forces refer to the proportion of CO 2 emissions changes caused by each factor. Each factor presents a different effect in the previous year. During the period 2008-2020, the overall CO 2 emission in Shenzhen's manufacturing decreased by 41.87%. CO 2 emission intensity decreased by 107.04%, followed by manufacturing structure (−11.24%) and energy emission structure (−0.71%). Manufacturing economic activity emissions increased by 77.13% and is the only positive driver. The period from 2008 to 2012 was the most important structural adjustment stage for Shenzhen's manufacturing CO 2 emission changes. It can be seen that the reduction of manufacturing emission intensity is the most important factor for the decrease [18,19,24]. All along, Shenzhen has taken the strategy of a high-quality economy as a long-term strategy for urban development, strived to become an important role in global innovation and development, and treated innovation as its primary driving force for sustainable urban growth. Shenzhen vigorously lays out seven "strategic emerging industries" (next-generation information technology industry, high-end equipment manufacturing industry, green low-carbon industry, biomedical industry, digital Economy Industry, new materials industry, and marine economy industry) and four "pillar industries" (cultural and creative industries, hightech industry, modern logistics, and financial industry), promotes the transformation of the economy into high-tech industries, strives to improve the quality of economic development, and continuously enhances the added value of the industry. The highquality development of the manufacturing industry is the top priority of Shenzhen's industrial development strategy. In 2020, the added value of Shenzhen's high-tech manufacturing and advanced manufacturing accounted for 66.1% and 72.5% of the added value of industrial enterprises above designated size, respectively [43]. The development of high-tech, high-value-added, low-carbon industries is an important reason for the decline in manufacturing emissions [26]. Therefore, it can be seen that there are two significantly different stages in the change of manufacturing CO 2 emissions in Shenzhen, and the driving factors are also different in those two stages. (2) Decomposition at different stages. Since there is a clear trend difference between the two stages of 2008-2012 and 2012-2020 in Shenzhen's manufacturing CO 2 emissions, it is necessary to examine the impact of different factors on these two periods. From Figure 4, each factor presents a different effect in different stages. Overall, the manufacturing activity effect (∆C Y ) remains positive during these two periods, which is the primary driving force of the changes in emission. However, the carbon emission intensity effect (∆C CI ) is the key factor that offsets the increase of manufacturing emissions at the two stages. In addition, the energy structure effect (∆C CS ) also contributes a small decrease in emissions during 2008-2012 but changes to nearly none during 2012-2020. Relative to the carbon intensity effect, the energy structure effect is insignificant. The manufacturing structure effect (∆C MS ) is negative in both stages. From 2008 to 2012, the key factor for the decrease in CO 2 emissions was carbon intensity, which decreases 21.92 million tons of emissions accumulatively in Shenzhen's manufacturing sectors. After the financial crisis in 2008, external demand slowed down and costs increased. Shenzhen firmly grasped the mechanism of the international financial crisis and took the initiative of industrial upgrading. With the help of the market, Shenzhen phased out of backward production capacity and released part of the industrial land to make room for the development of newly emerging industries. At the same time, it actively deployed strategic emerging industries and improved the industrial level, covering a wide range of high-tech fields such as electronic information, aviation, and microchips. From 2012 to 2020, the key driving factor of CO 2 emissions changes to manufacturing activity, which caused 8.95 million tons of emissions in Shenzhen's manufacturing sectors. In fact, CO 2 emissions are closely linked to the growth of GDP [24]. In general, countries with larger economic output have larger CO 2 emissions, and vice versa [44]. In addition to the impact of total economic activity, CO 2 emissions are also closely related to the industrial structure and emission intensity. The emission intensity reduction effect offset 6.43 million tons of CO 2 emissions, and industrial structure changes offset 0.94 million tons of emissions. However, the incremental effect of manufacturing activities exceeds the deductive effects of other factors. In recent years, Shenzhen has accelerated industrial transition and upgrading processes, and used market instruments, such as CO 2 emissions allowance trading, to promote the reduction of emissions among its manufacturing companies. The annual emission compliance rate of Shenzhen's CO 2 emission control units has been relatively high. From 2013 to 2020, the compliance rates of enterprises included in the CO 2 emissions allowance trading market were over 99.0% each year. Therefore, as a post-industrial and innovative megacity, Shenzhen first experienced a stage in which low-end and backward industries were eliminated due to the rapid transition and upgrading of the manufacturing industry, manifesting as a rapid reduction in CO 2 emissions (2008-2012). Then, with the rapid development and growth of various strategic emerging industries and innovative technology industries with low carbon intensity, the second stage of a slow increase in carbon emissions was formed (2012-2020). The optimization of industrial structure can reduce local CO 2 emissions or pollution [26] but can cause the transferal of emissions and pollution to other places [21,22]. One of the most important ways for a megacity to transfer direct CO 2 emissions is to import large amounts of electricity [26,45]. The essence of this is the relocation of the fossil energy generation industry. a stage in which low-end and backward industries were eliminated due to the rapid transition and upgrading of the manufacturing industry, manifesting as a rapid reduction in CO2 emissions (2008)(2009)(2010)(2011)(2012). Then, with the rapid development and growth of various strategic emerging industries and innovative technology industries with low carbon intensity, the second stage of a slow increase in carbon emissions was formed (2012-2020). The optimization of industrial structure can reduce local CO2 emissions or pollution [26] but can cause the transferal of emissions and pollution to other places [21,22]. One of the most important ways for a megacity to transfer direct CO2 emissions is to import large amounts of electricity [26,45]. The essence of this is the relocation of the fossil energy generation industry.

