A dynamic analysis of air pollution emissions in China: Evidence from nonparametric additive regression models

Ecological Indicators 63 (2016) 346–358

Contents lists available at ScienceDirect

Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind

A dynamic analysis of air pollution emissions in China: Evidence from nonparametric additive regression models Bin Xu a,b , Liangqing Luo a,b , Boqiang Lin c,∗ a

School of Statistics, Jiangxi University of Finance and Economics, Nanchang, Jiangxi 330013, PR China Research Center of Applied Statistics, Jiangxi University of Finance and Economics, Nanchang, Jiangxi 330013, PR China Collaborative Innovation Center for Energy Economics and Energy Policy, China Institute for Studies in Energy Policy, Xiamen University, Xiamen, Fujian 361005, PR China b c

a r t i c l e

i n f o

Article history: Received 20 May 2015 Received in revised form 5 October 2015 Accepted 10 November 2015 Keywords: PM2.5 emissions STIRPAT model Nonparametric additive regression models

a b s t r a c t PM2.5 emissions not only have serious adverse health effects, but also impede transportation activities, especially in air and highway transport. As a result, PM2.5 emissions have become a public policy concern in China in recent years. Currently, the vast majority of existing researches on PM2.5 are based on natural science perspective. Very few economic studies on the subject have been conducted with linear models. This paper adopts provincial panel data from 2001 to 2012, and uses the STIRPAT model and nonparametric additive regression models to examine the key driving forces of PM2.5 emissions in China. The results show that the nonlinear effect of economic growth on PM2.5 emissions is consistent with the Environmental Kuznets Curve (EKC) hypothesis. The nonlinear impact of urbanization exhibits an inverted “U-shaped” pattern due to the rapid development of urban real estate in the early stages and the strengthening of environmental protection measures in the latter stage. Coal consumption follows an inverted “U-shaped” relationship with PM2.5 emissions owing to massive coal consumption at the beginning and efforts to optimize the energy structure as well as technological progress in clean energy in the latter stages. The nonlinear inverted “U-shaped” impact of private vehicles may be due to the different roles of scale, structural and technical effects at different stages. However, energy efficiency improvement follows a positive “U-shaped” pattern in relation to PM2.5 emissions because of differences in the scale of the economy and the speed of technological progress at different times. As a result, the differential dynamic effects of the driving forces of PM2.5 emissions at different times should be taken into consideration when initiating policies to reduce PM2.5 emissions in China. © 2015 Elsevier Ltd. All rights reserved.

1. Introduction China is currently in a rapid urbanization and industrialization process (Xue et al., 2015). There is rapid increase in the number of motor vehicles as well as energy consumption (e.g., coal), coupled with large-scale fixed asset building activities (Zhang and Cao, 2015). These activities are exposing China to largescale, severe and persistent air pollution problems (Meng et al., 2015). Numerous studies have demonstrated that PM2.5 (fine particles) is the main cause of air pollution (Wang et al., 2015a). PM2.5 not only includes many small particles such as organic

∗ Corresponding author. Collaborative Innovation Center for Energy Economics and Energy Policy, China Institute for Studies in Energy Policy, Xiamen University, Fujian, 361005, China. Tel.: +86 5922186076; fax: +86 5922186075. E-mail addresses: [email protected], [email protected] (B. Lin). http://dx.doi.org/10.1016/j.ecolind.2015.11.012 1470-160X/© 2015 Elsevier Ltd. All rights reserved.

carbon, nitrates, ammonium salts and sulfates, but also includes various metal elements such as sodium, magnesium, calcium, aluminum, zinc, arsenic, cadmium and copper (Zhao et al., 2014; Huang et al., 2014). These small particles are easily inhaled, and may pose health risks by contaminating the blood (Yang et al., 2013). PM2.5 pollution caused about 1.25 million deaths in China in 2010, accounting for nearly 40% of the global total premature deaths (Wang et al., 2012). Therefore, PM2.5 emissions have become an urgent issue of public concern in recent years. PM2.5 emissions have been analyzed extensively in the literature (Mallia et al., 2015; Mardones and Sanhueza, 2015; Meng et al., 2015; Zhang and Cao, 2015). Most of the existing studies only use linear models to analyze the influences of the driving forces of PM2.5 emissions. In fact, there are a large number of linear and nonlinear relationships between economic variables (Catalano and Figliola, 2015).

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

The paper is concerned with the impacts of the driving forces of PM2.5 emissions in China at the aggregate level. Using a panel data set covering 29 provinces over the period 2001–2012, we employ the STIRPAT model and nonparametric additive regression models to explore the effects of the influencing factors of PM2.5 emissions in China. Non-parametric additive can capture the linear and nonlinear links between economic variables. The remaining parts of the paper are organized as follows. Section 2 briefly reviews the related literature and previous studies on PM2.5 emissions. Section 3 describes the applied method and the model. Section 4 presents the empirical results. Section 5 discusses the results of the empirical analysis. Conclusions and policy suggestions are provided in Section 6.

2. Literature review PM2.5 pollution is a natural phenomenon, but it is a man-made contamination caused by human economic activities (Mardones and Sanhueza, 2015). Therefore, from an economic and social point of view, analyzing the main driving forces of PM2.5 emissions is conducive for reducing PM2.5 emissions and stemming its hazardous impacts on human health. The existing literature has extensively studied PM2.5 emissions with different methods. Firstly, the classical method is structural decomposition analysis. PM2.5 emissions are decomposed into economic growth, capital formation, exports, production structure and emission intensity. Guan et al. (2014) research PM2.5 emissions growth in China by analyzing economic growth, fixed asset investment and exports, and Djalalova et al. (2010) extend the research to emission intensity and decompose PM2.5 emissions into energy intensity, energy structure and conversion efficiency for Europe. The second method is bottom-up sector-based analysis. Zhang et al. (2015) investigate PM2.5 pollutants emissions in the metropolis using a top-down approach. Liu et al. (2014) develop an integrated assessment model to analyze air pollutants from China’s iron and steel industry, and find that command-and-control instrument has excellent impact in controlling pollutants emissions. The method is also used in the study of energy consumption (Cai et al., 2015; Farzan et al., 2015) and CO2 emissions (Zhang and Chen, 2014; Tang et al., 2015). The third method is dynamic factor analysis (DFA). It has been widely applied in the analysis of the influencing factors of air pollutants (Shi et al., 2008; Ielpo et al., 2013; Yu et al., 2015) and in forecasting macroeconomic variables (Brauning and Koopman, 2014; Palardy and Ovaska, 2015). The fourth method is econometric model. Using panel econometric models, Loftus et al. (2015) study PM2.5 emissions in American agricultural areas; Olson and Burke (2006) research the seasonal variation and factors of PM2.5 emissions; and Sica and Susnik (2014) investigate the relationships between economic growth and PM2.5 contaminants with provincial data in Italy. Patel et al. (2013), Bozlaker et al. (2014) and Shen et al. (2014) examine the effects of different means of transport on PM2.5 emissions, and found the extensive use of motor vehicles has become one of the main sources of PM2.5 emissions. Khan et al. (2015) determine the major factors of PM2.5 emissions using time series models. Though PM2.5 emissions have been discussed extensively in the literature, there is a major shortcoming. In other words, most of these studies only use linear models to analyze the influence of the driving forces of PM2.5 emissions. Nonlinear relationships embodied in economic variables are largely ignored. Granger (1988) pointed out that the world is almost certainly constituted by nonlinear relationships. In this paper, we investigate the linear and nonlinear effects of the influencing factors of PM2.5 emissions using nonparametric additive regression model, since it can capture the linear and nonlinear linkages between economic variables.

347

3. Methodology and model specification 3.1. Nonparametric additive regression models Nonlinear thinking refers to a kind of thinking that differs from linear thinking. That is, it looks at things from an unconventional perspective, such as systems thinking, fuzzy thinking and so on (Costigan and Brink, 2015). It probably will not follow normal logical thinking. In a system, if the output is not proportional to its input, it is nonlinear. In fact, almost all known systems, regardless of whether they are natural sciences or social sciences, are non-linear, if the input amount is large enough (Islam et al., 2015). Therefore, non-linear systems are much more than linear systems. The real world has always been a non-linear system, and a linear system is only an approximate simulation (Cheung et al., 2015). For a nonlinear system, even a small disturbance, such as a small change in the initial conditions, is likely to cause great difference in system behavior in the next moment (Ide and Wiggins, 2015). Similarly, every 1% increase in urbanization, economic growth, private cars, technological progress and coal consumption does not guarantee the same amplitude changes in PM2.5 emissions. That is, there may be non-linear relationships between these factors and PM2.5 emissions. This has been confirmed by numerous studies (Sueyoshi and Yuan, 2015; Salazar et al., 2011; Ning et al., 2015; Zhou et al., 2014). Nonparametric regression model is a data-driven model, and the relationships between the variables are portrayed by the sample data itself (Lee and Robinson, 2015). Compared to the linear models, the advantages of nonparametric additive regression model are obvious. First, it does not require pre-set relationships between the variables, and the regression functional form is not constrained (Curtis et al., 2014). Second, it has strong adaptability and high robustness, and the specific forms of the regression models are completely determined by the sample data itself, i.e., the non-parametric regression models are data-driven models (Piegorsch et al., 2014). Third, for non-linear non-homogeneous problems, non-parametric regression models have very good simulation results (Zhou et al., 2011). Fourth, the nonparametric additive regression models belong to a data-driven model. It does not make any prior assumptions on the relationships between economic variables. So there is a low possibility for over-fitting problem in this model, which will reduce or destroy the generalization ability of the model application (Farias et al., 2013). Consequently, we employ nonparametric additive regression models to capture the linear and nonlinear impacts of the driving forces of PM2.5 emissions in China in this paper. Additive regression models were first proposed by Stone in 1985. In additive regression models, the dependent variable Yi (i = 1, 2, n) is the sum of arbitrary functions fj (j = 1, 2,. . .,p), which are the functions of the independent variables Xi1 , Xi2 . . .,Xip , respectively. Its specific form is: Yi =

p 

f (xij ) + i , i ∼iid(0,  2 )

