The effect of urbanization on carbon dioxide emissions efficiency in the Yangtze River Delta, China.

Cities have been one of the most important areas of CO2 emissions. It is increasingly important to research the effect of urbanization on CO2 emissions, especially in large emerging and developing economies, due to the indispensable need for understanding the effect of urbanization on CO2 emissions, evaluating carbon reduction tasks and providing the scientific basis for low-carbon urbanization. Utilizing a balanced panel dataset in the Yangtze River Delta (YRD), China, during the period of 2000-2010, this paper employed data envelopment analysis (DEA) window analysis and a spatial lag panel Tobit model to investigate the effect of urbanization on CO2 emissions efficiency (the ratio of the target CO2 emissions to the actual CO2 emissions). The results show that the average CO2 emissions efficiency was 0.959 in 2010, and CO2 emissions efficiency ranged from 0.816 to 1 and exhibited spatial clustering in the region. The larger potential of CO2 emissions reduction appeared in Zhenjiang and Yangzhou, indicating that more CO2 emissions reduction tasks should be allocated to these two cities. Urbanization has negative effects on improving CO2 emissions efficiency, and there is a U-curve relation between CO2 emissions efficiency and urbanization, indicating that CO2 emissions efficiency decreases at the early stage of urbanization, then increases when urbanization reach a high level. There is spatial spillover effect among the prefecture-level cities, suggesting that different prefecture-level governments should coordinate with each other to improve CO2 emissions efficiency in the whole area. Gross domestic product (GDP) per capita also plays a markedly positive role in improving CO2 emissions efficiency. This research highlights the effect of urbanization on CO2 emissions efficiency and the importance of improving CO2 emissions efficiency in developing countries.

Introduction

Cities are important CO2 source (Liu et al., 2011) and core areas of CO2 emissions reduction in China (Wu et al., 2017b, Zhu et al., 2017). Urbanization increases energy consumption and becomes one of the main contributors to CO2 emissions. China has become the country with the highest CO2 emissions since 2006 (Gregg et al., 2008). Urbanization level in China increased rapidly from 35.87% in 2000 to 55.61% in 2015. It has exceeded the world average since 2013 (World Bank, 2017). Rapid urbanization has led to a shortage of resources and worsened environmental pollution (Yang et al., 2014). In order to solve environmental problems, accelerating environmental protection and the construction of ecological civilization (e.g., reducing carbon emission) has been put forward since the Eighteenth National Congress of the Communist Party of China (Yang et al., 2013). China's new urbanization plan (2014–2020) calls for the promotion of low carbon development as well as the adoption of low-carbon lifestyles and urban construction modes (Bai et al., 2014). Chinese central government proposed to implement low-carbon urbanization, which attracted the attention of local governments and research scholars (Liu and Qin, 2016, Mittal et al., 2016, Wang et al., 2015).

CO2 emissions efficiency (CEE) is the ratio of the target CO2 emissions to the actual CO2 emissions, and the target CO2 emissions can be calculated by data envelopment analysis (DEA) model (Choi et al., 2012). CEE plays an important role in the reduction of CO2 emissions and low-carbon urbanization. However, there has been little research on CEE from the perspective of the impact of urbanization, which may help reveal the mechanisms of how urbanization impacts on CEE and provide a scientific basis for low-carbon urbanization.

Different methods have been developed to investigate environmental efficiency (the ratio of the target environment output to the actual environment output) or CEE. For instance, Charnes et al. (1978) first proposed the data envelopment analysis (DEA) model, which has been widely employed in research on environmental and energy efficiency (Dyckhoff and Allen, 2001, Qin et al., 2017, Zhou et al., 2017). Charnes and Cooper (1985) combined the DEA with window analysis and proposed the DEA window analysis to estimate the dynamic effect of data. The fuzzy DEA model was proposed by Ignatius et al. (2016) to evaluate CEE in 23 European Union (EU) countries. Slack-based measure (SBM) DEA model was applied by Choi et al. (2012) to calculate the efficiency and potential CO2 emissions reduction (PCR) (the slack of CO2 emissions) in China. The average PCR was 56.1 million tons in each province, and the CEE in eastern China is better than that in other regions in China.

Many studies investigated energy efficiency or CEE at the country level (Beltrán-Esteve and Picazo-Tadeo, 2017, Girod et al., 2017, Sakamoto and Managi, 2017, Zhou et al., 2017) or province level (Liu et al., 2016a, Liu et al., 2016b, Zhang et al., 2016b, Zhou and Liu, 2016). Due to the spatial variations in economic development and the source of carbon emissions inside the country or province, there is obvious spatial difference in CEE across its prefecture-level cities inside the country or province. And therefore, prefecture-level city data can provide more detailed information and make more accurate estimation, whereas national or provincial CEE analyses may have bias, at least due to the lack of spatial details. Meanwhile, many studies ignored the influence of spatial effect (Li et al., 2017). They assumed that study units were independent, but each spatial unit can actually affect its neighbors by technical diffusion and industrial transfer. For instance, Zhou et al. (2017) applied a two-stage DEA model to evaluate energy efficiency in Asia-Pacific Economic Cooperation (APEC) countries from 1995 to 2013. The study found that the energy efficiency of developed countries was usually higher than those of developing countries. Liu et al. (2016a) utilized the SBM model to analyze CEE of Chinese provinces and found that the CEE of most Chinese provinces showed a downward trend from 2000 to 2011.

