Social and Economic Factors of Industrial Carbon Dioxide in China: From the Perspective of Spatiotemporal Transition

: The reduction of CO 2 emission has become one of the signiﬁcant tasks to control climate change in China. This study employs Exploratory Spatial Data Analysis (ESDA) to identify the provinces in China with different types of spatiotemporal transition, and applies the Logarithmic Mean Divisia Index (LMDI) to analyze the inﬂuencing factors of industrial CO 2 emissions. Spatial autocorrelation of provincial industrial CO 2 emissions from 2003 to 2017 has been demonstrated. The results are as follows: (1) 30 provinces in China are categorized into 8 types of spatiotemporal transition, among which 24 provinces are characterized by stable spatial structure and 6 provinces show signiﬁcant spatiotemporal transition; (2) For all types of spatiotemporal transition, economic scale effect is mostly contributed to industrial CO 2 emission, while energy intensity effect is the most crucial driving force to reduce industrial carbon dioxide emission; (3) provinces of type HH-HH, HL-HL and HL-HH are most vital for CO 2 emission reduction, while the potential CO 2 emission increase of developing provinces in LL-LL, LH-LH and LL-LH should also be taken into account. Speciﬁc measures for CO 2 emission reduction are suggested accordingly. the classiﬁcation of time-space transition of each province, which helps to identify key factors of each type and provide more targeted policy suggestions. This study evaluates the spatiotemporal transitions of industrial CO 2 emissions in China’s 30 provinces from 2003 to 2017, and explores the main driving forces of CO 2 emissions. 8 kinds of terminal industrial energy consumption at the provincial level are considered to calculate the