(1)

j=1

where f (xi ) is a nonparametric function and can be estimated with nonparametric regression methods. In order to make the estimation feasible, it is assumed that E(fj ) = 0 (j = 1, 2,. . .,p) and fj are smooth. In addition, the additive regression models can be expressed as: E(Yi |xi1 , xi2 , . . ., xip ) =

p 

f (xij )

(2)

j=1

From Eq. (2), it can be seen that additive regression model is an improvement on linear models, where each explanatory variable is

348

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

represented with more general form fj (xij ) rather than traditional linear form ˇi xi . Once the additive regression models are used to fit the data, we can obtain p parallel functions that can be used to explain and predict the dependent variable. In order to provide a more visual interpretation of the effects of the influencing factors of PM2.5 emissions, and better connect and compare with the existing results obtained with linear regression models, the linear part is added to the additive regression models as follow: E(Yi |xi1 , xi2 , . . ., xip ) = a +

p 

ˇj xij +

j=1

p 

fj (xij )

(3)

j=1

where a and ˇi are the parameters of linear regression. Other parts are the same as in Eq. (2). Since the additive regression models have been widely used in economic (Linton and Hardle, 1996), political (Beck and Jackman, 1998), medical (Piegorsch et al., 2014) and environmental fields (Schwartz, 1994; Stauch and Jarvis, 2006; Gupta, 2012), many scholars have devoted time to research the estimation methods for additive regression models. Taking into account that the back-fitting algorithm has the advantages of ingenious iterative methods and simplified calculation, this paper utilizes it to perform the parameter estimation. The back-fitting algorithm is briefly described as follows: If the linear part of Eq. (3) is taken p as a special nonparametric function, denoted by g(xi )↔a + ˇ x , the estimations j=1 j ij reduce to estimating the function g(·) and fj (·). When one fi is estimated, assuming other fj (·) and g(·)  are known, the partial residual can be defined as rik = yi − g(xi ) − f (x ). By minimizing j= / k k ik the partial residual, we can get fˆk (xik ) = E(rik |xi ). Continuously repeating the above estimation process, we obtain the estimated values fˆ1 , fˆ2 , . . ., fˆp . Similarly, assuming all fj are fixed, we can ˆ when estimating g(xi ). The specific get the optimal estimate ˇ estimation procedure is as follows. First, there is need to initialize the functions gˆ 0 (xi ), fˆ10 (xi1 ), fˆ20 (xi2 ), . . ., fˆp0 (xip ). Assuming g(xi ) and fˆ 0 (xi2 ), . . ., fˆ 0 (xip ) are fixed, we can obtain fˆ 2 (xi1 ) through the 2

p

1

above estimation method. Second, repeating the above estimation process, we can further obtain gˆ 1 (xi ), fˆ11 (xi1 ), fˆ21 (xi2 ), . . ., fˆp1 (xip ). We continually perform the above iterative process until RSS =

n 

p

2

yj − g(xi ) − f (x ) reaches the preset convergence j=1 j ij criteria. Finally, we get the parameter estimates of the linear part and the nonparametric functions in Eq. (3). i=1

3.2. Model specification The IPAT identity (I = PAT) (Eq. (4)) is often used as a basis for investigating the role of the various factors driving environmental pollution (Chertow, 2001): I =P·A·T

(4)

where I represents the emission level of a pollutant, P denotes the population size, A represents a country’s affluence and T is technological progress. In order to fully examine the factors affecting environmental change, the IPAT model is simple and has limitations. First, the IPAT model is just a mathematical formula, which cannot directly test the hypothesis on how the various factors affect environmental change (Li and Yuan, 2014; Tursun et al., 2015). Second, the IPAT model simply assumes that the elasticities of population, prosperity and technology on environment change are unity, which is in conflict with the EKC hypothesis (Huo et al., 2015; Wang and Zhao, 2015). Thus, using this model as a basis, Dietz and Rosa (1997) proposed the STIRPAT model as follows: It = aPtb Act Ttd et

(5)

where a represents the intercept term, P, A and T are the same as in Eq. (4), b, c and d represent the elasticities of environmental impacts with respect to P, A and T respectively, et is the random disturbance and the subscript t denotes the year as it is an annual data analysis. The STIRPAT model has been applied to analyze the influence of the impacting factors of environmental pollution (Wang et al., 2011; Liddle, 2013; Xu and Lin, 2015a). In order to eliminate possible heteroscedasticity, all variables take logarithmic form. Given the panel data of 29 Chinese provinces, Eq. (5) can be written as below: LIit = La + b(LPit ) + c(LAit ) + d(LTit ) + eit

(6)

where P represents population size (104 persons), A is measured by the per capita GDP (RMB), T is a technology index and is measured by energy intensity (ton of energy use per 104 yuan) and i denotes the province as we conduct a regional analysis. In order to investigate the impacts of the driving forces of China’s PM2.5 emissions, Eq. (6) can be rewritten as follows: LPM2.5it = La + b(LPOPit ) + c(LGDPit ) + d(LEIit ) + eit

(7)

where PM2.5 represents PM2.5 emissions intensity in China (micrograms per cubic meter-␮g/m3 ), POP is population size (104 persons), and GDP denotes economic development level measured in real per capita GDP of 2001 constant yuan. EI represents energy intensity and is proxied by energy consumption divided by total outputs (ton of energy use per 104 yuan). This has been used to identify the changes in pollution emissions by Lin and Du (2014), Li and Lin (2015) and Shahbaz et al. (2015). A and e are the same as in Eq. (6). To further our analysis of the driving forces of China’s PM2.5 emissions, we expand the STIRPAT model by incorporating urbanization level, private vehicle inventory and coal consumption into the model to take cognizance of the specific situation in China for the following reasons. First, China is currently in a process of rapid urbanization (Wang et al., 2014; Zhang et al., 2014). It is predicted that by 2020 60% of the population will be living in urban areas (Normile, 2008). It has been recognized that rapid urbanization would lead to an increase in PM2.5 emissions (Chan and Yao, 2008; Niu et al., 2013). Consequently, it is very necessary to introduce urbanization level (URB) into the model to analyze the impact of urbanization on China’s PM2.5 emissions. Second, owing to an increase in residents’ income in recent years, private vehicle demand continues to rise. Private vehicle stock increased from 6.25 million units in 2000 to 88.39 million units in 2012 in China, an average annual growth rate of 25%. The rapid growth of road vehicles, particularly private cars, has resulted in continued increase in energy consumption. For instance, total energy demand of the transport sector was estimated to have increased from 57 million tons of standard coal equivalent (Mtce) in 2000 to 86 Mtce in 2005, and private cars contributed nearly half of the demand growth (Yan and Crookes, 2009). Moreover, private vehicle is widely acknowledged as a major contributor to the increase in PM2.5 emissions in China (Du et al., 2012; Walsh, 2014). Thus, we include private vehicle population (PC) in the model. Finally, China is already the current largest energy-consuming country in the world (since 2010) (Jiang and Lin, 2012). Moreover, China’s energy consumption comes mainly from coal (Michieka and Fletcher, 2012) which accounted for an average of 75% of China’s total energy consumption over 1990–2013 period. China is already the world’s largest coal consumer (Bloch et al., 2015) and coal will continue to be a main source of China’s energy consumption for a long time (Li and Leung, 2012). But coal combustion produces large quantities of pollutants such as PM2.5 emissions, which have significant adverse impacts on the environment and health (Giere et al., 2006). Hence,

B. Xu et al. / Ecological Indicators 63 (2016) 346–358 Table 1 Distribution of the 29 administrative regions in the three areas of China.

3.3. Data source and description

Area

Administrative regions

Eastern

Liaoning, Shanghai, Jiangsu, Zhejiang, Tianjin, Fujian, Shandong, Hebei, Guangdong, Hainan, Beijing Shanxi, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei, Hunan, Neimenggu Sichuan, Guizhou, Yunnan, Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang, Guangxi

Central Western

349

Table 2 Definition of all relevant variables used in the study. Variable

Definition

Units of measurement

PM2.5 GDP EI URB PC COAL POP

PM2.5 emissions intensity Per capita GDP Energy intensity Urbanization level Private vehicles inventory Coal consumption Population size

␮g per m3 Yuan Tce per 10,000 Yuan Percent 104 units 104 tce 104 people

k PM2.5 =

Notes: The data of Tibet province is excluded due to incomplete data. The data for Chongqing Municipality is merged together with that of Sichuan Province.

coal consumption (COAL) is introduced into the econometric model in this study. Based on the above analysis and STIRPAT model, we establish the econometric model of PM2.5 emissions in China as follows: LPM2.5it = La + ˇ1 LPOPit + ˇ2 LGDPit + ˇ3 LEIit + ˇ4 LURBit + ˇ5 LPCit + ˇ6 LCOALit + it

(8)

where PM2.5 represents PM2.5 emissions intensity in China (␮g/m3 ), POP, GDP and EI are the same as in Eq. (7), URB denotes urbanization level, and is measured by the percentage of urban population to total population (%), PC represents private vehicle inventory (104 units) and COAL indicates coal consumption (104 standard coal equivalent). Given that the priori model form could easily cause specification bias, this paper uses the nonparametric additive regression models to fit the dynamic impacts of the influencing factors of PM2.5 emissions. Specifically, we introduce the nonparametric part into the econometric model (8) and take first differences of all variables in the equation in order to eliminate the multicollinearity problem that may exist in the model (Xu and Lin, 2015b; Zhang and Liu, 2015). Thus, model (8) becomes: LPM2.5it = La + ˇ1 LPOPit + ˇ2 LGDPit + ˇ3 LEIit + ˇ4 LURBit + ˇ5 LPCit +ˇ6 LCOALit + g1 (LPOPit ) + g2 (LGDPit ) + g3 (LEIit ) + g4 (LURBit ) + g5 (LPCit ) + g6 (LCOALit ) + it

3.3.1. Data source The panel data set consists of cross-province observations for 29 provinces covering the period 2001–2012. The 29 administrative regions are shown in Table 1. China provincial PM2.5 panel data (2001–2010) are obtained from Battelle Memorial Institute and the Center for International Earth Science Information Network (CIESIN) at Columbia University. The remaining provincial PM2.5 panel data (2011–2012) are calculated based on the PM2.5 data compiled by the Chinese Environmental Monitoring Center (CNEMC). The formula for provincial PM2.5 pollution is as follows:

(9)

i=1

PM2.5i × Si

k

S i=1 i

(i = 1, 2, . . .K)

(10)

where PM2.5 represents the annual average value of PM2.5 pollution in one province, i means the ith city publishing PM2.5 data in one province, S denotes the size of the city area. The annual panel data of the other economic variables are collected from China Statistical Yearbook (2002–2013) and the provincial statistical yearbooks (2002–2013). The definitions of the variables are shown in Table 2 and Table 3 shows the statistical description of all the variables in the model.