Some scholars have investigated the factors that influence CEE (Cui and Li, 2015, Lin and Du, 2015, Qin et al., 2017). Nevertheless, few have examined the influence of urbanization on CEE. Lin and Du (2015) assessed China's CEE during the period of 1997–2009. The study found that market-oriented reforms were beneficial to improving CEE. Cui and Li (2015) employed virtual frontier DEA to analyze carbon efficiency of transportation sector, and they found that technology and management factors had bigger influence than structural factor in CEE. Qin et al. (2017) utilized global Malmquist-Luenberger productivity index to evaluate the energy efficiency from 2000 to 2012, and they found that technological improvement was the main influencing factors of energy efficiency. Zhao et al. (2015) investigated the impact of environmental regulations on the efficiency and CO2 emissions and their results indicated that market-based regulation and government subsidies were helpful for improving efficiency and CO2 emission reduction.

While there are some studies on CEE, there are three limitations in previous research. First, the study scale was mainly the country (Beltrán-Esteve and Picazo-Tadeo, 2017, Girod et al., 2017, Sakamoto and Managi, 2017, Zhou et al., 2017) or the province (Liu et al., 2016a, Liu et al., 2016b, Zhang et al., 2016b, Zhou and Liu, 2016), and few studies have examined the CEE of prefecture-level cities (Yuan et al., 2013). Second, previous studies on the factors that influence CEE mainly focus on market-oriented reforms (Lin and Du, 2015), environmental regulations (Zhao et al., 2015), technology and management factor (Cui and Li, 2015, Qin et al., 2017), while few of them focus on urbanization. Third, most studies ignored the effect of spatial spillover on CEE and assumed that adjacent regions were independent. However, ignoring spatial spillover can exacerbate the biases of the results.

To address these limitations of previous studies, this paper includes undesirable outputs into DEA window analysis to calculate CEE by utilizing a linear transformation in the Yangtze River Delta (YRD), China, during the period of 2000–2010. Urbanization has increased moderately, while CO2 emission has grown rapidly since 2000 in the YRD (Song et al., 2015, Tian et al., 2011). Thus, we investigated the effect of urbanization on CEE since 2000. Due to data availability, the study covered the period of 2000–2010. Different from most previous studies, this study takes the prefecture-level city as the basic study unit. Then, we analyze the relationship between urbanization and CEE. Finally, to take into account spatial spillover effect, a spatial lag panel Tobit model is constructed to analyze the impact of urbanization on CEE. The results are helpful for evaluating the carbon reduction task in the YRD, guiding policymakers to establish a plan to improve CEE, and providing a scientific basis for the development of low carbon urbanization.

This study investigates the following questions: (1) What are the carbon emissions efficiency and reduction potential in the study area? (2) What is the relationship between urbanization and CEE? (3) Is urbanization one of the main factors influencing CEE? (4) What is the suitable policy to improve CEE? To address these issues, the paper is arranged as follows. Section 2 describes the study area, data source, and methods, Section 3 presents the results, Section 4 discusses the results, and Section 5 concludes the study and proposes some policies.

Study area, data source, and methods

Study area

The Yangtze River Delta (YRD) mainly consists of Shanghai City, Jiangsu Province, and Zhejiang Province. Zhejiang Province includes 11 prefecture-level cities, while Jiangsu Province includes 13 prefecture-level cities (Fig. 1 ). The Yangtze River Delta Urban Agglomeration is one of the six largest metropolitan areas in the world and the largest metropolitan area in China (Tian et al., 2011). Urbanization level in the YRD increased from 47.95% in 2000 to 65.09% in 2010 (Fig. 2 ). The urbanization level of Shanghai is higher than those of Zhejiang and Jiangsu Provinces. The urbanization level of Jiangsu is the lowest in the YRD, but it has increased quickly. The gap of urbanization level between Jiangsu and Zhejiang has narrowed, and the urbanization level of Jiangsu was equal to that of Zhejiang in 2010.

Fig. 1

Study area in the Yangtze River Delta, China.

Fig. 2

Urbanization levels in the whole Yangtze River Delta and different provinces from 2000 to 2010.

Data source

In this study, we utilized workers (employed persons) and capital stock as the inputs (Choi et al., 2012). Capital stock was calculated by the perpetual inventory method (Feng et al., 2017a). CO2 emission was utilized as the undesirable output, and gross domestic product (GDP) was utilized as the desirable output. CO2 data were collected from the Emission Database for Global Atmospheric Research (EDGAR) (for more detail description of EDGAR, CO2 data and processing steps, see Li et al. (2017)). Workers and fixed asset investment were obtained from the China City Statistical Yearbook (2001–2011) (National Bureau Statistics of China, 2001–2011). Total population, non-agricultural population, the area, GDP, and secondary industry values of each prefecture-level city were obtained from the China Statistical Yearbook for Regional Economy (2001–2011) (National Bureau Statistics of China, 2001–2011). GDP, GDP per capita (GDPP) and capital stock were converted to constant prices in 2000. Table 1 presents the descriptive statistics of all variables.

Table 1

Descriptive statistics of variables during 2000–2010.