Introduction
Consistent growth of energy consumption in China has resulted in the massive emission of CO 2 , which contributes to one of the major challenges since the 21st centuryclimate change [1][2][3]. To deal with the threat of climate change, low-carbon development has become a global consensus. China plays an important role in mitigating global climate change [4]. In 2017, the CO 2 emissions in China increased by 105.57% from 4.523 billion tons in 2003 to 9.298 billion tons [5]. At the 2009 Copenhagen Climate Change Conference, China clearly proposed that its CO 2 emission per unit of gross domestic product (GDP) in 2020 would be reduced by 40-45% compared with that in 2005. This goal was taken as a medium-and long-term plan for its national economic development [6]. Besides, at the United Nation General Assembly in 2020, China's realization of carbon peak in 2030 and carbon neutrality in 2060 have been highly proposed. China promised its self-contribution to carbon abatement and has taken effective measures for these targets. The control of CO 2 emission, as to get a win-win achievement between the environment and economic development, has become an important goal for the sustainable economic development of China [7]. Carbon abatement in the industrial sector is crucial to the realization of China's CO 2 emission target. The industrial sector in China has high energy consumption and CO 2 emission [8]. In 2018, the total energy consumption of China's industrial sector was 3.11 billion tons of standard coal, accounting for 65.93% of that in China. With the rapid increase of the industry scale, its effect on CO 2 emissions is expected to continue to expand [9]. It is of great importance to explore issues of CO 2 emission in China's industrial sectors.
In recent years, the issue of CO 2 emission has been widely discussed. Owing to different development levels, CO 2 emissions have significant spatial differences across regions, which is necessary to discuss from spatial perspective [5,10,11]. Studies have shown that CO 2 emission has spatial spillover effects [12,13].  found significant spatial autocorrelation and spatial agglomeration effects of CO 2 emission in 30 Chinese provinces during the period of 2004-2016 [14].  demonstrated positive spatial autocorrelation of carbon emission intensity among 281 prefecture-level cities in China [15]. Some scholars began to investigate the temporal and spatial variation of CO 2 emission [16,17]. However, most of them only analyze from temporal or spatial perspective, few traced the spatiotemporal evolution of CO 2 emission with precision [18]. The Exploratory Spatial Data Analysis (ESDA) model provides solutions to quantitively capture the dynamic changes of CO 2 emission from both temporal and spatial perspective, which has been applied into various fields, such as CO 2 emission from agriculture and water use [19,20]. Rey proposed a space-time transition classification under the framework of ESDA, which is suitable to discuss the temporal and spatial changes of CO 2 emission [21]. Zhao et al., (2017) used ESDA model to classify the spatiotemporal transition of 30 provinces in China from 1997 to 2015, which found that the spatiotemporal evolution characteristics of carbon intensity among provinces show both "agglomeration'" and "differentiation" in the spatial distribution [22]. Only few researches have applied the space-time transition classification to industrial sectors in China. However, the CO 2 emission in industrial sectors is a massive part of China's carbon abatement [23,24].
It is also important to explore the influencing factors of CO 2 emissions to obtain better carbon abatement strategies [25]. Existing studies have pointed out that population size [26], economic growth [27,28], energy consumption [29,30], energy structure [31], and industry structure [32] are the main influencing factors of CO 2 emission. The impacts of different factors vary greatly on CO 2 emissions in different industries. Lin et al., (2014) demonstrated that industrial activity is the leading force to explain emission increase in the Chinese non-metallic mineral products industry, while energy intensity is the major contributor to the emission mitigation [33]. Ma et al., (2018) illustrated that population density contributed greatly to carbon abatement of China's commercial buildings [34]. Song et al., (2018) concluded that, in China's iron and steel industry, economic activity was the prominent contributor to increase CO 2 emission while technology progress was the main factor of mitigated CO 2 emission [35]. Quan et al., (2020) found that economic output, population size and energy structure play a positive role in CO 2 emission in China's logistics industry, while energy intensity plays a negative role [36]. Besides, the effects of related factors on CO 2 emissions may vary significantly in different provinces [37]. In order to mitigate industrial CO 2 emissions effectively, it is worthwhile to explore how these driving factors influence CO 2 emissions in industrial sectors of different provinces [16]. A variety of methods has been adopted to explore the influencing elements, such as Logarithmic Mean Divisia Index (LMDI) model, Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model, and Structural Decomposition Analysis (SDA) model, etc., [38][39][40][41]. LMDI has advantages on flexibility and easy access to data [42,43], which is especially suitable for provincial data in this study.
Based on existing studies, the contributions of this study are as follows. First, instead of only considering static distribution of CO 2 emission, this study adopts a ESDA method into the field of industrial CO 2 emission, which can capture the dynamic changes (both space and time) of CO 2 emission. Second, the effect decomposition of LMDI is conducted based on the classification of time-space transition of each province, which helps to identify key factors of each type and provide more targeted policy suggestions. This study evaluates the spatiotemporal transitions of industrial CO 2 emissions in China's 30 provinces from 2003 to 2017, and explores the main driving forces of CO 2 emissions. 8 kinds of terminal industrial energy consumption at the provincial level are considered to calculate the industrial CO 2 emissions, including coal, coke, crude oil, fuel oil, gasoline, kerosene, diesel oil and natural gas. Differentiated policy recommendations are proposed in line with the results and discussion.  [44][45][46][47]. The calculation is as follows:

Methodology and
where E(CO 2 ) is the industrial CO 2 emission; ω i refers to the CO 2 emission coefficient; E i denotes the total energy consumption; i represents the 8 types of fossil energy, including coal, coke, crude oil, fuel oil, gasoline, kerosene, diesel oil and natural gas. NCV i , CEF i and COF i denote the average low calorific value of energy resources, carbon emission coefficient (without conversion) and carbon oxidation factor, respectively. 44 and 22 are the molecular weights of carbon dioxide and carbon, respectively. c i is energy conversion coefficient of calorific value to standard coal. The above parameters are from the China Energy Statistical Yearbook (2017) and the IPCC Guidelines for National Greenhouse Gas Inventories (2006). The CO 2 emission coefficients (ω i ) of the 8 types of fossil energy are shown in Table 1. The ESDA method is introduced to measure the characteristics of spatial and temporal distributions of CO 2 emission. Moran's I index is used to examine the spatial agglomeration characteristics of regional carbon emission [48][49][50]. The calculations are as follows: (2) where n is the number of observed provinces; i and j denote city i and city j, respectively; X i and X j are the observed values in region i and region j; X is the average observed value of all provinces; W ij is the spatial weight matrix. A 0-1 matrix is used as the spatial weight matrix [51,52]. When i and j are adjacent, W ij = 1; otherwise W ij = 0. S 2 is the standard error of X i .