3.3.2. Data description Based on the annual data of the independent and dependent variables, we analyze the relative changes in PM2.5 emissions, per capita GDP, energy intensity, private vehicle inventory, urbanization level, coal consumption and population size in China over 2001–2012. As shown in Fig. 1, PM2.5 emissions showed a fluctuating growth trend, and it reached the highest value (29.31 ␮g per m3 ) in 2007. Energy intensity decreased significantly from 16.69 tce per 10,000 yuan in 2001 to 8.64 tce per 10,000 yuan in 2012. Private car ownership had the highest average annual growth rate, at 25% over the sample period. Economic growth also showed rapid growth characteristics, with an average annual growth rate of 15%. Urbanization showed a steady growth trend with the rate exceeding 50% since 2011. Numerous studies have confirmed that coal combustion is a major source of PM2.5 emission since high-polluting coal has been the main source of energy consumption for a long time. China’s total coal consumption continued to grow steadily, as the country has become the world’s largest coal consumer. As shown, the population size increased from 1.276 billion people in 2001 to 1.354 billion people in 2012, with the lowest growth rate of 0.54%. This is mainly due to two reasons. First, the Chinese government has been implementing strict family planning policy. Second, rising costs of child education and health care makes many families having fewer children.

Table 3 The statistical description of the dependent and explanatory variables in the model. Variable

Units of measurement

PM2.5 GDP EI URB PC COAL POP

␮g per m3 Yuan Tce per 10,000 yuan Percent 104 units 104 tce 104 people

Mean 26.98 11,520.39 1.34 41.89 135.09 6412.07 4496.64

Std. dev.

Min

11.52 7644.34 0.82 15.27 155.47 5749.71 2919.74

2.17 3000.00 0.17 16.24 3.20 53.60 523.10

Max 51.94 41,119.22 4.35 89.76 864.30 30,110.91 12,440.80

350

B. Xu et al. / Ecological Indicators 63 (2016) 346–358 30

17

PM2.5

EI

16 15 Tce per 10,000 yuan

ug per m3

29

28

27

14 13 12 11 10

26

9

25

8

01

02

03

04

05

06

07

08

09

10

11

12

01

02

03

04

05

06

Time (Year) 54

50

7,000

48

6,000 10,000 units

%

09

10

11

12

PC

8,000

46 44

5,000 4,000

42

3,000

40

2,000

38

1,000 0

36 01

02

03

04

05

06 07 Time (Year)

08

09

10

11

01

12

02

03

04

05

06 07 Time (Year)

08

09

10

11

12

40

2,600

COAL 2,400

36

2,200

32

2,000

28

1000 yuan

Million tce

08

9,000

URB

52

07

Time (Year)

1,800 1,600

GDP

24 20

1,400

16

1,200

12 8

1,000 01

02

03

04

05

06 07 Time (Year)

08

09

10

11

12

01

02

03

04

05

06 07 Time (Year)

08

09

10

11

12

136,000

POP

135,000

10,000 people

134,000 133,000 132,000 131,000 130,000 129,000 128,000 127,000 01

02

03

04

05

06 07 Time (Year)

08

09

10

11

12

Fig. 1. The trends of PM2.5 emissions, per capita GDP, energy intensity, urbanization level, private vehicle inventory, coal consumption and population size over 2001–2012.

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

351

Table 4 Results of panel unit root tests. Series

Fisher ADF

Fisher PP

IPS

Constant

Trend and intercept

Constant

Trend and intercept

Constant

Trend and intercept

Levels

PM2.5 GDP EI URB COAL PC POP

48.566 9.128 3.115 125.218*** 99.473*** 57.486 21.002

82.068** 123.681*** 74.232* 63.809 56.854 80.219** 50.787

44.867 8.343 2.503 158.533*** 147.593*** 35.306 37.261

113.858*** 230.418*** 154.211*** 76.424** 36.4855 92.128*** 70.373

−4.990*** 7.484 11.541 −18.463*** −2.325*** 3.864 5.113

−2.452*** −5.174*** −1.721** −12.066* 1.1049 −1.965 0.7113

First difference

PM2.5 GDP EI URB COAL PC POP

244.768*** 268.856*** 158.579*** 146.544*** 140.156*** 198.377*** 116.091***

165.394*** 232.336*** 132.337*** 142.850*** 151.395*** 165.546*** 128.478***

385.907*** 362.086*** 200.554*** 154.959*** 143.754*** 220.059*** 130.400***

290.185*** 328.565*** 210.076*** 180.785*** 200.853*** 244.752*** 290.326***

−12.659*** −13.751*** −7.685*** −17.490*** −6.181*** −9.954*** −4.053***

−6.570*** −8.372*** −4.262*** −10.380*** −5.866*** −6.110*** −2.634***

Note: Lags are all selected automatically by AIC and SC standard. * Represents p < 0.10. ** Represents p < 0.05. *** Represents p < 0.01. Table 5 Testing for bivariate co-integration between PM2.5 emission and its influencing factors. Test statistics

COAL

EI

URB

GDP

PC

POP

Panel v-statistic Panel rho-statistic Panel PP-statistic Panel ADF-statistic Group rho-statistic Group PP-statistic Group ADF-statistic

1.160 −2.215** −7.118*** −7.490*** 0.214 −10.427*** −8.464***

0.699 −2.309** −7.850*** −8.136*** 0.059 −9.928*** −8.600***

0.157 −2.280** −6.990*** −7.387*** 0.147 −8.710*** −7.589***

0.530 −2.254** −7.604 −7.821* 0.144 −9.565*** −8.114***

1.233 −2.146** −7.170*** −7.575*** 0.197 −9.192*** −8.115***

−0.249 −2.238** −7.383*** −6.910*** 0.207 −9.345*** −5.718***

Note: Lags are all selected automatically by AIC and SC standard. * Represents p < 0.10. ** Represents p < 0.05. *** Represents p < 0.01.

4. Empirical results 4.1. Unit root test and co-integration test Generally speaking, most sequences of economic variables are non-stationary. If a non-stationary sequence is used to conduct a regression analysis, it would generate spurious regression. The approach to non-stationary sequence is generally to transform it into a stationary sequence so that the corresponding methodology can be applied to conduct empirical analysis. The stability test of the panel data is investigated using panel unit root test, and the results are shown in Table 4. The results suggest that the majority of the variables are non-stationary, but their first difference series are stationary. Based on the panel data, all the independent variables are bivariate-cointegrated with PM2.5 emissions (Table 5). From Table 5, it can be seen that there is a significant cointegration relationship between each independent variable and PM2.5 emissions at confidence levels of 1%, 5% or 10%. Furthermore, based on the KAO panel test (Kao, 1999), the ADF (Augmented Dickey-Fuller) stat (−5.437, p = 0.0000) and the HAC (heteroskedasticity and autocorrelation-consistent) variance (0.008), there is a co-integration relationships between PM2.5 emissions and all explanatory variables. 4.2. Test the relationship between the independent and dependent variables The scatter chart visualizes the specific form of the relationship between the independent and dependent variables. Using panel data across 29 provinces, we obtain the scatter plot of the link

between PM2.5 emissions and its influencing factors. As shown in Fig. 2, there are lots of linear and non-linear relationships between PM2.5 emissions and its influencing factors (i.e., LCOAL, LEI, LURB, LGDP, LPC and LPOP). It also verifies that investigating the effects of the driving forces of PM2.5 emissions using nonparametric additive regression model is reasonable and applicable. 4.3. Robustness test of nonparametric additive regression models In order to confirm that the nonparametric additive regression models used in this study is robust, this paper uses linear panel data models to fit the sample data with the same dependent and explanatory variables. The Hausman test result (21.859, p = 0.0013) and Likelihood ratio test (230.719, p = 0.0000) suggest that the fixed effects model is the appropriate model to analyze the effects of the driving forces of PM2.5 emissions. Table 6 reports the estimation results of the linear fixed effects model (Eq. (8)) and the linear part of nonparametric additive regression models (Eq. (9)). It can be seen that the conclusions of the nonparametric additive regression models is consistent with those of the linear fixed effects models, with only a slight difference in the significance. The basic conclusions however remain unchanged. In order to compare the advantages and disadvantages of the two estimation models, we calculate the residual sum of squares. The results in Table 6 show that the residual sum of squares in the nonparametric additive regression models is less than that in the traditional linear panel data models. This indicates that the fitting effects of the nonparametric additive regression models are better. Based on the above results, there are reasons to believe that the nonparametric additive regression models not only accurately

352

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

grasp the linear relationship, but can also depict the nonlinear contribution of the factors to PM2.5 emissions. More importantly, the nonparametric additive regression models have better goodness of fit than the traditional linear regression models. Therefore, we employ it to implement the empirical analysis, making the results more accurate and scientific. 4.0

4.4. Linear effects analysis Table 6 provides the estimated results of the linear effects of the driving forces of PM2.5 emissions. It can be seen that all the estimated coefficients are statistically significant at the level of 1%, 5% or 10% (Eq. (9)).

