Spatial Evolution of Green Total Factor Carbon Productivity in the Transportation Sector and Its Energy-Driven Mechanisms

Abstract

Achieving carbon reduction is essential in advancing China’s dual carbon goals and promoting a green transformation in the transportation sector. Changes in energy structure and intensity constitute key drivers for sustainable and low-carbon development in this field. To explore the spatial spillover effects of the energy structure and intensity on the green transition of transportation, this study constructs a panel dataset of 30 Chinese provinces from 2007 to 2020. It employs a super-efficiency SBM model, non-parametric kernel density estimation, and a spatial Markov chain to verify and quantify the spatial spillover effects of green total factor productivity (GTFP) in the transportation sector. A dynamic spatial Durbin model is then used for empirical estimation. The main findings are as follows: (1) GTFP in China’s transportation sector exhibits a distinct spatial pattern of “high in the east, low in the west”, with an evident path dependence and structural divergence in its evolution; (2) GTFP displays spatial clustering characteristics, with “high–high” and “low–low” agglomeration patterns, and the spatial Markov chain confirms that the GTFP levels of neighboring regions significantly influence local transitions; (3) the optimization of the energy structure significantly promotes both local and neighboring GTFP in the short term, although the effect weakens over the long term; (4) a reduction in energy intensity also exerts a significant positive effect on GTFP, but with clear regional heterogeneity: the effects are more pronounced in the eastern and central regions, whereas the western and northeastern regions face risks of negative spillovers. Drawing on the empirical findings, several policy recommendations are proposed, including implementing regionally differentiated strategies for energy structure adjustment, enhancing transportation’s energy efficiency, strengthening cross-regional policy coordination, and establishing green development incentive mechanisms, with the aim of supporting the green and low-carbon transformation of the transportation sector both theoretically and practically.

1. Introduction

1.1. Background and Related Studies

With the rapid development of the economy and society, the transportation sector plays a pivotal role in facilitating factor mobility and regional connectivity. However, its high energy intensity and substantial carbon emissions make it a critical focus in China’s low-carbon transition. Balancing transport efficiency with carbon reduction has thus become a central concern for both policymakers and researchers. Against this backdrop, green total factor productivity (GTFP) has been widely employed to assess the low-carbon development performance of different sectors. GTFP not only captures the efficiency relationship between carbon emissions and output but also reflects the combined effects of technological progress, energy utilization efficiency, and regional coordination. Therefore, examining the measurement and spatial disparities of GTFP in the transportation sector can help to uncover the mechanisms through which the energy structure and energy intensity influence green transformation, thereby providing a scientific basis for region-specific emission reduction policies.

Therefore, accurately measuring GTFP and examining the presence and magnitude of its spatial spillover effects at both the provincial and regional levels in China—particularly the influence of the energy structure and energy intensity—holds important implications for the formulation of differentiated carbon reduction policies in the transportation sector and in promoting the overall advancement of low-carbon development.

In recent years, carbon emission efficiency (CEE) has become a central focus of academic inquiry, with analyses generally adopting either a single-factor or a multi-factor perspective. Single-factor methods primarily assess how a single determinant affects aggregate carbon emissions, commonly expressed as carbon dioxide emissions relative to economic output or energy consumption [1,2]. However, such indicators are unable to capture the combined effects of multiple inputs in production activities and tend to overlook the joint influences of factors such as economic development and energy use. In contrast, multi-factor approaches better reflect the systemic nature of CEE. GTFP incorporates multiple inputs—including technology, energy, labor, and capital—to assess the level of output per unit of carbon emissions. The key lies in embedding carbon emissions into the traditional GTFP framework as an environmental cost, thereby reflecting the sector’s low-carbon production efficiency.

Research on GTFP has been conducted from a variety of perspectives [3,4]. In terms of measurement methods, scholars have widely adopted data envelopment analysis (DEA) [5], its index-based extensions [6], and stochastic frontier analysis (SFA) as common modeling approaches.

Regarding research content, studies have evolved from basic empirical assessments of GTFP [7] toward more advanced topics such as spatial correlation analysis [8] and spatiotemporal convergence [9]. The spatial scope of analysis has also expanded across different geographic levels, including national [10], provincial [11], and city-level [12] studies.

Moreover, some scholars have explored the effects of various drivers on GTFP within unified analytical frameworks, including industrial intelligence [13], environmental regulation [14], the development of new infrastructure [15], the digital economy [16], renewable energy adoption [17], and green innovation [18].

Regarding the energy structure and energy intensity, considerable attention has been paid by many scholars, with analyses conducted from multiple perspectives. Some studies focus on the interrelationships between the energy structure, economic structure, and energy intensity [19], while others employ empirical approaches to examine the evolution of the energy intensity [20], as well as its temporal and spatial heterogeneity [21], and the global patterns of energy consumption [22]. At the same time, research has also been extended to broader analytical frameworks, exploring the impact of energy development on carbon dioxide emissions [23], the implications of energy structure adjustment for China’s environmental sustainability [24], the potential spillover effects of renewable energy consumption [25], and the relationship between energy use and carbon emissions in the context of the digital economy [26].

Despite growing research on GTFP, significant gaps remain. Existing studies largely neglect the spatial spillover mechanisms of the energy structure and intensity, rarely examine intertemporal transitions via Markov chains, and seldom apply dynamic spatial Durbin models to capture local and spillover effects. To address these gaps, this study employs a super-efficiency SBM model with undesirable outputs, a dynamic spatial Markov chain, and a dynamic spatial Durbin model using panel data from 30 Chinese provinces (2007–2020). The analysis depicts the spatiotemporal evolution of GTFP, identifies the effects of the energy structure and intensity, and highlights their role in fostering regional coordination. These findings provide theoretical insights for the optimization of low-carbon transition pathways in China’s transportation sector.

1.2. Mechanism Pathways

To further investigate how the energy structure and energy intensity indirectly affect GTFP through multiple transmission channels—and to clarify the spatial transmission logic and feedback mechanisms—this study develops a conceptual framework for mechanism analysis (Figure 1).

In terms of impact mechanisms, the existing literature suggests that energy optimization not only influences firms’ green behavior through cost constraints [27] but may also generate significant spatial spillover effects [28]. The research perspective has expanded beyond direct effects to include multi-dimensional mechanisms such as infrastructure coordination [29], digitalization support [30], and the diffusion of green innovation [31], contributing to both theoretical advancement and technological innovation in the field of green efficiency.