Variable	Unit	Minimum	Maximum	Mean	Standard deviation	Observation
Capital	108 dollars	28.626	4387.550	506.533	601.719	275
Labor	104 persons	10.160	929.200	211.438	190.277	275
GDP	108 dollars	13.775	1587.584	186.545	214.760	275
CO2	million tons	0.339	425.308	41.784	56.752	275
Urbanization	%	14.136	93.131	37.262	17.124	275
GDPP	dollars	479.751	11503.801	3353.194	2280.215	275
Industrial structure	%	33.254	65.209	53.774	6.376	275
Population density	Persons/km2	143.704	2227.251	695.790	359.730	275

Note: Capital, GDP and GDPP are converted into dollars ($) based on 2000 exchange rate (1$ = 8.278 Chinese Yuan (CNY)).

Methods

Traditional DEA models have been widely utilized in the study of environmental and energy efficiency (Dyckhoff and Allen, 2001, Qin et al., 2017, Zhou et al., 2017), but they ignored undesirable outputs (Seiford and Zhu, 2002). CO2 is an undesirable output in the calculation of CEE, and linear transformation method can be utilized to add undesirable outputs into the traditional DEA model (Seiford and Zhu, 2002). Combining DEA and window analysis, DEA window analysis can be utilized to estimate the dynamic effect of efficiency (Charnes and Cooper, 1985) and is suitable to estimate the CEE during the period of 2000–2010 in the YRD.

Data envelopment analysis

Data envelopment analysis (DEA) models include input-oriented models and output-oriented models. Input-oriented DEA models assume that output is fixed and input is minimum. However, output-oriented DEA models assume that input is fixed and output is maximum (Woo et al., 2015). In this study, we focus on improving CEE by increasing the desirable output (GDP) and decreasing the undesirable output (CO2). Thus, we chose output-oriented DEA models using the variable returns to scale (VRS) condition, and the equation of output-oriented DEA models is as follows (Seiford and Zhu, 2002):

Max.

Subject to where j denotes the number of decision making units (DMU); z denotes relative weights; jand yc j denote the capital, labor, GDP and CO2 of decision making unit j (DMU), respectively; denotes the slack associated with the capital, labor, GDP and CO2; denote the capital, labor, GDP and CO2 of decision making unit 0 (DMU0), respectively..

CEE was calculated as follows (Choi et al., 2012):

Y represents CEE; c 0 denotes the target of CO2 emission; c represents the actual value of CO2 emission.

DEA can simulate the actual production process well (Feng et al., 2017b). Based on the DEA theory, the inefficient areas can be improved and become the efficient areas by adjusting the slack (Shi et al., 2010). Potential carbon dioxide emission reduction (PCR) was calculated as follows (Guo et al., 2011, Wu et al., 2017a):where PCR denotes the potential carbon dioxide emission reduction; other variables share the same meaning in the Equation (2).

Linear transformation method

When there is undesirable output (CO2), the DEA data domain can be expressed as follows (Seiford and Zhu, 2002):where y denotes the outputs; x denotes the inputs; y and y denote the desirable output (GDP) and undesirable output (CO2), respectively; x and x denote the capital and labor, respectively.

CEE can be improved by increasing the desirable output (GDP) and decreasing the undesirable output (CO2), thus the undesirable output (CO2) was calculated as follows (Woo et al., 2015):where denotes CO2 emissions; denotes the value of CO2 emissions after linear transformation.

The model (4) can be expressed as follows:where the variables share the same meanings in Equations (4), (5).

Based on Equation (6), Equation (1) can be expressed as follows:

Max.

Subject to where the variables share the same meanings as in (1) and (6).

The DEA window analysis

DEA window analysis was utilized to calculate the CEE during 2000–2010. A window includes n × t observations, where n denotes the numbers of observations for one year and t represents the numbers of years in one window. The model can achieve the best performance for efficiency when t = 3 or t = 4 (Charnes et al., 2013). In this study, n = 25, t = 3, the first window covers the year 2000, 2001, and 2002, and the step of the window is one year. Thus, the second window covers the year 2001, 2002, and 2003. Nine windows with 25 × 3 observations are estimated in the study. Each year has three efficiency values, except 2000 and 2010 (which have one efficiency value), and 2001 and 2009 (which have two efficiency values). The average efficiency of each year was utilized as the final efficiency in the year.

Spatial lag panel Tobit model

CEE was taken as the dependent variable for building the econometric model. Since CEE values change between 0 and 1, the Tobit model is used (McDonald and Moffitt, 1980). The Tobit model has been widely used to investigate the influencing factors of environmental efficiency (Feng et al., 2017a, Lv et al., 2015, Yuan et al., 2013). The spatial spillover effect of CEE is the impact of some prefecture-level cities on their neighboring prefecture-level cities. The influences mainly derive from the demonstration effect in higher CEE prefecture-level cities through low-carbon technology diffusion and industrial transfer. Thus, a spatial lag model was constructed to take into account the spatial spillover effects of CEE. Meanwhile, this study explored whether there was environmental Kuznets curve (EKC) for CEE and urbanization. Based on the above considerations, the model is as follows (Anselin, 1988, Feng et al., 2017a, Yuan et al., 2013):where Y is the CEE; i and t represent the ith prefecture-level city and tth year, respectively; α is the constant term; ρ is the spatial autoregressive coefficient; W is the spatial weight matrix (here we choose queen contiguity weight matrix due to prefecture-level cities with one common edge or corner (Anselin, 1988)); α 1 is the coefficient of urbanization level; α 2 is the coefficient of squared-term urbanization level; β denotes the coefficient of control variables; X stands for the control variables; and ε represents the error term.