LPM2.5

3.5 3.0 2.5 2.0 1.5 1.0 LGDP

0.5 7.5

4.0

8.0

8.5

9.0

9.5

10.0

10.5

(LGDP_AN,LPM_AN) (LGDP_FJ,LPM_FJ) (LGDP_GS,LPM_GS) (LGDP_GXI,LPM_GXI) (LGDP_HEB,LPM_HEB) (LGDP_HLJ,LPM_HLJ) (LGDP_HUN,LPM_HUN) (LGDP_JSU,LPM_JSU) (LGDP_LIAON,LPM_LIAON) (LGDP_NX,LPM_NX) (LGDP_SC,LPM_SC) (LGDP_SHAND,LPM_SHAND) (LGDP_SHANXI,LPM_SHANXI) (LGDP_XJ,LPM_XJ) (LGDP_ZJ,LPM_ZJ)

(LGDP_BJ,LPM_BJ) (LGDP_GD,LPM_GD) (LGDP_GUIZ,LPM_GUIZ) (LGDP_HAIN,LPM_HAIN) (LGDP_HENAN,LPM_HENAN) (LGDP_HUB,LPM_HUB) (LGDP_JIL,LPM_JIL) (LGDP_JXI,LPM_JXI) (LGDP_NMG,LPM_NMG) (LGDP_QHAI,LPM_QHAI) (LGDP_SHAI,LPM_SHAI) (LGDP_SHANX,LPM_SHANX) (LGDP_TJIN,LPM_TJIN) (LGDP_YUN,LPM_YUN)

(LURB_AN,LPM_AN) (LURB_FJ,LPM_FJ) (LURB_GS,LPM_GS) (LURB_GXI,LPM_GXI) (LURB_HEB,LPM_HEB) (LURB_HLJ,LPM_HLJ) (LURB_HUN,LPM_HUN) (LURB_JSU,LPM_JSU) (LURB_LIAON,LPM_LIAON) (LURB_NX,LPM_NX) (LURB_SC,LPM_SC) (LURB_SHAND,LPM_SHAND) (LURB_SHANXI,LPM_SHANXI) (LURB_XJ,LPM_XJ) (LURB_ZJ,LPM_ZJ)

(LURB_BJ,LPM_BJ) (LURB_GD,LPM_GD) (LURB_GUIZ,LPM_GUIZ) (LURB_HAIN,LPM_HAIN) (LURB_HENAN,LPM_HENAN) (LURB_HUB,LPM_HUB) (LURB_JIL,LPM_JIL) (LURB_JXI,LPM_JXI) (LURB_NMG,LPM_NMG) (LURB_QHAI,LPM_QHAI) (LURB_SHAI,LPM_SHAI) (LURB_SHANX,LPM_SHANX) (LURB_TJIN,LPM_TJIN) (LURB_YUN,LPM_YUN)

(LPC_AN,LPM_AN) (LPC_FJ,LPM_FJ) (LPC_GS,LPM_GS) (LPC_GXI,LPM_GXI) (LPC_HEB,LPM_HEB) (LPC_HLJ,LPM_HLJ) (LPC_HUN,LPM_HUN) (LPC_JSU,LPM_JSU) (LPC_LIAON,LPM_LIAON) (LPC_NX,LPM_NX) (LPC_SC,LPM_SC) (LPC_SHAND,LPM_SHAND) (LPC_SHANXI,LPM_SHANXI) (LPC_XJ,LPM_XJ) (LPC_ZJ,LPM_ZJ)

(LPC_BJ,LPM_BJ) (LPC_GD,LPM_GD) (LPC_GUIZ,LPM_GUIZ) (LPC_HAIN,LPM_HAIN) (LPC_HENAN,LPM_HENAN) (LPC_HUB,LPM_HUB) (LPC_JIL,LPM_JIL) (LPC_JXI,LPM_JXI) (LPC_NMG,LPM_NMG) (LPC_QHAI,LPM_QHAI) (LPC_SHAI,LPM_SHAI) (LPC_SHANX,LPM_SHANX) (LPC_TJIN,LPM_TJIN) (LPC_YUN,LPM_YUN)

11.0

LPM2.5

3.5 3.0 2.5 2.0 1.5 1.0 LURB

0.5 2.4

2.8

3.2

3.6

4.0

4.4

4.8

4.0 LPM2.5 3.5 3.0 2.5 2.0 1.5 1.0 LPC

0.5 1

2

3

4

5

6

7

Fig. 2. Scatter chart of the relationships between PM2.5 emissions and its influencing factors (i.e., economic growth, urbanization, private cars, energy intensity, coal consumption, population size). Notes: The scatter charts are drawn based on 29 provincial panel data (AH, BJ, FJ, GD, GS, GUIZ, GXI, HAIN, HEB, HENAN, HLJ, HUB, HUN, JIL, JSU, JXI, LIAON, NMG, NX, QHAI, SC, SHAI, SHAND, SHANX, SHANXI, JJIN, XJ, YUN, ZJ represent Anhui, Beijing, Fujian, Guangdong, Gansu, Guizhou, Guangxi, Hainan, Hebei, Henan, Heilongjiang, Hubei, Hunan, Jilin, Jiangsu, Jiangxi, Liaoning, Neimenggu, Ningxia, Qinghai, Sichuan, Shanghai, Shandong, Shaanxi, Shanxi, Jilin, Xinjiang, Yunnan, Zhejiang provinces, respectively). Each line in the figure represents the relationship between two variables in a province. Therefore, each graph has 29 lines.

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

353

4.0 LPM2.5

3.5

(LEI_AN,LPM_AN) (LEI_FJ,LPM_FJ) (LEI_GS,LPM_GS) (LEI_GXI,LPM_GXI) (LEI_HEB,LPM_HEB) (LEI_HLJ,LPM_HLJ) (LEI_HUN,LPM_HUN) (LEI_JSU,LPM_JSU) (LEI_LIAON,LPM_LIAON) (LEI_NX,LPM_NX) (LEI_SC,LPM_SC) (LEI_SHAND,LPM_SHAND) (LEI_SHANXI,LPM_SHANXI) (LEI_XJ,LPM_XJ) (LEI_ZJ,LPM_ZJ)

3.0 2.5 2.0 1.5 1.0 LEI

0.5 -2 4.0

-1

0

1

(LEI_BJ,LPM_BJ) (LEI_GD,LPM_GD) (LEI_GUIZ,LPM_GUIZ) (LEI_HAIN,LPM_HAIN) (LEI_HENAN,LPM_HENAN) (LEI_HUB,LPM_HUB) (LEI_JIL,LPM_JIL) (LEI_JXI,LPM_JXI) (LEI_NMG,LPM_NMG) (LEI_QHAI,LPM_QHAI) (LEI_SHAI,LPM_SHAI) (LEI_SHANX,LPM_SHANX) (LEI_TJIN,LPM_TJIN) (LEI_YUN,LPM_YUN)

2

LPM2.5

3.5

(LCOAL_AN,LPM_AN) (LCOAL_FJ,LPM_FJ) (LCOAL_GS,LPM_GS) (LCOAL_GXI,LPM_GXI) (LCOAL_HEB,LPM_HEB) (LCOAL_HLJ,LPM_HLJ) (LCOAL_HUN,LPM_HUN) (LCOAL_JSU,LPM_JSU) (LCOAL_LIAON,LPM_LIAON) (LCOAL_NX,LPM_NX) (LCOAL_SC,LPM_SC) (LCOAL_SHAND,LPM_SHAND) (LCOAL_SHANXI,LPM_SHANXI) (LCOAL_XJ,LPM_XJ) (LCOAL_ZJ,LPM_ZJ)

3.0 2.5 2.0 1.5 1.0 LCOAL

0.5 3

4

5

6

7

8

9

10

(LCOAL_BJ,LPM_BJ) (LCOAL_GD,LPM_GD) (LCOAL_GUIZ,LPM_GUIZ) (LCOAL_HAIN,LPM_HAIN) (LCOAL_HENAN,LPM_HENAN) (LCOAL_HUB,LPM_HUB) (LCOAL_JIL,LPM_JIL) (LCOAL_JXI,LPM_JXI) (LCOAL_NMG,LPM_NMG) (LCOAL_QHAI,LPM_QHAI) (LCOAL_SHAI,LPM_SHAI) (LCOAL_SHANX,LPM_SHANX) (LCOAL_TJIN,LPM_TJIN) (LCOAL_YUN,LPM_YUN)

11

4.0 LPM2.5 3.5 3.0 2.5 2.0 1.5 1.0 LPOP

0.5 6.0

6.5

7.0

7.5

8.0

8.5

9.0

(LPOP_AN,LPM_AN) (LPOP_FJ,LPM_FJ) (LPOP_GS,LPM_GS) (LPOP_GXI,LPM_GXI) (LPOP_HEB,LPM_HEB) (LPOP_HLJ,LPM_HLJ) (LPOP_HUN,LPM_HUN) (LPOP_JSU,LPM_JSU) (LPOP_LIAON,LPM_LIAON) (LPOP_NX,LPM_NX) (LPOP_SC,LPM_SC) (LPOP_SHAND,LPM_SHAND) (LPOP_SHANXI,LPM_SHANXI) (LPOP_XJ,LPM_XJ) (LPOP_ZJ,LPM_ZJ)