Changes of CO2 Emissions Intensity and Manufacturing Structure
(1) Evolution of carbon intensity. Figure 5 shows the trends of CO2 emissions intensity of Shenzhen's manufacturing by four categories during 2008-2020. We can see that all four of the categories' CO2 emission intensities declined by more than 2/3 during these years. CO2 intensity represents the total amount of CO2 emissions per unit of GDP, and its reciprocal is CO2 productivity, which reflects the efficiency of CO2 emissions. As early as 2010, Shenzhen was listed as one of the first low-carbon pilot areas.
In 2012, Shenzhen planned the CO2 emission allowance trading market. In 2013, Shenzhen's emission allowance trading system was officially launched. The first batch of 621 manufacturing enterprises was included. These manufacturing enterprises are decreasing their emissions while increasing industrial value added. Therefore, the sharp decline in the CO2 intensity of the Shenzhen manufacturing industry is likely to indicate that the emission efficiency of the sub-sectors has been significantly improved. Efficiency improvement is an important way for cities to slow the

Changes of CO 2 Emissions Intensity and Manufacturing Structure
(1) Evolution of carbon intensity. Figure 5 shows the trends of CO 2 emissions intensity of Shenzhen's manufacturing by four categories during 2008-2020. We can see that all four of the categories' CO 2 emission intensities declined by more than 2/3 during these years. CO 2 intensity represents the total amount of CO 2 emissions per unit of GDP, and its reciprocal is CO 2 productivity, which reflects the efficiency of CO 2 emissions. As early as 2010, Shenzhen was listed as one of the first low-carbon pilot areas. In 2012, Shenzhen planned the CO 2 emission allowance trading market. In 2013, Shenzhen's emission allowance trading system was officially launched. The first batch of 621 manufacturing enterprises was included. These manufacturing enterprises are decreasing their emissions while increasing industrial value added. Therefore, the sharp decline in the CO 2 intensity of the Shenzhen manufacturing industry is likely to indicate that the emission efficiency of the sub-sectors has been significantly improved. Efficiency improvement is an important way for cities to slow the increase in their CO 2 emissions [21]. The manufacturing technological level is improved, and the potential for emission reduction is fully released. In addition to the changing trends of CO 2 emission intensity, Figure 5 also shows the difference in CO 2 intensity between different industries. Electronic equipment manufacturing with the highest CO 2 emissions has the lowest CO 2 intensity among these four categories of manufacturing. Manufacturing industries that process raw materials have the highest carbon intensity. While light industry and equipment manufacturing and others are in the middle level. The deep transition of the manufacturing industry is also one of the bases for the peak of total CO 2 emissions [26]. Therefore, the industries with the lowest CO 2 emission intensity are already the most important sectors in Shenzhen's manufacturing, which means that the potential for emission reduction by lowering its CO 2 intensity is exhausted unless Shenzhen increases the share of electricity generated from renewable sources and the efficiency with which it is used. of the bases for the peak of total CO2 emissions [26]. Therefore, the industries with the lowest CO2 emission intensity are already the most important sectors in Shenzhen's manufacturing, which means that the potential for emission reduction by lowering its CO2 intensity is exhausted unless Shenzhen increases the share of electricity generated from renewable sources and the efficiency with which it is used. (2) Trend in manufacturing value added. Optimizing industrial structure is an important way to promote direct CO2 emission reduction and pollution control [22,28]. Figure 6 shows the proportion trend of the added value of Shenzhen's manufacturing sector by four categories during 2008-2020. The proportion of the added value of electronic equipment manufacturing in the total added value of the manufacturing increased from 55.71% in 2008 to about 63.72% in 2020. According to the National Economic Industry Classification Standard published by the National Bureau of Statistics in 2017 [46], manufacturing of communication equipment, computers, and other electronic equipment (manufacturing sector ID_27) includes computer manufacturing, communication equipment manufacturing, radio and