Spatiotemporal Evolution and Influencing Factors of Carbon Emission Efficiency in China’s Resource-Based Cities Based on Super-Efficiency SBM-GML Measurement and Spatial Econometric Tests

Abstract

To advance global climate governance, this study investigates the carbon emission efficiency (CEE) of 110 Chinese resource-based cities (RBCs) using a super-efficiency SBM-GML model combined with kernel density estimation and spatial analysis (2006–2022). Spatial Durbin model (SDM) and geographically and temporally weighted regression (GTWR) further elucidate the driving mechanisms. The results show that (1) RBCs achieved modest CEE growth (3.8% annual average), driven primarily by regenerative cities (4.8% growth). Regional disparities persisted due to decoupling between technological efficiency and technological progress, causing fluctuating growth rates; (2) CEE exhibited high-value clustering in the northeastern and eastern regions, contrasting with low-value continuity in the central and western areas. Regional convergence emerged through technology diffusion, narrowing spatial disparities; (3) energy intensity and government intervention directly hinder CEE improvement, while rigid industrial structures and expanded production cause negative spatial spillovers, increasing regional carbon lock-in risks. Conversely, trade openness and innovation level promote cross-regional emission reductions; (4) the influencing factors exhibit strong spatiotemporal heterogeneity, with varying magnitudes and directions across regions and development stages. The findings provide a spatial governance framework to facilitate improvements in CEE in RBCs, emphasizing industrial structure optimization, inter-regional technological alliances, and policy coordination to accelerate low-carbon transitions.

1. Introduction

As global climate governance intensifies, low-carbon development has emerged as a critical pathway for achieving the Paris Agreement’s temperature targets [1]. The IPCC Sixth Assessment Report underscores the urgency of reducing global carbon emissions by 43% by 2030 to limit warming to 1.5 °C, a scientific imperative that has redefined international climate action as a “responsibility reallocation” challenge. Over 130 countries and regions have announced carbon neutrality targets by 2023 [2]. As the world’s largest developing country, China accounted for 31.8% of global carbon emissions in 2023, making its actions crucial to global climate progress [3]. In response to the climate crisis, China has incorporated carbon reduction into its national sustainable development strategy framework and announced its “dual carbon” goals in 2020: peak carbon emissions by 2030 and achieving carbon neutrality by 2060 [4].

Resource-based cities (RBCs) play a pivotal role in both energy production and consumption, making them key nodes in China’s energy security and high-carbon lock-in strategies [5]. There are 262 RBCs in China, accounting for 40% of all cities nationwide; yet, their carbon intensity per unit of GDP is 1.8 times the national average [6]. From 2000 to 2020, total carbon emissions in these cities grew at an average annual rate of 4.5%, far exceeding the national rate of 2.8% [7], while about 68% of RBCs have experienced slower economic growth than the national average [8]. Although resource extraction contributes between 32% and 67% of local fiscal revenues, this economic reliance often comes at a cost [9]. It is often accompanied by a narrow industrial structure, ecological degradation, and intensified social tensions. In some cases, RBCs have faced systemic decline following resource depletion, and “resource exhaustion–urban recession” poses direct challenges to sustainable and high-quality urban development [10,11].

At the critical stage of low-carbon transformation, RBCs must find a balance between economic growth and carbon emissions reduction. Carbon emission efficiency (CEE), defined as the ratio of actual to optimal output per unit of CO₂ emissions, integrates factors such as economic scale, population density, and resource endowment, revealing the trade-off between growth and environmental impact. Unlike total emission metrics [12], CEE incorporates CO₂ emissions as an undesirable output within a production efficiency framework, offering a more nuanced measure of progress toward carbon neutrality and urban sustainability. In addition, RBCs exhibit significant CEE differentiation due to geographic and developmental differences [13]. These variations contribute to differing spatiotemporal patterns of CEE, including generally higher levels in eastern regions [14], localized spatial clustering [15], and distinct regional gradients [16,17]. Moreover, findings differ on whether these regional disparities are widening or narrowing over time [18,19]. Thus, addressing the imbalance in CEE has become an inherent requirement for RBCs to achieve low-carbon development.

The previous studies on CEE mainly involved efficiency measurement [20,21], spatiotemporal evolution patterns [22], and driving factor analysis [23,24]. Regarding the measurement of CEE, early studies mostly used single-factor analyses (e.g., carbon emission intensity or energy input) to estimate emissions per unit of GDP or energy consumption [25,26]. Subsequent studies have shifted toward total-factor frameworks, including stochastic frontier analysis (SFA) [27] and data envelopment analysis (DEA) [28], which integrate multidimensional inputs (e.g., capital, labor, energy) to construct comprehensive evaluation systems [16,29]. Chung et al. [30] further incorporated the directional distance function (DDF) into DEA models, such as the CCR and BCC models, enabling the simultaneous improvement of desirable outputs and reduction in undesirable outputs. However, these models are essentially radial or angular in nature and tend to overlook the impact of input or output slack variables on actual efficiency, which may lead to overestimated efficiency scores. To overcome this limitation, the super-efficiency SBM model containing undesirable outputs has been widely applied because it effectively addresses the radial and angular biases of traditional DEA models and overcomes measurement limitations caused by insufficient decision-making units [31].

In the analysis of the spatiotemporal evolution of CEE, studies typically explore different spatial scales, including national, regional, and provincial levels, aiming to reveal the spatial aggregation and disparity of CEE [21,32]. Commonly used methods include the following. Regional inequality is examined using the Dagum Gini coefficient [33]. To further capture dynamic transition patterns and long-term trends, Markov chains are commonly used to analyze the evolution of regional CEE distributions over time [15]. To avoid the subjectivity of regional classification in the Markov approach, kernel density estimation combined with Moran’s I is employed to examine spatial and temporal variations in CEE in RBCs [34]. In addition, geographic information systems (GIS) are used to enable intuitive spatiotemporal visualization [35].

To further explore the mechanisms influencing CEE, many studies have investigated diverse factors, such as technological progress, economic development, and low-carbon policies [36,37]. Various methodologies have been applied in these studies, including the least squares (OLS) method [38], factor decomposition [39], and spatial econometrics [40]. For instance, the DID model is used to analyze policy effects on CEE in China’s RBCs [41]. Regression models, including censored and limited dependent variable approaches, are often applied to examine how various factors affect CEE [42]. Compared with conventional panel models, the spatial Durbin model (SDM) and the geographically and temporally weighted regression (GTWR) model can effectively capture the spatial spillover effects and spatial heterogeneity of factors influencing the CEE of RBCs [17], thereby addressing the limitations of traditional methods in capturing spatial dependence mechanisms. For example, SDM-based studies have shown that economic, demographic, and social factors can generate significant negative spatial spillover effects on CEE [40,43]. Furthermore, the impacts of these drivers have been found to differ substantially across cities and regions [44,45].

Although CEE has been widely studied, several important aspects remain underexplored. First, while regional disparities in urban CEE have been documented, the role of RBCs in global climate governance has attracted relatively limited attention, and distinctions among different RBC types are seldom examined in depth. Second, most existing studies rely on static efficiency measurements, which may overlook dynamic features such as technical catch-up effects reflected in both radial and angular changes. Third, many spatiotemporal analyses focus either on the temporal evolution or the spatial distribution of CEE in RBCs, but they rarely consider how these two dimensions interact. Moreover, most studies examine the impacts of internal factors on a city’s own CEE while largely ignoring the potential influence of neighboring cities. In reality, geographic proximity, environmental externalities, and regional policy diffusion can significantly affect local CEE outcomes. However, the traditional econometric models commonly used to explore influencing factors often fail to capture such spatial effects, resulting in an incomplete understanding of interregional dynamics.