2. Methods and Data

2.1. Methodological Framework

2.1.1. Measurement of GTFP

This study adopts a super-efficiency SBM model with undesirable outputs to measure the GTFP of regional transportation systems. The approach excludes the evaluated DMU from the reference set, assumes variable returns to scale (VRS), and applies a non-oriented specification [32].

$$\begin{array}{cc}\hfill min\rho & ={\displaystyle \frac{1+\frac{1}{m}{\sum }_{i=1}^m\frac{s_i^{-}}{x_{ik}}}{1-\frac{1}{p+q}\left({\sum }_{r=1}^q\frac{s_r^{-}}{y_{rk}}+{\sum }_{b=1}^p\frac{s_b^{-}}{b_{uk}}\right)}}\hfill \\ \hfill \mathrm{s}.\mathrm{t}.& \left\{\begin{array}{cc}{\displaystyle \sum _{j=1,j\ne k}^n}{x}_{ij}{\lambda }_j-s_i^{-}\le x_{ik}\hfill & (i=1,2,\cdots ,m)\hfill \\ {\displaystyle \sum _{j=1,j\ne k}^n}{y}_{rj}{\lambda }_j+s_r^{+}\ge y_{rk}\hfill & (r=1,2,\cdots ,q)\hfill \\ {\displaystyle \sum _{j=1,j\ne k}^n}{u}_{bj}{\lambda }_j-s_b^{-}\ge u_{bk}\hfill & (b=1,2,\cdots ,p)\hfill \\ {\lambda }_j\ge 0\hfill & (j=1,2,\cdots ,n;j\ne k)\hfill \\ s_i^{-},s_r^{+},s_b^{-}\ge 0\hfill \end{array}\right.\hfill \end{array}$$

(1)

Here, the vector ${\lambda }_j$ represents the weights. $x_{ik}$, $y_{rk}$, and $b_{uk}$ represent the i-th input, the r-th desirable output, and the u-th undesirable output of the k-th decision-making unit (DMU), respectively. The variables $s_i^{-}$, $s_r^{+}$, and $s_u^b$ denote the slack variables for inputs, desirable outputs, and undesirable outputs, respectively [33].

2.1.2. Kernel-Based Distribution Analysis

As a non-parametric technique, kernel density estimation (KDE) is applied to illustrate the temporal dynamics of a variable’s distribution. The general expression is

$$f\left(x\right)={\displaystyle \frac{1}{nd}}\sum _{i=1}^{n}K\left({\displaystyle \frac{X_i-x}{d}}\right),$$

(2)

where K is the kernel function, d is the bandwidth controlling smoothness, and n is the sample size. A larger d produces smoother curves but may increase bias.

2.1.3. Spatial Characteristic Analysis

(1)

Markov Chain

We apply a Markov chain model to analyze the dynamic evolution of GTFP. Provinces are divided into four quartile-based categories ($k=1,2,3,4$), where a higher k indicates higher GTFP. Upward (downward) transitions occur when a province moves to a higher (lower) quartile. The transition probability is defined as

$$m_{ij}={\displaystyle \frac{n_{ij}}{n_i}},$$

(3)

where $n_{ij}$ is the number of transitions from state i to j, and $n_i$ is the total frequency of state i.

(2)

Spatial Markov Chain

Traditional Markov chain models do not account for spatial dependence and therefore fail to capture the spatial spillover effects inherent in socioeconomic phenomena. To address this limitation, we adopt the spatial Markov chain model. The specific formulation is as follows:

$$\mathrm{Laga}=\sum Y_b{W}_{ab}$$

(4)

In the equation, $W_{ab}$ represents the spatial relationship between region a and region b, which, in this study, $Y_b$ is defined based on the contiguity (adjacency) principle. By comparing the spatial Markov transition probability matrices, one can assess the extent to which the spatial context influences the transition dynamics of socioeconomic phenomena.

2.1.4. Global Moran’s I

This statistic is adopted to measure the spatial clustering of GTFP among Chinese provinces, given by

$$I={\displaystyle \frac{{\displaystyle \sum _{i=1}^n}{\displaystyle \sum _{j=1}^n}{w}_{ij}(x_i-\overline{x})(x_j-\overline{x})}{s^2{\displaystyle \sum _{i=1}^n}{\displaystyle \sum _{j=1}^n}{w}_{ij}}},$$

(5)

$$s^2={\displaystyle \frac{1}{n}}\sum _{i=1}^n{(x_i-\overline{x})}^2.$$

(6)

Here, $w_{ij}$ denotes elements of the spatial weight matrix, $x_i$ and $x_j$ are the GTFP values of provinces i and j, $\overline{x}$ is the overall mean, and n is the number of provinces. Moran’s I ranges from $-1$ to 1, with positive values indicating positive spatial autocorrelation and values near zero suggesting spatial randomness.

2.1.5. Spatial and Dynamic Spatial Durbin Model Specification

Given the observed spatial autocorrelation in GTFP, it is necessary to adopt an econometric specification that explicitly accounts for spatial dependence. The spatial Durbin model (SDM) is particularly suitable in this context, as it extends the conventional spatial regression framework by including both the spatially lagged dependent variable and spatially lagged explanatory variables, thereby capturing endogenous interactions across regions, as well as potential indirect spillover effects. The general form of the SDM is expressed as follows [34]:

$$Y=\alpha +\rho WY+X\beta +WX\theta +\epsilon ,\epsilon \sim \mathcal{N}(0,{\sigma }^2{I}_n)$$

(7)

In this equation, Y represents the dependent variable, X is the matrix of independent variables, and $\alpha$ is the intercept term. W is the matrix of spatial weights reflecting the geographical or economic connectivity among provinces. The coefficient $\rho$ measures the degree of spatial dependence in the dependent variable through its spatial lag $WY$, while $\beta$ captures the impact of local explanatory variables. The parameter vector $\theta$ accounts for the spillover effects of explanatory variables from neighboring regions via $WX$. The error term $\epsilon$ is assumed to follow a normal distribution with mean zero and variance ${\sigma }^2{I}_n$.

To further extend the analysis and incorporate temporal dynamics, this study adopts the dynamic spatial Durbin model (DSDM). By including lagged dependent variables, the DSDM not only mitigates potential endogeneity but also allows for the more precise characterization of short- and long-term effects. The specification of the dynamic model is given as:

$$Y_t=\alpha +\sum _{k=1}^p{\varphi }_k{Y}_{t-k}+\rho W{Y}_t+X\beta +WX\theta +\mu +{\tau }_t{1}_n+{\epsilon }_t$$

(8)

Here, $Y_t$ denotes the dependent variable at time t, and ${\varphi }_k$ captures the influence of its lagged terms $Y_{t-k}$. The coefficient $\rho$ reflects contemporaneous spatial dependence, while $\beta$ and $\theta$ represent the effects of the local explanatory variables and their spatial lags, respectively. $\mu$ denotes province-specific fixed effects, and ${\tau }_t{1}_n$ accounts for time fixed effects common to all provinces. The disturbance term ${\epsilon }_t$ is assumed to follow a normal distribution with variance ${\sigma }^2{I}_n$.

By combining spatial dependence, temporal persistence, and spillover effects, the SDM and DSDM provide a comprehensive framework to analyze the spatial–temporal evolution of GTFP in China’s transportation sector.

2.1.6. Specification of the Spatial Weight Matrix

To capture spatial interactions, the spatial weight matrix $W_{ij}$ is specified using geographical distances. Let $d_{ij}$ denote the actual distance between province i and province j. Then, $W_{ij}$ is defined as the reciprocal of $d_{ij}$, with diagonal elements constrained to zero, as shown in Equation (9):

$$W_{ij}=\left\{\begin{array}{cc}{\displaystyle \frac{1}{d_{ij}}},\hfill & i\ne j\hfill \\ 0,\hfill & i=j\hfill \end{array}\right.$$

(9)

2.2. Data Sources and Variable Selection

2.2.1. Variable Selection

The evaluation of GTFP in the transportation sector requires a clear definition of the production system. In this study, the production units (DMUs), along with their input factors and both expected and undesired outputs, are characterized as follows.

Energy input: The estimation of carbon emissions is conducted using the conversion coefficients recommended in the 2006 IPCC Guidelines for National Greenhouse Gas Inventories [35], with the detailed values reported in

Table 1. Reference parameters for standard coal equivalency and CO₂ emissions.

