Data envelopment analysis (DEA) is favored by researchers in the field of efficiency evaluation, because it does not need any weight assumption, does not need to determine the input-output function relationship in advance, does not need dimensionless data processing, and can quickly deal with the comprehensive evaluation of complex systems with multiple inputs and multiple outputs (Charnes et al., 1978). Among many DEA models, the slacks-based measure (SBM) model based on slack variables can not only bring the unexpected outputs into the analysis framework, but also effectively solve the problem of input-output slack and improve the accuracy of measurement results (Guan and Xu, 2016; Sun et al., 2018). Therefore, we choose the SBM model to measure CTGE. As the SBM model has developed quite mature and widely used, the specific calculation formulas are not repeated here. Please refer to Ma et al. (2021) and Zhou et al. (2020) for details.

The spatial Markov chain combines the traditional Markov method with the concept of “spatial lag” (Su et al., 2018; Wang et al., 2019). Based on the spatial auto-correlation theory, the spatial weight matrix was introduced to calculate the weighted average attribute value of adjacent units (spatial lag), and judge the spatial lag state of regional units (Rey and Montouri, 1999). It makes up for the deficiency that the traditional Markov chain ignores the interaction between regions, and can quantitatively analyze the spillover effect of the regional spatial lag on CTGE. Based on the spatial lag of the CTGE types in the initial year, the traditional Markov transition matrix of CTGE is decomposed into k K*K conditional transition probability matrices (see formula (1)). For the k-th conditional matrix, the element m_ij|k represents the spatial transition probability with the spatial lag type i of the region of year t as the background condition, and the time t belongs to the spatial transition probability from type i to the type j at the time t+1.

To obtain the spatial Markov transition matrix of CTGE in China, we calculate the spatial lag value of region i and neighborhood j in t years, and the formula is as follows.

where Lag_j is spatial lag value of region j. x_i is the efficiency value of region i. N is the number of regions. ω_ij is spatial weight matrix, represents the spatial relationship between region i and region j. This paper applies neighboring rook spatial weight matrix to calculate the spatial lag value. That is to say, if the areas i and j are adjacent, ω_ij = 1, otherwise ω_ij = 0. It is worth noting that in this study, we consider that Hainan Province is adjacent to Guangdong and Guangxi.

Exploratory spatio-temporal data analysis (ESTDA) was developed from exploratory spatial data analysis (ESDA). ESDA can obtain the spatial distribution description and visual atlas of the research objects through multi-angle spatial measurement, but it ignores the time factor. In order to realize spatio-temporal coupling, Rey (2010) fully considered the characteristics of time variation and proposed the ESTDA model, which can realize the visualization of spatio-temporal interaction. Therefore, this paper analyses the spatio-temporal dynamics of CTGE by means of global spatial autocorrelation, local spatial autocorrelation, LISA time path and spatio-temporal transition methods.

3.3.1 Local indicator of spatial association time path

The local indicator of spatial association (LISA) time path integrates the time dimension, which makes the traditional static LISA dynamic. Through the time migration of the LISA coordinates in the Moran scatter plot, the paired movement of spatial unit attribute values and their spatial lag are visualized (Rey et al., 2010), thus revealing the spatio-temporal evolution of the regional CTGE in local regions, and determining the spatio-temporal dynamics of the local spatial difference in CTGE. The moving path of LISA coordinates in the Moran scatter plot of region i can be expressed as a set of vectors [(y_i,1, yL_i,1), (y_i,2, yL_i,2), …, (y_i,t, yL_i,t)]. y_i,t represents the normalized value of CTGE in year t of region i. yL_i,t is the spatial lag of region i in year t. The geometric features of LISA’s time path are usually expressed by relative length (L_i) and tortuosity (δ_i) (Rey et al., 2011). The calculation method is as follows:

where L_i and δ_i denote relative length and tortuosity respectively; n is the number of space units; T is the annual time interval; L_i,t are LISA coordinates of unit i at time t. d(L_i,t, L_i,t+1) represents the movement distance of unit i from time t to t+1. d(L_i,1, L_i,T) represents the distance of unit i from the first year to the last year. The larger L_i is, the longer the moving length is, which indicates that the local spatial structure and local spatial dependence of CTGE is more dynamic. Similarly, the larger δ_i is, the more curved the LISA time path is, which indicates that CTGE has a more dynamic spatial dependence direction and a more fluctuating growth process.

