Stepwise Improvement for Environmental Performance of Transportation Industry in China: A DEA Approach Based on Closest Targets

Transportation is regarded as an industry with high energy consumption and high CO 2 emissions. Little attention has been paid to the environmental performance improvement of China’s transportation industry, especially in a stepwise improvement way. In this study, we ﬁrst apply the closest targets DEA method to evaluate the environmental performance in the transportation industry of 30 provincial-level regions in China’s mainland from 2010 to 2017. Then, we incorporate the closest targets and context-dependent DEA model and thus conform a stepwise projection path for each ineﬃcient province to improve environmental performance with less eﬀort by the way of identifying a sequence of intermediate closest targets. The empirical study shows that the environmental performance of the transportation industry obtained from the closest targets model is greater than that obtained from the SBM model for each province. Among the three areas, the eastern area performs the best in environmental performance followed by the central region and western region. Shanghai has the best environmental performance. Additionally, compared with conventional DEA models, the proposed stepwise improvement method can generate easier and closer achieved targets for the ineﬃcient provinces. Hainan, Yunnan, and Xinjiang provinces have the lowest environmental performance, which need four steps to achieve eﬃciency.


Introduction
With the benefit from the "reform and opening up" launched in 1978, China's economy has rapidly improved. China's GDP (gross domestic product) increased from 0.37 trillion yuan in 1978 to 90.03 trillion yuan in 2018, with an average growth rate of 9.4% [1]. However, the rapid economic growth has brought about problems of huge energy consumption and pollution emissions (e.g., CO 2 emissions) [2,3]. In 2018, the total energy consumption of China reached 4.64 billion tons of standard coal equivalent (tce), 7.7 times that of 1978, ranking first in the world [4]. e transportation industry has been the third-largest energy consumer industry in China [5], and it has become one of China's main sources of CO 2 emissions [6,7]. With the rapid development of the transportation industry, the energy consumption and CO 2 emissions of the transportation industry will continue to grow [8]. In 2050, the energy consumption of China's transportation industry would increase to 636 million tons of oil equivalent and produce 16.02 billion tons of CO 2 [9]. Hence, it is of great importance to effectively assess the environmental performance of the transportation industry in China. By doing this, it may provide helpful information to decision-makers to achieve a balance between economic growth and sustainable development and finally to improve environmental performance.
To properly evaluate the environmental performance of the transportation industry, many scholars have proposed approaches based on data envelopment analysis (DEA) [10,11]. DEA, which was first proposed by Charnes et al. [12], has been extensively used to evaluate the relative performance of a set of homogeneous decisionmaking units (DMUs) which consume multiple inputs to produce multiple outputs [13,14]. In addition to evaluating performance, DEA can also provide benchmarking information or targets to guide inefficient DMUs to improve their performance [15,16]. Given its advantages, DEA has been applied as an analytical technique in the fields of agriculture, banking, transportation, supply chain, and others [17].
us, we suggest the use of the DEA methodology as the main tool to measure the environmental performance of the transportation industry in China's mainland.
Various DEA models have been applied to the environmental performance evaluation of China's transportation industry (see [18][19][20][21][22]). However, the studies on the environmental performance improvement of the transportation industry are largely lacking. On the other hand, prior research applying DEA approaches usually yields a "furthest" target or benchmark for any inefficient DMU. Under such circumstances, it is difficult for the inefficient DMU to achieve efficiency along the direction determined by its "furthest" target or benchmark in a single step because of the large difference in the inputs and/or outputs between it and the targets. To avoid this problem, one effective way is to find the closest targets for the inefficient DMU. e closest targets have values for inputs and/or outputs similar to the current values of the inefficient DMU; thus, it can achieve such targets with less effort [15,23]. However, when there is a large performance gap between the inefficient DMU and its corresponding closest target, it is still hard for such inefficient DMU to achieve the closest targets in a single step or in a short time [24].
To fill the gaps in the prior literature, this paper proposes a new stepwise improvement method that incorporates closest targets and context-dependent DEA model. In contrast to the traditional DEA models (e.g., SBM) that yield the "furthest" targets for the inefficient DMUs, the proposed approach in this study generates the closest targets that have the inputs and outputs similar to the assessed inefficient DMUs, which means that the inefficient DMUs can improve to the efficient frontier with less effort along the direction to the corresponding closest targets. In particular, to help an inefficient DMU that is far away from its closest targets achieve efficiency, our approach provides a stepwise improvement path that consists of several intermediate closest targets on different levels of efficient frontier identified by the context-dependent DEA approach, thus ensuring the inefficient DMU improve to the efficient frontier by following this path. e rest of the paper is organized as follows. e following section reviews the literature on environmental performance evaluation of the transportation industry based on DEA methods and the closest targets approaches in DEA. In Section 3, we provide the preliminaries of relevant DEA models. e stepwise improvement approach that incorporates the closest targets and contextdependent DEA model is proposed in Section 4. In Section 5, we apply our approach to the transportation industry at provincial administrative regions in mainland China. e last section gives the conclusion and several possible research directions.