To address these issues, this study makes two key contributions. First, it provides a comprehensive assessment of CEE across different types of Chinese RBCs by integrating static and dynamic perspectives to capture their spatiotemporal evolution, offering a nuanced and type-specific understanding of development trajectories. Second, it innovatively integrates the SDM with the GTWR model, addressing the limitations of traditional spatial analyses and revealing localized and differentiated drivers of CEE across city types. Specifically, this study applies a super-efficient SBM-GML approach to assess the CEE of Chinese RBCs from 2006 to 2022. Static efficiency is measured using the super-efficient SBM model, while dynamic productivity changes are captured through the global Malmquist–Luenberger (GML) index. This combination provides a more accurate depiction of CEE dynamics in RBCs. Spatiotemporal dynamics are analyzed using kernel density estimation, standard deviational ellipse, and Moran’s index to quantify temporal trends, spatial agglomeration, and autocorrelation, revealing systemic CEE imbalances. Finally, SDM and GTWR disentangle local effects from cross-regional spillovers while capturing spatiotemporal heterogeneity in drivers. This approach offers more focused theoretical guidance and a solid foundation for decision making in the low-carbon transformation of RBCs.

2. Materials and Methods

2.1. Study Area

China’s National Sustainable Development Plan for Resource-Based Cities (2013–2020) explicitly identifies 262 RBCs [46]. Among them, 126 are prefecture-level administrative units. This study focuses on 110 prefecture-level RBCs (Figure 1), comprising 14 growing cities, 58 mature cities, 23 declining cities, and 15 regenerative cities.

2.2. Research Framework

This study proposes a three-phase analytical framework (efficiency assessment, spatiotemporal evolution, and influencing factors) to systematically examine the multidimensional characteristics of CEE in RBCs, as illustrated in Figure 2.

2.3. CEE Measurement

2.3.1. Super-SBM Model for Measuring Static CEE

To address the limitations of traditional DEA in handling input/output slack and measurement errors, this study employs the super-SBM model with undesirable outputs [47]. This non-radial, non-oriented approach measures CEE by allowing efficiency scores above 1, thereby effectively differentiating high-performance decision-making units. Building on previous studies [11,29], this study applies the constant returns to scale (CRS)-based super-SBM model to evaluate the CEE of RBCs. The CRS assumption ensures that efficiency comparisons are made under the same scale conditions, providing a more consistent and clearer assessment of overall technical efficiency. Additionally, this avoids the variable returns to scale (VRS) framework, which may lead to infeasible solutions in super-efficiency models. A comparison of the results from both approaches finds that CRS better ensures the robustness and applicability of the evaluation. The model can be expressed as

$${\rho }^{\ast }=\min {\displaystyle \frac{1+\frac{1}{m}{\displaystyle {\sum }_{i=1}^m\frac{s_i^{-}}{x_{io}}}}{1-\frac{1}{s_1+s_2}\left({\displaystyle {\sum }_{r=1}^{s_1}\frac{s_r^{+}}{y_{ro}}+{\displaystyle {\sum }_{k=1}^{s_2}\frac{s_k^{-}}{z_{ko}}}}\right)}}$$

(1)

$$s.t.\left\{\begin{array}{l}{x}_{io}\ge {\displaystyle {\sum }_{j=1,j\ne o}^n{\lambda }_j{x}_{ij}-s_i^{-},i=1,2,\dots ,m}\\ y_{ro}^g\le {\displaystyle {\sum }_{j=1,j\ne o}^n{\lambda }_j{y}_{rj}+s_r^{+},r=1,2,\dots ,s_1}\\ z_{ko}^b\ge {\displaystyle {\sum }_{j=1,j\ne o}^n{\lambda }_j{z}_{kj}-s_k^{-},k=1,2,\dots ,s_2}\\ s_i^{-}\ge 0,s_r^{+}\ge 0,s_k^{-}\ge 0,{\lambda }_j\ge 0,j=1,2,\dots ,n\end{array}\right.$$

(2)

where x, y^g, z^b denote the input, desired output, and non-desired output variables, respectively; $s_i^{-}, s_r^{+}, s_k^{-}$ denote the slack variables for the ith input, rth desired output, and kth non-desired output, respectively; m, s₁, s₂ denote the number of input, desired output, and non-desired output indicators; j denotes the index for traversing all decision-making units (DMUs); o denotes the specific DMU targeted for efficiency evaluation; λ_i denotes the indicator weight; ρ* denotes the value of super efficiency.

2.3.2. Selection of CEE Indicators

The CEE evaluation system integrates input–output indicators tailored to RBCs, as summarized in

Table 1. Indicator system for measuring CEE.

Indicator Type	Specific Indicators	Indicator Characterization
Input	Labor	Refers to the total number of individuals employed in the city during the year.
	Capital	Expressed as capital stock. Equation: $S_{i,t}=S_{i,t-1}\left(1-{\delta }_{i,t}\right)+I_{i,t}$ where Si,t denotes the capital stock of city i in year t (CNY billion); δi,t denotes the depreciation rate, set at 9.6% [49]; Ii,t denotes the capital investment (CNY billion).
	Energy	Total energy consumption. Using the city’s calendar year natural gas consumption, liquefied petroleum gas consumption, the city’s annual electricity consumption, and the amount of raw coal needed for city heating, converted to standard coal.
Desirable output	Economic benefits	Expressed as city GDP. To overcome the effect of price factors, GDP is calculated at constant 2006 prices [50].
Undesirable output	Carbon emissions	Expressed in terms of urban carbon emissions; the IPCC accounting method was used [8]. Equation: $C{E}_i={\displaystyle {\sum }_{j=1}^n\left(E_{i,j}\times F_j\times {\omega }_j\right)}$ where CEi denotes the carbon emissions of city i; Ei,j denotes the consumption of energy jth in city i; Fj denotes the standard coal conversion factor of energy j; ωj denotes the carbon emission factor of energy j.

. The input indicators include labor, capital, and energy consumption, informed by RBCs’ developmental contexts and prior studies [16,48]. The outputs are categorized into desirable outputs and undesirable outputs [30].

2.3.3. GML Index for Measuring Dynamic CEE

To measure dynamic changes in CEE, this study employs the GML index, which overcomes the Malmquist index’s inability to account for undesirable outputs. The GML index, grounded in the global directional distance function, offers advantages over other decomposition models by enabling cyclical cumulativeness and cross-period comparability [51]. Its formulation derives from the global production possibility set P^G(x). According to Meng et al. [52], $P^G(x)=P^1(X^1)\cup P^2(X^2)\cup \dots \cup P^T(X^T)$ is defined as a global set containing all the contemporaneous benchmark technologies. And its mathematical expression is given by the following equation:

$$S^G\left(x,y,z\right)=\max \left\{\beta \left|\left(y+\beta y,z-\beta z\right)\right.\in P_G\left(x\right)\right\}$$

(3)

where S^G(x, y, z) denotes the directional distance function; x, y, z denote the input variables, desirable outputs, and undesirable outputs of each decision unit, respectively; β denotes the value of the distance function, which maximizes the desired output and minimizes the non-desired output.