Energy Category (104 t)	Coefficient for Standard Coal Equivalence	CO2 Emission Factor
BC	0.7143	0.7559
PR	1.4714	0.5538
KF	1.4714	0.5714
DO	1.4571	0.5921
HFO	1.4286	0.6185
LPG	1.7143	0.5042
NG	1.3300	0.4483

Note: The data are sourced from the 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Energy use is measured in units of 10⁴ tons. Standard coal equivalents are shown per 10⁴ t, while CO₂ emission factors are reported as tCO₂/10⁴ t. BC = Bituminous Coal, PR = Petrol, KF = Kerosene Fuel, DO = Diesel Oil, HFO = Heavy Fuel Oil, LPG = Liquefied Petroleum Gas, NG = Natural Gas.

. Eight types of energy—coal, oil, natural gas, etc.—are unified in terms of standard coal equivalents and combined to obtain the overall energy input. The values of the electricity conversion coefficients are obtained from the 2011 and 2012 Regional Grid Average CO₂ Emission Factors in China, published by the Department of Climate Change, National Development and Reform Commission (NDRC) [36], and are listed in Table 2.

Capital input: Capital input is measured using fixed asset investments in the transportation sector [37], and the perpetual inventory method is employed to calculate the capital stock.

Labor input: Labor input is represented by the number of employed persons in the transportation sector in urban areas.

Table 2. Average CO₂ emission factors of regional power grids.

Grid Region	Included Provinces	CO2 Emission Factor
NG	Inner Mongolia, Tianjin, Shanxi, Hebei, Beijing, Shandong	0.8843
NE	Jilin, Liaoning, Heilongjiang	0.7769
EG	Fujian, Shanghai, Zhejiang, Jiangsu, Anhui	0.7035
CG	Chongqing, Hubei, Hunan, Henan, Sichuan, Jiangxi	0.5257
NWG	Shaanxi, Qinghai, Xinjiang, Gansu, Ningxia	0.6671
SG	Guangxi, Guizhou, Yunnan, Guangdong, Hainan	0.5271

Note: The data are sourced from the “2011 and 2012 Regional Grid Average CO2 Emission Factors in China”, compiled by the Department of Climate Change, National Development and Reform Commission (NDRC) and widely used in carbon accounting studies [38]. Abbreviations: NG = North Grid, NE = Northeast Grid, EG = East Grid, CG = Central Grid, NWG = Northwest Grid, SG = South Grid. The coefficient of the carbon emission unit is kg CO₂/kWh.

Table 2. Average CO₂ emission factors of regional power grids.

Grid Region	Included Provinces	CO2 Emission Factor
NG	Inner Mongolia, Tianjin, Shanxi, Hebei, Beijing, Shandong	0.8843
NE	Jilin, Liaoning, Heilongjiang	0.7769
EG	Fujian, Shanghai, Zhejiang, Jiangsu, Anhui	0.7035
CG	Chongqing, Hubei, Hunan, Henan, Sichuan, Jiangxi	0.5257
NWG	Shaanxi, Qinghai, Xinjiang, Gansu, Ningxia	0.6671
SG	Guangxi, Guizhou, Yunnan, Guangdong, Hainan	0.5271

Note: The data are sourced from the “2011 and 2012 Regional Grid Average CO2 Emission Factors in China”, compiled by the Department of Climate Change, National Development and Reform Commission (NDRC) and widely used in carbon accounting studies [38]. Abbreviations: NG = North Grid, NE = Northeast Grid, EG = East Grid, CG = Central Grid, NWG = Northwest Grid, SG = South Grid. The coefficient of the carbon emission unit is kg CO₂/kWh.

The outputs comprise both positive and negative indicators, with the sector’s value added capturing the desirable component, reflecting the economic benefit of transport activities. The undesirable output is the total amount of carbon dioxide emissions [39].

There are two primary methods for the calculation of CO₂ emissions: the top-down approach and the bottom-up approach. The top-down method estimates emissions based on the total energy consumption of the transportation sector. It is widely used due to its data availability and relatively high accuracy. In contrast, the bottom-up approach calculates CO₂ emissions based on specific types of transport modes, travel distances, and the corresponding energy consumption. While this method allows for a more detailed analysis of the contributions from different transport modes, it requires large amounts of disaggregated data—such as vehicle types and travel distances—often scattered across various agencies and enterprises, making it difficult to implement.

Therefore, this study adopts the top-down approach and estimates CO₂ emissions using the model specified in Equation (10):

$${\mathrm{CO}}_2=\sum _{i=1}^{14}{\mathrm{CO}}_{2i}=\sum _{i=1}^{14}{E}_{ti}·{CEF}_i$$

(10)

In this expression, $E_{ti}$ stands for energy use, where t indicates the energy category and i the time period, and $CE{F}_i$ denotes the associated emission coefficient emission coefficient. The coefficients for all energy types are documented in Table 1 and Table 2.

2.2.2. Data Sources

Due to data unavailability, Tibet, Hong Kong, Macao, and Taiwan are excluded from the analysis. The country is further divided into four regions: Eastern, Central, Western, and Northeastern China.

Transportation energy consumption is measured by converting the use of eight major energy types—including coal, gasoline, diesel, and natural gas—into standard coal equivalents using standard coal conversion coefficients [40]. The relevant data are obtained from the China Statistical Yearbook, China Transportation Yearbook, China Energy Statistical Yearbook, China Environmental Yearbook, and National Bureau of Statistics database [41].

2.3. Variable Selection

2.3.1. Core Explanatory Variables

(1)

Energy Structure of Transportation Industry (ESTI)

The energy structure determines the carbon emission baseline and reduction potential. A higher share of clean energy can significantly reduce the environmental burden per unit of transportation activity. In this study, the energy structure of transportation industry (ESTI) variable is measured by the proportion of electricity consumption in the total energy consumption of the transportation sector.

(2)

Energy Intensity of Transportation Industry (EITI)

The energy intensity of transportation industry (EITI) variable reflects the energy efficiency level of the transport system. Enhancing vehicle performance and optimizing transport organization can lower the energy demand per unit of transport turnover. This indicator is calculated as the ratio of total energy consumption to total transport turnover (10,000 tons of standard coal per 100 million ton-kilometers). The calculation follows the method proposed by Yu et al. [42], in which the total transport turnover is obtained by converting passenger and freight volumes across different modes using a comprehensive turnover coefficient.

Table 3. Transport mode conversion coefficients.

Transport Mode	Highway	Waterway	Railway
Coefficient	0.10	0.33	1.00

Note: Data source: conversion coefficients for passenger and freight turnover provided by the Ministry of Transport of China.

presents the adopted conversion coefficients in detail.

2.3.2. Control Variables

This study includes five control variables: the economic development level, R&D intensity, express delivery volume, transportation intensity, and environmental regulation (

Table 4. Description of variables.

Category	Variable	Symbol	Measurement Method
Explained Variable	Green total factor productivity in transportation	GTFP	Measured by super-efficiency SBM model
Core Explanatory Variables	Energy structure of transportation industry	ESTI	Percentage of electricity consumption in total industry energy consumption
	Energy intensity of transportation industry	EITI	Total industry energy consumption/comprehensive transport turnover
Control Variables	Economic development level	EDL	Per capita GDP
	R&D intensity	RDI	Internal expenditure on R&D as % of GDP
	Express delivery volume	EDV	Annual total express delivery business volume
	Transportation intensity	TITI	Comprehensive transport turnover/value added of transportation industry
	Environmental regulation	ER	Completed industrial pollution control investment/industrial value added

Note: CNY refers to Chinese Yuan. For reference, CNY 1 billion ≈ USD 145 million based on the average exchange rate in 2020 (USD 1 ≈ CNY 6.9). All GDP values are expressed in constant prices and reported in CNY to maintain consistency with official Chinese statistical data.