Variable and software selection

Urbanization may play an important role in improving CEE. People may need more infrastructure at a low urbanization level, which may increase carbon emissions and decrease CEE (Zi et al., 2016). However, when urbanization reaches a certain level, resource management and economies of scale may become better, and urbanization may become conducive to improving CEE (Martínez-Zarzoso et al., 2007, York, 2007). Furthermore, since GDP per capita (GDPP) is an important influence on energy consumption and CO2 emissions (Li et al., 2017), GDPP is also found to be closely related to CEE (Yuan et al., 2013). In addition, CEE may be affected by industrial structure and population density (Feng et al., 2017a, Yuan et al., 2013). Secondary industry is the main source of CO2 emissions, and it is more difficult to improve the CEE of secondary industry (Hao et al., 2015). There is an uncertain relationship between population density and CEE. Liu et al. (2016b) suggested that higher population density may urge the government and population to pay more attention to the environment, and it may improve CEE. However, Cropper and Griffiths (1994) thought that higher population density may lead to higher pressure on the environment, which can in turn lead to a decrease in CEE.

In this study, urbanization (UR) was chosen as the main independent variable (Dong et al., 2016), and it was calculated as the percentage of the non-agricultural population to represent the urbanization level. The following variables were selected as control variables: GDP per capita (GDP divided by the population) (GDPP) (Feng et al., 2017a), industrial structure (the ratio of secondary industry GDP to total GDP) (IS) (Bian et al., 2016) and population density (the ratio of population to the area) (PD) (Liu et al., 2016b).

To analyze the spatial autocorrelation of CEE, Moran's I and local indicator of spatial association (LISA) were calculated by GeoDa 1.4.1. LISA cluster map was drawn at the 5% significance level. Bivariate local Moran's I for CEE and urbanization was also calculated using GeoDa 1.4.1, and bivariate cluster map was drawn at the 5% level of significance. All correlation analyses were conducted using Matlab 2014a (MathWorks, Natick, Massachusetts, USA).

Results

Carbon emissions efficiency

Correlations between the inputs and outputs were analyzed. The correlation coefficients show positive correlation at the 1% significance level (Table 2 ), indicating that the more the inputs, the more the outputs. In particular, the relationship between GDP and CO2 presents significant positive correlation, indicating that reducing carbon emissions may restrain economic development in the YRD.

Table 2

Correlation matrixes of inputs and outputs.

	Capital	Labor	GDP	CO2
Capital	1
Labor	0.701∗∗∗	1
GDP	0.981∗∗∗	0.689∗∗∗	1
CO2	0.913∗∗∗	0.735∗∗∗	0.932∗∗∗	1

Notes: ∗∗∗ denotes two-tailed significance at 1% level.

Based on the DEA window analysis, we calculated the CEE and PCR, and the results are shown in Table A1, Fig. 3 and Table A2. In 2010, the average CEE was 0.959 in the YRD, indicating that if the YRD can reach the current desirable outputs, the average CO2 emissions can reduce by approximately 4.1%. Shanghai had the highest CEE of 1.000, followed by Zhejiang Province with a CEE of 0.990, while Jiangsu Province was at the bottom with the minimum CEE of 0.929. At the prefecture-level city scale in 2010, Suzhou, Lishui, Ningbo, Jinhua, Wuxi, Zhoushan, Tai'zhou, Wenzhou, Hangzhou, and Quzhou had the highest CEE of 1, while Zhenjiang, Yangzhou and Xuzhou had the smaller CEE of 0.816, 0.818 and 0.855, respectively. Excluding these three cities with small CEE, the CEE of the other prefecture-level cities are more than 0.9, indicating that CEE is relatively high in the YRD.

Fig. 3

CO2 emissions efficiency in the Yangtze River Delta from 2000 to 2010.

In general, CEE exhibited an overall increasing trend from 2000 to 2010, whilst it can be roughly divided into three stages: the first stage is from 2000 to 2003, a decreasing first and then increasing trend; the second stage is from 2003 to 2005, a decreasing trend; and the third stage is from 2005 to 2010, an increasing trend. The trends of CEE in Jiangsu and Zhejiang are similar to that of the whole YRD region. From 2000 to 2010, Shanghai had the highest CEE, followed by Zhejiang and Jiangsu, while the gap in CEE between Zhejiang and Jiangsu had narrowed in these ten years.

Moran's I values of CEE are in column 2 of Table 3 . They are positive and significant at the 5% level except for 2004 and 2005. This indicates that there is significantly positive spatial autocorrelation across adjacent prefecture-level cities in the YRD except for 2004 and 2005 and the CEE at one prefecture-level city tends to be similar to those of their neighboring prefecture-level cities.

Table 3

The temporal characteristics of CO2 emission efficiency (CEE) in the Yangtze River Delta, China, during the period of 2000–2010.

Year	Moran's I	Pearson	Bivariate Moran's I
2000	0.234∗∗	−0.044	−0.173∗
2001	0.323∗∗∗	0.122	−0.087
2002	0.301∗∗	0.180	−0.055
2003	0.347∗∗∗	0.084	−0.068
2004	0.111	−0.052	−0.198∗∗
2005	0.067	−0.118	−0.273∗∗∗
2006	0.261∗∗	−0.079	−0.300∗∗∗
2007	0.519∗∗∗	−0.117	−0.364∗∗∗
2008	0.546∗∗∗	−0.172	−0.373∗∗∗
2009	0.490∗∗∗	−0.200	−0.354∗∗∗
2010	0.434∗∗∗	−0.206	−0.348∗∗∗

Notes: ∗, ∗∗, ∗∗∗ represent coefficients are significant at the 10%, 5%, 1% levels, respectively.