Environmental Performance Evaluation of Transportation
Industry. Considering the importance of reducing CO 2 emissions and energy consumption, a large body of the literature has used DEA methods to evaluate the environmental performance of transport sectors. Egilmez and Park [25] integrated EIO-LCA and DEA to access the environmental performance of the U.S. transportation industry. Beltrán-Esteve and Picazo-Tadeo [26] used a directional distance function approach to measure the environmental performance changes in the transportation industry of 38 countries/regions from 1995 to 2009. ey found that the improvement of environmental performance is mainly driven by eco-innovation. Park et al. [11] applied a nonradial SBM-DEA model to evaluate the environmental efficiency and potential CO 2 reduction of the transportation sectors in the U.S. from years 2004 to 2012. eir findings revealed that the transportation sectors in the U.S. were environmentally inefficient with an average environmental efficiency score below 0.64. Mavi et al. [27] applied a common set of weights double frontier DEA-based Malmquist productivity index method to track the changes of the environmental performance of the transportation industry in Iran. e results indicated that the environmental performance of the transportation industry in Iran had a constant or declining trend from 2014 to 2017. Omrani et al. [28] used a DEAcooperative game approach to evaluate the energy efficiency in transportation sector at the provincial level in Iran. ey found that smaller provinces have higher energy efficiency.
In recent years, transportation industry has become one of the industries with high energy consumption and high CO 2 emissions in China, and the environmental performance evaluation of China's transportation industry has received widespread attention. Chang et al. [10] used a nonradial SBM model to analyze the environmental efficiency of China's transportation sectors at the provincial regional level. eir results showed that the environmental efficiency of China's transportation industry is very low, and the environmental efficiencies of most provinces are below 50% of the target level. Cui and Li [29] employed a threestage virtual frontier DEA model to evaluate energy efficiencies in the transportation industry of 30 Chinese provincial administrative regions. ey found that structure and management measures have impacts on transportation energy performance. Wu et al. [18] measured the energy and environmental performance of transportation systems at the provincial level in China based on a parallel DEA approach. e result showed that there are large efficiency differences between the passenger and freight transportation subsystems. Stefaniec et al. [30] proposed a triple bottom linebased network DEA approach to evaluate the environmental performance of inland transportation in China. e results indicated that the overall efficiency of the transportation industry shows an upward trend. Zhu et al. [31] developed a new equilibrium efficient frontier DEA approach to assess the environmental performance of transportation sectors in China under the constraints of energy consumption and environmental pollutions. e findings revealed that there exist large disparities in environmental performance among regions. Also, some scholars have paid attention to the environmental performance of the transportation industry focusing one region; for example, Tian et al. [32] utilized an improved super-efficiency SBM-DEA model to measure the sustainable development of the transportation industry in Shaanxi province.