3.3.2 Spatio-temporal transition

Spatio-temporal transition is based on LISA time path, which embeds the moving distance, direction, concentration degree and other attributes of each spatial unit in a specific time period into the Markov chain, and further reveals the temporal change of local neighborhood spatial relationship (Ye and Rey, 2013). Spatio-temporal transition is usually divided into four types, namely, Type 0, Type 1, Type 2 and Type 3 (see Table 1), which respectively represent the local and neighbouring CTGE as stable, only local CTGE in transition, only neighbouring CTGE in transition and both local and neighbouring CTGE in transition. Type 3 can also be subdivided into two types, Type 3A and Type 3B, which respectively represent the situation of co-migration and reverse migration of local and neighbouring CTGE.

Rey (2010) defines space-time flux (SF) and space-time cohesion (SC) as the ratio of the number of transitions of a certain type to the number of all transitions (n) in the study period. The formula is as follows:

where Type 0, Type 1, Type 2 and Type 3A represent the transition number of Type 0, Type 1, Type 2 and Type 3A, respectively. n denotes the total number of transition, n=(2018-2003)×30=450.

On the basis of previous studies, we constructed an index system that can reflect the social development, and regarded it as an expected output in the evaluation framework of transportation efficiency, thus reflecting the coordinated development of transportation and economic-social-environmental system. Table 2 shows the evaluation index system of CTGE. The input indicators include infrastructure, capital stock, labor force and energy consumption, while output indicators include expected output (traffic added value and social development index¹) (1 The social development index consists of four first-level indicators (living standard, urbanization level, transportation level and scientific technological & educational level) and 15 second-level indicators. Among them, the living standard includes GDP per capita, urban Engel’s coefficient, and rural Engel’s coefficient. Urbanization level includes the proportion of non-agricultural population, disposable income of urban residents’ families and built areas. Transportation level includes four secondary indicators: passenger turnover, cargo turnover, traffic accident and property loss in traffic accidents. The level of science and education is measured by the turnover of technology market, the number of patents granted, the proportion of science and education expenditure to government budget expenditure, the number of 10,000 college students, and illiteracy rate.)and unexpected output (carbon dioxide emission from transportation sector). Among them, the added value of transportation reflects the economic connotation of CTGE, that is, obtaining the maximum economic output with the least input of resources. Carbon emission index shows the environmental connotation of CTGE, which requires that the production process of transportation system should be based on protecting and improving the natural environment and maintaining the ecological balance, and gradually reduce the damage of unexpected output to the ecological environment. The social development index shows the social connotation of CTGE, which is people-oriented, realizing the sharing and fair distribution of transportation resources, continuously meeting the transportation needs of human development, and enhancing human well-being.

The data of infrastructure, capital stock, labor force, traffic added value and social development index are from China Statistical Yearbook, and the data of energy consumption and carbon emission are from China Energy Statistical Yearbook. Due to limited data, our research objects are 30 regions in China, excluding Hong Kong, Macao, Taiwan and Tibet, and the research period is from 2003 to 2018. Please refer to Ma et al. (2021) for the detailed calculation process of relevant indicators.

Figures 1 and 2 show the measurement results of CTGE. Overall, CTGE shows a trend of first decreasing and then increasing in China, with an average annual efficiency below 0.65, in a low overall level (Figure 1). Specifically, from 2004 to 2011, The CTGE declined rapidly, dropped to the lowest level (0.494) in 2011, and then steadily increased. Since the 21st century, with the accelerated development of China’s market economy, the demand for transportation has increased rapidly, and input elements such as infrastructure construction, capital, manpower, and energy in the transportation industry have increased rapidly. The management mode of “heavy scale but light quality” leads to the imbalance of supply-side resource allocation in the transportation industry, and the rapid growth of unexpected output leads to a rapid decline in CTGE (Ma et al., 2021). After 2011, China attached great importance to green transportation development, accelerated structural reform on the supply side of transportation, optimized resource allocation, and significantly promoted CTGE.