Further, the GML index is the relative change in CEE from period t + 1 to period t, and it can be decomposed into two components, the combined technical efficiency change (GEC) and the technical progress change (GTC), as follows:

$$\begin{array}{l}GM{L}_k^{t,t+1}={\displaystyle \frac{1+S^G\left(x_k^t,y_k^t,z_k^t\right)}{1+S^G\left(x_k^{t+1},y_k^{t+1},z_k^{t+1}\right)}}\\ ={\displaystyle \frac{1+S^t\left(x_k^t,y_k^t,z_k^t\right)}{1+S^{t+1}\left(x_k^{t+1},y_k^{t+1},z_k^{t+1}\right)}}\times \left[{\displaystyle \frac{1+S^G\left(x_k^t,y_k^t,z_k^t\right)}{1+S^t\left(x_k^t,y_k^t,z_k^t\right)}}\times {\displaystyle \frac{1+S^{t+1}\left(x_k^{t+1},y_k^{t+1},z_k^{t+1}\right)}{1+S^G\left(x_k^{t+1},y_k^{t+1},z_k^{t+1}\right)}}\right]\\ =GE{C}_k^{t,t+1}\ast GT{C}_k^{t,t+1}\end{array}$$

(4)

where $GM{L}_k^{t,t+1}$ denotes the change in total factor efficiency in the two phases of DMU_k; S^G, S^t, S^t+1 denote the directional distance functions based on the global frontier, t, t + 1 period, respectively. According to the literature [52], $GM{L}_k^{t,t+1}=1$ denotes no change in CEE; $GM{L}_k^{t,t+1}>1$ denotes an increase in CEE, and vice versa. $GE{C}_k^{t,t+1}=1$ and $GT{C}_k^{t,t+1}=1$ denote no change in technical efficiency and technological progress; $GE{C}_k^{t,t+1}>1$ and $GT{C}_k^{t,t+1}>1$ denote an increase in technical efficiency and technological progress, and vice versa.

2.4. Modeling of Spatial and Temporal Evolution

2.4.1. Kernel Density Function

Kernel density estimation transforms discrete spatial data into continuous surfaces, facilitating the visualization of agglomeration patterns [21]. Applied to CEE in RBCs, this method reveals temporal evolution trends through its kernel density function:

$$f\left(x\right)={\displaystyle \frac{1}{Nh}}{\displaystyle \sum _{i=1}^{N}K\left(\frac{{\alpha }_i-\overline{\alpha }}{h}\right)}$$

(5)

where N denotes the number of observations; h denotes the width; α_i denotes independently and identically distributed observations; $\overline{\alpha }$ denotes the mean of the observations; K(·) denotes the kernel density function.

2.4.2. Standard Deviational Ellipse

The standard deviational ellipse quantifies spatial patterns of geospatial data by converting discrete point distributions into ellipse metrics that reveal central tendency, directional bias, and dispersion patterns [53]. The analysis process usually includes the elements of distribution center of gravity, long-axis standard deviation, short-axis standard deviation, and azimuth angle, which can be used to determine the distribution center of gravity, the main direction of distribution, and the evolution trend of the CEE space of RBCs. This analysis was performed in ArcGIS 10.8 with the following calculation formulae:

$$\overline{X}={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n{w}_i{x}_i}}{{\displaystyle {\sum }_{i=1}^n{w}_i}}},\overline{Y}={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n{w}_i{y}_i}}{{\displaystyle {\sum }_{i=1}^n{w}_i}}}$$

(6)

where $(\overline{X},\overline{Y})$ denote the latitude and longitude coordinates of the weighted centroid of the ellipse, and w_i denotes the weight at spatial element i.

$${\tilde{x}}_i=x_i-\overline{X},{\tilde{y}}_i=y_i-\overline{Y}$$

(7)

where ${\tilde{x}}_i,{\tilde{y}}_i$ denote the coordinate deviations from the spatial centroid for each city.

$$\tan \theta ={\displaystyle \frac{({\displaystyle {\sum }_{i=1}^m{w}_i^2{\tilde{x}}_i^2}-{\displaystyle {\sum }_{i=1}^m{w}_i^2}{\tilde{y}}_i^2)+\sqrt{{({\displaystyle {\sum }_{i=1}^m{w}_i^2{\tilde{x}}_i^2}-{\displaystyle {\sum }_{i=1}^m{w}_i^2}{\tilde{y}}_i^2)}^2+4{\displaystyle {\sum }_{i=1}^m{w}_i^2{\tilde{x}}_i{\tilde{y}}_i}}}{{\displaystyle {\sum }_{i=1}^{m}2w_i^2{\tilde{x}}_i{\tilde{y}}_i}}}$$

(8)

where θ denotes the azimuth angle.

$$\left\{\begin{array}{l}{\sigma }_x=\sqrt{{(2{\displaystyle {\sum }_{i=1}^m(q_i{\tilde{x}}_i\cos \theta }-q_i{\tilde{y}}_i\sin \theta )}^2)/({\displaystyle {\sum }_{i=1}^m{q}_i^2)}},\\ {\sigma }_y=\sqrt{{(2{\displaystyle {\sum }_{i=1}^m(q_i{\tilde{x}}_i\sin \theta }+q_i{\tilde{y}}_i\cos \theta )}^2)/({\displaystyle {\sum }_{i=1}^m{q}_i^2)}}\end{array}\right.$$

(9)

where σ_x and σ_y denote the standard deviations along the X and Y axes.

$$S=\pi {\sigma }_x{\sigma }_y$$

(10)

where S denotes the area of the ellipse.

2.4.3. Moran’s Index

This study employs spatial autocorrelation analysis to examine the spatial interdependencies and disparities in CEE across RBCs [54]. The global Moran’s I quantifies the overall spatial association and its statistical significance. With values of I > 0, there is a positive correlation between regions; with values of I < 0, there is a negative correlation; when I = 0, there is no correlation. The calculations are as follows:

$$I={\displaystyle \frac{n{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}\left(CE{E}_i-\overline{CEE}\right)\left(CE{E}_j-\overline{CEE}\right)}}}{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}{\displaystyle \sum _{i=1}^n{\left(CE{E}_i-\overline{CEE}\right)}^2}}}}}$$

(11)

where $\overline{CEE}$ denotes the sample mean; n denotes the number of RBCs; W_ij denotes an element in the spatial weighting matrix.

Furthermore, local Moran’s I examines spatial autocorrelation at finer scales, revealing localized clustering or dispersion patterns and interactions among neighboring regions. It is thus employed to capture the agglomeration and diffusion characteristics of CEE in RBCs at a more refined spatial scale. The formula is defined as

$$I_i={\displaystyle \frac{n\left(CE{E}_i-\overline{CEE}\right)}{{\displaystyle \sum _{i=1}^n{\left(CE{E}_j-\overline{CEE}\right)}^2}}}{\displaystyle \sum _{j=1}^n{W}_{ij}\left(CE{E}_j-\overline{CEE}\right)}$$

(12)

2.5. Analysis of Influencing Factors

2.5.1. The Spatial Durbin Model

To explore the spatial dependence in CEE drivers, this study employs spatial econometric modeling [54]. The spatial error model (SEM), spatial lag model (SLM), and SDM are the three mainstream spatial econometric models at present. In this study, through a series of statistical tests, such as Hausman, LM, and LR, the findings indicate that the SDM with double fixed effects best fits the data. The SDM simultaneously captures temporal and spatial spillovers and is formulated as

$$CEE=\mu {\displaystyle \sum _{j=1}^n{W}_{ij}\cdot CEE+\phi \ln x_{ij}+\psi \cdot {\displaystyle \sum _{j=1}^n{W}_{ij}\cdot \ln x_{ij}+{\tau }_i}}+{\gamma }_t+u_{it}$$

(13)

where subscripts i and j both denote cities; subscript t denotes the year; μ denotes the spatial lag coefficient; φ and ψ denote the parameters to be estimated; x_ij denotes a series of explanatory variables, including the influences of energy intensity, economic level, and governmental interventions; τ_i and γ_t denote the spatial and temporal fixed effects; u_it denotes the normally distributed error term; W_ij denotes the spatiotemporal weight matrix.