).

Economic Development Level (EDL): Higher economic development lays the groundwork for green technological innovation and enhances the resource allocation efficiency via industrial upgrading. Consistent with prior studies, regional economic development is represented by the per capita GDP in this paper.

R&D Intensity (RDI): RDI reflects the density of regional input into innovation. A stronger technological innovation capacity can drive green transformation through improved production processes and enhanced energy efficiency. In this study, RDI is measured by the ratio of intramural R&D expenditure to GDP.

Express Delivery Volume (EDV): EDV directly reflects the scale of demand for transport services. While rapid growth in express delivery can improve the efficiency of transport networks, it may also increase energy consumption due to intensified pressure from last-mile distribution. This study uses the total annual volume of express deliveries as the indicator.

Transportation Intensity (TITI): This variable indicates the degree of dependence of the economic output on transport services. The close coupling between the transport demand and economic activity may exacerbate energy consumption and environmental pollution. TITI is measured as the quotient of total transport turnover and value added in the transportation industry, standardized in terms of 100 million ton-kilometers per CNY 100 million.

Environmental Regulation (ER): Environmental regulation strengthens ecological constraints, forcing firms to improve their pollution control technologies and upgrade to cleaner production processes, thereby reducing the environmental impact of economic activities. ER is quantified as the proportion of completed investment in industrial pollution control relative to industrial value added.

3. Empirical Analysis and Results

3.1. Temporal Trends of Transportation GTFP

This study employs the super-efficiency SBM model to calculate the GTFP of 30 Chinese provinces from 2007 to 2020.

Table 5. GTFP calculated values.

Province	2007	2014	2020	Province	2007	2014	2020
Beijing	0.267	0.361	0.297	Henan	0.589	0.486	0.557
Tianjin	0.310	0.521	0.536	Hubei	0.247	0.295	0.362
Hebei	0.783	0.911	1.187	Hunan	0.370	0.423	0.373
Shanxi	0.473	0.351	0.578	Guangdong	0.384	0.397	0.363
Inner Mongolia	0.332	0.496	0.664	Guangxi	0.265	0.283	0.253
Liaoning	0.282	0.357	0.964	Hainan	0.267	0.255	0.344
Jilin	0.313	0.301	0.351	Chongqing	0.254	0.276	0.274
Heilongjiang	0.313	0.295	0.197	Sichuan	0.327	0.224	0.231
Shanghai	0.243	0.485	0.566	Guizhou	0.244	0.437	0.261
Jiangsu	0.516	0.565	0.616	Yunnan	0.147	0.098	0.269
Zhejiang	0.356	0.422	0.346	Shaanxi	0.231	0.278	0.355
Anhui	0.497	0.312	0.649	Gansu	0.405	0.169	0.217
Fujian	0.552	0.441	0.433	Qinghai	0.162	0.125	0.150
Jiangxi	0.366	0.412	0.513	Ningxia	0.283	0.429	0.432
Shandong	0.724	0.492	0.571	Xinjiang	0.253	0.249	0.238

and Figure 2 report the results, with GTFP evaluated using the SBM model.

According to the data and visualizations, the GTFP across Chinese provinces exhibited diverse trends from 2007 to 2020. To further analyze its temporal evolution, the provinces are classified into four regions—Eastern, Central, Western, and Northeastern China—following the standard division of the National Bureau of Statistics. The regional changes are illustrated in Figure 2. Throughout the study period, the national GTFP fluctuated within the range of 0.20–0.60. Regional heterogeneity became more evident and gradually widened, exhibiting spatial stratification characterized by “East > National Average > Central > Northeast > West”.

The eastern region consistently outperformed the national average in GTFP, with its advantage gradually increasing. This indicates a strong foundation in green efficiency and greater responsiveness to environmental policies. The central region generally lagged slightly behind the national average, although, around 2014, it temporarily caught up or slightly exceeded the average, providing a short-term boost. Notably, the GTFP in this region surged after 2016, which may have been closely related to the implementation of national policies promoting transport restructuring and green infrastructure—such as increased railway electrification and pilot projects for green highway transport [43].

The northeastern region followed a trajectory similar to the national trend, with GTFP values consistently below the national average. However, a clear upward shift occurred after 2016, reflecting progress in transport electrification and improvements in logistics organization. These developments suggest that policy transformations had begun to take effect. In contrast, the western region persistently recorded the lowest GTFP among all regions. A pronounced decline around 2014, followed by only a modest recovery, highlights its lagging performance and underscores the region as a critical target for future green transport development.

Kernel density estimation was applied to assess the distributional dynamics of GTFP across provinces. In this context, the height of the density peak at each time point represents the degree of temporal concentration in GTFP, as indicated by the kernel density value. The results are shown in Figure 3.

Based on the kernel density estimation results shown in Figure 3, the dynamic distribution of GTFP in China’s transportation sector from 2007 to 2020 exhibits the following characteristics:

Over the study period, the kernel density curves demonstrate a fluctuating trend—shifting rightward initially and then leftward. From 2007 to 2012, the main peak gradually moved to the right, indicating a general improvement in GTFP across provinces. This rightward trend continued between 2012 and 2016, reflecting sustained efficiency gains. However, by 2020, the main peak shifted sharply to the left, accompanied by a pronounced leftward tail, suggesting an overall decline in GTFP at the national level, although high-efficiency provinces maintained relatively stable performance.

The height and width of the density peaks fluctuated in stages. Between 2007 and 2012, the peak height declined and the distribution range widened, indicating increasing interprovincial disparities and the initial emergence of regional divergence. In 2016, the peak height rebounded and the distribution range narrowed, reflecting short-term convergence as some central and western provinces caught up under regional coordination policies. By 2019, the peak had become significantly taller and narrower, showing growing convergence in GTFP across provinces. However, the accumulation of density in the left tail revealed a persistent structural gap between low-efficiency western provinces and high-performing regions.

From 2007 to 2016, the density curves featured coexisting primary and secondary peaks, reflecting a polarized spatial structure of high- and low-efficiency regions. In 2019, the secondary peak disappeared and the distribution developed a unimodal, right-skewed shape. Nonetheless, the extended left tail highlighted the lack of inclusive efficiency improvements and suggested a shift from explicit bimodality to implicit segmentation. This transformation indicates that, while phased policy interventions can temporarily narrow efficiency gaps, long-term structural disparities persist—particularly in low-efficiency regions constrained by limited technological absorptive capacity and strong path dependence. Addressing these entrenched differences will require deeper structural reforms.

3.2. Spatial Clustering Characteristics

This study conducts a spatial autocorrelation analysis of GTFP in the transportation sector across provinces using the Moran’s I index. The results show that Moran’s I was 0.303 (p = 0.005) in 2007 and increased to 0.428 (p = 0.000) in 2020, indicating a strong and rising positive spatial correlation in provincial GTFP.