television equipment manufacturing, radar and ancillary equipment manufacturing, non-professional audiovisual equipment manufacturing, manufacturing of smart consumer devices, manufacturing of electronic devices, manufacturing of electronic components and electronic materials, and manufacturing of other electronic devices. This industry has relatively high technological intensiveness and industrial added value with low carbon intensity, and plays an irreplaceable and fundamental role in improving both the industrial economy and the quality of life of ordinary citizens. The second largest added value share is equipment manufacturing and others. Its proportion of added value in the total added value of manufacturing is around 20%, which is a slight increase of about 1% in 2020 compared to its value in 2008. The value added share of light manufacturing decreased from 11% in 2008 to 6.22% in 2020. The pro- (2) Trend in manufacturing value added. Optimizing industrial structure is an important way to promote direct CO 2 emission reduction and pollution control [22,28]. Figure 6 shows the proportion trend of the added value of Shenzhen's manufacturing sector by four categories during 2008-2020. The proportion of the added value of electronic equipment manufacturing in the total added value of the manufacturing increased from 55.71% in 2008 to about 63.72% in 2020. According to the National Economic Industry Classification Standard published by the National Bureau of Statistics in 2017 [46], manufacturing of communication equipment, computers, and other electronic equipment (manufacturing sector ID_27) includes computer manufacturing, communication equipment manufacturing, radio and television equipment manufacturing, radar and ancillary equipment manufacturing, non-professional audiovisual equipment manufacturing, manufacturing of smart consumer devices, manufacturing of electronic devices, manufacturing of electronic components and electronic materials, and manufacturing of other electronic devices. This industry has relatively high technological intensiveness and industrial added value with low carbon intensity, and plays an irreplaceable and fundamental role in improving both the industrial economy and the quality of life of ordinary citizens. The second largest added value share is equipment manufacturing and others. Its proportion of added value in the total added value of manufacturing is around 20%, which is a slight increase of about 1% in 2020 compared to its value in 2008. The value added share of light manufacturing decreased from 11% in 2008 to 6.22% in 2020. The proportion of the raw material processing industry decreased from 12.47% in 2008 to 8.45% in 2020. The process of optimizing the industrial structure of a city with a strict environmental policy can lead to the transfer of pollution and heavy industry to other cities. Collaborative control of CO 2 emissions and pollution with neighboring regions is also important [21,22,30]. The impact of Shenzhen's industrial relocation on CO 2 emissions on other surrounding cities needs to be deeply analyzed to reduce carbon leakage in the future.

Conclusions and Implications
In order to analyze the current situation of CO2 emissions from manufacturing subsectors with four categories in Shenzhen during 2008-2020 and the driving factors behind these changes, this paper used CO2 emission coefficients to calculate the CO2 emissions of different energy types in various industries and conduct structural analysis. The LMDI method was introduced to decompose and analyze the impact of manufacturing economic activities, manufacturing structure, emission intensity, and energy structure of emissions on the changes in its total manufacturing emissions. Based on the above findings and discussion, we make the following conclusions: First, the successful transition and upgrading of the industry led to a cliff-like decline in the total CO2 emissions of Shenzhen's manufacturing industry from 2008 to 2012, which has been well maintained and controlled since then. Furthermore, from the perspective of energy structure, electricity is the main energy source of emissions in Shenzhen's manufacturing industry with a continuous and stable trend, followed by diesel and then fuel oil, whose proportion has dropped sharply since 2011. That is, the CO2 emissions from the direct use of fossil energy in manufacturing enterprises in Shenzhen account for a very low proportion of total CO2 manufacturing emissions, but the proportion of CO2 emissions embodied in electricity rose to about 95%.