2.5.2. The Geographically and Temporally Weighted Regression Model

While the SDM captures global spatial effects, it fails to account for localized spatiotemporal heterogeneity. To address this limitation, GTWR extends traditional GWR by incorporating temporal dynamics. By integrating both spatial and temporal non-stationarity into regression coefficients, GTWR enables simultaneous analysis of evolving spatial patterns and time-dependent variations. This approach rigorously quantifies the mechanisms driving spatiotemporal disparities in geographical phenomena, formulated as

$$CE{E}_i={\eta }_0\left(z_i,v_i,t_i\right)+{\displaystyle \sum _{k=1}^P{\eta }_k\left(z_i,v_i,t_i\right)}\cdot \ln X_{ik}+{\epsilon }_i$$

(14)

where CEE_i denotes the dependent variable for observation i; η₀(z_i, v_i, t_i) denotes the intercept term at the ith point as a function of the geographic coordinates (u_i, v_i) and time t_i; η_k(z_i, v_i, t_i) denotes the coefficient of the kth explanatory variable at observation i; lnX_ik denotes the value of the kth explanatory variable for observation i; ε_i denotes the random error term; P denotes the total count of explanatory variables.

2.5.3. Selection of Influencing Factors

To comprehensively identify the determinants of CEE in RBCs, this study selects seven representative variables: energy intensity (X1), industrial structure (X2), government intervention (X3), economic level (X4), trade openness (X5), innovation level (X6), and production scale (X7), as detailed in

Table 2. Influencing factors of CEE in China’s RBCs.

Factor	Number	Meaning
Energy intensity	X1	Total energy consumption per unit of regional GDP
Industrial structure	X2	Share of secondary industry value added in regional GDP
Government intervention	X3	Local government expenditure as a share of the general budget
Economic level	X4	Per capita gross regional product
Trade openness	X5	Ratio of total trade to regional GDP
Innovation level	X6	Share of regional GDP allocated to R&D and innovation
Production scale	X7	Share of fixed-asset investment in regional GDP

. Meanwhile, we find that the VIF test results are all below 5, indicating no serious multicollinearity issues.

X1 reflects RBCs’ long-standing reliance on fossil fuels and the low efficiency of their energy conversion, both of which contribute to increased carbon emissions. X2 in RBCs is dominated by energy-intensive sectors, although technological progress may help mitigate their impact [48]. X3 influences CEE depending on fiscal priorities: support for clean energy and ecological infrastructure boosts efficiency, while subsidies for high-emission sectors may worsen carbon lock-in. X4 follows the environmental Kuznets curve, with early growth stages typically increasing emissions and later stages driving efficiency through structural shifts and innovation. X5 can facilitate low-carbon technology transfer but may also increase emissions by attracting pollution-intensive industries [55]. X6 helps improve efficiency but may also cause rebound effects when increased productivity leads to higher energy demand. X7 supports CEE when focused on green infrastructure and modernization, but unchecked growth in resource-heavy sectors often raises emissions and reduces efficiency [56]. Collectively, these variables offer a comprehensive framework for analyzing the drivers of CEE in RBCs.

2.6. Data Sources

Since CEE indicators are not uniformly available across all prefecture-level cities nationwide, excluding cities with incomplete data is a practice supported by previous literature works [29,35]. Based on the availability of CEE-related data from 2006 to 2022, we selected 110 out of the 126 officially designated prefecture-level RBCs as the sample for this study [46]. In terms of regional distribution, it comprises 19 cities from the northeastern region, 20 from the eastern region, 35 from the central region, and 36 from the western region. Primary data sources include the China Urban Statistical Yearbook, China Urban Construction Statistical Yearbook, and provincial or municipal statistical yearbooks, supplemented by municipal statistical bulletins for fixed-asset investment records. Missing values were imputed using linear interpolation in accordance with established panel data imputation procedures [37].

3. Empirical Analysis

3.1. CEE of RBCs

3.1.1. Analysis of Static Efficiency Results

This study quantifies the CEE of 110 Chinese RBCs from 2006 to 2022 using the super-SBM model (Figure 3), based on results from Equations (1) and (2). Overall, 84% of RBCs have shown CEE improvements, indicating initial progress in national decarbonization efforts. The CEE trajectory of Chinese RBCs can be categorized into three distinct patterns. (1) Sustained-growth cities, primarily in energy-rich regions like the Bohai Rim and Inner Mongolia, demonstrate consistent CEE improvements, reflecting favorable coordination between economic development and resource utilization. (2) Fluctuating-growth cities exhibit high volatility in CEE, likely due to energy price fluctuations and insufficient coordination between policy and technology [50]. (3) Stagnant/regressive cities maintain low or declining CEE, hindered by structural issues, such as resource dependence, weak innovation, and limited capital accumulation [18]. As a result, these cities require urgent strategic reforms to overcome decarbonization barriers.

To further highlight the differences in CEE across RBCs at various stages of development, Figure 4 illustrates the time trends of CEE for each typology. All categories show upward trends, with regenerative cities leading at an impressive 4.8% annual growth rate. By 2014, their CEE surpassed that of mature cities and accelerated further after 2020, reflecting a shift away from resource dependence toward more sustainable development, positioning them as role models for other RBCs. Mature cities, with a steady annual growth rate of 3.6%, have seen their CEE plateau. Despite their industrial strength, these cities face constraints due to technological bottlenecks and path dependence [57], resulting in stable resource extraction but insufficient momentum for further gains. In declining cities, CEE steadily increased from 0.23 in 2006 to 0.43 in 2022, although the rate of improvement slowed significantly after 2018. This indicates that resource depletion and industrial transitions are becoming more pronounced, with current resource extraction increasingly unable to meet societal demands. Conversely, growing cities recorded an average annual CEE growth rate of 3.8% but exhibited significant fluctuations, likely driven by spikes in energy demand amid rapid economic expansion [42]. Despite the growth potential, these cities remain locked in resource-intensive development models, hindering effective energy and emissions management. These differences highlight the need for RBCs to adjust their low-carbon transformation strategies based on the unique development stages and structural conditions.