LMDI Model
To further explore the contribution of factors influencing industrial CO 2 emissions, a LMDI model is adopted based on the classical Kaya identity proposed by Yoichi Kaya [53]. The effects of population, economy and energy can be evaluated in the Kaya identity: where CO 2 represents total CO 2 emissions, POP represents total population, GDP denotes gross regional product, and E denotes total energy consumption. Under the framework of Kaya identity, LMDI model is established to analyze the contribution of each influencing factor [54][55][56]. Due to similar properties, crude oil, fuel oil, gasoline, kerosene, and diesel oil are integrated into oil products. The 8 energy categories are simplified into 4 types. In the following calculation, Equation (4) can be extended as follows: where CO 2 refers to the total CO 2 emission of industrial sector; CO 2im refers to the CO 2 emission generated by i (energy consumption of the industrial sector) in province m; P m represents the total population of region m; GDP m represents the GDP of region m; V m is the industrial added value in region m; E m is the total industrial energy consumption in province m, and E im is the industrial energy consumption of energy type i in province m. P is resident population at the end of the year as a proxy for population size effect; G represents GDP per capita as a proxy for economic scale effect; M is the proportion of industrial output value in GDP, which measures the effect of industrial structure; K is the energy consumption per unit industrial output value to indicate the effect of energy intensity; N reflects the effect of energy structure, expressed by the proportion of energy i in the total industrial energy consumption; S is the CO 2 emission coefficient, which is indicated by the CO 2 emission per unit terminal energy consumption. Given that S is assumed to be a constant value, only five influencing factors are decomposed, which are shown in Equations (6)- (10). where is the contribution of energy intensity effect, and ∆C N t,t−1 is the contribution of energy structure effect. A positive value indicates that the influencing factor is conducive to the increase of CO 2 emissions, whereas a negative value shows that the influencing factor reduces of CO 2 emissions.

Sources of Data
The data of GDP, resident population at the end of the year and industrial output value are from the China Statistical Yearbook (2004-2018). GDP is converted into real price (base year = 2003). Per capita GDP is calculated by the proportion of real GDP and resident population at the end of the year. The industrial energy consumption of coal, coke, crude oil, fuel oil, gasoline, kerosene, diesel oil and natural gas and the total province are from China Energy Statistical Yearbooks (2004-2018). Owing to data limitations, Tibet Autonomous Region, Hong Kong, Macao, and Taiwan Special Administrative Regions are excluded.

Regional Distribution of Industrial CO 2 Emissions
The trend of total industrial CO 2 Figure 2. The CO 2 emissions of most provinces in China raised first and then decreased at the second stage, which is consistent with the overall national level in Figure 1. However, the level and distribution of CO 2 emissions in provinces display regional dynamic characteristics. The CO 2 emissions in Hebei, Shandong, Henan, and Jiangsu provinces were constantly the top 4 among provinces. The rapid economic development and flourishing industrial development in these areas lead to high carbon dioxide emissions. These high carbon emission areas with developed industries are the key areas of CO 2 emission reduction. Beijing and Hainan have always been low CO 2 emission areas compared with others. Although economy in these regions develops fast, it is driven by the tertiary industry, rather than secondary industries. Therefore, the industrial CO 2 emissions in the two provinces are relatively small. The CO 2 emissions of Yunnan, Shanxi, Gansu, and Qinghai provinces were lower than the national average. Low economic level in these provinces leads to low industrial CO 2 emissions.