From the regional perspective (Figure 2), only five regions (Tianjin, Hebei, Shanghai, Qinghai, Ningxia) have an efficiency value of 1, which shows that their input and output are completely effective. Only 12 regions have efficiency values are higher than the average level. Most of these regions are located in eastern part of China, which shows that there are great regional differences in China’s CTGE. Qinghai and Ningxia are located in west China, and their economic level is far lower than that of Shanghai and Tianjin in east China. There are two major reasons. One is that Qinghai and Ningxia are sparsely populated, low in economic level, backward in transportation development and inadequate in transportation resources. The other is that the unexpected output caused by less transportation demand is far lower than that in the developed eastern regions. The combination of less input and unexpected output leads to higher CTGE. It should be noted that the efficiency measured by the DEA model is relatively effective rather than the actual one, which explains why the CTGE in the underdeveloped areas is effective.

4.2.1 Result analysis on Markov state transition matrix

The quartile method is used to divide the measurement results of CTGE into four states, which are I, II, III and IV from low to high. The change of CTGE from a low-value state to a high-value state is considered to be an upward movement, and from high-value state to low-value state considered to be a downward movement. If the state of CTGE does not change, it is considered to be stable. The results are shown in Table 3 and Figure 3.

(1) Table 3 shows that the value on the diagonal is much higher than that on the non-diagonal, with the maximum value of 0.927 and the minimum value of 0.639, which indicates that the transfer of CTGE has strong stability, and the minimum probability of no transfer during the study period is 63.9%. In other words, the probability of state III changing into other states is 36.1%. In addition, the probability of CTGE at states I, II and IV remaining unchanged in the future is 89.3%, 92.2% and 92.7%, respectively, which indicates that “club integration” exists in CTGE.

(2) The transition probability between different types of CTGE is very small, and the maximum value on the non-diagonal line is 0.125, which indicates that if the CTGE of a region is Type 3A at the beginning of the study, the maximum probability of switching to other types in the following years is 12.5%. This phenomenon indicates that the probability of transition from state III to state IV is 12.5%, which shows that the CTGE in some regions has improved, and the transition from the high state to the higher state occurred. It can also be found that the transition probability between two adjacent types is greater than that of cross type, which also indicates that the transition type of CTGE is stable and gradual, but does not cross type.

(3) The CTGE is polarized between upper and lower levels. For regions with high-value at the beginning of the study, the probability of maintaining high-value in the next few years is 92.7%, and the probability of downward transition is only 7.3%. However, it is located in Type I region at the initial stage of the study, and the probability of maintaining this type is 89.3% in the next few years, and the probability of upward transition is only 10.7%, and it only transitions to Type II. This shows that the status of most low-level areas has not changed, and the regions with low-level are in urgent need of promoting, so as to promote the overall promotion of CTGE in China

(4) From the perspective of spatial distribution (Figure 3), during the study period, 20 regions (accounting for 2/3 of the total) maintained a stable state of CTGE, among which Beijing, Tianjin, Hebei, Shanghai, Qinghai and Ningxia maintained a relatively high level. Downward regions include Inner Mongolia, Jilin, Zhejiang, Anhui, Fujian, and Chongqing. Jiangsu, Jiangxi, Henan, and Guizhou are regions where the status of CTGE has moved up, which indicates that there are regional differences in the spatial and temporal evolution trend of CTGE.

4.2.2 Result analysis on spatial Markov state transition matrix

The traditional Markov chain can calculate the time evolution characteristics of China’s CTGE, but it cannot detect the effect of CTGE in the neighboring regions on this region. Therefore, this paper calculates the spatial Markov transition probability of CTGE from 2003 to 2018 in China (Table 4), takes the average level of CTGE in the surrounding regions as the spatial lag type, and draws the spatial pattern distribution map of China’s CTGE type transfer and neighborhood transfer (Figure 4).

Table 4 shows that there is no region with Type I as the regional background (spatial lag), so the CTGE state transition with Type I as the spatial lag condition is not considered in this paper.