Given the spatial distribution disparities among RBCs and their influence on CEE, the trends of CEE across different regions are illustrated in Figure 5. From 2006 to 2022, national CEE steadily increased from 0.25 to 0.46, with an average annual growth rate of 3.8%, indicating the overall effectiveness of emission reduction policies. Moreover, the growth rate of CEE was higher before 2016 but slowed in subsequent years. Regionally, the northeastern and northern regions exhibit relatively higher CEE levels than the central and western regions. The northeast, with its robust industrial base and targeted policies, achieved a 3.9% annual growth rate, particularly accelerating during the “Revitalization of the Old Northeastern Industrial Bases” green policy [58]. The eastern region leads the country with a stable annual growth rate of 5.0%, driven by its strategic location along rivers, coasts, and borders, which facilitate economic activity and trade. In contrast, CEE improvements in the central and western regions lag behind the national average. The central region has the lowest CEE, attributable to insufficient economic and labor inputs that limit economic output [11]. Additionally, the high concentration of energy-intensive industries in this region drives elevated carbon emissions, further hindering CEE improvement. In the western region, despite the large number of RBCs and a vast land area, economic development lags due to reliance on outdated extraction technologies and inefficient resource utilization. These regional disparities underscore the need for more targeted policies to enhance CEE and promote a low-carbon transition across Chinese RBCs.

3.1.2. Analysis of Dynamic Efficiency Results

The dynamic CEE of RBCs from 2006 to 2022 is decomposed using the GML index, as shown in Figure 6, based on the calculations from Equation (4). Most RBCs have a GML index greater than 1 (Figure 6a), indicating continuous CEE improvements and increasing returns to scale. In contrast, cities with a GML index below 1 are concentrated in the central and western regions, reflecting regional disparities in efficiency gains. This spatial pattern is further supported by the distributions of GEC and GTC in Figure 6b,c. Specifically, eastern cities tend to have higher GEC values, reflecting notable improvements in technical efficiency, while coastal and economically advanced areas perform better in GTC, driven by technological progress. For example, Huaian reports a high CEE of 1.11, with a GEC of 1.07 and a GTC of 1.04, demonstrating the combined benefits of enhanced technical efficiency and progress. This synergy is supported by Huaian’s significant investment in industrial modernization and its location in the Yangtze River Delta [42], where favorable policies and active technology diffusion facilitate coordinated development. In contrast, western cities generally report lower GEC and GTC values. Zhangye, for instance, shows a lower CEE of 0.97, with GEC and GTC values of 0.94 and 0.97, both below 1. As a representative underdeveloped city [54], Zhangye faces structural and fiscal challenges that constrain industrial upgrading and access to innovation resources, thereby hindering technological progress and efficiency improvements.

At the national level (Figure 6d), the average GML index of CEE in RBCs is 1.04, indicating overall improvement. However, the trend shows a gradual decline in growth momentum over time. Notably, the GML index fell below 1 in 2017 and 2022, suggesting a temporary decrease in average CEE, likely linked to cyclical economic fluctuations [27]. Decomposition further reveals that the annual averages of the global GEC and GTC are both above 1, confirming their positive contributions to CEE improvements. However, GEC and GTC often move in opposite directions across most years, implying a lack of synergy between efficiency gains and technological progress. On the one hand, overreliance on either efficiency improvements or technological innovation causes the other to lag behind and become a constraint [11]. On the other hand, their effects may offset each other [51], thereby inhibiting overall efficiency gains. For instance, some RBCs may have achieved short-term efficiency gains through administrative interventions or resource optimization while lacking long-term technological innovation capacity. Conversely, others may have invested in new technologies that have yet to be fully absorbed into local production systems, leading to a disconnect between innovation and practical efficiency gains. Such inconsistencies have caused notable fluctuations in CEE evolution across RBCs.

3.2. Spatiotemporal Evolutionary Analysis

3.2.1. Time Series Analysis

This study examines the temporal evolution of CEE in RBCs using kernel density estimation (Figure 7), based on the method outlined in Equation (5). The height of the kernel density curve reflects the concentration of cities around the average CEE value, with a higher peak indicating that more cities are clustered around similar efficiency levels. In contrast, the width of the curve represents the dispersion of CEE, with a wider curve suggesting greater variation. The kernel density distribution of national CEE is illustrated in Figure 7a. The peak is skewed to the left, suggesting that many RBCs exhibit low efficiency. The distribution also shows a long right tail, indicating the presence of cities with relatively high CEE. The peak remains relatively stable over time, implying that no clear trend of polarization in CEE has emerged among RBCs.

The kernel density distribution of growing cities (Figure 7b) exhibits multiple peaks, indicating strong heterogeneity and dynamic shifts in CEE. The emergence of a bimodal structure and narrowing peak widths highlights increasing polarization, with some cities achieving efficiency gains while others face transition challenges. In mature cities (Figure 7c), a lower peak and broader width signal increased dispersion and decentralization. Over time, the rightward shift of the peak demonstrates steady progress in CEE, primarily driven by technological and industrial advances. Declining cities (Figure 7d) show the peak evolving from wide and low to sharp and high, gradually shifting leftward, reflecting an increasing concentration of cities trapped in low-efficiency states. These cities face mounting structural pressures [5], including limited reform incentives and rigid development models. In contrast, regenerative cities (Figure 7e) exhibit a high peak positioned to the right, indicating overall higher efficiency. The shift from a single peak to multiple peaks suggests the emergence of stratified efficiency tiers. Although the efficiency values within each tier have become more concentrated, the widening gap between peaks signals deepening inequality. This reflects a growing “Matthew effect” [59], where high-CEE cities continue to reinforce their advantages, further widening disparities both within and across city types.

3.2.2. Analysis of Spatial Evolution Pattern

By tracing the centroid of CEE over different periods, the trajectory and evolution of national CEE are clearly captured, as shown in Figure 8. These results are derived from the calculations outlined in Equations (6)–(10). Overall, the centroid shifted from the central region toward the northeast, with a cumulative displacement of 69.62 km, indicating a spatial shift of higher-efficiency areas toward northeastern China. The spatial coverage contracted from 2.42 million km² to 2.38 million km², suggesting an increasingly concentrated distribution of CEE among RBCs. Additionally, the slight contraction of both the major and minor axes of the standard deviational ellipse reflects a gradual convergence of spatial heterogeneity. Taken together, these changes indicate a relatively stable spatial pattern and a narrowing regional gap in CEE.

From a phase-dynamic perspective, the center of gravity of CEE shifted markedly by 46.71 km northeast (64.36°) from 2006 to 2011, reflecting higher CEE in the northeastern RBCs. Between 2011 and 2015, the shift reversed toward the southwest (197.22°), but the distance narrowed sharply to 5.77 km. This suggests enhanced CEE in the central and western regions, likely related to policies promoting technology transfer to less-developed areas for enhanced energy efficiency. From 2017 to 2022, the center of gravity again moved 17.15 km northeast (66.06°), with the azimuthal expansion indicating increasing divergence between the northeast’s diffusion of low-carbon technologies and the more rigid development in the central and western regions. This spatial trajectory underscores the uneven progression of CEE across China’s RBCs, driven by the northeast but hindered by the lagging central and western areas. Nevertheless, with ongoing policy support in the west, such as the steady implementation of the Western Development Strategy, the promotion of coordinated regional development policies, and the expansion of clean energy projects [60], new momentum has been injected into the transformation of central and western cities. These policies are promoting technology diffusion, improving energy efficiency, and strengthening regional cooperation, thereby accelerating development and narrowing the CEE gap across regions.

3.2.3. Spatial Autocorrelation Analysis

The spatial correlation of CEE among RBCs is examined using the global Moran’s I index, which is calculated based on Equation (11), as shown in

Table 3. Global Moran’s I index of CEE of Chinese RBCs (2006–2022).