(LPOP_BJ,LPM_BJ) (LPOP_GD,LPM_GD) (LPOP_GUIZ,LPM_GUIZ) (LPOP_HAIN,LPM_HAIN) (LPOP_HENAN,LPM_HENAN) (LPOP_HUB,LPM_HUB) (LPOP_JIL,LPM_JIL) (LPOP_JXI,LPM_JXI) (LPOP_NMG,LPM_NMG) (LPOP_QHAI,LPM_QHAI) (LPOP_SHAI,LPM_SHAI) (LPOP_SHANX,LPM_SHANX) (LPOP_TJIN,LPM_TJIN) (LPOP_YUN,LPM_YUN)

9.5 Fig. 2. (Continued ).

The elasticity of economic growth is the greatest (0.527), indicating that economic activities are a major driver of PM2.5 emissions in China. The estimated result is supported by Guan et al. (2014), whose results reveal that economic development are a major contributor to PM2.5 emissions growth. Economic growth increases PM2.5 emissions mainly through the following two channels. First, rapid economic growth requires a lot of energy (e.g., coal, electricity), and China’s electrical energy is mainly sourced from thermal

power. Between 2001 and 2011, the average annual rate of thermal power in total electricity was 72.6%, which requires large amounts of coal. It is well known that coal combustion produces large amounts of PM2.5 . The second factor has to do with changes in China’s economic structure. In recent years, China’s secondary and tertiary industries have been rapidly growing. The secondary industry still accounts for nearly 50% of GDP (Gross domestic product) over the study period, consuming lots of fossil fuels and emitting

354

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

Table 6 Estimation results: linear part of nonparametric additive regression models and linear fixed effects model during 2001–2012. Variables

LGDP LEI LURB LPC LCOAL LPOP Intercept R2 SSR F-statistic Observations Individuals

Linear part of nonparametric additive regression model 0.527** −0.188*** 0.429*** 0.080*** 0.095*** 0.372*** 5.908*** 0.837 19.793 – 319 29

Linear fixed effects model 0.611*** −0.214*** 0.560*** 0.030 0.078*** 0.495*** −4.963*** 0.903 58.896 59.912*** 319 29

SSR represents residual sum of squares. ** Significant at 5% level. *** Significant at 1% level.

large amounts of PM2.5 . Urbanization passes the t-test with elasticity of 0.429. That is to say a 1% increase in urbanization level would lead to 0.429% increase in PM2.5 emissions when other factors remain constant. This means that urban development, rapid growth of urban population and diffusion expansion of urban area lead to a rapid increase in PM2.5 emissions. This verifies the real situation that urban areas are the main outbreak area of PM2.5 emissions. The estimated results conform to the results of Lin et al. (2014a), Lin et al. (2014b), which showed that urban expansion is primarily responsible for PM2.5 emissions growth. Coal consumption is significant with a coefficient of 0.095. A 1% growth in coal consumption would generate 0.095% increase in PM2.5 emissions. China is now the world’s largest coal consumer (Bhattacharya et al., 2015). According to the “World Energy Statistics Yearbook”, the world’s total coal consumption was 6.79 billion tons in 2014, and China accounted for 51.7% of total consumption. Although, as China’s air pollution problem get worse, the government has introduced new policies and regulations, such as enhancing the efficiency of coal-fired electricity generation and developing alternative energy and clean coal technologies. However, it cannot significantly reduce PM2.5 emissions in the short term due to the huge coal consumption, and coal consumption will continue to be a major source of PM2.5 emissions (Chen et al., 2015). The elasticity of private cars on PM2.5 emissions is positive (0.080%), indicating that private car exhaust emissions are a major factor of PM2.5 pollution. With rising incomes and improvement in living standard, private car ownership is rapidly rising. This will lead to an increase in PM2.5 emissions. Furthermore, PM2.5 pollution is exacerbated by urban traffic congestion (Ercan and Tatari, 2015). The negative sign of energy intensity indicates that higher energy efficiency is inclined to reduce PM2.5 emissions, corroborating the work of Lin et al. (2015). In order to reduce PM2.5 emissions, the government has gradually introduced a series of measures, such as developing oil-gas hybrid, clean energy, and pure electric vehicle technologies. These measures will help to mitigate PM2.5 emissions. Population has a positive relationship with PM2.5 emissions. A 1% increase in population size leads to an increase of 0.372% in PM2.5 emissions. PM2.5 pollution is mainly caused by anthropogenic activities, such as coal burning, large-scale infrastructure construction, straw burning and fireworks during festivities (Seidel and Birnbaum, 2015; Dekoninck et al., 2015). The greater the population, the more intense anthropogenic economic activities will be. This will produce more PM2.5 emissions.

4.5. Nonlinear effects analysis The estimated results of the nonlinear impacts of the driving factors of PM2.5 emissions are shown in Fig. 3. All the explanatory variables are statistically significant at the confidence level of 10% or higher. The nonlinear effects of economic growth on PM2.5 emissions show an inverted “U-shape” pattern, which supports the Environmental Kuznets Curve (EKC) hypothesis. This can be explained by the fact that in China’s early stages of economic growth, coal energy accounted for a much greater proportion of total energy consumption and thus produces more PM2.5 . With the optimization of the economic structure and the development of alternative energy, PM2.5 emissions intensity is experiencing a gradual decline. Energy intensity follows a positive “U-shaped” pattern in relation to PM2.5 emissions. This finding is consistent with Chang (2015). It indicates that technical progress can offset the negative environmental impact resulting from economic growth in the early stage. This effect disappears in the later stage of economic development due to the limited technological progress and rapid increase in total energy consumption (Recalde and Ramos-Martin, 2012). The nonlinear influence of urbanization on PM2.5 emissions shows an inverted “U-shaped” pattern, meaning that in the early stages of urbanization, the emissions intensity of urbanization gradually increased. However, when urbanization level surpasses a certain point, PM2.5 intensity of urbanization gradually declines. The nonlinear impact of private vehicles also shows an inverted “U-shaped” pattern, suggesting that in the early stages of private car growth, fossil fuel-powered private vehicles lead to a surge in PM2.5 emissions (scale effect). However, with the promotion of new energy cars, electric vehicles and stringent emission standards of vehicle exhaust, PM2.5 emission intensity of private vehicles experience a gradual decline (structure and technical effects). The relationship between coal consumption and PM2.5 emissions shows an inverted “U-shape” pattern. This result is in accordance with the findings of Menegaki and Tsagarakis (2015). The growth rate of coal consumption was faster than the advancements in clean technologies in the early stages, leading to a rapid increase in PM2.5 emissions (scale effect). But, this effect gradually disappears in the later stage due to increased R&D investment in clean energy technologies (technical effect) and rising coal prices (Saboori and Sulaiman, 2013). Population has a positive “U-shaped” influence on PM2.5 emissions. This result is supported by Lee et al. (2014). This implies that there are low-intensity anthropogenic activities due to low income and small population in the early stages, so energy consumption and PM2.5 emissions are not intensive. However, with increase in income, residents consume more energy (e.g., coal, electricity), construct more buildings and buy more vehicles in recent years. This inevitably leads to the rapid increase in PM2.5 emissions.

5. Discussion According to the above empirical results, we found several major phenomena. The first finding is that the nonlinear effect of economic growth on PM2.5 emissions demonstrates an inverted “U-shaped” pattern. This finding is supported by Jiao et al. (2014) and Kasman and Duman (2015). This can be explained by China’s economic development path. In the early stages of extensive economic growth, China’s relied mainly on investment and exports (Sun et al., 2015; Zhao and Tang, 2015). Expansion of investment and increase in exports lead to an increase in PM2.5 emissions (Guan et al., 2014). Fixed asset investment activities generate a lot of PM2.5 emissions (Panepinto et al., 2014). Building construction activities caused by fixed assets investment yield large amounts of construction dust in

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

355

Fig. 3. The nonlinear effects of driving forces on PM2.5 emissions. Notes: Shaded areas correspond to 95% confidence intervals.

two main ways – building construction-related transport activities and large areas of bare land. Dust is a major element in the formation of PM2.5 . From 2001 to 2012, the average annual growth rate of fixed-asset investment in China was 23%, which leads to a rapid increase in PM2.5 emissions. Also, high energy consumption associated with export production results in increase in PM2.5 emissions (Guan et al., 2009). Since the reform and opening up, following the international trade “comparative advantage” law, China vigorously develops labor-intensive industries (e.g., textiles, clothing and primary products industries) and processing industries (e.g., electronic products) (Claro, 2006). These industries consume large amounts of electrical energy. Because the vast majority of China’s electrical energy comes from thermal power (Zhong et al., 2015), it requires large amounts of coal consumption, which increase the intensity of PM2.5 emissions (Kong et al., 2015). With further economic development, the tertiary industry will be the main driving force of economic growth (Zhang and Wang, 2013). This has been confirmed by the history of economic development in developed countries. The rapid development of the tertiary industry will offset the effects of investment activities and exports on PM2.5 emissions. Thus, economic growth helps to reduce PM2.5 emissions at the later stage, which is consistent with the Environmental Kuznets Curve. The second finding is that energy efficiency improvement follows a positive “U-shaped” pattern in relation to PM2.5 emissions.