(1) The condition of spatial lag has an important influence on the state transition of CTGE in China. There is synergy between regional CTGE types and neighborhood types. When the neighborhood type is I, the number of regions with Type I CTGE is obviously higher than that of other types. When the domain type is II, the number of regions with Type IV CTGE is obviously higher than other types of regions. It shows that China’s CTGE has significant spatial spillover effect, that is, CTGE depends not only on its own development level, but also on its neighboring areas.

(2) Generally speaking, the transferring downward probability of CTGE types in the region increases when it is adjacent to the region with low-value CTGE, and the transferring upward probability of CTGE types in the region increases when it is adjacent to the region with high value. That is to say, when a region is adjacent to a high-value region, the probability of the CTGE level in the region will increase. For example, Table 4 shows that the downward transition probability of Type IV is 0.042 when the neighborhood is Type IV, and increases to 0.057 and 0.106 when Type III and Type II are the neighborhood, respectively. Under the neighborhood conditions of Type IV, Type III and Type II, the upward transfer probabilities of Type III regions are 0.086, 0.161 and 0.167, respectively, which indicates that there is a significant spatial spillover effect in the development of CTGE. If the high-value regions are adjacent, the spillover effect is positive, and the probability of CTGE shifting upward increases. Similarly, if low-value regions are adjacent, the spillover effect will be negative, and the probability of CTGE downward transition will increase.

(3) For regions with Type II, under the neighborhood conditions of Type II, Type III and Type IV, the transition probabilities to Type III are 0.084, 0.064 and 0.500, respectively, while the transition probabilities to Type I are 0.031, 0 and 0, respectively. This phenomenon suggests that Type II regions tend to transition to Type III. Especially under the condition of Type IV neighborhood, this trend is more prominent. For regions with Type III, the probabilities of transition to Type IV are 0.086, 0.161 and 0.167, respectively under the conditions of Type II, Type III and Type IV of neighborhoods, and the transition probabilities to Type II are 0.228, 0.258 and 0.167 respectively. This indicates that neighboring regions have higher CTGE levels, the higher the transition probability to high-value regions, and conversely, the higher the transition probability to lower types.

The spatial spillover effect of CTGE is the result of many factors such as market, technology and policy (Wang et al., 2020). Marketization deepens the free flow of people and goods between regions, and the open economy enables regions to learn advanced technologies from other regions, especially neighbouring regions. The marginal utility of policy innovation has prompted inter-governmental institutional learning and imitation, especially the development of informatization and the continuous improvement of regional transportation infrastructure, which has reduced the flow cost of elements between regions, thus making the spatial spillover effect of CTGE more significant.

Spatial relation is the fundamental reason for the formation and evolution of regional spatial structure. Therefore, it is necessary to consider the relevance of geospatial effects in the study of regional CTGE spatio-temporal evolution.

4.3.1 Global spatial autocorrelation analysis

This paper calculates and investigates the global Moran’s I and the global spatial correlation of CTGE from 2003 to 2018 by using the GeoDa software. Table 5 shows that the global Moran’s I of CTGE in China is positive from 2003 to 2018, and it is significant in other years except 2009, which indicates that CTGE has a strong agglomeration effect (positive correlation) in space. That is to say, the high (low) value regions of the CTGE tend to be adjacent in space.

4.3.2 Local spatial autocorrelation analysis

Since the global autocorrelation Moran’s I can not reflect the spatial agglomeration characteristics of a specific region, this paper draws the LISA agglomeration map of the CTGE in 2003, 2008, 2013 and 2018 by combining Moran scatter plot and local Moran’s I to explore whether there is a local agglomeration phenomenon of CTGE in China (Figure 5).

Through the comparative analysis of the four agglomeration types, we found that the high-value agglomeration centers (H-H agglomeration) of CTGE are relatively stable in north China and the Yangtze River Delta, while the low-value agglomeration centers (L-L agglomeration) gradually spread from the south of the Yangtze River to the northeast. During the research period, the number of L-H agglomeration increased slightly, while the spatial distribution pattern changed greatly. In 2003, the regions of L-H agglomeration were scattered in eastern, northeastern and northwestern China. In 2008, we found that the low-value isolated centers of L-H agglomeration were mainly in north China and the Yangtze River Delta. In 2013, low-value isolation centers were concentrated in Xinjiang, Heilongjiang and Yellow River Basin. In 2018, two isolated centers with low value were formed in the middle reaches of the Yellow River and the Yangtze River. The number of high-value isolated centers (H-L clusters) decreased slightly during the study period, but their spatial distribution pattern did not change much, and spread in eastern, central and western China in a divergent state. Generally speaking, the H-H concentration and L-L concentration are concentrated and distributed in a large number, while the L-H concentration and H-L concentration are divergent and distributed in a small number.