Year	I	E(I)	sd(I)	z	p-Value
2006	0.469	−0.009	0.095	5.048	0.000
2007	0.391	−0.009	0.094	4.258	0.000
2008	0.383	−0.009	0.088	4.441	0.000
2009	0.435	−0.009	0.094	4.722	0.000
2010	0.492	−0.009	0.094	5.319	0.000
2011	0.496	−0.009	0.092	5.477	0.000
2012	0.546	−0.009	0.093	5.990	0.000
2013	0.494	−0.009	0.092	5.442	0.000
2014	0.470	−0.009	0.093	5.130	0.000
2015	0.446	−0.009	0.094	4.828	0.000
2016	0.372	−0.009	0.094	4.063	0.000
2017	0.419	−0.009	0.094	4.543	0.000
2018	0.419	−0.009	0.095	4.533	0.000
2019	0.339	−0.009	0.095	3.668	0.000
2020	0.270	−0.009	0.095	2.931	0.002
2021	0.225	−0.009	0.095	2.474	0.007
2022	0.227	−0.009	0.095	2.480	0.007

.

From 2006 to 2022, Moran’s I remained consistently positive and significant at the 1% level, demonstrating robust spatial autocorrelation and notable spatial spillover effects. Dynamically, Moran’s I increased from 2006 to 2012, peaking during this period. This suggests that spatial clustering increased as high-efficiency regions consolidated their advantages through scale effects, thereby intensifying regional disparities. However, from 2013 to 2022, the index gradually declined, reflecting a shift toward spatial diffusion. This trend underscores the need for enhanced policy-driven regional coordination, such as industrial relocation and ecological compensation [61], to support low-carbon transitions in central and western regions. Moreover, the COVID-19 pandemic likely contributed to narrowing regional disparities in CEE by unevenly impacting the national economy across RBCs [62], thereby mitigating the differences in energy use and carbon emissions across RBCs.

To examine the spatial clustering and temporal dynamics of CEE in RBCs, Moran’s scatter plots for 2006, 2011, 2017, and 2022 are constructed based on Equation (12), as shown in Figure 9. Most RBCs cluster in the first (“high–high” clusters) and third (“low–low” clusters) quadrants, indicating that both cities and their neighboring areas share relatively high or low levels of CEE, respectively. Over time, spatial clustering has weakened. In 2006 (Figure 9a), RBCs were predominantly located in the “low–low” quadrant, indicating strong clustering of cities with low CEE and similarly low-performing neighbors. This clustering effect intensified by 2011 (Figure 9b), with a further increase in the number of RBCs in this quadrant. By 2017 (Figure 9c), this pattern began to ease, with weaker spatial clustering and a shift of some RBCs toward the second quadrant (“low–high” clusters), suggesting weaker spatial dependence as low-efficiency cities became surrounded by high-efficiency regions. Meanwhile, a few cities emerged as isolated “high-efficiency islands”. By 2022 (Figure 9d), the distribution was characterized by greater diffusion and weakened spatial dependence, reflecting a trend toward more balanced development.

3.3. Analysis of Influencing Factors

3.3.1. Model Testing

To select the appropriate spatial econometric model, this study conducts LM, LR, and Wald tests, as shown in

Table 4. Results of spatial econometric modeling tests.

Test Method	Test Indicator	Statistical Value
LM test	LM (error) test	141.035 ***
	Robust LM (error) test	167.142 ***
	LM (lag) test	18.706 ***
	Robust LM (lag) test	44.813 ***
LR test	LR test spatial error	84.94 ***
	LR test spatial lag	87.88 ***
Wald test	Wald test spatial error	78.05 ***
	Wald test spatial lag	89.35 ***
Hausman test	—	93.13 ***

*** p < 0.01.

. The results support the use of SDM. Additionally, the Hausman test result (93.13, significant at the 1% level) further supports the fixed-effects specification. Therefore, the SDM with two-way fixed effects is adopted for the analysis.

3.3.2. Analysis of the Results of the Spatial Durbin Model Estimation

To mitigate heteroskedasticity, the spatial econometric analysis is conducted using the logarithmic forms of the explanatory variables [63], as shown in Equation (13).

Table 5. Spatial measurement model regression results.

Variables	SAR	SEM	SDM	W*x
lnX1	−0.107 ***	−0.112 ***	−0.121 ***	W*lnX1	0.065 *
	(0.006)	(0.006)	(0.007)		(0.039)
lnX2	0.048 ***	0.050 ***	0.056 ***	W*lnX2	−0.374 ***
	(0.013)	(0.013)	(0.013)		(0.105)
lnX3	−0.011 **	−0.012 ***	−0.012 ***	W*lnX3	0.065 *
	(0.004)	(0.004)	(0.004)		(0.034)
lnX4	−0.007	−0.004	0.026 ***	W*lnX4	−0.101 *
	(0.008)	(0.009)	(0.010)		(0.054)
lnX5	0.007 ***	0.006 ***	0.007 ***	W*lnX5	0.058 ***
	(0.002)	(0.002)	(0.002)		(0.021)
lnX6	−0.004 *	−0.004 *	−0.005 *	W*lnX6	0.043 **
	(0.003)	(0.003)	(0.003)		(0.019)
lnX7	−0.169 ***	−0.166 ***	−0.139 ***	W*lnX7	−0.212 ***
	(0.008)	(0.008)	(0.009)		(0.065)
N	1870.000	1870.000	1870.000
R2	0.445	0.481	0.461
ρ/λ	0.381 ***	0.471 ***	0.214 *
	(0.099)	(0.102)	(0.124)
sigma2_e	0.003 ***	0.003 ***	0.002 ***
	(0.000)	(0.000)	(0.000)

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

presents the results of the SDM, alongside those of the SAR and SEM, allowing for a comparative assessment of model performance and estimation outcomes.

The SDM yields an R² of 0.461 and a spatial autoregressive coefficient (ρ) of 0.214, significant at the 10% level, indicating spatial spillover effects in the CEE of RBCs. Furthermore, the spatial lag coefficients of W*lnX1, W*lnX3, W*lnX5, and W*lnX6 are positive, suggesting that improvements in energy intensity, government intervention, trade openness, and innovation in one city positively influence neighboring cities. In contrast, the coefficients of W*lnX2, W*lnX4, and W*lnX7 are negative, implying that industrial structure, economic level, and production scale exert negative spatial spillover effects.

The estimated coefficients from the SDM do not directly reflect the marginal effects of explanatory variables on CEE [64]. Therefore, the direct, indirect, and total effects are calculated to capture the full impact, as shown in

Table 6. Decomposition results of spatial effects.

Variables	Direct	Indirect	Total
lnX1	−0.120 ***	0.048	−0.072 *
	(0.007)	(0.046)	(0.043)
lnX2	0.054 ***	−0.454 ***	−0.400 ***
	(0.012)	(0.151)	(0.150)
lnX3	−0.012 ***	0.082 *	0.070
	(0.004)	(0.046)	(0.046)
lnX4	0.026 ***	−0.128 *	−0.102
	(0.009)	(0.071)	(0.068)
lnX5	0.008 ***	0.079 **	0.086 ***
	(0.002)	(0.032)	(0.032)
lnX6	−0.004	0.051 **	0.047 **
	(0.003)	(0.024)	(0.024)
lnX7	−0.140 ***	−0.303 ***	−0.443 ***
	(0.009)	(0.083)	(0.081)

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

. The direct effect measures the influence of changes in each factor on CEE within the same region. The indirect effect captures the impact of these changes on CEE in neighboring regions. The total effect represents the overall impact across all regions.