In order to estimate the spatial autocorrelation of CEE, the local indicator of spatial association (LISA) was calculated. High-High type represents the prefecture-level cities with high CEE (their CEE are higher than the average CEE in the YRD) being surrounded by cities with high CEE, while Low-Low type represents the prefecture-level cities with low CEE (their CEE are lower than the average CEE in the YRD) being surrounded by cities with low CEE. As shown in Fig. 4 , in 2010, the prefecture-level cities for the High-High type are mainly concentrated in Shaoxing, Jinhua, Tai'zhou, Lishui, Quzhou, and the prefecture-level cities for the Low-Low type mainly appeared in Yangzhou, Zhenjiang, and Nanjing.

Fig. 4

Local indicator of spatial association (LISA) cluster map for CO2 emission efficiency in the Yangtze River Delta, China, in 2010.

The relationship between urbanization and carbon dioxide emissions efficiency

The Pearson correlation coefficients between urbanization and CEE are shown in column 3 of Table 3. They are negative except for the periods 2001–2003, with p values higher than 0.10, indicating that the relationship between urbanization and CEE is negative but insignificant in the YRD.

In the fourth column of Table 3, the Bivariate Moran's I values show a negative and decreasing trend during the period 2004–2010, and p values are <0.05. They indicate that there was a significant negative spatial correlation between CEE at a prefecture-level city and the urbanization level of its neighboring prefecture-level cities, and the negative spatial correlation increased during the period 2004–2010.

In order to analyze the local relationship between CEE at a prefecture-level city and urbanization of its neighbors, the bivariate local Moran's I values for CEE and urbanization were calculated. The first variable is the CEE of a prefecture-level city and the second variable is the urbanization level in the neighboring cities. As shown in Fig. 5 , there are three types including the Low-High, High-Low and High-High types. For example, the Low-High type means that these prefecture-level cities with low CEE (their CEE are lower than the average CEE in the YRD) are surrounded by cities with high urbanization level (their urbanization level are higher than the average urbanization level in the YRD). In 2010, the prefecture-level cities with the Low-High type were mainly concentrated in Taizhou, Zhenjiang, Changzhou, and Jiaxing. The prefecture-level cities for the High-Low type were mainly concentrated in Jinhua, Tai'zhou, Lishui, and Wenzhou. The prefecture-level cities for the High-High type were mainly concentrated in Suzhou and Huzhou.

Fig. 5

Bivariate cluster map of CO2 emission efficiency (CEE) and urbanization in the Yangtze River Delta, China, in 2010.

Potential carbon dioxide emission reduction (PCR)

PCR varied from 0 ton to 75.30 million tons in 2010 (Table A2), with the average PCR of 16.75 million tons in the YRD, 28.88 million tons in Jiangsu Province, 3.93 million tons in Zhejiang Province, and 0 ton in Shanghai City. Thus, the average PCR in Jiangsu is higher than those in Shanghai and Zhejiang. At the prefecture-level cities scale, PCR was 0 in Quzhou, Hangzhou, Wenzhou, Zhoushan, Tai'zhou, Wuxi, Jinhua, Ningbo, Suzhou, and Lishui, while larger PCR appeared in Zhenjiang and Yangzhou, with the values of 75.30 million tons and 74.55 million tons respectively.

The main explanatory factors of CO2 emissions efficiency (CEE)

Moran's I of CEE is 0.4056 during the period 2000–2001, with a p value less than 0.01, indicating that there is significant positive spatial autocorrelation across neighboring prefecture-level cities in the YRD, and the CEE between prefecture-level cities and their neighboring cities are similar. Thus, the influence of spatial factor should not be ignored and the spatial econometric model was constructed to analyze the main influencing factors of CEE.

Maximum likelihood estimation was utilized to estimate a spatial lag panel Tobit regression model specified in Equation (4), and the results are shown in Table 4 . The coefficients of urbanization and quadratic item are −1.103 and 0.789, which represent a U-curve relationship between CEE and urbanization (Feng et al., 2017a). The influence of GDP per capita (GDPP) is significant and positive on CEE. Both industrial structure (IS) and population density (PD) are not significant (p > 0.05), while spatial autoregressive coefficient (ρ) is 0.153 and significant at the 5% level, indicating a significant spatial spillover effect of CEE across different prefecture-level cities.

Table 4

Results of spatial lag panel Tobit regression.

Variable	Coefficient	p
Constant	0.735	0.000
UR	−1.103	0.000
UR2	0.789	0.001
GDPP	0.745	0.000
IS	−0.067	0.455
PD	0.007	0.952
ρ	0.153	0.026
R2	0.670
Adj R2	0.631
Log-likelihood	26.762

Discussion

CEE exhibited considerable spatial variation in the YRD and this may be due to different urbanization and economic development levels. In 2010, Zhenjiang had the minimum CEE value of 0.816. Shanghai, Suzhou, Lishui, Ningbo, Jinhua, Wuxi, Zhoushan, Tai'zhou, Wenzhou, Hangzhou, and Quzhou had the maximum CEE value of 1 in 2010, and these cities can be selected as the best demonstration cities for carbon emission reduction. Although they are in the frontier of better carbon efficiency in the YRD, there is still room for further improvement of their CEE.

