Analyzing the carbon emission efficiency and influencing factors of China’s thermal power generation sector based on super-SBM and ESTDA models

Abstract

The progress of carbon emission reduction and the effectiveness of the energy transition in the thermal power generation industry (TPI) directly impact both the quality of the implementation of China’s dual carbon goals and the broader landscape of sustainable development. To precisely analyze the core patterns of low-carbon transformation in the industry, this study overcame the limitations of existing research on the thermal power sector’s carbon emission efficiency (TPCEE) indicator system. These limitations include insufficient industry adaptability, an inadequate characterization of efficiency evolution dynamics, and an insufficient representation of regional differences. It innovatively constructed a TPCEE indicator system, focusing on the spatiotemporal evolution mechanisms and influencing factors of TPCEE. An integrated research framework of “efficiency measurement, spatiotemporal analysis, influencing factor exploration” was established. In addition, based on panel data from 30 Chinese provinces covering the period 2005–2022, empirical research was conducted using the Super-SBM model, exploratory spatiotemporal data analysis, and the Tobit model. The findings indicated that: (1) the TPCEE showed an overall fluctuating downward trend during the period of 2005–2022, and high-TPCEE areas were located primarily in North China and coastal provinces, while low-TPCEE regions were scattered in Northwest, Northeast, Central, and Southwest China. (2) Given the probability of spatiotemporal coalescence exceeding 70%, the spatial structure of the TPCEE was comparatively stable, showing distinct path dependence. (3) At the national level, the industrial structure, power generation mix, energy intensity, and degree of government intervention contributed to overall efficiency improvements. From a regional perspective, the impact of these factors on the TPCEE exhibited significant regional heterogeneity. The government may use the results as a foundation for building regional energy-saving and emission-reduction plans, as well as to encourage low-carbon transition and sustainable development in the Chinese TPI.

Introduction

Global warming and carbon dioxide (CO₂) overproduction are closely linked, leading to a cascade of serious effects such as the melting of glaciers, rising sea levels, and a reduction in biodiversity [1]. In 2006 saw China replaced the US as the largest emitter of CO₂ and experienced substantial demands to decrease emissions [2]. Therefore, in 2020, China’s government set goals for becoming “carbon neutral” by 2060 and reaching “carbon peaking” by 2030. This reflects China’s responsibility in global climate governance and its firm pursuit of sustainable development.

Currently, despite the rapid rise of clean-energy-centered power generation, it does not fully satisfy China’s industrial electrical consumption and daily electricity demand, and thermal power generation remains the primary mode of power supply [3]. In addition, the carbon emissions from the power sector account for approximately 40% of China’s total carbon emissions [4, 5], of which the thermal power industry (TPI) is a major contributor. Promoting a reduction of carbon emissions in the TPI is the key to realizing the “dual carbon” goal. However, along with the promotion of the “dual carbon” goal, the high carbon emission problem of the TPI is becoming intensifying, and there is an urgent need to improve the carbon emission efficiency (CEE) to realize low-carbon transformation.

Against this background, conducting a scientific and reasonable evaluation of the carbon emission efficiency of the thermal power industry (TPCEE), performing a comparative analysis of the spatiotemporal dynamics, examining the main elements influencing the TPCEE, and providing practical countermeasures and are important. This is of great practical importance for fostering an ongoing advancement of carbon emission reduction in the Chinese power industry and to realizing sustainable development.

The rest of this article as follows: Sect. "Literature review" summarizes the relevant literature. Section "Data and methodology" describes the data indicators and methodology. Section "Analysis" presents the results of the analyses, Sect. "Discussion" presents a discussion, and, finally, an overview of the findings, as well as policy recommendations, are provided.

Literature review

In the literature, there is limited documentation of the TPCEE in the existing literature, however results on CEE have provided some insights. CEE measured include single-factor CEE and total-factor CEE [3]. The former is generally calculated as the ratios of carbon emissions to a factor such as carbon emission intensity [6, 7], and the advantage of its measurement lies in the fact that relevant data can be obtained using simple methods and the indicators are easy to understand. However, the generation and impacts of carbon emissions are related to human production and lifestyle, as well as other Earth organisms. Meanwhile, this single measurement method suffers from the problem of being relatively one-sided and inconsistent with actual conditions. The-total factor CEE measurements consider the influence of social, economic, ecological and scientific, and technological factors on carbon emissions, making it more comprehensive and scientific.