Spatial Autocorrelation of Industrial CO 2 Emission
As shown in Table 2, the global Moran's I indices of China's industrial CO 2 emission from 2003 to 2017 have all passed the significance test, which ranges from 0.127 to 0.305. The results demonstrate positive spatial autocorrelation and a clustering trend of CO 2 emission among provinces. The industrial CO 2 emission of each region will be influenced by the neighboring regions, showing an "agglomeration" trend in space. Therefore, it is necessary to distinguish and analyze the geographic distribution of CO 2 emission.     Table 3 displays spatial agglomeration types of industrial CO 2 emissions according to Moran's I scatter plot. Provinces of HH (high CO 2 emission in the local and neighborhood), LH (low CO 2 emission in the local while high CO 2 emission in the neighborhood), and LL (low CO 2 emission in the local and neighborhood) clusters account for the majority in China, while the HL (high CO 2 emission in the local while low CO 2 emission in the neighborhood) agglomeration area constitutes the minority. Specifically, provinces in the HH agglomeration area have larger economic scale and higher population density, which causes more energy consumption and CO 2 emission in the local. Besides, the regional agglomeration enhances economy in surrounding areas, which leads to high CO 2 emissions in neighboring provinces. This type mainly consists of Hebei, Shandong, and other provinces. The LH agglomeration area is mainly concentrated in Beijing, Shanghai, Tianjin, and other regions. After rapid economic development, high carbon dioxide emission industries in local areas will be transferred to neighboring areas, demonstrating the characteristics of LH accumulation: low CO 2 emissions locally and high CO 2 emissions in the adjacent areas. The LL agglomeration area includes Yunnan, Gansu, Qinghai and other provinces, which has low CO 2 emissions both in the local and surrounding areas. Only Guangdong, Sichuan and Hunan are always in the HL agglomeration area, which has high industrial CO 2 emission in the local but low industrial CO 2 emission in the surrounding provinces. The distribution of each agglomeration type differs between 2003 and 2017, which means that some provinces transfer to other quadrants. For instance, Hubei changed from HL cluster in 2003 into HH cluster in 2017. Heilongjiang and Guizhou changed from LH cluster in 2003 to LL cluster in 2017. The spatiotemporal transition of industrial carbon emissions is necessary to be further evaluated.

Spatiotemporal Transition of Industrial CO 2 Emission
The spatial agglomeration characteristics of industrial CO 2 emissions in each province are evaluated according to the classification of spatiotemporal transition [19,20]. By observing the changes of 4 spatial agglomeration types (HH, LH, LL and HL) in starting and ending years (2003 and 2017), the spatial and temporal variation of CO 2 emission in different provinces can be categorized into 16 spatiotemporal transition types. It is shown in Table 4 that LH→LL, LL→LH, LH→HH, LL→HL, LH→HL, LL→HH, HH→HL, HL→HH, HH→LH, HL→LL, HH→LL and HL→LH type have spatiotemporal transition, while LH→LH, LL→LL, HH→HH and HL→HL type do not. The 30 provinces in China belong to 8 types: LH→LH, LL→LL, LH→LL, LL→LH, HH→HH, HL→HL, HL→HH and HH→HL. To be more precise, provinces of LH→LH or LL→LL type have low CO 2 emission with no spatiotemporal transition between local and neighborhood regions. Regions in LH→LL or LL→LH type regions show low local CO 2 emission, while the spatiotemporal transition only occurs in the neighboring provinces rather than in the observed province. HH→HH and HL→HL type have high local CO 2 emission, but no significant spatiotemporal transition in the local and neighborhoods. HL→HH type areas show high local CO 2 emission, and the spatiotemporal transition only occurs in the neighboring provinces. For HH→LH type provinces, the local CO 2 emission changes from high to low and there is no spatiotemporal transition in neighboring areas.
It is concluded that the CO 2 emission in local provinces is easily influenced by that of neighboring provinces. Therefore, the provinces in LH→LL, LL→LH, HL→HH and HH→LH type should be paid more attention to promote the transition trend of CO 2 emission in China. Besides, HH→HH type provinces are supposed to have stable and high CO 2 emission in the future, which illustrates the necessity to control CO 2 emission in provinces with this transition path.

Driving Factors of National Industrial CO 2 Emission
To further study what impacts carbon dioxide emission and how to effectively reduce carbon emission, driving factors of CO 2 emission are analyzed on the basis of the LMDI mode (see