4.4.1 Local indicator of spatial association time path geometry features

The relative length and tortuosity of LISA time path can better reveal the dynamic and spatial correlation fluctuation of the local spatial structure of CTGE. In this paper, the length and tortuosity of LISA time path for CTGE from 2003 to 2018 are divided into four grades: low-relative length (low-tortuosity), medium-relative length (medium-tortuosity), higher-relative length (higher-tortuosity), and high-relative length (high-tortuosity). The relative length and tortuosity of LISA time path are visualized by ArcGIS 10.2 software, and the results are shown in Figure 6.

According to the relative length of LISA time path (Figure 6a), there are 20 regions below the average level (L_i<1), accounting for 66.67% of the regional proportion, which shows that the local spatial structure of CTGE has strong stability. In terms of spatial distribution, the long-path (1.080-2.438) regions are scattered, including Beijing, Heilongjiang, Inner Mongolia, Shandong and Henan in the north and Guizhou, Hainan and Fujian in the south. The short-path (0.352-1.079) regions are relatively concentrated in spatial distribution, mainly in north, northwest and southeast China. The reason is that the transportation level in these places is generally low, and the transportation service is mainly road transportation with high consumption and high emission, which leads to low CTGE and hard to change. Specifically, the relative length between Shandong and Hainan is 1.644-2.438, which is a high-relative length region, and its local space structure is extremely unstable. The seven provinces of Beijing, Heilongjiang, Inner Mongolia, Jiangsu, Fujian, Henan and Guizhou are higher-relative length, ranging from 1.080 to 1.643, and its local spatial pattern is dynamic. The relative length of Hebei, Tianjin, Shanxi, Liaoning, Jilin, Guangdong, Shanghai and Zhejiang is between 0.595 and 1.079, which is a medium-relative length region, indicating that the local spatial structure is relatively stable. These regions are mainly concentrated in northeast, north China and south of the Yangtze River. Xinjiang, Gansu, Sichuan, Shaanxi, Hunan and Hubei are low-relative length, with path length ranging from 0.353 to 0.874, which is mainly concentrated in western China and the middle reaches of the Yangtze River.

According to the tortuosity of LISA time paths (Figure 6b), 19 regions are smaller than the average tortuosity (8.262), accounting for 63.33% of the regional proportion, indicating that the local spatial dependence of CTGE is relatively stable. From the spatial distribution of tortuosity, the tortuosity of LISA time path of CTGE is in a central distribution structure, and it is decreasing from Beijing to Tianjin. Specifically, the regions with low-tortuosity (1.610-4.607) are mainly distributed in the northeast and southeast of China, and have the greatest stability in the direction of spatial dependence, which is related to the low level of CTGE in this region. The similar development level leads to weak fluctuation of local spatial dependence change process. The regions with high-tortuosity (16.867-32.444) are Beijing, Tianjin and Qinghai, which reflects that these regions have the most significant fluctuation and very dynamic spatial change process in the direction of spatial dependence on their neighbouring regions. The regions with medium-tortuosity (4.608-8.287) are mainly concentrated in the northwest and southwest regions of China. The provinces with higher- tortuosity (8.288-16.866) are mainly concentrated in north China and the middle reaches of the Yangtze River, which include Hebei, Shanxi, Shandong, Sichuan and Hubei provinces, indicating that CTGE has a strong local spatial dependence.