To more accurately examine the spatial clustering patterns and temporal evolution of GTFP among provinces, Moran scatter plots for the years 2007, 2012, 2016, and 2020 were generated using the Stata 17 software, as shown in Figure 4.

This study employs Moran’s I index to analyze the spatial autocorrelation of GTFP across provinces. The results show that the Moran’s I values were 0.3034 in 2007, 0.1442 in 2012, 0.3763 in 2016, and 0.4280 in 2020, indicating a generally positive spatial correlation in provincial GTFP, albeit with some fluctuations over time.

To provide a clearer view of the spatial clustering and its evolution, Moran scatter plots were produced for 2007, 2012, 2016, and 2020. As shown in Figure 3, most provinces are concentrated in quadrants I and III, corresponding to “high–high” (H-H) and “low–low” (L-L) agglomerations, while only a minority appear in quadrants II and IV, corresponding to “low–high” (L-H) and “high–low” (H-L) configurations.

In the four selected years, the proportion of provinces falling into the H-H and L-L quadrants accounted for 60%, 60%, 73%, and 86% of the total 30 provinces, respectively (

Table 6. Spatial clustering types of provincial GTFP based on Moran scatter plots (2007–2020).

Year	H-H	H-L	L-H	L-L
2007	Shandong, Henan, Hebei, Shanxi, Anhui, Jiangsu, Jiangxi, Guangxi	Guangdong, Hunan, Gansu, Fujian	Beijing, Tianjin, Liaoning, Shanghai, Zhejiang, Inner Mongolia, Hubei, Hainan	Qinghai, Shaanxi, Ningxia, Heilongjiang, Jilin, Xinjiang, Chongqing, Guizhou, Sichuan, Yunnan
2012	Shandong, Tianjin, Henan, Inner Mongolia, Fujian, Jiangsu	Guangdong, Hunan, Jiangxi, Ningxia, Guizhou, Hebei	Beijing, Shanghai, Liaoning, Shanxi, Anhui, Zhejiang	Hainan, Hubei, Jilin, Heilongjiang, Shaanxi, Guangxi, Chongqing, Gansu, Yunnan, Qinghai, Sichuan, Xinjiang
2016	Beijing, Tianjin, Shandong, Liaoning, Shanxi, Zhejiang, Henan, Fujian, Hebei, Jiangsu, Inner Mongolia	Jiangxi, Hunan, Guangdong, Guizhou	Shanghai, Hainan, Anhui, Jilin	Heilongjiang, Hubei, Guangxi, Shaanxi, Chongqing, Ningxia, Sichuan, Xinjiang, Yunnan, Gansu, Qinghai
2020	Liaoning, Hebei, Shandong, Tianjin, Shanxi, Henan, Jiangsu, Inner Mongolia, Anhui, Shanghai	Jiangxi	Beijing, Jilin, Zhejiang, Heilongjiang	Hubei, Shaanxi, Guangdong, Ningxia, Fujian, Hainan, Hunan, Gansu, Guangxi, Chongqing, Guizhou, Sichuan, Yunnan, Qinghai, Xinjiang

). From 2007 to 2020, GTFP at the provincial level was predominantly characterized by spatial clustering, with strong evidence of positive spatial dependence among neighboring regions.

The H-H clusters are concentrated in eastern coastal provinces like Shanghai and Jiangsu, supported by solid economic bases and advances in energy structure innovation. Conversely, L-L clusters appear mainly in western and northeastern areas, such as Gansu, Qinghai, and Heilongjiang, where the elevated energy intensity and inflexible industrial structures limit GTFP growth.

3.3. Spatiotemporal Transitions in GTFP

To reveal the dynamic spatiotemporal evolution of GTFP across regions under spatial dependence, this study applies the Markov chain approach. Based on the quartile method, all regions in China are classified into four categories: Type I (low productivity), Type II (lower-middle productivity), Type III (upper-middle productivity), and Type IV (high productivity).

By comparing the traditional Markov transition probability matrix (

Table 7. Conventional Markov chain transition probability matrix.

Type	I	II	III	IV	N
I	0.7551	0.2245	0.0204	0.0000	98
II	0.1939	0.6020	0.1837	0.0204	98
III	0.0300	0.1300	0.6700	0.1700	100
IV	0.0000	0.0106	0.1489	0.8404	94

) with the spatial Markov transition probability matrix (

Table 8. Transition probability matrix of the spatial Markov chain.

Lag	t/(t + 1)	I	II	III	IV	N
I	I	0.8519	0.1111	0.0370	0.0000	27
	II	0.5000	0.3333	0.1667	0.0000	6
	III	0.1111	0.1111	0.6667	0.1111	9
	IV	0.0000	0.0000	0.0000	1.0000	1
II	I	0.8200	0.1800	0.0000	0.0000	50
	II	0.1667	0.7333	0.1000	0.0000	30
	III	0.0667	0.2000	0.6667	0.0667	15
	IV	0.0000	0.0000	0.5000	0.5000	4
III	I	0.5000	0.4286	0.0714	0.0000	14
	II	0.1860	0.6279	0.1628	0.0233	43
	III	0.0000	0.0476	0.7619	0.1905	42
	IV	0.0000	0.0278	0.2222	0.7500	36
IV	I	0.4286	0.5714	0.0000	0.0000	7
	II	0.1579	0.4211	0.3684	0.0526	19
	III	0.0294	0.2059	0.5588	0.2059	34
	IV	0.0000	0.0000	0.0755	0.9245	53

), we further investigate the spatial and temporal transfer patterns of GTFP across provinces, aiming to uncover the underlying regularities in its dynamic evolution.

As shown in Table 7, the Markov transition probability matrix reveals a strong path dependence in the distribution of GTFP levels across China. The diagonal elements (0.7551, 0.6020, 0.6700, 0.8404) are consistently higher than the off-diagonal values, indicating strong self-reinforcing inertia within each category. In particular, the probabilities of remaining in the lowest (Type I) and highest (Type IV) categories reach 75.51% and 84.04%, respectively, forming a “low-level lock-in” and “high-level convergence” club phenomenon.

The transitions exhibit a gradual evolutionary pattern, with the highest probabilities occurring between adjacent categories. For example, the probability of upgrading from Type I to Type II is 22.45%; the transition probabilities from Type II to Type I and Type III are 19.39% and 18.37%, respectively; and the probabilities of moving from Type III to Type II and Type IV are 13.00% and 17.00%, respectively. No cross-level transitions were observed.

The mobility pattern is asymmetric. The upward transition probability from Type I is 24.49%, that from Type II is 18.37%, and that from Type III is only 17.00%, indicating that upward mobility becomes increasingly difficult at higher productivity levels. It is also noteworthy that the downward transition probability from Type IV is 15.95%, with 14.89% shifting to Type III, suggesting that even high-performing regions face the risk of regression.

The spatial Markov chain analysis highlights several notable patterns. First, the transition probability matrices exhibit significant variation across different spatial lag specifications, implying that disparities in GTFP levels among neighboring provinces shape the likelihood of shifts between productivity categories.

Second, the diagonal values of the matrices do not always dominate the off-diagonal ones, which indicates that spatial spillover effects weaken the persistence of GTFP “lock-in”. This phenomenon is most evident under the Type IV spatial lag.

Third, the appearance of non-zero probabilities both above and below the diagonal reflects instability in GTFP transitions. While upward mobility to higher categories exists, downward shifts are also possible. Importantly, these changes occur only between adjacent categories, with no evidence of jumps across multiple levels.