Energy intensity (X1) and production scale (X7) both have significantly negative direct and total effects. Higher energy consumption increases the dependence on fossil fuels, directly raising carbon emissions. The excessive expansion of production scale leads to resource overuse and waste, further degrading the environment and hindering improvements in CEE. However, the indirect effect of X1 is not statistically significant, while X7 shows strong negative spillover effects. This pattern reflects the broader regional consequences of unsustainable industrial growth driven by energy-intensive and high-emission traditional industries [65]. Focusing solely on output growth while continually expanding production scale inevitably undermines low-carbon development efforts. This extensive production mode forces neighboring cities to bear unnecessary environmental pressures, such as increased pollution and resource depletion, which in turn hinder their own CEE improvements. These findings underscore the urgent need for RBCs to prioritize cleaner and more energy-efficient production strategies to break away from high-carbon development paths.

In contrast, industrial structure (X2) and economic level (X4) have significantly positive direct effects but significantly negative indirect effects. This highlights a conflict between local benefits and regional costs, a typical manifestation of the siphon effect [38]. For X2, although shifting toward cleaner and more advanced industries, integrating green technologies, and optimizing heavy industries may improve local CEE, these positive impacts are smaller than the negative spillover effects on neighboring regions. This is largely due to the deeply entrenched industrial structure, still dominated by energy-intensive and high-emission sectors in RBCs. Negative spillovers occur mainly through two channels: the transfer of highly polluting industries to nearby areas and continued dependence on traditional heavy industries, which hinders the spread of green technologies and sustainable practices across regions. These factors exacerbate cross-regional environmental burdens, ultimately reducing CEE in neighboring areas. Similarly, improvements in X4 can indirectly strain neighboring regions by intensifying the competition for labor and capital and attracting high-quality industries, which may limit their capacity for low-carbon development [51]. However, the total effect of X4 is negative but statistically insignificant, indicating a weak or uncertain overall impact. This finding contrasts with the positive and statistically significant coefficient of X4 in the SDM, further confirming that such model estimates do not directly reflect the marginal effects of explanatory variables on CEE.

Trade openness (X5), innovation level (X6), and government intervention (X3) all have positive total effects on CEE, although only X3′s effect is statistically insignificant. Specifically, X3 has a significantly negative direct effect but a significantly positive indirect effect. This reflects inefficiencies in local government intervention, which may result from a preference for short-term economic growth or administrative expansion in many RBCs. Nevertheless, the surrounding cities may benefit from spatial policy spillovers, such as demonstration effects or fiscal transfers, leading to indirect improvements in their CEE. X5 has significant positive effects both locally and regionally. It facilitates the introduction of low-carbon technologies and green investments [66] while also fostering regional cooperation on emission reduction. X6 has a significant positive indirect effect but an insignificant direct effect. This means that local efforts mainly focus on expanding production efficiency, which limits the short-term ability to translate innovation into improvements in CEE [36]. However, the surrounding cities benefit from knowledge spillovers, as innovations and technological advances spread, enhancing CEE in neighboring regions.

3.3.3. GTWR Model Estimation Results

The factors influencing CEE were examined holistically, but the significant spatiotemporal imbalance in CEE among China’s RBCs confirms the presence of systemic heterogeneity. Therefore, analyzing this heterogeneity using the GTWR model, based on Equation (14), provides a deeper understanding of how these factors affect CEE across different RBCs. The GTWR model outperforms both the OLS and GWR models, with an R² of 0.7557 and an adjusted R² of 0.7548, as shown in

Table 7. Parameter estimates for each model.

	OLS	GWR			GWTR
Variables		Average	Max	Min	Average	Max	Min
lnX1	−0.0663 ***	−0.8746	−0.0734	−2.6344	−0.9492	0.2636	−4.3133
lnX2	−0.0022 ***	−0.2065	0.0359	−0.5144	−0.2319	0.1864	−0.8497
lnX3	−0.0013 **	−0.1598	−0.0172	−0.3453	−0.1768	0.0779	−0.5842
lnX4	0.0224 ***	0.4463	1.3226	−0.3455	0.4506	1.4197	−0.5737
lnX5	−0.0212 ***	−0.0368	0.4285	−0.4421	−0.0643	0.5531	−0.9779
lnX6	0.0113 ***	0.2072	1.2018	−0.4203	0.1915	4.2816	−0.5363
lnX7	0.0048	−0.3256	0.4826	−1.7617	−0.2027	2.2782	−2.1144
Intercept	0.6120 ***	−0.2332	0.1815	−1.0218	−0.2227	3.1983	−1.8878
Bandwidth	-		0.1150			0.1150
AICc	4059.8526		3118.6400			2905.6400
RSS	951.7522		540.4230			456.8460
R2	0.4908		0.7110			0.7557
Adj-R2	0.4889		0.7099			0.7548

** p < 0.05, *** p < 0.01.

. Furthermore, reductions in AICc and residual sum of squares (RSS) confirm the model’s superior fit, indicating that spatiotemporal embedding enhances explanatory power. The average regression coefficients for energy intensity (X1), industrial structure (X2), government intervention (X3), trade openness (X5), and production scale (X7) are negative, while those for economic level (X4) and innovation level (X6) are positive.

The temporal variations in factors impacting CEE in RBCs from 2006 to 2022 are shown in Figure 10. Notably, energy intensity (X1) has the most significant negative impact, with its coefficient dropping from −0.68 to −1.22, a decline of 79.4%. Energy efficiency improvements have lagged behind economic growth. The effect of production scale (X7) has shifted from a positive driver (0.07) to a negative constraint (−0.44), indicating a reversal in the marginal contribution of extensive expansion to carbon emission pressures. The negative elasticity of industrial structure (X2) has deepened, with an average annual reduction of 1.6 percentage points, highlighting the growing lock-in effect of heavy industries. Government intervention (X3) has consistently shown a negative effect with little variation over time, signaling the limited effectiveness of administrative measures in promoting carbon emission reductions and requiring further improvement. Although the negative impact of trade openness (X5) has gradually declined, structural contradictions remain. A high reliance on traditional resource exports and limited adoption of green technologies (only 0.12%) continue to hinder the low-carbon transition [67]. On the other hand, the positive drivers exhibit diminishing momentum. Although economic level (X4) maintains the largest positive coefficient (0.41), its cumulative contribution has decreased by 31%. The coefficient of innovation level (X6) remains below 0.25 and has stagnated, revealing systemic bottlenecks in green technology transformation.

Using the local estimation from the GTWR model, the regression coefficients of factors influencing CEE in Chinese RBCs from 2006 to 2022 were averaged. The visualization results (Figure 11) show that notable regional differences exist in the effects of these factors, highlighting the need to consider local spatiotemporal heterogeneity in the analysis.