This finding has been proved by Chang (2015). The main reason is that progressive energy-saving and clean energy technologies help to reduce PM2.5 emissions in the early stages due to smaller production activities (technical effect) (Xu et al., 2009). At this stage, the number of manufacturers emitting dust and their production scales are small. Thus, technology effect in favor of PM2.5 emissions intensity decline at an early stage. As production scale in China expands and there is limited room for technological progress, production activities offset the influence of technological progress at the later stage (scale effect) (Li and Lin, 2015). For example, the average annual growth rate of the overall economy was 15.2% over 2001–2012, while that of technological progress was only 6%. The third finding is that the nonlinear impact of urbanization on PM2.5 emissions exhibits an inverted “U-shaped” pattern. This finding is consistent with Zhu et al. (2012) and Wang et al. (2015b). This is because in the early stages, rapid urbanization resulted in a sharp increase in China’s urban population (Li et al., 2015), which ultimately led to increase in demand for housing and rapid development of urban real estates (Zhan, 2015). For instance, real estate investment increased from 421.7 billion yuan in 2001 to 4937.4 billion yuan in 2012, an average annual growth of about 25%. It is well known that housing construction and real estate activities produce a lot of dust, and dust is the main element in the formation of PM2.5 (Meng et al., 2015). Thus, large-scale construction activities caused by urbanization generate a lot PM2.5 . However, at a

356

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

later stage, increasing environmental awareness and worsening air pollution force the government to develop stringent measures to control dust pollution from real estate construction (Schifano et al., 2013). Some of these measures include ensuring that bare construction sites are covered with vegetation and only clean-energy transport vehicles are allowed on the road. Thus, PM2.5 pollution intensity of urbanization gradually decreases. The fourth finding is that coal consumption follows an inverted “U-shaped” relationship with PM2.5 emissions. This finding is supported by Pao and Tsai (2010). In our opinion, it can be explained as follows. In the early stages, most manufacturers ignored research and use of clean coal technology, and residential energy consumption mainly comes from highly polluting coal (Lin et al., 2014a,b). According to the China Statistical Yearbook, the annual average share of coal consumption in total energy consumption was about 75% from 1980 to 2006. This leads to a rapid increase in PM2.5 emissions (scale effect). With the continued rise in energy prices, increasing public environmental awareness and increasingly stringent government environmental regulations, on the one hand, manufacturers take the initiative to optimize their energy consumption structure and increase the proportion of non-polluting hydroelectric, nuclear and natural gas consumption (composition effect) (Zhao et al., 2012). The proportion of hydro, nuclear and wind power in total energy increased from 7.9% in 2005 to 9.4% in 2012. Over the same period, the share of coal consumption declined from 68.0% to 66.6%. On the other hand, since the prices of fossil energy (e.g., coal and oil) are increasing, hydro and nuclear power are increasingly being used relative to high-polluting coal (Zheng et al., 2014). The share of electricity, liquefied petroleum gas, liquefied natural gas and natural gas in residential energy consumption per capita increased from 65.3% in 2005 to 77.8% in 2012. Thus, at the later stage, the effect of coal consumption on PM2.5 emissions gradually declines. The fifth finding is that the nonlinear impact of private vehicles also shows an inverted “U-shaped” pattern. This result is supported by Hua et al. (2010). In the early stages, more residents acquire private cars as income grows. Therefore, there is a surge in the number of private cars (Wu et al., 2014). According to the China Statistical Yearbook, private car ownership increased from 0.29 million units in 1985 to 88.4 million units in 2012. Moreover, at this stage, private cars are powered by fossil-fuel energy (e.g., gasoline and diesel oil), and motor vehicle emission standards are low (Jin et al., 2015). This leads to a rapid increase in PM2.5 emissions. With the surge in vehicle ownership, vehicle’s exhaust fumes have become an important source of PM2.5 pollution (Lu and Cao, 2015). Especially in some large cities (e.g., Beijing, Tianjin and Shanghai), automobile exhaust has become the largest contributor to PM2.5 pollution (Wang et al., 2015b). In order to reduce PM2.5 emissions, the government adopts a series of measures to reduce PM2.5 pollution. Some of these measures include encouragement of research and application of energy-saving technology, optimization of motor vehicles’ energy structure, and improvement in vehicle exhaust standards (Cao et al., 2014; Zhang et al., 2015). Hence, the impact of private cars on PM2.5 emissions gradually declines.

relation to PM2.5 emissions because of differences in the scale of the economy and the speed of technological progress at different times. The nonlinear impact of urbanization exhibits an inverted “U-shaped” pattern on account of the rapid development of urban real estate in the early stages and the strengthening of environmental protection measures in the later stage. Coal consumption follows an inverted “U-shaped” relationship with PM2.5 emissions owing to massive coal consumption in the early stages and energy structure optimization and technological progress in clean energy in the later stage. Moreover, the nonlinear effect of private vehicles also exhibits an inverted “U-shaped” pattern. The above results have important policy implications. Firstly, China should implement targeted measures to reduce PM2.5 emissions at the different stages of economic growth. On one hand, government should take measures to reduce building activities – and fixed asset investment-related dust in the early stage of economic development. This can be done by closing construction sites, encouraging car wash-in-site system and construction site watering. On the other hand, there is need to constantly optimize export product structure, and reduce the proportion of high energy-consuming export products. Some related policies such as subsidizing high-tech export products and imposing levy on high energy-consuming export products are crucial for mitigating PM2.5 emissions associated with exports. In the later part of economic development, China should reduce PM2.5 emissions from the tertiary industry. In the tertiary industry, the transport sector is the most important sources of PM2.5 emissions. Therefore, the government should take effective measures to reduce PM2.5 emissions from the transport sector by expanding new energy automobile consumption (electricity energy and bio-energy vehicles). Secondly, energy-saving technology research and development should be further strengthened. In order to improve energy efficiency and reduce PM2.5 emissions, the government must improve energy-saving technology using fiscal instruments such as the establishment of energy-saving technology research centers in local colleges and universities and supporting of enterprises to engage in R&D of energy-saving technology through preferential policies. Thirdly, different measures should be taken to reduce PM2.5 emissions intensity of coal consumption at different stages. In the early stages of coal consumption, the government should encourage research and application of clean coal combustion technology to reduce PM2.5 emissions intensity. However, in the later stages of coal consumption, the government should encourage the use of alternative energy sources, such as nuclear, bio-energy and gas energy. Finally, policies aimed at reducing PM2.5 emissions of private cars should not be the same at different stages. Owing to the fact that vehicle fuel technology is difficult to significantly improve in the short term, the government should increase the use of alternative energy, such as oil and gas hybrid energy. In the long run, given that the margin of improving automobile engine and refining technology is limited, research and application of electric vehicle technology should be strengthened while providing subsidies for the purchase and use of new energy vehicles.

Acknowledgements 6. Conclusions and policy implications Using panel data of 29 Chinese provinces during 2001–2012, this paper explores the driving forces and reduction potentials of China’s PM2.5 emissions using nonparametric additive regression models. The nonlinear effect of economic growth on PM2.5 emissions demonstrates an inverted “U-shaped” pattern due to the fact that economic growth rely mainly on investment and exports in the early stage and on domestic demand at the later stage. Energy efficiency improvement follows a positive “U-shaped” pattern in

The paper is supported by Xiamen University–Newcastle University Joint Strategic Partnership Fund, the Grant for Collaborative Innovation Center for Energy Economics and Energy Policy (No. 1260-Z0210011), Xiamen University Flourish Plan Special Funding (No. 1260-Y07200), and the China Sustainable Energy Program (No. G-1506-23315), National Social Science Foundation of China (No. 15BTJ022), the National Natural Science Foundation of China (No. 71563014), Jiangxi Science and Technology Fund in Jiangxi Province (No. GJJ14324), The Jiangxi Natural Science Foundation of