CEE showed an increasing trend during the period of 2001–2003, which may be due to the implementation of the strict environmental protection policies in the Tenth Five-Year Plan Period (2001–2005) (Yuan and Zuo, 2011). CEE showed an obvious decreasing trend in 2004. This may be due to the outbreak of Severe Acute Respiratory Syndrome (SARS) which seriously influenced the economy in the YRD in 2003 and continued into 2004 (Zhong et al., 2003). CEE exhibited a distinctly increasing trend from 2005 to 2010. This is because of the implementation of sustainable development policies which paid more attention to energy conservation, pollutant reduction and green development (Li and Lin, 2016). In 2009, the Chinese government set a target to reduce carbon intensity to 40–45% of the 2005 level by 2020 (Qiu, 2009), and each prefecture-level government put more emphasis on reducing carbon emission. Meanwhile, the global financial crisis of 2008 seriously impacts China's economy. The central government accelerated the economic adjustment in the post-financial crisis period after 2008 (Li et al., 2016) and thus CEE significantly increased after 2009. CEE increased from 2005 to 2010, which does not mean that CO2 emission decreased from 2005 to 2010. The reason is that GDP increased during this period and GDP growth was usually accompanied by the increase in total CO2 emission. The Pearson correlation analysis indicates an insignificant negative relationship between CEE and urbanization (Table 3). When spatial factors were considered, bivariate Moran's I value is significant and negative during the period 2004–2010, indicating that there is a significant negative relationship between CEE in one prefecture-level city and their neighbors' urbanization.

Some prefecture-level cities, including Shaoxing, Jinhua, Tai'zhou, Lishui, and Quzhou, stay in the High-High type based on the LISA value of CEE (Fig 4). The reason is that prefecture-level cities with high CEE have positive radiation driving effect on its neighbors. In addition, these neighboring cities improve their CEE by introducing low carbon technology from the prefecture-level cities, and these prefecture-level cities form the high CEE core. These prefecture-level cities should spread advanced low carbon technology to other regions and help other cities in their carbon reduction efforts (Zhang et al., 2016b). The prefecture-level cities including Yangzhou, Zhenjiang, and Nanjing fall into the Low-Low type, that may be due to their high energy-intensive and carbon-intensive industries. Thus, these prefecture-level cities should adjust their industrial structure, introduce advanced low-carbon technology, and strengthen their cooperation with the High-High type prefecture-level cities. Meanwhile, local governments should provide more technology and financial aids to help improve their CEE.

Fig. 5 shows bivariate cluster map for CEE and urbanization. In terms of the Low-High type of prefecture-level cities with low CEE, their nearby cities with high urbanization are more competitive compared to these cities with low CEE. Without sufficient capital and technology, it is difficult for cities with low CEE to further improve their CEE (Choi et al., 2012, Qin et al., 2017, Zhang et al., 2016b). The High-Low type prefecture-level cities own advanced low-carbon technology and attract more high-quality resources from their neighboring cities with low urbanization level. Thus, these cities had high carbon emissions efficiency. For the prefecture-level cities of the High-High type, their nearby cities have a better radiation effect to these cities, and they can attract advanced low-carbon technology to improve their CEE. The PCR showed variation from 0 ton to 75.30 million tons in 2010 (Table A2). According to the PCR, different CO2 emission reduction tasks should be allocated to prefecture-level cities. Zhenjiang and Yangzhou are the two prefecture-level cities with the largest PCR because their leading industries are still heavy industries that consume more energy and produce more carbon emissions. Thus, Zhenjiang and Yangzhou are the key location for reducing CO2 emissions. More CO2 emission reduction tasks should be allocated to these two cities. In the future, Zhenjiang and Yangzhou should promote the adjustment of their industrial structure and technological innovation. Meanwhile, measures are needed to introduce advanced low-carbon technologies and strengthen interregional cooperation (Zhang et al., 2016a).

Urbanization and CEE shows a significant negative and U-curve relationship, indicating that CEE decreases at the early stage of urbanization and then increases at higher urbanization levels. More infrastructures and resources are needed at the early stage of urbanization, leading to more carbon emissions (Zhou and Liu, 2016). However, when urbanization reaches a higher level, the prefecture-level cities may have more resource to improve management efficiency and make better use of public infrastructures. This can improve CEE (Liddle, 2013, Martínez-Zarzoso et al., 2007, York, 2007), forming a win-win relationship between CEE and urbanization. People and governments are willing to pay more attention to the quality of the environment, so governments may invest more in the development and innovation of low-carbon technologies (Zhang et al., 2016b). In order to ensure the quality of urbanization and improve CEE, governments should develop different development strategies according to the stage of urbanization, optimize spatial layout, adjust urbanization develop model, and carry out low-carbon urbanization (Zhou and Liu, 2016).

GDP per capita (GDPP) plays a significant positive role in improving CEE, consistent with the results of Choi et al. (2012). When GDP per capita is higher, people have more resources to protect the environment, and local governments can have more money to invest in environmental protection and low-carbon technology. In the future, local governments should pay close attention to developing a low-carbon economy, as well as upgrading economic structure and growth manner (Li and Lin, 2016). The correlation coefficient between industrial structure and CEE is negative but not significant, similar to the result of Bian et al. (2016). Secondary industries are relatively inefficient than tertiary industries (Bian et al., 2016). When the proportion of secondary industries is higher, it is more difficult to improve CEE (Hao et al., 2015). Some effective measures have been adopted to adjust and optimize industrial structure, and the proportion of secondary industry has decreased in some cities (Liu and Li, 2015). In the future, local governments should put more emphasis on the transformation and upgrading of industrial structure. Meanwhile, byproducts should be reused more and industrial cooperation needs further improvement.