Stochastic frontier analysis (SFA) and data envelopment analysis (DEA) are the most commonly used methods for assessing total-factor CEE [8]. SFA sets an exact production function and encounters certain constraints [9]. DEA is an input-output- based efficiency evaluation model that has been widely used to evaluate factors such as CEE [1], environmental performance [10], and ecological effectiveness [11], DEA modeling promotes the rationalization of efficiency evaluations and provides a key indicator for quantifying the responses to climate shifts [12]. In the study of the TPCEE, scholars generally use DEA models as the core measurement tool. Yan et al. determined provincial TPCEE of China using the Undesirable-SBM model [13]. Xu et al. conducted a spatiotemporal analysis of the TPCEE, using a three-stage SBM-DEA model [3]. In these studies, capital, labor, energy, and thermal power outputs are generally considered input variables and desired output indicators. Meanwhile, CO₂ is often seen as an undesired output [14].

Fang et al. conducted a study on the TPCEE from the micro perspective of power plants [15]. They screened indicators from energy, economic, and environmental perspectives, and constructed a multi-indicator evaluation system of energy efficiency for sustainable development by combining the operational characteristics of power generation enterprises. The present analyzed and compared the advantages, disadvantages and applicability of previous studies, proposes the adoption of the DEA model, incorporated the economic technology and economic output dimensions of power generation into the TPCEE evaluation system.

Based on the dimensions of geographical differences and temporal evolution, each region exhibits unique societal, environmental, and economic characteristics at different stages of development [16]. Numerous scholars have analyzed the temporal and spatial features of CEE after it has been measured. Common approaches include the use of a standard deviation ellipse, the Terre index, and the spatial autocorrelation model and so on [17, 18]. However, these methods do not merge time and space. In a thorough exploration of dynamic spatiotemporal characteristics, Rey and Janikas systematically discussed and proposed an innovative method, that is, exploratory spatial-temporal data analysis (ESTDA) [19]. This method incorporates an analysis of spatially correlated local variables with the direction of movement in the distribution of CEE [20], and it effectively captures the complex evolution patterns of CEE in the geospatial and temporal dimensions through the precise characterization of data at different spatial and temporal scales, providing a brand new perspective and a powerful tool for studies in related fields. Therefore, this study used ESTDA to explore the spatiotemporal evolutionary components of the TPCEE.

The factors influencing CEE have attracted considerable attention from scholars. Influencing factors that have been extensively studied include urbanization level [1, 21], industrial structure (IS) [22], renewable energy [23], trade openness [14, 24], R&D investment [25], green innovation [26], and technological progress [27]. Studies have shown that IS negatively affects CEE, whereas technological progress has a significantly positive effect. However, existing research results have not revealed the influence of factors such as urbanization level. Most studies on the factors influencing CEE have focused on a single variable, though the formation mechanism of CEE is complexity and is a result of several factors interacting with one another [14]. In addition, the TPCEE is influenced by specific industry factors. Therefore, a systematic, multi-dimensional analysis of the set of factors influencing the TPCEE is urgently required so as to reveal the interaction mechanism and entire effect of each factor.

A review of the literature revealed that, while scholars have made some progress in studying CEE and its influencing factors, research focused on the TPI remains limited. The key gaps in this research field include the following. First, the lack of a three-dimensional TPCEE indicator system integrating technology, economy, and the environment has prevented an accurate representation of the industry’s technical characteristics and efficiency objectives. Second, the insufficient exploration of the dynamic spatiotemporal interaction mechanisms of the TPCEE has hindered the integration of temporal and spatial factors. Third, no analytical framework has integrated macro-environmental factors with industry characteristics while accounting for regional heterogeneity, making it difficult to determine regionally differentiated impact mechanisms. Building upon these gaps, the innovations and contributions of this study include the following: (1) It expands the traditional DEA input-output framework to incorporate power generation economics and corporate profitability dimensions, thereby obtaining a more comprehensive TPCEE indicator system. This addresses the shortcomings of current total-factor indicators, which prioritize general applicability over industry-specific relevance, and, thus, better aligns efficiency measurements with the operational realities of TPI. (2) This study did not perform a traditional fragmented spatiotemporal analysis. Instead, it employed ESTDA to thoroughly examine the spatiotemporal characteristics of the TPCEE. Using tools such as spatiotemporal transition matrices, it integrated the spatial agglomeration features of the TPCEE with temporal evolution trends, precisely determining the spatiotemporal dynamics of efficiency evolution, and filling the gaps in existing research on efficiency dynamics. (3) Based on industry characteristics, this study expanded the dimensions of influencing factors, incorporating both macro variables (such as economic development level and EI) and industry-specific factors (such as electricity consumption scale). It also accounted for regional differences by using the Tobit model to determine how each factor affects the TPCEE across different regions.