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

Jiangxi Province (No. 20142BAB201014; 20142BAB201010), Jiangxi Soft Science projects in Jiangxi Province (No. 20151BBA10037), and the monitoring, early warning and decision support of strategic emerging industries fund in Jiangxi Province (No. 2015–07). Appendix A. Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ecolind.2015. 11.012. References Beck, N., Jackman, S., 1998. Beyond linearity by default: generalized additive model. Am. J. Polit. Sci. 42 (2), 596–627. Bhattacharya, M., Rafiq, S., Bhattacharya, S., 2015. The role of technology on the dynamics of coal consumption-economic growth: new evidence from China. Appl. Energy 154, 686–695. Bloch, H., Rafiq, S., Salim, R., 2015. Economic growth with coal, oil and renewable energy consumption in China: prospects for fuel substitution. Econ. Model. 44, 104–115. Bozlaker, A., et al., 2014. Elemental characterization of PM2.5 and PM10 emitted from light duty vehicles in the washburn tunnel of Houston, Texas: release of rhodium, palladium, and platinum. Environ. Sci. Technol. 48 (1), 54–62. Brauning, F., Koopman, S.J., 2014. Forecasting macroeconomic variables using collapsed dynamic factor analysis. Int. J. Forecast. 30 (3), 572–584. Cai, Y.Y., Newth, D., Finnigan, J., Gunasekera, D., 2015. A hybrid energy-economy model for global integrated assessment of climate change, carbon mitigation and energy transformation. Appl. Energy 148, 381–395. Cao, J.J., et al., 2014. 1st UMN-CAS Bilateral Seminar on PM2.5 science, health effects and control technology Xi’an, China, May 27–28, 2014. Particuology 16, 227–229. Catalano, L., Figliola, A., 2015. Analysis of the nonlinear relationship between commodity prices in the last two decades. Qual. Quant. 49 (4), 1553–1558. Chan, C.K., Yao, X., 2008. Air pollution in mega cities in China. Atmos. Environ. 42, 142. Chang, M.C., 2015. Room for improvement in low carbon economies of G7 and BRICS countries based on the analysis of energy efficiency and environmental Kuznets curves. J. Clean. Prod. 99, 140–151. Chen, W., Hong, J.L., Xu, C.Q., 2015. Pollutants generated by cement production in China, their impacts, and the potential for environmental improvement. J. Clean. Prod. 103, 61–69. Chertow, M.R., 2001. The IPAT equation and its variants: changing views of technology and environmental impact. J. Ind. Ecol. 4, 13–29. Cheung, N.J., Xu, Z.K., Ding, X.M., Shen, H.B., 2015. Modeling nonlinear dynamic biological systems with human-readable fuzzy rules optimized by convergent heterogeneous particle swarm. Eur. J. Oper. Res. 247 (2), 349–358. Claro, S., 2006. Why does China protect its labour-intensive industries more? Econ. Transit. 14 (2), 289–319. Costigan, R.D., Brink, K.E., 2015. On the prevalence of linear versus nonlinear thinking in undergraduate business education: a lot of rhetoric, not enough evidence. J. Manage. Organ. 21 (4), 535–547. Curtis, S.M., Banerjee, S., Ghosal, S., 2014. Fast Bayesian model assessment for nonparametric additive regression. Comput. Stat. Data Anal. 71, 347–358. Dekoninck, L., Botteldooren, D., Panis, L.I., 2015. Using city-wide mobile noise assessments to estimate bicycle trip annual exposure to black carbon. Environ. Int. 83, 192–201. Dietz, T., Rosa, E.A., 1997. Effects of population and affluence on CO2 emissions. Proc. Natl. Acad. Sci. U. S. A. 94, 175–179. Djalalova, I., et al., 2010. Ensemble and bias-correction techniques for air quality model forecasts of surface O3 and PM2.5 during the TEXAQS-II experiment of 2006. Atmos. Environ. 44 (4), 455–467. Du, L.M., Wei, C., Cai, S.H., 2012. Economic development and carbon dioxide emissions in China: provincial panel data analysis. China Econ. Rev. 23, 371–384. Ercan, T., Tatari, O., 2015. A hybrid life cycle assessment of public transportation buses with alternative fuel options. Int. J. Life Cycle Assess. 20 (9), 1213–1231. Farias, V.F., Jagabathula, S., Shah, D., 2013. A nonparametric approach to modeling choice with limited data. Manage. Sci. 59 (2), 305–322. Farzan, F., Jafari, M.A., Gong, J., Farzan, F., Stryker, A., 2015. A multi-scale adaptive model of residential energy demand. Appl. Energy 150, 258–273. Giere, R., Blackford, M., Smith, K., 2006. TEM study of PM2.5 emitted from coal and tire combustion in a thermal power station. Environ. Sci. Technol. 40 (20), 6235–6240. Granger, C.W.J., 1988. Some recent developments in a concept of causality. J. Econom. 139 (1/2), 199–211. Guan, D., et al., 2009. Journey to world top emitter: an analysis of the driving forces of China’s recent CO2 emissions surge. Geophys. Res. Lett. 36, 15. Guan, D.B., et al., 2014. The socioeconomic drivers of China’s primary PM2.5 emissions. Environ. Res. Lett. 9 (2), 19. Gupta, E., 2012. Global warming and electricity demand in the rapidly growing city of Delhi: a semi-parametric variable coefficient approach. Energy Econ. 34 (5), 1407–1421.

357

Hua, L.T., Noland, R.B., Evans, A.W., 2010. The direct and indirect effects of corruption on motor vehicle crash deaths. Accid. Anal. Prev. 42 (6), 1934–1942. Huang, G.H., et al., 2014. Optical properties and chemical composition of PM2.5 in Shanghai in the spring of 2012. Particuology 13, 52–59. Huo, J.W., et al., 2015. Analysis of influencing factors of CO2 emissions in Xinjiang under the context of different policies. Environ. Sci. Policy 45, 20–29. Ide, K., Wiggins, S., 2015. The role of variability in transport for large-scale flow dynamics. Commun. Nonlinear Sci. Numer. Simul. 29 (1–3), 459–481. Ielpo, P., et al., 2013. Identification of pollution sources and classification of Apulia region ground waters by multivariate statistical methods and neural networks. Trans. ASABE 56 (6), 1377–1386. Islam, S., Liu, P.X., El Saddik, A., 2015. Nonlinear adaptive control for teleoperation systems with symmetrical and unsymmetrical time-varying delay. Int. J. Syst. Sci. 46 (16), 2928–2938. Jiang, Z.J., Lin, B.Q., 2012. China’s energy demand and its characteristics in the industrialization and urbanization process. Energy Policy 49, 608–615. Jiao, R.J., Zhang, P.D., Zhu, S., He, L., Niu, H.P., 2014. Identification and implications of relationships among pollutant emission, economic structure and economic growth in china through multivariate analysis. J. Environ. Sci. Manage. 17 (1), 1–11. Jin, Y.F., et al., 2015. Review and evaluation of China’s standards and regulations on the fuel consumption of motor vehicles. Mitig. Adapt. Strateg. Global Change 20 (5), 735–753. Kasman, A., Duman, Y.S., 2015. CO2 emissions, economic growth, energy consumption, trade and urbanization in new EU member and candidate countries: a panel data analysis. Econ. Model. 44, 97–103. Kao, C., 1999. Spurious regression and residual-based tests for cointegration in panel data. J. Econom. 90, 1–44. Khan, M.F., et al., 2015. Seasonal effect and source apportionment of polycyclic aromatic hydrocarbons in PM2.5 . Atmos. Environ. 106, 178–190. Kong, S.F., et al., 2015. Variation of polycyclic aromatic hydrocarbons in atmospheric PM2.5 during winter haze period around 2014 Chinese Spring Festival at Nanjing: insights of source changes, air mass direction and firework particle injection. Sci. Total Environ. 520, 59–72. Lee, D.H., Porta, M., Jacobs, D.R., Vandenberg, L.N., 2014. Chlorinated persistent organic pollutants, obesity, and type 2 diabetes. Endocr. Rev. 35 (4), 557–601. Lee, J., Robinson, P.M., 2015. Panel nonparametric regression with fixed effects. J. Econom. 188 (2), 346–362. Li, G.P., Yuan, Y., 2014. Impact of regional development on carbon emission: empirical evidence across countries. Chin. Geogr. Sci. 24 (5), 499–510. Li, K., Lin, B.Q., 2015. Metafroniter energy efficiency with CO2 emissions and its convergence analysis for China. Energy Econ. 48, 230–241. Li, R., Leung, G.C.K., 2012. Coal consumption and economic growth in China. Energy Policy 40, 438–443. Li, Y.M., Zhao, R., Liu, T.S., Zhao, J.F., 2015. Does urbanization lead to more direct and indirect household carbon dioxide emissions? Evidence from China during 1996–2012. J. Clean. Prod. 102, 103–114. Liddle, B., 2013. Urban density and climate change: a STIRPAT analysis using citylevel data. J. Transp. Geogr. 28, 22–29. Lin, B.Q., Du, K.R., 2014. Decomposing energy intensity change: a combination of index decomposition analysis and production-theoretical decomposition analysis. Appl. Energy 129, 158–165. Lin, B.Q., Omoju, O.E., Okonkwo, U.J., 2015. Impact of industrialisation on CO2 emissions in Nigeria. Renew. Sustain. Energy Rev. 52, 1228–1239. Lin, G., et al., 2014a. Spatio-temporal variation of PM2.5 concentrations and their relationship with geographic and socioeconomic factors in China. Int. J. Environ. Res. Public Health 11 (1), 173–186. Lin, W.B., Chen, B., Luo, S.C., Liang, L., 2014b. Factor analysis of residential energy consumption at the provincial level in China. Sustainability 6 (11), 7710–7724. Linton, O.B., Hardle, W., 1996. Estimation of additive regression models with known links. Biometrika 83, 529–540. Liu, Z.Y., Mao, X.Q., Tu, J.J., Jaccard, M., 2014. A comparative assessment of economic-incentive and command-and-control instruments for air pollution and CO2 control in China’s iron and steel sector. J. Environ. Manage. 144, 135–142. Loftus, C., et al., 2015. Regional PM2.5 and asthma morbidity in an agricultural community: a panel study. Environ. Res. 136, 505–512. Lu, J.Y., Cao, X., 2015. PM2.5 Pollution in major cities in china: pollution status, emission sources and control measures. Fresenius Environ. Bull. 24 (4A), 1338–1349. Mallia, D.V., Lin, J.C., Urbanski, S., Ehleringer, J., Nehrkorn, T., 2015. Impacts of upwind wildfire emissions on CO, CO2 , and PM2.5 concentrations in Salt Lake City, Utah. J. Geophys. Res.: Atmos. 120 (1), 147–166. Mardones, C., Sanhueza, L., 2015. Tradable permit system for PM2.5 emissions from residential and industrial sources. J. Environ. Manage. 157, 326–331. Menegaki, A.N., Tsagarakis, K.P., 2015. Rich enough to go renewable, but too early to leave fossil energy? Renew. Sustain. Energy Rev. 41, 1465–1477. Meng, J., Liu, J.F., Xu, Y., Tao, S., 2015. Tracing primary PM2.5 emissions via Chinese supply chains. Environ. Res. Lett. 10 (5), http://dx.doi.org/10.1088/1748-9326/ 10/5/054005. Michieka, N.M., Fletcher, J.J., 2012. An investigation of the role of China’s urban population on coal consumption. Energy Policy 48, 668–676. Ning, D.Z., Shi, J., Zou, Q.P., Teng, B., 2015. Investigation of hydrodynamic performance of an OWC (oscillating water column) wave energy device using a fully nonlinear HOBEM (higher-order boundary element method). Energy 83, 177–188.