Closest Targets Approach in DEA.
Finding the closest targets for inefficient DMUs to help them achieve efficiency with less effort has been a hot issue in the DEA area. In DEA research on finding closest targets, two primary ways have attracted attention. One way is to identify all efficient facets, calculate the least distance from the inefficient DMU to each efficient facet, and finally choose the minimum distance from these least distances. is type of method originated from Briec [41], who used the Hölder distance function to determine the least distance, and Frei and Harker [42], who used the Euclidean distance to the Pareto-efficient frontier to obtain the closest targets or benchmarks. Later, weighted Euclidean distance approaches were proposed by Amirteimoori and Kordrostami [43] and Aparicio and Pastor [44], among others. e other way is to find the closest targets for a certain inefficient DMU based on similarity criteria is an approach based originally on the mixed-integer linear program proposed by Aparicio et al. [15], which can obtain the closest target on the Pareto-efficient frontier for a given inefficient DMU. In line with Aparicio et al. [15], Pastor and Aparicio [45], Ando et al. [46], Aparicio and Pastor [47], and Fukuyama et al. [48], the properties of such methodologies were further improved (see [49] for a detailed discussion).
In addition to the research on property improvement, the closest targets approach has also been applied to various areas. An et al. [50] used the closest targets model based on the enhanced Russell measure to evaluate the environmental performance of 20 thermal power enterprises in Anhui province of China. By using the closest targets method, Li et al. [20] provided benchmarking information for primary freight transportation seaports in China to improve their environmental performance. Wu et al. [19] incorporated the closest targets technique into carbon emissions abatement allocation and applied it for carbon emissions abatement target setting and allocation for 20 APEC economies. Razipour-GhalehJough et al. [51] proposed a closest targets model in the presence of weight restrictions to evaluate and improve the efficiency of Iranian banks.
In reviewing the above discussed literature, we find that although various DEA models have been used in the environmental performance evaluation of the transportation industry, the studies on the environmental performance improvement of the transportation industry are largely requiring. Additionally, prior research applying DEA approaches usually yields a "furthest" target or benchmark for the inefficient DMU. As a result, it is difficult for the inefficient DMU to achieve efficiency along the direction determined by its "furthest" target or benchmark in a single step because of the large difference in the inputs and outputs between it and the targets. erefore, in this study, we incorporate closest targets and context-dependent DEA model and thus conform a stepwise projection path for each inefficient province to improve environmental performance with less effort by the way of identifying a sequence of intermediate closest targets.

Slacks-Based Measure (SBM) considering Undesirable
Outputs. Suppose that there are n DMUs, and each DM U j (j � 1, 2, · · · , n) consumes m inputs to produce s desirable outputs accompanied by h undesirable outputs. Variables x ij (i � 1, 2, · · · , m), y rj (r � 1, 2, · · · , s), and z qj (q � 1, 2, · · · , h) represent the i-th input, r-th desirable output, and q-th undesirable output of DM U j (j � 1, 2, · · · , n), respectively. e production possibility set which is constructed by these n DMUs is defined as follows [52,53]: λ j y rj ≥ y r , r � 1, 2, · · · , s, n j�1 λ j z qj � z q , q � 1, 2, · · · , h, λ j ≥ 0, j � 1, 2, · · · , n (1) Given an evaluated DM U k , the following linear program, namely, nonradial and nonoriented slacksbased measure (SBM) based on production possibility set (1) can be used to measure its relative environmental efficiency [54][55][56][57]. Because the SBM model encompasses the excesses of inputs and undesirable outputs and the Mathematical Problems in Engineering shortfalls of desirable outputs simultaneously, this technique has been widely used in environmental performance evaluation: (2) In model (2), θ measures the relative environmental efficiency of DM U k ; it ranges from 0 to 1, i.e., the higher value of θ is, the better environmental efficiency DM U k achieves. s − i and s − q , respectively, represent the potential reductions of i-th input and q-th undesirable output, while s + r indicates the potential expansion of r-th desirable output [54]. Additionally, λ j (j � 1, 2, · · · , n) are intensity variables which connect inputs and outputs. Denoting the optimal solution of model (2) , we have the following two remarks.