Fourth, the response of the same GTFP category differs under alternative spatial lag structures. For example, under the Type IV lag, the probability of moving from low to lower-middle GTFP reaches 57.14%, compared to only 11.11% under the Type I lag.

Finally, the effect of a specific lag type varies depending on the initial GTFP position. Under the Type II lag, the probabilities of moving upward by one level are 18.00%, 10.00%, and 6.67% for the low, lower-middle, and upper-middle categories, respectively, showing a downward gradient. This indicates that both the spatial lag structure and the starting GTFP level jointly influence the transition probabilities.

3.4. Selection and Analysis of Spatial Econometric Models

When the time dimension (T) is less than 20, unit root and cointegration tests for panel data are generally not required. In this study, $T=14$, so neither unit root nor cointegration testing is conducted.

Following the approach of [44], to ensure the accuracy of the GTFP determinant analysis, we conduct a series of tests. In

Table 9. Results of LM, Wald, and LR diagnostic tests.

Test	Stat.	p
LM-Err	4.552 ***	0.033
RLM-Err	8.257 ***	0.004
LM-Lag	14.078 ***	0.000
RLM-Err	17.783 ***	0.000
W-Err	12.54 ***	0.0281
W-Lag	16.54 ***	0.0206
LR-Err	17.21 ***	0.0086
LR-Lag	15.94 ***	0.0141

Note: Stat. = Test Statistic; p = p-value; RLM = Robust LM, LR = Likelihood Ratio; *** significant at 1% levels.

, the test results are reported.

As shown in Table 9, the tests all return significant results. In addition, the Hausman test rejects the null hypothesis of random effects at the 1% level, indicating that the SDM with both spatial and time fixed effects is the appropriate specification.

3.4.1. Benchmark Regression Result Analysis

Regression analyses are conducted using the traditional panel model, the SDM, and the dynamic SDM, both with fixed effects.

As shown in

Table 10. Regression estimates: traditional panel model and static and dynamic spatial Durbin models (FE).

Variable	Traditional Panel Model	Static SDM (FE)	Dynamic SDM (FE)
L.GTFP	—	—	−0.200 ***
ESTI	0.0815 *	0.235 ***	0.263 ***
EITI	−0.1524 ***	−0.290 ***	−0.212 **
EDL	0.1707	0.243 **	0.262 ***
RDI	−0.1701	−0.317 ***	−0.310 ***
EDV	−0.0188 ***	0.172 ***	0.203 ***
TITI	−0.8771	−0.0991	−0.0248
ER	−0.6290 *	−1.071 ***	−1.288 ***
$R^2$	0.3252	0.157	0.093

Note: FE = fixed effects. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.

, the regression results based on GTFP are compared across the traditional fixed-effects panel model, the SDM, and the dynamic SDM. The findings indicate that, after controlling for both individual and time effects and explicitly accounting for spatial and temporal dependence, the impacts of core energy variables are systematically amplified.

Specifically, the positive effect of the energy structure in the transportation sector (ESTI) increases from 0.0815 in the traditional model to 0.235 and 0.263 in the static and dynamic SDMs, respectively. The negative impact of energy intensity (EITI) becomes more pronounced, from $-0.1524$ in the traditional model to $-0.290$ in the static SDM. Although it slightly decreases to $-0.212$ in the dynamic model, it remains significantly negative. These results suggest that the promoting effect of cleaner energy and the suppressing effect of the energy intensity on GTFP are more accurately identified under a well-specified spatiotemporal framework.

Regarding control variables, the coefficient for the economic development level (EDL) shifts from insignificant in the traditional model to significantly positive in the spatial models. The negative impact of the R&D intensity (RDI) deepens from $-0.1701$ to $-0.317$, highlighting that innovation inputs have not yet translated into improvements in green efficiency. The coefficient for the express delivery volume (EDV) changes from $-0.0188$ to significantly positive in both spatial and dynamic settings, suggesting that, once spatial and temporal dependencies are considered, the expansion of the business scale and network integration may contribute to overall improvements in green efficiency.

The transportation intensity (TITI) remains insignificant across all three models, indicating that its marginal explanatory power may be absorbed by energy efficiency and structural variables. The coefficient for environmental regulation (ER) decreases further from $-0.6290$ to $-1.071$ and $-1.288$, implying that, during the study period, environmental regulation primarily exerted a compliance cost effect. Finally, the coefficient for the lagged dependent variable ($L.\mathrm{GTFP}$) in the dynamic SDM is $-0.200$, indicating a significant negative adjustment effect, i.e., GTFP exhibits a tendency for downward correction in its temporal dynamics.

3.4.2. Spatial Effect Decomposition

Within the spatial Durbin model, the coefficients obtained from estimation cannot be viewed as straightforward marginal effects. Instead, following the approach of LeSage and Pace [45], these impacts are separated into direct effects that remain within a region and indirect effects that diffuse across neighboring regions.

The direct effect represents the impact of an explanatory variable on the local GTFP, while the indirect effect captures the spillover influence of the same variable in neighboring regions on the local GTFP.

Given that this study employs a dynamic SDM, both the direct and indirect effects are further divided into short-term and long-term effects. The short-term effect reflects the immediate impact of a variable, while the long-term effect incorporates time-lagged dynamics. The corresponding empirical results are provided in detail in

Table 11. Short-term and long-term direct and indirect effects from the dynamic SDM model.

Variable	Short Term (S)			Long Term (L)
	D	I	T	D	I	T
ESTI	0.253 ***	1.195 ***	1.447 ***	0.203 ***	0.307 **	0.510 ***
	(4.95)	(2.72)	(3.17)	(3.51)	(2.40)	(4.67)
EITI	−0.197 **	−1.287 **	−1.484 ***	−0.132	−0.396 **	0.528 ***
	(−2.41)	(−2.43)	(−2.71)	(−1.36)	(−1.97)	(−3.20)
EDL	0.249 ***	1.236 **	1.484 ***	0.195 *	0.332	0.527 ***
	(3.19)	(2.43)	(2.94)	(1.92)	(1.60)	(3.57)
RDI	−0.311 ***	0.352	0.041	−0.383 ***	0.398 *	0.015
	(−6.07)	(0.66)	(0.08)	(−5.31)	(1.77)	(0.08)
EDV	0.201 ***	0.646 **	0.847 ***	0.184 ***	0.115	0.299 ***
	(8.01)	(2.36)	(3.00)	(6.28)	(1.37)	(4.22)
TITI	−0.011	−0.997 **	−1.008 *	0.061	−0.419 **	−0.359 **
	(−0.14)	(−1.96)	(−1.91)	(0.65)	(−2.07)	(−2.08)
ER	−1.251 ***	−5.779 ***	−7.030 ***	−1.016 ***	−1.471 ***	−2.486 ***
	(−5.06)	(−3.04)	(−3.47)	(−4.00)	(−3.11)	(−5.33)

Note: S = short-term effects; L = long-term effects; D = direct effect; I = indirect effect; T = total effect. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

.

(1) Energy Structure in the Transportation Sector (ESTI): As shown in the decomposition results, the short-term direct, indirect, and total effects of ESTI are all significantly positive at the 1% level. Notably, the coefficient of the indirect effect (1.195) is substantially larger than that of the direct effect (0.253), suggesting that the positive impact of optimizing the energy structure in the transportation sector on GTFP is primarily manifested through strong spatial spillover effects.