4.4.2 Local indicator of spatial association time path movement direction

The direction of the LISA time path can represent the comprehensive characteristics of the evolution of local spatial pattern of an element. This paper compares the specific positions of LISA scatter plot of CTGE in 2003 and 2018, calculates the moving direction of the LISA coordinates in various regions, and divides them into four types: win-win type (0°-90°), loss-win type (90°-180°), loss-loss type (180°-270°) and win-loss type (270°-360°). Among them, the win-win type indicates that CTGE of the province and its neighbors shows a positive cooperative growth. The lost-win type indicates that the province itself shows a low growth trend, while the neighborhood offers a high growth trend, and vice versa.

In terms of LISA time path transition direction (Figure 7), there are 13 regions with coordinated transition of CTGE, accounting for 43.33% of the total number of regions, which indicates that the integration of spatial pattern evolution of CTGE is weak. In the process of spatial evolution, coordinated spatial growth and spatial competition coexist. Among them, there are eight regions which are growing together, mainly distributed in the North China Plain and the Yangtze River Delta. The CTGE in these areas is generally high, showing obvious characteristics of coordinated and high-speed growth. There are five growing regions with negative cooperativity, which are concentrated in northeast China, Inner Mongolia, and Gansu, showing the characteristics of coordinated low-speed growth.

4.4.3 Analysis of spatio-temporal transition

The physical change characteristics of the LISA time path reveal the movement trend of the regional CTGE’s LISA coordinates, but it can not reflect the mutual transformation of local spatial association types in CTGE’s LISA coordinates. Therefore, this paper uses the spatio-temporal transition method proposed by Rey (2010) to further describe the transfer characteristics and evolution types of the local spatial association types of CTGE in China. The results are shown in Table 6.

Table 6 shows that the diagonal value of the probability transfer matrix is much larger than that on the non-diagonal, with the maximum value of 0.907 and the minimum value of 0.827 on the diagonal, indicating that the spatial structure between regions of CTGE is relatively stable, and the transition possibility different types are small. China’s CTGE has obvious transfer inertness. Among the different types of transitions, HH_t→LH_t+1, LH_t→ HH_t+1 and LH_t→LL_t+1 have the highest transition probabilities, which are 0.101, 0.095 and 0.069, respectively. The transition probabilities of the other types are low, which indicates that the spatial structure of CTGE is relatively stable. Turning to the spatio-temporal transition probability, Moran’s I scatter points with Type 0 is the highest during the research period, reaching 387 times, accounting for 86%. That is to say, the probability of Moran’s I scatter points staying in the same quadrant (Type 0) is 86%. The transition frequency of Type 3 is the least, only two times, accounting for only 0.4%, while the transition frequency of Type 1 and Type 2 is 34 times and 27 times, respectively, with probabilities of 0.076 and 0.060. According to formulas (4) and (5), the spatio-temporal conversion and spatio-temporal convergence probability of CTGE are calculated. The spatio-temporal cohesion (SC) of Moran’s I is 0.864, which indicates that the probability of no spatio-temporal transition in China’s comprehensive transportation green efficiency is 86.4%, and the space-time flux (SF) probability of Moran’s I is only 13.6%, indicating that the spatial agglomeration of CTGE has strong path dependence and locking characteristics.

Using the SBM, spatial Markov and ESTDA model, we calculated CTGE of 30 provinces from 2003 to 2018 in China, and explored the spatio-temporal interaction characteristics of its spatial pattern. The major conclusions are as follows:

(1) The green efficiency of China’s comprehensive transportation shows a U-shaped trend (first decreasing and then increasing), but the overall level is low. The efficiency values are ranging from 0.179 to 1, which is quite different in regions. The spatial distribution roughly presented eastern > western > central regions, without economic spatial regularity.

(2) The transfer of CTGE in China showed a steady and gradual feature, but there is also a polarization phenomenon of “the better, the worse”. The spatial spillover effect of CTGE in China is obvious. Generally speaking, if it is close to a high-level region, the upward transfer probability of CTGE will increase, while if it is close to the low-level region, the upward transfer probability of CTGE will decrease.

(3) During the study period, the CTGE has significant positive spatial correlation and spatial agglomeration characteristics, and spatial convergence (degree of agglomeration) shows a trend of decreasing at first and then increasing. H-H agglomeration and L-L agglomeration are large in number and concentrated in distribution, while L-H agglomeration and H-L agglomeration are small in number and dispersed.