Remark 1. DM U k is strongly efficient if and only if
DM U k is weakly efficient if and only if θ * � 1 and s − * i ≠ 0, s + * r ≠ 0, and s − * q ≠ 0 for some inputs and outputs.
Remark 2. θ * < 1 means DM U k is inefficient, and the target on the efficient frontier can be calculated by

Closest Targets Model considering Undesirable Outputs.
To help the inefficient DMUs become efficient with the least effort (minimizing the contraction of inputs and/or augmentation of outputs), the closest target and minimum distance to the Pareto-efficient frontier approach have been proposed and investigated by many scholars (see Aparicio et al. [49] for details). Denoting E as the set of strongly efficient units, Aparicio et al. [15] constructed the following strongly efficient frontier without considering undesirable outputs under the assumption of constant returns to scale (CRS): where M is a large non-negative constant. e first two constraints ensure that each DMU in T is a linear combination of strongly efficient DMUs. Constraints s r�1 u r y rj − m i�1 w i x ij + φ j � 0, ∀j ∈ E, and w i , u r ≥ 1, ∀i, r construct the hyperplanes that the units belonging to the production possibility set either lie on or away from. If λ j > 0, that is, b j � 0 and φ j � 0, ∀j ∈ E, then DM U j is a peer for other DMUs.
Considering the undesirable outputs, the following model is proposed to evaluate the environmental efficiencies for the inefficient DMUs on the basis of (3). e evaluated inefficient DMU is denoted as DM U p .
In model (4), the maximization of the objective function helps the decision-makers discover the closest targets for inefficient DM U p .
e optimal solution of model (4) is , and DM U p can achieve the closest target on the efficient frontier by Compared with the target obtained from a conventional DEA model, such as the SBM mentioned above, model (4) generates a closest Pareto-efficient target on the efficient frontier for any inefficient DMU. Such closest targets are as similar as possible to the evaluated DMUs' observed inputs and outputs. erefore, each inefficient DMU can improve its performance by moving toward efficiency along the direction to its closest target with less effort than along the projection direction used in the SBM model. Figure 1 clearly illustrates the SBM projection and closest target projection.

Stepwise Improvement Based on Closest
Targets in DEA

4.2.
Stepwise-Closest Targets Model. Model (4) can help the inefficient DMUs to improve to the Pareto-efficient frontier with less effort than conventional DEA models. When an inefficient DMU is far away from the efficient frontier, it may be hard to improve to be efficient in one step because of its current limited technology. In this section, we propose the stepwise-closest targets model which incorporates the closest targets and the context-dependent DEA model to seek a stepwise improvement path for the inefficient DMUs, as illustrated in Figure 3.
e distinct advantage of the stepwise-closest targets model is that it generates several intermediate closest targets thus helping the inefficient DMUs, especially the inefficient DMUs far away from the Pareto-efficient frontier improve to the ultimate Pareto-efficient frontier step by step. Taking the DMU C in Figure 3 as an example, for example, C 1 , C 2 , and C 3 , are the three closest targets located on the first-level,   Mathematical Problems in Engineering second-level, and third-level frontiers, respectively. We should note that C 3 is the easiest target for DMU C to achieve, followed by C 2 and C 1 .