Second, significant heterogeneity in CO2 emissions across different manufacturing sectors was revealed. Specifically, regarding the 27 sub-sectors of manufacturing, the manufacturing of communications equipment, computers, and other electronic equipment accounts for the highest proportion of emissions, followed by the manufacturing of electrical machinery and equipment, and then the manufacturing of rubber. Considering the four categories, the emissions from high to low are electronic equipment manufacturing, equipment manufacturing and others, the raw material processing industry, and light manufacturing. In addition, in the observed years, the emissions embodied in electricity accounted for at least 65% of the total CO2 emissions in various manufacturing industries. In the future, saving electricity may become the most important measure to reduce the

Conclusions and Implications
In order to analyze the current situation of CO 2 emissions from manufacturing subsectors with four categories in Shenzhen during 2008-2020 and the driving factors behind these changes, this paper used CO 2 emission coefficients to calculate the CO 2 emissions of different energy types in various industries and conduct structural analysis. The LMDI method was introduced to decompose and analyze the impact of manufacturing economic activities, manufacturing structure, emission intensity, and energy structure of emissions on the changes in its total manufacturing emissions. Based on the above findings and discussion, we make the following conclusions: First, the successful transition and upgrading of the industry led to a cliff-like decline in the total CO 2 emissions of Shenzhen's manufacturing industry from 2008 to 2012, which has been well maintained and controlled since then. Furthermore, from the perspective of energy structure, electricity is the main energy source of emissions in Shenzhen's manufacturing industry with a continuous and stable trend, followed by diesel and then fuel oil, whose proportion has dropped sharply since 2011. That is, the CO 2 emissions from the direct use of fossil energy in manufacturing enterprises in Shenzhen account for a very low proportion of total CO 2 manufacturing emissions, but the proportion of CO 2 emissions embodied in electricity rose to about 95%.
Second, significant heterogeneity in CO 2 emissions across different manufacturing sectors was revealed. Specifically, regarding the 27 sub-sectors of manufacturing, the manufacturing of communications equipment, computers, and other electronic equipment accounts for the highest proportion of emissions, followed by the manufacturing of electrical machinery and equipment, and then the manufacturing of rubber. Considering the four categories, the emissions from high to low are electronic equipment manufacturing, equipment manufacturing and others, the raw material processing industry, and light manufacturing. In addition, in the observed years, the emissions embodied in electricity accounted for at least 65% of the total CO 2 emissions in various manufacturing industries. In the future, saving electricity may become the most important measure to reduce the CO 2 emissions of Shenzhen's manufacturing. The application, dissemination, and promotion of electricity-saving technologies for production equipment in manufacturing enterprises should be strengthened. At the same time, manufacturing enterprises should be encouraged to consume renewable electricity and improve the efficiency of electricity.
Third, CO 2 intensity, manufacturing structure, and energy structure contributed to the decline in CO 2 emissions from Shenzhen's manufacturing, while economic activities led to its increase. Moreover, Shenzhen's manufacturing emissions have gone through two significant stages, namely 2008-2012 and 2012-2020. The first stage is to rapidly reduce CO 2 emissions due to the rapid elimination of low-end backward manufacturing industries with high CO 2 intensity. The second stage is the period of slow CO 2 emissions increase caused by the rapid growth of high-value-added manufacturing industries with low CO 2 intensity. The reduction in manufacturing emissions intensity is the most important factor offsetting the increase in CO 2 emissions from Shenzhen's manufacturing industry activities during the two stages. This suggests that governments and manufacturing companies should continue to increase research and development and investment in energy-saving technologies. This also demonstrates that the manufacturing emission intensity targets and measures formulated by the Shenzhen municipal government are effective, and that it should continue to strengthen the control of CO 2 intensity, especially the decrease in the intensity of electricity use. Furthermore, the Shenzhen municipal government should continue to make full use of market mechanisms such as controlled CO 2 emission trading, the gradual reduction of emission quotas and the raising of allowance prices of the trading market. Then companies that pursue profit maximization will actively upgrade their products, develop high value-added products with higher technology, and pursue the potential of structural upgrading in the overall manufacturing industry.
Finally, the manufacturing structure and energy structure of Shenzhen have already been upgraded to a relatively low carbon level, and the marginal impact of manufacturing restructuring on CO 2 emissions has been relatively insignificant up to now. Through continuous industrial upgrading, Shenzhen's manufacturing industry has undergone a process from processing to manufacturing and then to creation. Previously, there were many enterprises that processed raw materials on a client's behalf, assemble parts for those clients and then process according to the client's samples or engage in compensatory trade. Shenzhen's economy then basically went through labor-intensive manufacturing, such as original equipment manufacturing, before finally entering the product design and technology research and development field at the top of the value chain. Shenzhen has been striving to develop strategic emerging industries with high added value, high technology content, strong innovation, and low carbon intensity. These manufacturing industries have gradually become the pillar industries of Shenzhen, especially the manufacturing of communication equipment, computers, and other electronic equipment. Manufacturing structural optimization is one of the important determinants of Shenzhen's manufacturing emissions. Therefore, in the future, the potential of emission reduction through the optimization of manufacturing industry structure and energy structure is expected to be gradually exhausted.

Conflicts of Interest:
The authors declare no conflict of interest.