Data and methodology

Research framework

The main research approach of this study comprised four components (see Fig. 1). (1) Using the Super-SBM model to calculate the TPCEE: Based on panel data from 30 provinces covering the period 2005–2022, the Super-SBM model was used to calculate the TPCEE, providing robust baseline data for subsequent spatiotemporal analysis. (2) Spatiotemporal evolution analysis of the TPCEE: Building upon the efficiency measurements, this component involved characterizing efficiency evolution based on temporal trends and spatial differentiation, preliminarily revealing the dynamic patterns and regional imbalances in the TPCEE distribution. (3) In-depth analysis of spatiotemporal interaction mechanisms: Utilizing ESTDA, particularly LISA time paths and spatiotemporal transition matrices, this study examined the interactive characteristics and dependencies of the TPCEE across spatiotemporal dimensions. It assessed the interconnectedness of TPCEE spatial patterns, addressing the limitations of traditional static spatial analysis. (4) Identifying influencing factors based on the Tobit model: We divided the 30 provinces into seven major geographic regions and established a Tobit panel regression model with the TPCEE (calculated using the Super-SBM model) as the dependent variable. This approach identified the key factors influencing efficiency across regions, along with their differences, providing empirical evidence that can support differentiated policy formulation.

Fig. 1.

Technology roadmap

Through the progressive logic of the “Super-SBM model providing reliable data; spatio-temporal evolution analysis establishing preliminary insights; ESTDA deepening spatio-temporal interaction patterns; Tobit model analyzing regional heterogeneity mechanisms,” this study constructed a research framework encompassing foundational data generation, evolutionary pattern discovery, and impact mechanism validation. This approach ensured that the research outcomes possessed quantitative support and a dynamic perspective while maintaining policy relevance.

Indicators selection

Data source

Our study population covered 30 provinces in China, Hong Kong, Macau, Taiwan and Tibet were not included in the study owing to missing data, and the time span was 2005–2022. The study subjects were categorized into North China, East China, Central China, South China, Northeast China, Southwest China, and Northwest China according to the geographical division of China [28].

The data were primarily obtained from the China Statistical Yearbook, China Financial Yearbook, China Electric Power Yearbook, China Energy Statistical Yearbook, China Science and Technology Statistical Yearbook, and various local statistical yearbooks. The missing values in this study were supplemented via linear interpolation.

Carbon emission efficiency indicators for the thermal power generation sector

This study measured the total-factor TPCEE of China. The measurement indicators included input and output indicators, which are listed in Table 1.

Table 1.

Indicator system for carbon emission efficiency in thermal power generation sector

Dimension	Indicator	Explanation	Unit
Inputs	Capital	Installed thermal power capacity	10,000 kw
	Labor force	Number of workers engaged in electricity and heat production and supply	People
	Energy	Thermal power generation multiplied by standard coal consumption for power generation	10,000 tons
	Hours of equipment utilization	Hours of equipment utilization	Hours
	Electricity consumption rate	Ratio of self-consumption of power plant to total power generation	%
Expected output	Thermal power generation	Thermal power generation	billion kWh
	Gross industrial output value	Gross industrial output value of the electricity and heat production and supply sector for the 2005-based period	Billion yuan
Undesired output	CO2 emissions	IPCC Carbon Emission Factor Method	10,000 tons