Variation of Sub-Index Contribution Degree to Different Spatiotemporal Transition Types
The effects of sub-index on carbon dioxide emission have significant differences between provinces classified by spatiotemporal transition types. Figure 5  For all spatiotemporal transition types, economic scale and energy intensity present strongest effect on CO 2 emission among the five influencing factors. The effect intensity of economic scale has the same trend with that of energy intensity. The spatiotemporal transition types with strong (weak) economic scale effect also have strong (weak) energy intensity effect. However, the two effects display opposite contribution to carbon dioxide emission. The economic scale effect for all spatiotemporal transition types promotes CO 2 emission, while energy intensity effect shows an inhibition to CO 2 emission. It is owing to the contradiction of regional economic development and national goal of emission reduction. For a long term, economic development has been the priority for local governments. The national goal of emission reduction will more or less impede economic growth in a short term. Therefore, the inconsistent targets between the central government and local governments result to a counterbalance of the economic scale effect and energy intensity effect, which finally weaken the effectiveness of carbon dioxide emission. Population size, industrial structure and energy structure have less effect on CO 2 emission compared with economic scale and energy intensity. Among the three effects, the industrial structure effect decreases the CO 2 emission in all spatiotemporal transition types, while energy structure effect and population size effect (except for LL-LH type) contribute to CO 2 emissions.   Among the 8 types of spatiotemporal transition, types of HH-LH and LL-LL are typical with the highest and lowest accumulative contribution degrees respectively. The accumulative contribution degrees of each influencing factors in HH-LH type are significantly higher than that in other types. The corresponding region of this type is Zhejiang Province. Although the rapid economic growth and large population size in this province contribute a lot to CO 2 emission, the efficient energy utilization and rational industrial structure promote emission reduction. It achieves a win-win development of economy and carbon dioxide emission reduction, as maintain the CO 2 emission in a stable and relatively low level. LL-LL type regions typically have low accumulative contribution degree of each influencing factors, especially for economic scale effect and energy intensity effect. This type of provinces concentrates on Western regions of China, such as Yunnan, Gansu, Qinghai, Guangxi, Ningxia and Xinjiang. The carbon dioxide emission has still been at a low level in these provinces due to its undeveloped economy, which leads to the low accumulative contribution degrees. To better examine the difference between region types and distinguish the particularity of provinces in each type, the accumulative contribution values of each province are further discussed in the next section.