In the long term, both the direct and indirect effects remain positive, being significant at the 1% and 5% levels, respectively. However, their magnitudes (0.203 and 0.307) are lower than those in the short term, indicating that, while energy structure optimization continues to enhance GTFP over time, its spatial spillover effect tends to diminish.

(2) Energy Intensity in the Transportation Sector (EITI): For EITI, the short-term direct and indirect effects are both significantly negative, and the total effect is significant. This indicates that a reduction in energy intensity (i.e., improvements in energy efficiency) can significantly enhance green productivity both locally and in neighboring regions.

In the long term, the direct effect becomes statistically insignificant, while the indirect and total effects remain significant at the 5% and 1% levels, respectively. Furthermore, the magnitude of the long-term indirect effect (0.396) is notably smaller than that of its short-term counterpart (1.287), implying that the spatial spillover effects of reduced energy intensity weaken over time.

(3) Control Variables:

For the economic development level (EDL), the short-term direct, indirect, and total effects are all positive and statistically significant at the 1%, 5%, and 1% levels, respectively. Notably, the spatial spillover effect (1.236) is considerably larger than the direct effect (0.249), suggesting that improvements in economic development can effectively promote green efficiency in neighboring regions. In the long term, the direct effect (0.195) is significant at the 10% level, while the indirect effect is insignificant, indicating that the spatial spillover of economic development may gradually weaken or even disappear over time.

The R&D intensity (RDI) exerts a significantly negative direct effect in both the short ($-0.311$) and long run ($-0.383$), with the impact strengthening over time, suggesting that R&D investment has not effectively enhanced green efficiency. The indirect effects are generally weak, as only the long-term spillover is marginally significant at the 10% level.

Regarding the express delivery volume (EDV), both the short-term direct and indirect effects are significantly positive at the 1% and 5% levels, respectively. In the long term, the direct effect remains significantly positive, but the indirect effect becomes insignificant. This indicates that expanding express delivery services sustainably enhances local green efficiency, but its spillover benefits tend to diminish over time.

For the transportation intensity (TITI), the short-term direct effect is not statistically significant, whereas the short-term indirect effect is significantly negative. In the long run, the direct effect remains insignificant, while the indirect effect continues to be significantly negative, implying that increases in transport intensity may exert a negative spillover effect on neighboring regions’ green productivity through interregional industrial linkages. Moreover, this adverse spillover appears to weaken over time.

Finally, both the short- and long-term direct and indirect effects of ER are significantly negative at the 1% level. The magnitude of the short-term indirect effect is notably larger than that of the direct effect, and the same pattern holds in the long term. This result may reflect that, at the current stage, China’s transportation sector is still in the early phase of green transition, where regulatory policies are more associated with rising compliance costs rather than generating positive innovation incentives.

On one hand, the dominance of command-and-control regulatory tools, often lacking adequate supportive incentives in some regions, has led enterprises to prioritize cost containment over proactive innovation, thus limiting the effectiveness of environmental regulation in driving efficiency gains [46]. On the other hand, the spatial decomposition results reveal significant negative spillover effects, suggesting that, in some highly regulated regions, polluting industries may relocate to surrounding areas, thereby undermining the overall regional green efficiency [47]. Additionally, the relatively limited absorptive capacity for green technologies and weak policy adaptability in resource-dependent regions such as Central, Western, and Northeastern China may further exacerbate the “compliance cost suppression” effect of environmental regulation.

3.4.3. Robustness Checks

To confirm that the empirical conclusions are not dependent on a particular specification, two robustness checks are designed. The initial test reconsiders the spatial weight matrix, exploring whether variations in spatial structure affect the results. For this purpose, the model is recalculated with two alternative settings: an economic–geographic nested weight matrix and an inverse-square distance matrix derived from provincial coordinates, replacing the original distance matrix.

Second, we test the robustness by adjusting the sample size. In particular, we exclude the data for the years 2007 and 2008 and re-estimate the model. The corresponding results are reported in

Table 12. Robustness analysis of ESTI and EITI.

Method	Var	Short Term (S)			Long Term (L)
		D	I	T	D	I	T
Economic-Geographic	ESTI	0.195 ***	0.771 **	0.966 **	0.163 ***	0.288 *	0.450 ***
	EITI	−0.239 ***	−1.732 ***	−1.972 ***	−0.150	−0.776 ***	0.926 ***
Inverse Square Distance	ESTI	0.226 ***	0.061	0.287 **	0.225 ***	0.042	0.268 **
	EITI	−0.259 ***	−0.653 ***	−0.912 ***	−0.251 ***	−0.601 ***	−0.852 ***
Sample Adjustment	ESTI	0.303 ***	1.405 ***	1.708 ***	0.240 ***	0.300 **	0.540 ***
	EITI	−0.078	−0.994 *	−1.072 **	−0.000	−0.342 *	−0.342 **

Note: S = short-term effects; L = long-term effects; D = direct effect; I = indirect effect; T = total effect. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

.

The comparison shows that the direction, magnitude, and significance of the direct, indirect, and total effects of the explanatory variables remain largely consistent across alternative specifications. Although the statistical significance of some variables changes slightly, the sign of the coefficients remains stable. These findings remain consistent under alternative spatial weight matrices and sample periods.

3.4.4. Heterogeneity Analysis

This section further investigates whether the effects of the energy structure and energy intensity in the transportation sector on GTFP exhibit regional heterogeneity.

According to the decomposition results presented in

Table 13. Regional heterogeneity in the short-term and long-term effects of ESTI and EITI.

Region	Var	Short Term (S)			Long Term (L)
		D	I	T	D	I	T
Eastern	ESTI	−0.047	0.736 ***	0.690 ***	−1.564	1.891	0.327 ***
	EITI	−0.704 ***	0.057	−0.647 **	−2.683	2.374	−0.309 **
Central	ESTI	−0.165	0.898 ***	0.733 **	−0.569 **	1.038 ***	0.470 **
	EITI	−0.482 **	$1.171$	$0.689$	−1.142 **	1.581 **	0.439
Western	ESTI	−0.156	−1.045 *	−1.201 *	−0.682	0.285	−0.398 *
	EITI	−0.307 *	−1.993 *	−2.300 **	−0.733	−0.037	−0.770 **
Northeast	ESTI	−4.975 ***	−5.734 ***	−10.708 ***	−1.872 ***	−2.034 ***	−3.906 ***
	EITI	−5.497 ***	8.593 ***	3.097 ***	−2.347 ***	3.476 ***	1.130 ***

Note: S = short-term effects; L = long-term effects; D = direct effect; I = indirect effect; T = total effect. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

, the positive impact of the energy structure in the transportation sector (ESTI) is most prominent in the eastern and central regions. In the east, the short-term indirect effect (0.736) substantially exceeds the small negative direct effect ($-0.047$), and the long-term total effect remains positive at 0.327. In the central region, the short-term indirect effect (0.898) is significant at the 1% level, and both the long-term direct and indirect effects are statistically significant, with a total long-term effect of 0.470. These results suggest that, in these two regions, the penetration of clean energy significantly enhances GTFP through persistent spatial spillovers.

In contrast, the western region exhibits significantly negative total effects of ESTI in both the short and long term, indicating that, under conditions of resource constraints and lagging industrial transformation, the benefits of energy structure optimization may be offset. The northeast displays an even stronger inhibitory effect: the short- and long-term total effects of ESTI are $-10.708$ and $-3.906$, respectively, both significant at the 5% level. This reflects the limited spillover benefits of clean energy transformation under the outdated industrial structure of the region.