358

B. Xu et al. / Ecological Indicators 63 (2016) 346–358

Niu, Z.C., et al., 2013. Source contributions to carbonaceous species in PM2.5 and their uncertainty analysis at typical urban, peri-urban and background sites in southeast China. Environ. Pollut. 181, 107–114. Normile, D., 2008. China’s living laboratory in urbanization. Science 319, 740–743. Olson, D.A., Burke, J.M., 2006. Distributions of PM2.5 source strengths for cooking from the research triangle park particulate matter panel study. Environ. Sci. Technol. 40 (1), 163–169. Palardy, J., Ovaska, T., 2015. Decomposing household, professional and market forecasts on inflation: a dynamic factor model analysis. Appl. Econ. 47 (20), 2092–2101. Panepinto, D., Brizio, E., Genon, G., 2014. Atmospheric pollutants and air quality effects: limitation costs and environmental advantages (a cost–benefit approach). Clean Technol. Environ. Policy 16 (8), 1805–1813. Pao, H.T., Tsai, C.M., 2010. CO2 , emissions, energy consumption and economic growth in BRIC countries. Energy Policy 38, 7850–7860. Patel, M.M., et al., 2013. Traffic-related air pollutants and exhaled markers of airway inflammation and oxidative stress in New York City adolescents. Environ. Res. 121, 71–78. Piegorsch, W.W., Xiong, H., Bhattacharya, R.N., Lin, L.Z., 2014. Benchmark dose analysis via nonparametric regression modeling. Risk Anal. 34 (1), 135–151. Recalde, M., Ramos-Martin, J., 2012. Going beyond energy intensity to understand the energy metabolism of nations: the case of Argentina. Energy 37 (1), 122–132. Saboori, B., Sulaiman, J., 2013. Environmental degradation, economic growth and energy consumption: evidence of the environmental Kuznets curve in Malaysia. Energy Policy 60, 892–905. Salazar, J.M., Zitney, S.E., Diwekar, U.M., 2011. Minimization of water consumption under uncertainty for a pulverized coal power plant. Environ. Sci. Technol. 45 (10), 4645–4651. Schifano, P., et al., 2013. Effect of ambient temperature and air pollutants on the risk of preterm birth, Rome 2001–2010. Environ. Int. 61, 77–87. Schwartz, J., 1994. Generalized additive models in epidemiology. In: 17th International Biometric Conference, pp. 55–80. Seidel, D.J., Birnbaum, A.N., 2015. Effects of Independence Day fireworks on atmospheric concentrations of fine particulate matter in the United States. Atmos. Environ. 115, 192–198. Shahbaz, M., Solarin, S.A., Sbia, R., Bibi, S., 2015. Does energy intensity contribute to CO2 emissions? A trivariate analysis in selected African countries. Ecol. Indic. 50, 215–224. Shen, X.B., et al., 2014. PM2.5 emissions from light-duty gasoline vehicles in Beijing, China. Sci. Total Environ. 487, 521–527. Shi, K., Liu, C.Q., Ai, N.S., Zhang, X.H., 2008. Using three methods to investigate timescaling properties in air pollution indexes time series. Nonlinear Anal.: Real World Appl. 9 (2), 693–707. Sica, E., Susnik, S., 2014. Geographical dimension and environmental Kuznets curve: the case of some less investigated air pollutants. Appl. Econ. Lett. 21 (14), 1010–1016. Stauch, V.J., Jarvis, A.J., 2006. A semi-parametric gap-filling model for eddy covariance CO2 flux time series data. Global Change Biol. 12 (9), 1707–1716. Stone, C.J., 1985. Additive regression and other nonparametric models. Ann. Stat. 113, 689–705. Sueyoshi, T., Yuan, Y., 2015. China’s regional sustainability and diversified resource allocation: DEA environmental assessment on economic development and air pollution. Energy Econ. 49, 239–256. Sun, C.Z., Yang, Y.D., Zhao, L.S., 2015. Economic spillover effects in the Bohai Rim Region of China: is the economic growth of coastal counties beneficial for the whole area? China Econ. Rev. 33, 123–136. Tang, L., Wu, J.Q., Yu, L., Bao, Q., 2015. Carbon emissions trading scheme exploration in China: a multi-agent-based model. Energy Policy 81, 152–169. Tursun, H., et al., 2015. Contribution weight of engineering technology on pollutant emission reduction based on IPAT and LMDI methods. Clean Technol. Environ. Policy 17 (1), 225–235. Walsh, M.P., 2014. PM2.5 global progress in controlling the motor vehicle contribution. Front. Environ. Sci. Eng. 8 (1), 1–17. Wang, H., et al., 2012. Age-specific and sex-specific mortality in 187 countries, 1970–2010: a systematic analysis for the global burden of disease study 2010. Lancet 380, 2071–2094. Wang, L.L., Liu, Z.R., Sun, Y., Ji, D.S., Wang, Y.S., 2015a. Long-range transport and regional sources of PM2.5 in Beijing based on long-term observations from 2005 to 2010. Atmos. Res. 157, 37–48.

Wang, M.W., Che, Y., Yang, K., Wang, M., Xiong, L.J., Huang, Y.C., 2011. A local-scale low-carbon plan based on the STIRPAT model and the scenario method: the case of Minhang District, Shanghai, China. Energy Policy 39 (11), 6981–6990. Wang, S.J., Ma, H.T., Zhao, Y.B., 2014. Exploring the relationship between urbanization and the eco-environment – a case study of Beijing-Tianjin-Hebei region. Ecol. Indic. 45, 171–183. Wang, Y., Zhang, X., Kubota, J.P., Zhu, X.D., Lu, G.F., 2015b. A semi-parametric panel data analysis on the urbanization-carbon emissions nexus for OECD countries. Renew. Sustain. Energy Rev. 48, 704–709. Wang, Y.N., Zhao, T., 2015. Impacts of energy-related CO2 emissions: evidence from under developed, developing and highly developed regions in China. Ecol. Indic. 50, 186–195. Wu, T., Zhang, M.B., Ou, X.M., 2014. Analysis of future vehicle energy demand in China based on a Gompertz function method and computable general equilibrium model. Energies 7 (11), 7454–7482. Xu, B., Lin, B.Q., 2015a. How industrialization and urbanization process impacts on CO2 emissions in China: evidence from nonparametric additive regression models. Energy Econ. 48, 188–202. Xu, B., Lin, B.Q., 2015b. Factors affecting carbon dioxide (CO2 ) emissions in China’s transport sector: a dynamic nonparametric additive regression model. J. Clean. Prod. 101, 311–322. Xu, F., et al., 2009. Experimental investigation on charging characteristics and penetration efficiency of PM2.5 emitted from coal combustion enhanced by positive corona pulsed ESP. J. Electrostat. 67 (5), 799–806. Xue, C.L., Zheng, X.Q., Zhang, B., Yuan, Z.Y., 2015. Evolution of a multidimensional architectural landscape under urban regeneration: a case study of Jinan, China. Ecol. Indic. 55, 12–22. Yu, H.L., Lin, Y.C., Kuo, Y.M., 2015. A time series analysis of multiple ambient pollutants to investigate the underlying air pollution dynamics and interactions. Chemosphere 134, 571–580. Yang, L.X., et al., 2013. Source identification and health impact of PM2.5 in a heavily polluted urban atmosphere in China. Atmos. Environ. 75, 265–269. Yan, X., Crookes, R.J., 2009. Reduction potentials of energy demand and GHG emissions in China’s road transport sector. Energy Policy 37 (2), 658–668. Zhan, S.H., 2015. From local state corporatism to land revenue regime: urbanization and the recent transition of rural industry in China. J. Agrar. Change 15 (3), 413–432. Zhang, B., Chen, G.Q., 2014. China’s CH4 and CO2 emissions: bottom-up estimation and comparative analysis. Ecol. Indic. 47, 112–122. Zhang, C.G., Liu, C., 2015. The impact of ICT industry on CO2 emissions: a regional analysis in China. Renew. Sustain. Energy Rev. 44, 12–19. Zhang, F., et al., 2015. Seasonal variations and chemical characteristics of PM2.5 in Wuhan, central China. Sci. Total Environ. 518, 97–105. Zhang, M., Wang, W.W., 2013. Decouple indicators on the CO2 emission-economic growth linkage: the Jiangsu Province case. Ecol. Indic. 32, 239–244. Zhang, Y.J., et al., 2014. The impact of economic growth, industrial structure and urbanization on carbon emission intensity in China. Nat. Hazards 73 (2), 579–595. Zhang, Y.L., Cao, F., 2015. Is it time to tackle PM2.5 air pollutions in China from biomass-burning emissions? Environ. Pollut. 202, 217–219. Zhao, J.F., Tang, J.M., 2015. Industrial structural change and economic growth in China, 1987–2008. China World Econ. 23 (2), 1–21. Zhao, X.Y., et al., 2014. Compositions and sources of organic acids in fine particles (PM2.5 ) over the Pearl River Delta region, south China. J. Environ. Sci. – China 26 (1), 110–121. Zhao, Z.Y., Ling, W.J., Zillante, G., Zuo, J., 2012. Comparative assessment of performance of foreign and local wind turbine manufacturers in China. Renew. Energy 39 (1), 424–432. Zheng, X.Y., et al., 2014. Characteristics of residential energy consumption in China: findings from a household survey. Energy Policy 75, 126–135. Zhong, H.W., Xia, Q., Chen, Y.G., Kang, C.Q., 2015. Energy-saving generation dispatch toward a sustainable electric power industry in China. Energy Policy 83, 14–25. Zhou, B., Xu, Q.F., You, J.H., 2011. Efficient estimation for error component seemingly unrelated nonparametric regression models. Metrika 73 (1), 121–138. Zhou, Y., Hammitt, J., Fu, J.S., Gao, Y., Liu, Y., Levy, J.I., 2014. Major factors influencing the health impacts from controlling air pollutants with nonlinear chemistry: an application to China. Risk Anal. 34 (4), 683–697. Zhu, C.S., et al., 2012. Indoor and outdoor chemical components of PM2.5 in the rural areas of Northwestern China. Aerosol Air Qual. Res. 12 (6), 1157–1165.