Discussion on Regional CO 2 Emission Reduction based on Spatiotemporal Transition Types
The accumulative contribution value and total carbon dioxide emission of each province based on spatiotemporal transition types during 2003-2017 are shown in Figure 6. Detailed analysis and targeted suggestion to various spatiotemporal transition type will be discussed in the following.
LH-LH type includes Tianjin, Beijing, Shanghai, Chongqing, Jiangxi, Shaanxi and Jilin, the industrial CO 2 emission of which are all ranking low in China. However, different provinces show differentiated characteristics. 4 of them are municipalities (Tianjin, Beijing, Shanghai, Chongqing) in China. In these regions, the inhibition effect of energy intensity to CO 2 is stronger than the promotion effect of economic scale compared to the other provinces. In other words, due to the rapid economic development, high efficiency of energy utilization and the outward movement of related industries, the carbon dioxide abatement in these provinces has been carried out well and steadily during the fifteen years. Other three provinces (Jiangxi, Shaanxi and Jilin) are surrounded by several high CO 2 emission provinces, such as Hubei, Hunan, Guangdong, Anhui, Hebei, Shanxi and Liaoning. The industry of the three provinces has still been left behind the adjacent regions. Therefore, developed areas in LH-LH type, such as Tianjin, Beijing, Shanghai, Chongqing, should maintain their low emission levels as well as produce an impetus effect on neighboring provinces, as integrate resources to promote industrial transformation of the whole region; developing areas in LH-LH type, such as Jiangxi, Shaanxi and Jilin, should pay special attention for the potential environmental pollution caused by industrial undertaking from surrounding provinces.   The provinces in LL-LL type concentrate in the western region in China, which include Yunnan, Gansu, Qinghai, Guangxi, Ningxia and Xinjiang. Since these regions fall behind both in economic and industrial development, all of the industrial CO 2 emissions are lower than the average level in China, which gradually form a LL spatial agglomeration in these provinces. However, during the current transformation of industry from the east to the west (One Belt, One Road etc.), a coordinated development should be attached importance to in LL-LL type. For instance, developing environmental conservation industries and making use of energy with high quality in the industry production. LH-LL type includes Fujian and Hainan. The two provinces discharge low level of carbon dioxide, especially Hainan. This is because, both of them are highly dependent on the single industry of tourism, and do not take the second industry as the main development pathway. The LL-LH type consists of Heilongjiang and Guizhou. The CO 2 emission of two are not high, but the contribution of the economic scale and the industrial structure are the second highest compared to other provinces. On one side, the industry progressed fast in Guizhou. The greater the economic scale, the more CO 2 emission produced. On the other side, as a traditional industrial province, Heilongjiang has been adopting the extensive and inefficient development mode. Preventing Heilongjiang and Guizhou from becoming high CO 2 emission provinces, provinces of the LL-LH type should be emphasized as key provinces to the abatement carbon dioxide emission during economic development.
The provinces of type HH-HH (Hebei, Shandong, Shanxi, Liaoning, Henan, Anhui, Jiangsu, Inner Mongolia), HL-HL (Guangdong, Sichuan, Hunan) and HL-HH (Hubei) show higher industrial CO 2 emission over the national average, ranking the top 12 in China. Besides, the spatial agglomeration of high CO 2 emission is especially demonstrated in provinces of HH-HH type and HL-HH type during 2003-2017. As for the effects of five influencing factors, the highest promoting effect of economic scale on CO 2 emission is demonstrated in the HH-HH type, and followed by the HL-HL type. The population size positive effect in some provinces of type HH-HH (Hebei, Shandong, Shanxi) and HL-HL(Guangdong) are more intense than provinces of other types. For one thing, 75% among the 12 provinces both have large population size and economic scale, leading to more energy consumption and carbon dioxide emission. For another, many provinces are geographically adjacent, such as Huabei region (Hebei, Inner Mongolia, Shanxi), Huazhong region (Henan, Hunan, Hubei) and Huadong region (Jiangsu, Anhui, Shandong). The spatial agglomeration of adjacent provinces further strengthens the emission level of industrial CO 2 in these provinces. These provinces are supposed to break through provincial boundaries and build a coordinated mechanism, which can maximize the advantages of each province in CO 2 emission reduction, so as to realize the transformation and upgrading of low-carbon green industry, and jointly complete the target of carbon emission reduction. In HH-LH, the industrial CO 2 emission of Zhejiang has declined in 2003-2017 for the reason that, Zhejiang reinforces the energy usage and adjusts the industrial structure rationally with the economy prosperity.

Conclusions
By using the ESDA and LMDI methods, the study analyzes the spatiotemporal evolution of industrial CO 2 emissions among 30 provinces in China from 2003 to 2017 and further explore the impacts of five driving factors on CO 2 emissions. Corresponding conclusions and findings on the basis of different spatiotemporal transitions are as follows: (1) China's provinces with high industrial CO 2 emissions are mainly distributed in central and eastern regions. The industrial CO 2 emission in western provinces stays low but has an upward trend. In general, most provinces in China maintain steady spatial agglomeration types of industrial CO 2 emission. 24 provinces in China show no spatiotemporal transition of industrial CO 2 emission, while other 6 provinces have spatiotemporal transition of industrial CO 2 emission in the local or neighboring regions.
(2) For all the 8 types of spatiotemporal transition, economic scale effect is mostly contributed to industrial CO 2 emission; energy intensity effect is the most crucial factor in reducing carbon emission, which is followed by industrial structure effect; population size effect and energy structure effect both promote industrial CO 2 emission, but play little role compared with other effects.
(3) The most vital provinces for CO 2 emission abatement include provinces of type HH-HH, HL-HL and HL-HH. For provinces in HH-HH, HL-HL and HL-HH type, it is suggested to construct a particular mechanism of joint prevention and control for carbon emission reduction, so as to take the advantages of in each province, and further enhance the efficiency of emission abatement. The developing provinces in LL-LL, LH-LH and LL-LH should also pay attention to emission control to prevent potential increase of CO 2 emission. It is important to seize the opportunity of industrial transformation, technology innovation and energy efficiency improvement. HH-LH and LH-LL type have advantages in controlling CO 2 emission. HH-LH type can be taken as an example to achieve the balance between the economy and environment protection. Provinces in LH-LL are supposed to make full use of the local superior resources to achieve prosperity such as ecological agriculture, ecological tourism and renewable resources.