Input indicators: Existing studies have primarily selected input indicators for the TPCEE from the capital, labor, and energy dimensions, including the thermal power installed capacity, number of employees in the electricity and heat production and supply industry, and total energy consumption for thermal power generation [3]. This study drew from the approach of Fang et al. by introducing economic and technical indicators of power generation to capture the driving mechanisms of the TPCEE more comprehensively [15]. Equipment utilization hours directly reflect the actual operational intensity and asset-intensive utilization level of generating units. Higher utilization hours typically indicate a more rational load scheduling, reduced start-stop losses, and economies of scale, contributing to a lower carbon emission intensity per unit output. However, excessively high thermal power utilization hours suggest insufficient generation space for clean energy sources such as wind and solar power, hindering the green transition of the energy structure. The power plant electricity consumption rate can be used to directly characterize a facility’s energy consumption level. A lower rate indicates reduced self-consumption losses during energy conversion, as well as higher operational efficiency, enabling organizations or groups outside thermal power enterprises to receive more electricity with the same energy input. These two indicators directly reflect the technical and managerial effectiveness of the power generation process from the critical dimensions of operational intensity and internal system losses, serving as important supplements to traditional input metrics.

Output indicators: These were categorized into expected and unintended outputs. Beyond thermal power generation, expected outputs incorporate the economic viability of the TPI, adding the industrial output value indicator for the electricity and heat production and supply industry to balance environmental and economic benefits. The unexpected output is CO₂ emissions. As China currently lacks directly published carbon emissions data, this study adopted the calculation methods recommended in the 2006 IPCC National Greenhouse Gas Inventory Guidelines for estimation [3, 29], ensuring data authority and comparability.

Impact factor indicators

This study assessed the potential factors influencing the TPI in five dimensions: economic, social, energy, science and technology, and government (see Table 2).

Table 2.

Indicators of factors influencing carbon emission efficiency in the TPI

Dimension	Indicator	Interpretation	Unit
Economic	IS	Share of secondary sector in GDP	%
	LED	GDP per capita with 2005 as base period	10,000 yuan
	LEI	Share of total imports and exports in GDP	%
Social	ECS	Per-capita electricity consumption	kWh/person
Energy	EI	Total energy consumption/GDP for the 2005-based period	Tons of standard coal/10,000 yuan
	GM	Ratio of thermal power generation to total power generation	%
Technology	SSTS	R&D internal expenditure as a share of GDP	%
Government	LGI	Fiscal expenditure as a share of GDP	%

IS: Indicated by the share of secondary industries in the GDP [30]. An increase in the proportion of the secondary industries will increases the demand for electricity supply, and the TPI will expands its production scale. A moderate IS ratio promotes technological intensification of the TPI through a scale effect, but too high a proportion reinforces fossil-fuel dependence and inhibits the clean-energy transition of the TPI.

Level of economic development (LED): This indicator is generally measured in terms of the GDP per capita. A higher LED promotes technological progress and industrial upgrade, prompting thermal power generation enterprises to embrace advanced energy-saving and emission-lowering techniques to enhance the effectiveness of energy utilization [31]. Conversely, economic growth drives energy demand growth, and if clean energy substitution is insufficient, it may cause an increase in thermal power generation, which, in turn, has a dampening effect on CEE.

Level of exports and imports (LEI): Calculated by dividing the total value of imports and exports by the GDP. A high LEI may increase the demand for trade in energy-intensive products, prompting the expansion of thermal power generation and reducing CEE; concurrently, international trade may introduce advanced technologies and environmental standards to improve CEE [14].

lectricity consumption scale (ECS): The per-capita electricity consumption is an important component of the scale of electricity consumption. Similar to the IS, an increase in the ECS prompts thermal power producers to expand production, and total carbon emissions rise. However, the optimization of administration and technology, along with the enhancement of energy efficiency, by thermal power enterprises, performed to cope with the increase in the ECS will, in part, improve CEE.

Energy intensity (EI): This indicator reflects the level of energy usage per unit of GDP [30]. For the TPI, its rise suggests an increase in carbon emissions per unit of electricity output, which has a negative impact on CEE.

Generation mix (GM): Measured by a percentage of the thermal power generation relative to the overall power generation. A higher GM suggests a stronger dependency on high-carbon energy sources, limiting the substitution effect of cleaner energy sources and, thus, reducing the overall CEE of the TPI.

Strength of scientific and technological support (SSTS): A higher investment in scientific research can promote technological innovation, prompting thermal power generation enterprises to embrace more effective combustion technology and energy-saving and emission-reduction equipment, thereby reducing the carbon emissions per unit of power generation and improving CEE.