The correlation coefficient between population density and CEE is positive but not significant, similar to the finding of Yuan et al. (2013). Within a certain range, as population density increases, people can use resources more efficiently, which can improve CEE (Zhao et al., 2014). The spatial autoregressive coefficient (ρ) is significant and positive, indicating that there are spatial spillover effects across different prefecture-level cities. Local governments should consider the policies of neighboring prefecture-level cities and coordination with nearby cities is indispensable.

Conclusion and policy implications

The paper utilized DEA window analysis to investigate the efficiency and reduction potential of CO2 emissions during the period of 2000–2010 in the Yangtze River Delta (YRD). The influencing factors of CEE were also investigated using a spatial lag panel Tobit model. The results show that the average CEE was 0.959 in the YRD in 2010, most of the prefecture-level cities have relatively high CEE. Shanghai City exhibits the best CEE of 1, followed by Zhejiang Province (with CEE of 0.990) and Jiangsu Province (with CEE of 0.929). CEE shows spatial clustering across prefecture-level cities in the YRD. The prefecture-level cities of the Low-Low type need introduce advanced low-carbon technologies and cooperate with the High-High type prefecture-level cities.

The PCR of some prefecture-level cities are zero and their CEE are much better. With the development of new technologies, there is more room to reduce carbon emissions in those prefecture-level cities. Larger PCR appeared in Zhenjiang and Yangzhou, and therefore more CO2 emission reduction responsibilities should be allocated to these cities. Analysis of the PCR is conducive to allocate the quotas of carbon emission reduction. It is of great significance for successfully completing carbon emission reduction targets in the YRD.

There is a U-curve relationship between CEE and urbanization. CEE is explained mainly by urbanization level, GDP per capita, spatial effects.

To our knowledge, this study is the first attempt to investigate CEE across prefecture-level cities using grid data in the Yangtze River Delta (YRD). Different from many studies at the country or province level, this study used data on prefecture-level cities and provided more detailed information for policymakers. In addition, we investigated the effect of urbanization on CEE, which was often ignored in previous studies. With rapid urbanization, it has become an important factor influencing CEE in China. Furthermore, this study considered spatial spillover effect, which can improve the accuracy of the results.

China has started the national carbon emissions trading market (power generation industry) since 2017, allocating carbon emissions quota is the foundation of carbon trading and also key to carrying out the carbon trading (Zhang et al., 2016b). This paper provides a useful method for assessing the carbon emissions reduction demand in the YRD and other regions, such as the Pearl River Delta in South China (Qin et al., 2017).

Similar to many studies, some uncertainty existed in our current research. Due to data availability, the study only covers the period of 2000–2010. With more data available in the future, the analyses of the period after 2010 can provide more recent information about CEE in the YRD. The PCR was calculated based on the DEA window analysis, only considering the influence of efficiency but ignoring the need to also consider fairness (Miao et al., 2016). When fairness is considered in the new framework, the results can better guide the allocation of carbon emission reduction tasks.

This paper increases our understanding on urban carbon emission in the YRD. First, our results identified obvious disparities across the prefecture-level cities in these provinces, so more attention should be paid to the research of prefecture-level cities. Second, urbanization is one of the main factors influencing CEE through the U-curve relationship, and therefore more studies on CEE are needed from the perspective of urbanization. Last, spatial factor had significant impact on CEE, thus spatial spillover effect should receive more attention in future researches.

Based on our results, three policy suggestions were made for policy makers. First, our paper provided the information of CEE and PCR. Cities with high CEE own advanced low-carbon technologies, and they should share and export advanced low-carbon technologies to other cities. Meanwhile, in order to reduce the spatial disparities of CEE, the cooperation among different cities should be strengthened, and governments should tailor carbon emission reduction policies. According to the disparities of PCR, different CO2 emission tasks should be allocated to cities. Second, our results found both economic development and urbanization level had marked effects on CEE, and therefore low-carbon urbanization should be emphasized by controlling the improper urbanization speed, optimizing the spatial layout and making full use of infrastructure. Last, due to the existence of spatial spillover effect of CEE across different prefecture-level cities, when local governments make policies, they should consider and make effort to integrate the policies of their nearby cities.

Table A.1

CO2 emission efficiency (CEE) in the Yangtze River Delta, China, during the period of 2000–2010