(4) The LISA time path geometry features show that the evolution process of local spatial structure and local spatial dependence of CTGE is stable, but the integration of spatial evolution is weak, and spatial cooperation growth and spatial competition coexist. The results of LISA spatio-temporal transition show that CTGE has obvious transfer inertness, and has certain characteristics of path dependence and spatial locking, which will become a major difficulty that hinders the improvement of CTGE.

According to the conclusions, the following implications and enlightenment can be drawn.

(1) Optimize transportation structure to improve CTGE. Although CTGE has been rising in recent years, its overall level is still relatively low. The reason lies in the lack of effective links among various modes of transportation, which leads to unbalanced resource allocation and serious waste at the input end of comprehensive transportation system. Therefore, optimizing the transportation structure, strengthening the effective connection of various modes of transportation are the primary problems facing the construction of high-quality comprehensive transportation systems. Concrete measures are as follows. The development of new energy automobile industry using pure electricity, hybrid power, and hydrogen energy and having reduced consumption of fossil fuels, such as gasoline and diesel, should be promoted. New modes of transport organization, such as multimodal transport, swap trailer transport, etc., should be adopted to promote container transport and reduce transport intensity. Furthermore, it would be important to optimize the transport structure, vigorously promote highway to waterway and railway transport, and improve the green efficiency of comprehensive transportation.

(2) Making differentiated policy to improve CTGE. The evolution process of local spatial structure of CTGE in China has a strong stability. For low-efficiency areas, to break this stable process, it is necessary to optimize the traffic structure, scientifically plan the comprehensive traffic layout, and strengthen the connection between various modes of transportation. At the same time, it is necessary to intensify scientific and technological research and development, so as to make comprehensive transportation develop in the direction of intelligence, promote the development of transportation from resource input to technological innovation, and ensure that the expected output is increased and carbon emissions are reduced. For high-efficiency areas, it is necessary to keep the stability of CTGE local spatial structure and the evolution process of spatial dependence, and on this basis, continuously optimize the allocation of transportation production materials and improve the utilization rate of resources. On the other hand, it is necessary to give full play to the demonstration and leading role of efficient areas, improve the diffusion and spillover effects of advanced technology and management experience, and realize the cross-regional coordinated promotion of comprehensive transportation green efficiency.

(3) Narrow regional differences to improve CTGE. On the one hand, due to natural conditions, resource endowments, economic foundation and other factors, the speed and quality of traffic development in different areas are quite different. On the other hand, the difference in the degree and effect of policy implementation leads to the widening gap of CTGE in different regions. Therefore, under the guidance of the building national strength in transportation, local governments should strive to promote the development of comprehensive transportation system, formulate a coordinated development strategy of inter-regional comprehensive transportation, and establish a sound linkage mechanism for transportation cooperation, such as the integrated development of Beijing-Tianjin-Hebei, so as to achieve coordinated development of regional transportation while improving the green efficiency of their comprehensive transportation. In addition, it is necessary to strengthen the exchange and cooperation of talents, capital and technology among regions, give full play to the leading role of the developed regions around Bohai Sea, Yangtze River Delta and Pearl River Delta in the central and western regions and northeastern China, promote the spillover of new technologies, methods and knowledge, narrow regional differences, and realize the overall improvement of comprehensive transportation green efficiency.

In any case, this study has some limitations that merit mention. Firstly, there is no uniform standard for the construction of the social dimension index system at present. The index system established in this paper is not comprehensive, the calculation results are rough, and few indicators reflecting the government management decision-making level are involved. Further development and improvement are needed. Secondly, considering the availability and convenience of the data, all the data in this paper come from the National Bureau of Statistics, and the calculation results are relatively rough, so it is necessary to further refine the comprehensiveness and accuracy of the data in the future. Thirdly, this paper takes provincial units as the research object, and the research scale is large, so it is impossible to explore regional differences. In the future, it is necessary to refine the granularity and evaluate the CTGE in China with prefecture-level cities as evaluation units.

Nevertheless, we hope that our research results can provide sufficient theoretical support for the establishment of high-quality comprehensive transportation systems to realize the rationalization of transportation resources and the coordinated development of the regional economy.