Background and Dataset.
With the rapid economic growth and urbanization, China's transportation industry is developing rapidly. e passenger and freight turnover volumes in China have also grown with high-speed since 2000, i.e., the passenger turnover volumes increased from 1.23 trillion passenger-kilometers in 2000 to 3.42 trillion passenger-kilometers in 2018, and the freight turnover increased from 4.43 trillion ton-kilometers to 20.47 trillion ton-kilometers [1]. e transportation industry is in a period of rapid growth, the massive use of energy, and the generation of large amounts of CO 2 emissions which result in huge pressure on the environment. erefore, it is necessary to evaluate and improve the environmental performance of China's transportation industry. Specifically, in this section, we mainly focus on the evaluation and improvement of environmental performance in transportation industry of 30 provincial-level regions (this study refers provincial-level regions to provinces for convenience) in China's mainland from 2010 to 2017 (Tibet is excluded due to the missing data).
Referring to Cui and Li et al. [29] and Wu et al. [18], we select the number of staff working in the transportation industry (Labor), transportation fixed assets investment (Capital), and energy consumption of the transportation industry (Energy consumption) as three inputs. Freight turnover volume (FTV) and passenger turnover volume (PTV) are two desirable outputs, and CO 2 emissions are the undesirable output. e data related to inputs (Labor, Capital, and Energy consumption) and desirable outputs ( Because the official data of provincial CO 2 emissions in China are not directly provided, following Chang et al. [10] and Wu et al. [18], we use a fuel-based carbon footprint model to measure the CO 2 emissions in the regional transportation industry. According to the Intergovernmental Panel on Climate Change guidelines [59], we calculate the CO 2 emissions by the following equation: where CO 2 denotes the CO 2 emissions (unit: ten thousand tons); E represents the carbonaceous fuel; CCF i denotes the carbon content factor of fuel i; HE i is the heat equivalent of fuel i; COF i represents the carbon oxidation factor of fuel i; and 44/12 represents the ratio of the molecular weight of CO 2 to the molecular weight of carbon. For the standard of carbon dioxide emission factor, we applied National Development and Reform Commission (NDRC) (2007) in China which has been successfully used by Chang et al. [10] and Wu et al. [18]. Also, the amount of consumption of each fuel by each province in the transportation industry is from China Energy Statistical Yearbook 2011-2018. A statistical description of the inputs and outputs is reported in Table 1.  Table 2. To be specific, columns 1 and 2 present the three regions and the 30 provinces, respectively. Columns 3-10 provide the environmental performance from 2010 to 2017, and the last column presents the mean value of environmental performance across the whole observation period. We can draw the following conclusions from Table 2  from the closest targets model (4) is listed in Table 3. First, the average annual environmental performance of China's transportation industry was 0.6733 over the eight years. Two-third of provinces' environmental performance exceeds the average annual environmental performance. Second, the eastern area performed the best (0.7444), followed by the central area (0.7268) and the western area (0.5633). In the eastern area, the environmental performance of Shanghai's transportation industry was 1 while Beijing had the minimum (0.3723). In the central area, Anhui performed the best which was 0.9858 and the minimum was 0.4671 for Heilongjiang. In the western area, the maximum environmental performance was 0.7018 for Gansu and the minimum was 0.3718 for Yunnan.

Environmental Performance
Combining Tables 2 and 3 and Figure 4, we compare the environmental performance of the transportation industry of 30 provinces that, respectively, are obtained from the SBM model (2) and the closest targets model (4) and draw the following findings. First, the environmental performance calculated by the closest targets approach (4) was higher than that calculated by the SBM method (2) for each province. Second, there are 14 provinces that have environmental performance which exceeds the average in terms of the closest targets model (4), while 10 provinces in terms of the SBM model (2). ird, the central area performed the best in environmental performance under the SBM model (2) while the eastern area performed the best under the closest targets model (4), and the western area had the worst environmental  performance over the eight years in terms of both methods. Fourth, only Shanghai province performed the best in terms of the SBM model and closest targets approach while Yunnan province performed the worst.

Environmental Performance Improvement Projection
Based on the SBM and Closest Targets Models. In this section, we chose the year 2017 as an example to demonstrate the environmental performance improvement projection based       (2) and closest targets approach (4). From Tables 2 and 3, we find that 7 provinces are efficient in 2017 in terms of both methods. Here, we mainly pay attention to the improvement of inefficient provinces. Table 4 provides the environmental performance improvement projections of the inefficient provinces in terms of the SBM model (2) and closest targets model (4). To be specific, the first row shows the original values of the inputs and outputs of each province. e improvement targets obtained from the SBM model (2) and closest targets method (4) are presented in the second row and third row, respectively. To be more intuitive, the proportions of the original value relative to the projected targets that need to be increased/decreased (improvement percentages) are presented in parentheses. Taking (4). ese figures show that Jiangsu province can become efficient by reducing two inputs (labor by 32% and energy by 48%) and the undesirable output (CO 2 by 48%), while increasing one desirable output (FTV by 22%) based on the SBM model (2). However, in terms of the closest targets model (4), Jiangsu province achieves efficiency by reducing energy by 8% and CO 2 by 7%, while increasing PTV by 6%. It indicates that the closest targets may be more easily achieved with less improvement than a conventional radial model.