Prefecture-level city	2000	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010
Shanghai	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000
Nanjing	0.798	0.794	0.819	0.819	0.795	0.819	0.843	0.861	0.871	0.902	0.924
Wuxi	1.000	1.000	1.000	1.000	0.905	0.912	0.958	0.967	0.975	0.987	1.000
Xuzhou	0.769	0.748	0.769	0.789	0.808	0.817	0.814	0.820	0.818	0.839	0.855
Changzhou	1.000	0.998	0.995	0.961	0.926	0.887	0.875	0.881	0.894	0.924	0.944
Suzhou	0.799	1.000	1.000	0.997	0.897	0.905	0.953	0.967	0.972	0.986	1.000
Nantong	0.884	0.862	0.887	0.898	0.893	0.884	0.894	0.909	0.922	0.956	0.977
Lianyungang	0.858	0.802	0.807	0.815	0.824	0.852	0.883	0.896	0.915	0.943	0.961
Huaian	0.827	0.783	0.797	0.809	0.815	0.839	0.863	0.874	0.889	0.917	0.934
Yancheng	0.950	0.950	0.970	0.984	0.988	0.972	0.938	0.918	0.905	0.929	0.949
Yangzhou	0.820	0.816	0.836	0.862	0.873	0.889	0.883	0.847	0.818	0.810	0.818
Zhenjiang	0.841	0.847	0.861	0.880	0.887	0.902	0.904	0.880	0.866	0.840	0.816
Taizhou	0.867	0.856	0.873	0.898	0.912	0.919	0.908	0.903	0.900	0.924	0.941
Suqian	0.850	0.819	0.838	0.858	0.862	0.912	0.911	0.909	0.913	0.937	0.955
Hangzhou	0.923	0.904	0.915	0.929	1.000	0.926	0.934	1.000	0.988	0.994	1.000
Ningbo	1.000	1.000	1.000	1.000	1.000	0.912	0.924	1.000	0.960	0.972	1.000
Wenzhou	0.930	0.932	0.951	0.977	0.951	0.986	0.999	1.000	1.000	0.999	1.000
Jiaxing	0.828	0.794	0.830	0.873	0.807	0.827	0.855	0.872	0.886	0.914	0.933
Huzhou	1.000	1.000	0.993	1.000	0.994	0.931	0.955	0.964	0.964	0.976	0.974
Shaoxing	0.936	0.919	0.875	0.898	0.916	0.934	0.955	0.979	0.959	0.972	0.987
Jinhua	0.946	0.914	0.923	0.936	0.931	0.926	0.948	0.982	1.000	0.999	1.000
Quzhou	0.974	0.942	0.928	0.920	0.896	0.908	0.935	0.965	0.980	0.994	1.000
Zhoushan	1.000	1.000	0.997	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000
Tai'zhou	1.000	1.000	1.000	1.000	0.993	1.000	1.000	1.000	1.000	0.993	1.000
Lishui	1.000	0.962	0.937	0.918	0.902	0.915	0.938	0.960	0.989	0.999	1.000

Table A.2

Potential CO2 emissions reduction (PCR) in the Yangtze River Delta, China, during the period of 2000–2010 (million tons)

Prefecture-level city	2000	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010
Shanghai	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Nanjing	28.99	30.41	29.00	35.40	49.93	51.64	50.45	47.62	46.97	35.94	27.90
Wuxi	0.00	0.00	0.00	0.06	25.95	25.68	12.35	10.49	8.57	4.60	0.00
Xuzhou	34.24	39.47	38.40	41.65	44.00	50.84	60.42	63.99	69.75	63.41	57.45
Changzhou	0.00	0.23	0.75	7.21	16.70	32.90	43.07	44.68	41.84	30.54	22.77
Suzhou	25.85	0.00	0.00	0.41	24.03	24.57	11.80	9.19	8.23	4.30	0.00
Nantong	17.23	21.92	19.14	20.76	25.95	34.89	36.64	33.77	30.37	17.47	8.92
Lianyungang	23.02	36.67	40.67	46.47	51.04	49.19	42.69	40.55	34.98	23.52	16.32
Huaian	27.98	38.88	40.21	45.01	50.91	51.46	48.82	48.49	45.41	34.41	27.85
Yancheng	6.90	7.17	4.65	2.96	2.68	7.31	19.95	29.38	37.74	29.00	20.95
Yangzhou	25.06	26.12	24.66	24.60	26.55	27.87	34.38	51.69	68.33	76.18	74.55
Zhenjiang	21.49	20.86	20.31	20.93	23.24	24.16	27.48	38.99	47.52	61.42	75.30
Taizhou	20.47	23.11	22.07	20.78	20.96	22.85	30.63	35.86	40.28	31.21	24.42
Suqian	24.39	31.69	31.04	32.06	36.66	26.13	30.49	34.49	35.29	26.11	19.05
Hangzhou	10.06	13.48	13.01	13.02	0.00	22.25	20.02	0.00	4.20	2.21	0.00
Ningbo	0.00	0.00	0.00	0.00	0.00	26.62	22.73	0.00	14.25	9.95	0.00
Wenzhou	9.68	9.66	7.61	4.15	11.57	3.70	0.36	0.00	0.00	0.46	0.00
Jiaxing	27.33	35.86	30.93	26.33	53.14	55.95	51.60	48.76	45.42	34.96	27.34
Huzhou	0.00	0.00	1.10	0.00	1.35	20.69	15.04	12.99	14.00	9.76	10.75
Shaoxing	9.15	12.21	23.01	22.42	21.08	19.04	14.29	7.29	15.49	11.06	5.17
Jinhua	8.21	13.91	13.88	13.59	17.68	22.74	17.91	6.30	0.00	0.39	0.00
Quzhou	3.92	9.07	12.66	17.29	27.70	28.89	22.67	12.91	7.92	2.24	0.00
Zhoushan	0.00	0.00	0.57	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Tai'zhou	0.00	0.00	0.00	0.00	1.49	0.00	0.00	0.00	0.00	2.59	0.00
Lishui	0.00	5.82	10.64	17.29	25.22	25.78	21.04	14.33	4.11	0.43	0.00
YRD mean	12.96	15.06	15.37	16.50	22.31	26.21	25.39	23.67	24.83	20.49	16.75
Jiangsu mean	19.66	21.27	20.84	22.95	30.66	33.04	34.55	37.63	39.64	33.70	28.88
Zhejiang mean	6.21	9.09	10.31	10.37	14.47	20.51	16.88	9.32	9.58	6.73	3.93