Both energy intensity (X1) and government intervention (X3) have universally negative impacts across all cities (Figure 11a,c). X1 has the strongest suppressive effect in southeastern manufacturing hubs like Ganzhou and Hengyang. This is due to the large-scale transfers of high-energy-consuming and low-value-added manufacturing industries to these regions in earlier stages, leading to high energy consumption per unit of GDP and significantly hindering CEE. In contrast, X1′s effect is weaker in the northwest and northeast. In the northwest, lower industrialization and slower economies contribute to this, while in the northeast, mature industrial stability helps ease energy intensity pressures. These structural differences highlight regional disparities in energy efficiency. X3′s negative influence is most pronounced in the northeast, where deeply entrenched RBCs’ development patterns complicate policy enforcement [68]. Government investment in promoting low-carbon development has had limited effectiveness, and it is challenging to reverse the region’s long-standing inefficiencies in carbon emission governance. In contrast, the negative impacts of X3 are relatively weaker in central and western RBCs. This is likely due to market-oriented reforms and more flexible governance structures, which have facilitated relatively rapid development and partially offset the negative effects associated with carbon emissions [56].

Industrial structure (X2), trade openness (X5), and production scale (X7) generally exert negative effects on CEE across most RBCs. X2 (Figure 11b) particularly suppresses CEE in heavy-industry-dominated northeastern China, where energy-intensive sectors like steel mining and processing dominate, significantly constraining CEE improvements. However, slight improvements are observed in some western and central-western RBCs, such as Jinchang and Baoji, mainly due to early industrial restructuring that phased out outdated capacities and shifted the economy toward high-quality tertiary industries. This transition facilitated a greener industrial structure by reducing the dominance of heavy industries. X5 (Figure 11e) has a polarized effect in western RBCs. It hinders CEE in areas dependent on resource exports and lacking green technology integration yet benefits several cities in Yunnan and Sichuan, linked to green development. Yunnan’s strategic position as a gateway to southeast and south Asia supports active border trade and green technology exchange, while Sichuan leverages industrial clusters and urban internationalization to advance Belt and Road green trade and industry. X7 (Figure 11g) suppresses CEE in ecologically sensitive areas like Anshun and Heihe, where fragile ecosystems and limited environmental capacity make these regions particularly vulnerable [69]. Expanding production scale in these areas exacerbates ecological degradation and reduces green space, further increasing carbon emissions. Conversely, cities along the Yangtze River, including Ezhou and Huangshi, have stronger environmental resilience and have improved CEE through industrial cooperation and the expansion of green value chains.

Conversely, economic level (X4) and innovation level (X6) serve as the main positive drivers of CEE in most cities. X4 (Figure 11d) has a positive impact on CEE in nearly 90% of RBCs, indicating a widespread shift from solely growth-oriented development to green and sustainable development. The stepped distribution reveals stronger impacts in northern cities compared to southern ones, underlining the need to narrow regional economic gaps to support balanced low-carbon transitions. However, this relationship is weak or even negative in some cities, such as Zigong and Panzhihua, which reflects a structural mismatch between economic expansion and low-carbon goals [60]. X6 (Figure 11f) particularly benefits CEE in central cities like Ganzhou and Chenzhou, where they benefit from integration between industrial and innovation chains. In the southwest, cities like Lijiang show significant CEE gains due to breakthroughs in eco-industrial technologies. However, cities in the northwest like Baiyin face challenges related to path dependence. Their remote locations hinder effective technological collaboration with other cities, limiting the advancement of green technologies and CEE improvements. In the east, despite higher innovation inputs, the disconnect between imported technologies and local decarbonization efforts limits effectiveness [70].

4. Conclusions and Suggestions

4.1. Conclusions

(1)

This study employs the super-efficiency SBM-GML index model to evaluate the CEE of Chinese RBCs from both static and dynamic perspectives over the period 2006–2022. The results indicate a generally increasing trend in CEE. Specifically, CEE varies across city types, and the CEEs of regenerative cities, growing cities, mature cities, and declining cities decrease in sequence. Regionally, cities in the eastern and northeastern areas outperform those in the central and western regions. Decomposing the GML index into technological efficiency and technological progress indicates an average GML value of 1.04, with a volatile and gradually declining trend over time. This suggests that while overall CEE has improved, the growth momentum is weakening, largely due to insufficient synergy between efficiency gains and technological advancement. Overall, there remains significant room for improving CEE in these cities.

(2)

Spatial and temporal evolution analysis reveals that low-CEE cities still dominate Chinese RBCs, though high-CEE cities are gradually emerging. Growing cities are experiencing intensified polarization, while mature cities show efficiency improvements but increasing internal disparities. Declining cities face downward constraints on efficiency growth, and regenerative cities demonstrate relatively high efficiency but exhibit signs of multi-polarization. The national center of gravity for CEE has shifted toward the northeast, with strong spatial correlations among cities. The initial “high–high” and “low–low” clusters have gradually expanded to surrounding areas. Although intra-group disparities still exist among different types of RBCs, the overall spatial gap in CEE has narrowed, indicating gradual convergence and enhanced coordination at the national level. These trends reflect the positive impact of coordinated regional development policies.

(3)

The combined results of the SDM and GTWR models reveal significant spatial and temporal heterogeneity in the factors influencing CEE across Chinese RBCs. Except for innovation level, all examined variables exert significant direct effects on CEE. Total effects indicate that energy intensity, industrial structure, and production scale are significantly negatively associated with CEE, while trade openness and innovation level display significant positive effects. Notably, industrial structure and production scale exhibit strong negative spatial spillover effects. The GTWR results further highlight a clear north–south divide in energy intensity, with the strongest inhibitory effects observed in energy-intensive clusters in central and western regions. Production scale broadly suppresses CEE, while industrial structure and government intervention have deepened carbon lock-in in the northeast and southwest. However, in some central and western regions, a weak but positive effect on CEE has begun to emerge through industrial upgrading. Trade openness exhibits distinct polarization in these regions. Economic development and innovation generally promote CEE, but spatial disparities persist due to technological path dependence in the northwest and localization lag in the east.

4.2. Policy Suggestions

Under the pressure of climate change, improving CEE in RBCs is a crucial step toward achieving the dual carbon goals. This study analyzes the spatiotemporal evolution of CEE and the spatial heterogeneity of its influencing factors in China’s RBCs. Based on our findings, we propose the following targeted policy recommendations to promote the low-carbon transformation of different types of RBCs:

(1)

While CEE has improved overall in RBCs, some cities still face significant bottlenecks requiring targeted actions. As the center of CEE shifts northeastward and regional disparities gradually narrow, policymakers should leverage high-performing regenerative cities to demonstrate best practices and drive cross-regional coordination. These cities should take on a leading role, fostering coordinated development in surrounding areas. Concurrently, they must facilitate green technology transfer and industrial collaboration between eastern/northeastern and central/western RBCs while directing resources to regions showing significant efficiency gains. For growing and mature cities, it is essential to maximize the positive effects of economic development and technological innovation on CEE, gradually transforming economic progress into technological advancement to form a reinforcing positive feedback loop. For declining cities facing severe resource depletion and insufficient motivation to improve CEE, the focus should be on reducing excessive production scale and adjusting industrial structure to alleviate their suppressive effects on CEE.

(2)

The government should optimize the allocation based on varying impacts of key variables across different regions. Resources should be directed to factors most effective in improving CEE. Specifically, strengthening government intervention and improving policy implementation are crucial to accelerate industrial upgrading in northeastern China’s old industrial base, with a focus on streamlining heavy industries and overcoming path dependence and structural lock-in linked to high energy consumption. To harmonize economic growth with CEE improvements, RBCs in central and western regions should transition from extensive growth models to innovation-driven and green spillover development models. In regions where openness and production scale exert significant negative impacts on CEE, efforts should aim to break the scale effect trap by promoting low-carbon technologies and cultivating green trade advantages.