Stepwise Environmental Performance Improvement Projection Based on the SBM and Closest Targets Models.
Utilizing the calculation steps proposed by Seiford and Zhu [58], 30 provinces are divided into five different levels of efficient frontiers, and columns 1 and 2 in Table 5 report these levels of the efficient frontier and the provinces they contain. In addition, Table 5 gives the stepwise improvement targets for the inefficient provinces based on the SBM model and closest targets model. To be specific, for each province that is inefficient at a particular level, its original values of inputs and outputs are presented in the first row, and the stepwise improvement targets calculated by the SBM and closest targets model are listed in the second row and third row, i.e., row "E2" lists the targets located on the level 1 efficient frontier, namely, the ultimate frontier, row "E3" gives the targets located on the level 2 efficient frontier, and so on.
We still take Jiangsu province as an example. It is identified as efficient in the level 2 frontier, which means it can improve to the level 1 frontier in one step. e targets yielded by the stepwise-closest targets model prove that this province would achieve efficiency by reducing one input (energy by 8%) and one undesirable output (CO 2 by 7%) while increasing one desirable output (PTV by 7%). However, it becomes efficient by adjusting two inputs (reduce labor by 32% and energy by 48%), one desirable output (increase FTV by 40%), and one undesirable output (reduce CO 2 by 48%) with the SBM model. at is, Jiangsu province can achieve the closest targets more easily. Note that we assume that reducing the number of input/output variables to change makes a change easier without considering the improvement costs of inputs and/or outputs.
is is a reasonable simplifying assumption although in real-life situations it may be harder to change one variable by 1% than another variable by 10%.
Moreover, we choose Hainan province which is at the lowest efficiency level (E5) as another example. Table 6 reports the stepwise targets of Hainan based on the SBM model and closest targets model, and Figure 5 clearly demonstrates the corresponding improvement percentages  for different levels in terms of the stepwise-closest targets model. Hainan is located on the level 5 frontier, which means that it needs four steps to achieve efficiency. For example, Hainan can improve to the level 4 efficient frontier by reducing energy by 8% and CO 2 by 8% when using the closest targets model and achieve the environmental efficiency of 0.9463.
In summary, compared with the SBM model, the closest targets model can generate easier and closer achieved targets for the inefficient provinces. An inefficient province would become efficient with the minimization of reduction of inputs and undesirable outputs and/or augmentation of desirable outputs.

Conclusion
e transportation industry has greatly promoted China's economic development but also is a major source of CO 2 emissions, which hurt the environment. erefore, it is necessary to measure and improve its environmental performance. In the current study, using the data of the transportation industry of 30 provincial-level regions in China during the period of 2010-2017, we incorporate the closest targets and context-dependent DEA model to evaluate the environmental performance. Moreover, our proposed stepwise-closest targets method can identify a sequence of intermediate closest targets and form a stepwise projection path for each inefficient province so as to achieve the goal of improving environmental performance with less effort. We draw the following findings from the empirical study.
First, the environmental performance of the transportation industry obtained from the closest targets model is greater than that obtained from the SBM model for each province. Among the three areas, the eastern area performed best in environmental performance, followed by the central area, and the western area performed the worst. Only Shanghai province performed the best in terms of the SBM model and closest targets approach while Yunnan province performed the worst.
Second, compared with the conventional SBM model, the closest targets model can generate easier and closer achieved targets for the inefficient provinces. An inefficient province may not achieve efficiency in a short time when a large efficiency gap exists between it and efficient frontier, and the stepwise-closest targets model can help the inefficient province to improve to efficiency using several intermediate closest targets, each of which can encourage the province to continue its improvement efforts.
ere are three provinces (Hainan, Yunnan, and Xinjiang) with the lowest environmental performance, which need four steps to achieve efficiency.
is study is not free of limitations, and several future research directions should be considered. First, our study's method directs the inefficient DMUs to the Pareto-efficient frontier via a path consisting of several intermediate closest targets, and these targets are hypothetical DMUs, so we suggest limiting the intermediate targets to the existing DMUs in the future study. Second, future methods could also take varying improvement costs of inputs and outputs into consideration and find a path with the minimum improvement costs.
Data Availability e data generated or analyzed during this study are included in this published article (and its supplementary information file). e datasets generated during and/or analyzed during the current study are available in the National Bureau of Statistics of China (http://www.stats.gov.cn/ tjsj/ndsj/) and Ministry of Transport of the People's Republic of China (http://www.mot.gov.cn/shuju/).

Conflicts of Interest
e authors declare that they have no conflicts of interest.