Level of government intervention (LGI): Indicated by the ratio of government general budget expenditure to GDP. A higher LGI can incentivize enterprises to adopt energy-saving and emission-reduction measures as well as improve CEE through financial subsidies and policy guidance [2]. However, a less efficient allocation of government resources, or poor policy implementation, may also lead to resource mismatches and hinder the improvements in the CEE.

Methodology

Super-SBM model

The core requirement to measure China’s TPCEE is to accurately incorporate carbon emissions as an undesirable output while enabling a differentiated ranking of decision-making units (DMUs) across provinces and years. Traditional models struggle to satisfy both demands simultaneously. Specifically, the primary limitation of conventional DEA models lies in their output indicators, which typically represent desirable outputs, making them unsuitable as undesirable outputs. The Super-SBM model proposed by Tone, based on slack variable measurements, can calculate efficiency values as the input and output slack levels vary [32]. However, it encounters the issue of multiple DMUs simultaneously achieving perfect efficiency. The Super-SBM model cannot further differentiate or rank provinces with an efficiency level of 1, making it difficult to support subsequent detailed analyses of spatiotemporal evolution characteristics. To address this issue, Tone further proposed the Super-SBM model [33]. By optimizing slack variables, it allows efficiency values to exceed the 1 constraint, enabling the precise ranking of efficient units under the SBM framework. This model is suitable for panel data studies requiring “handling undesirable outputs and conducting detailed efficiency comparisons across multiple DMUs.” This study used the Super-SBM model for efficiency measurements based on TPI data from 30 provinces spanning 2005–2022. Given the need to incorporate carbon emissions as undesirable outputs while distinguishing efficiency variations across regions and years, the model adopted the following formula:

1

2

Where  denotes the efficiency value; , , and represent their corresponding slack variables respectively; , ,  and stand for inputs, desirable outputs, and undesirable outputs respectively;  and  indicate the number of desirable output indicators and undesirable output indicators respectively;  is the number of input indicators; and  is the weight vector.

Spatial autocorrelation

Spatial autocorrelation can be used to measure distribution characteristics and interrelationships of spatial data. The fundamental concept is that data values at spatially adjacent or proximate locations may exhibit some form of dependency or similarity that diminish or disappear as distance increases [34]. Given the significant cross-regional linkages in the TPI’s energy consumption, technology diffusion, and policy implementation, its CEE is highly likely to exhibit spatial dependency. When analyzing the spatial distribution patterns of China’s TPCEE, verifying whether its efficiency values display cross-regional spatial correlation characteristics is crucial step in spatial analysis [35]. The global Moran’s I index serves as a core tool for quantifying this characteristic.

The index determination logic is as follows: A Moran’s I > 0 indicates a positive spatial correlation. The higher the value, the more likely high-efficiency (low-efficiency) provinces are to be adjacent to other high-efficiency (low-efficiency) provinces, indicating a more pronounced spatial clustering pattern. A Moran’s I < 0 indicates a negative spatial correlation, where smaller values suggest that high-efficiency (low-efficiency) provinces tend to be adjacent to low-efficiency (high-efficiency) provinces; values close to 0 indicate a random spatial distribution. The formula for the global Moran’s I index is as follows:

3

Where n is the number of study units;  and  are the TPCEE in neighboring units, respectively;  is the mean efficiency value; and  denotes the spatial weighting matrix.

Local spatial autocorrelation provides an indicator for analyzing spatial data across different regions or units within a study area. It reflects the degree and significance of the spatial variation within each province or between provinces and their neighbors, that is, clustering characteristics. The formula is as follows:

4

Exploratory Spatiotemporal data analysis

Rey and Janikas proposed ESTDA, which integrates a dynamic evolutionary perspective of the temporal dimension into the traditional framework of Exploratory Spatial Data Analysis [36]. By combining spatial pattern characterization with temporal trend tracking, it enables a quantitative analysis of the “spatiotemporal interaction characteristics” of variables. This method is suitable for research scenarios involving long-term panel data and significant regional correlations, particularly for analyzing how the spatial patterns of variables evolve over time and whether temporal differences exist in interregional interactions [35]. By using tools such as LISA time paths and spatiotemporal transitions, ESTDA simultaneously captures both the intrinsic temporal variation in efficiency and the dynamics of spatial associations, thereby precisely addressing the core requirement of this study to unravel the spatiotemporal co-evolutionary mechanisms of the TPCEE.