As for the energy intensity (EITI), the eastern region shows significantly negative short-term direct and total effects, with the long-term total effect also remaining negative. This implies that improvements in energy efficiency steadily promote both local and neighboring green productivity. In the central region, only the short-term direct effect is negative, suggesting that energy efficiency improvements may require longer time lags or stronger policy support to influence GTFP.

In the western and northeastern regions, the overall effects are mixed. In the west, both the short- and long-term total effects are significantly negative, suggesting that improvements in energy efficiency have not translated into productivity gains, possibly due to structural and institutional barriers. In the northeast, both the short- and long-term total effects of EITI are positive and statistically significant. However, given the high energy consumption base of the region, this may be driven by short-term output expansion during a “pollute first, clean up later” style of industrial upgrading.

In summary, energy structure optimization and energy efficiency improvements exhibit robust positive effects on GTFP in the eastern and central regions. By contrast, the western and northeastern regions show clear heterogeneity, underscoring the importance of tailoring low-carbon and energy policies to the regional conditions.

4. Conclusions and Recommendations

4.1. Conclusions

This study employs a multi-method framework to capture the spatial evolution of GTFP in China’s transportation sector and arrives at the following conclusions.

(1) China’s transportation GTFP exhibits pronounced regional heterogeneity and dynamic spatiotemporal evolution, accompanied by strong spatial clustering and “club convergence” effects. During the study period, national GTFP fluctuations intensified, and interregional efficiency disparities expanded significantly, forming a gradient pattern of “high in the east, low in the west, and lagging in the northeast”. The kernel density results reveal an evolutionary trajectory from widespread improvement to increasing polarization, followed by brief convergence and structural solidification, reflecting the long-term and structural nature of regional divergence. Low-efficiency provinces in the west face a risk of “low-level lock-in”. The Moran’s I index and scatter plots further confirm the significant and intensifying spatial autocorrelation of GTFP, characterized by “high–high” clusters in the eastern coastal regions and “low–low” clusters in the western and northeastern regions. Both traditional and spatial Markov chain analyses verify the path dependence and transition inertia of green efficiency evolution. Neighboring provinces’ GTFP significantly influences local trajectories—proximity to efficient regions encourages upward transitions, whereas low-efficiency clusters may hinder progress.

(2) The optimization of the energy structure in the transportation sector has a significant positive impact on GTFP. In the short term, it exhibits strong spatial spillover effects—improvements in the local energy structure not only enhance local green productivity but also positively influence surrounding areas through spatial transmission mechanisms. However, the long-term spillover effects weaken over time, suggesting diminishing marginal returns in regional synergy as energy structure optimization progresses.

(3) Reducing the energy intensity, i.e., improving energy efficiency, significantly promotes GTFP and similarly demonstrates strong short-term spatial spillover effects. Although the magnitude of the long-term spillovers declines, energy efficiency improvements continue to contribute meaningfully to overall green productivity gains. The heterogeneity analysis reveals that these positive effects are most prominent in Eastern and Central China, whereas the western and northeastern regions display instability and potential risks of negative spillovers, underscoring the need for regionally tailored energy and low-carbon policies.

4.2. Policy Recommendations

From the findings, several policy implications can be derived.

Foremost, region-specific approaches to adjusting the energy structure are needed. For the eastern and central areas, priority should be given to accelerating renewable energy adoption, advancing transport electrification and intelligent transformation, and establishing interregional coordination systems to reinforce positive spatial spillovers. In the western region, greater emphasis should be placed on resolving the tension between delayed industrial transformation and resource constraints by fostering integration between emerging renewable sectors and traditional industries to mitigate potential negative spillovers. In the northeast, efforts should be made to accelerate the upgrading of outdated industrial structures and support the deployment of clean energy technologies and demonstration projects to address structural inefficiencies.

Second, it is necessary to improve transportation’s energy efficiency by accelerating the technological upgrading of transport equipment and systems. All regions should strengthen the enforcement of energy efficiency standards, eliminate outdated and highly energy-consuming transportation modes, and promote the adoption of energy-efficient alternatives. In particular, the western and northeastern regions should receive increased technical and financial support to drive the green transformation of traditional transport industries and mitigate negative spatial spillovers from energy efficiency improvements.

Third, it is necessary to enhance cross-regional coordination and policy integration. Provinces should work to overcome administrative fragmentation by establishing institutional mechanisms for policy coordination, facilitating the sharing of information, technologies, and resources across regions. Given the identified spatial spillover effects of the energy structure and efficiency, tailored regional cooperation policies should be designed to promote the interprovincial connectivity of energy infrastructure and to support the integration and optimization of low-carbon industrial chains.

Finally, it is recommended to establish long-term incentive and regulatory mechanisms. Governments should construct a green development system that balances incentives and constraints, accelerate the development of a market-oriented carbon emissions trading system, and expand green finance instruments to encourage enterprise-led green technology innovation and adoption. Moreover, long-term regulatory frameworks for environmental governance should be strengthened to avoid short-term negative spillover effects of environmental regulations and enhance the stability and effectiveness of policy implementation over time.

5. Discussion

5.1. Limitations

This study has several limitations that merit discussion.

First, the temporal scope of the analysis is limited to the period from 2006 to 2020. While data beyond 2020 are available in some domains, the COVID-19 pandemic introduced significant volatility and structural disruptions in energy consumption and transportation systems, making post-2020 data less representative of long-term regular patterns. By focusing on the pre-pandemic period, this study aims to capture stable and generalizable mechanisms under normal conditions.

Second, although the study defines ESTI as the proportion of electricity in the total energy consumption, it does not differentiate among electricity derived from renewable versus non-renewable sources. Ideally, a more accurate proxy for the greenness of the energy structure would consider only renewable electricity consumption. Provincial-level data on renewable electricity use remain unavailable, limiting the measurement accuracy. In contrast, the aggregate electricity consumption in the transport sector is available with consistent coverage and comparability, making it a practical and widely used proxy in empirical studies.

Third, while the conceptual framework (as illustrated in Figure 1) outlines several potential internal mechanisms—such as technological diffusion, cost channels, and industrial agglomeration—that may mediate the relationship between the energy structure and green productivity, these mediating pathways are not directly tested within the empirical model. The study focuses primarily on spatial effects and overall structural relationships, leaving detailed internal mechanisms for future exploration.

Fourth, due to data limitations, this study does not employ exogenous instrumental variable methods for deeper causal identification. Future research could improve the identification strategy by incorporating specific policy experiment data.

5.2. Future Research Directions

Future studies can be enhanced in three directions.

First, future studies could incorporate post-2020 data to explore the phase-specific dynamics of GTFP in the transportation sector and its influencing mechanisms, particularly in the context of the COVID-19 shock and subsequent recovery.

Second, as the data availability improves, more accurate indicators reflecting the degree of energy greening—such as the share of renewable electricity and renewable energy integration—may be introduced to enhance the precision and theoretical alignment of energy structure measurement.

Finally, methodological advancements such as structural equation modeling (SEM), mediation effect regression, and machine learning techniques can be employed to quantitatively identify the transmission mechanisms proposed in Figure 1—including cost constraints, technological diffusion, and spatial spillovers—thus providing improved theoretical insights and informing policy on the energy structure’s role in green efficiency.