LISA time path: This reflects the geometric characteristics of the movement of the geographic variables of the LISA coordinates in a Moran scatterplot. By analyzing the pairwise migration of the TPCEE in each province with a spatial lag value, it corresponds to explains the characteristics of the spatiotemporal synergistic change and dynamics of the TPCEE [37]. Specifically, relative length ( Γ \), curvature (), and direction () are included, as follows:

5

6

7

    

Where: N is the number of research units;  is the movement distance of research unit i between t and t + 1. If ,  is greater than 1, the research unit has a more dynamic local spatial structure with dependent direction during the study period, and vice versa.  is the direction of travel of the province. 0°−90° and 180°−270°, represent the synergistic positive or negative growth of the observation unit and the neighboring unit, respectively; 90°−180° indicates that the observation unit has a negative change and the neighboring unit has a positive change; the opposite is true for 270°−360°.

Spatiotemporal leap: To further investigate the transition state and dynamic path of the local spatial correlation types of the TPCEE, this study relied on the concept of spatiotemporal leap [36], and incorporated the LISA temporal path with the traditional Markov chain to create a spatiotemporal leap matrix. As shown in Table 3 [38], there are four main types. Spatial cohesion (SC) is measured according to the type of jump in the TPCEE across every area, the spatial cohesion (SC) will be measured.

Table 3.

LISA type of Temporal transfer

Type	Sort of spatiotemporal transition	Transition form
Type I	Self-stability, neighbor-stability	HHt→HHt + 1, HLt→HLt + 1, LLt→LLt + 1, LHt→LHt + 1
Type II	Self-transition, neighbor-stability	HHt→LHt + 1, HLt→LLt + 1, LLt→HLt + 1, LHt→HHt + 1
Type III	Self- stability, neighbor-transition	HHt→HLt + 1, HLt→HHt + 1, LLt→LHt + 1, LHt→LLt + 1
Type IV	Type IVa: Self and neighborhood, simultaneous transitions; Type IVb: Self and neighborhood reverse transitions	Type IVa: HHt→LLt + 1, LLt→HHt + 1, Type IVb: HLt→LHt + 1, LHt→HLt + 1

8

Where Type I represents the number of types that are stable for both itself and its neighbors. Type Iva is the number of types that are transferred for both itself and its neighbors, and m is the total number of jumps.

Tobit model

The Tobit model, first proposed by Tobin, is a regression analysis method designed for dependent variables with value constraints such as truncation, censoring, and upper/lower bounds [39]. The Tobit model is mainly applied in cases where the dependent variable has explicit value boundaries (e.g., the interval [0,1] or non-negative values) or where some observations are truncated owing to data characteristics. Specifically, in efficiency evaluation, the efficiency scores calculated using DEA methods typically fall between 0 and 1, representing classic two-sided truncated data. Traditional linear models struggle to fit such data structures, whereas the Tobit model can accurately handle them.

In this study, the TPCEE measurements derived from the Super-SBM model were constrained, whereas ordinary least squares assume that the dependent variable is unconstrained, which may lead to bias [14]. Therefore, when investigating the determinants of the TPCEE, it is essential to address the estimation biases arising from constrained dependent variable values. The Tobit model provides a classical econometric approach suited for such scenarios, with its formula presented as follows:

9

10

Where  is the observed variable,  is the TPCEE of the ith province in year t,  is the set of all influencing factors in the model,  is the regression coefficient of the explanatory variables, and  is the random error term.

Analysis

Temporal analysis of carbon emission efficiency in thermal power generation sector

Overall level

Table 4 lists the results of the TPCEE. Overall, the TPCEE of China showed a fluctuating and slow downward trend from 2005 to 2022 (except for that in 2012), with its value decreasing from 0.9032 to 0.7342, a difference of 18.70%. This shows that China’s large power demand and the inadequate realization of the effects of emission reduction technologies have hindered the fossil-energy-dominated TPI from achieving energy structure transformation.

Table 4.

Carbon emission efficiency results for the thermal power generation sector

Region/Province	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	Average efficiency (2005–2022)
North China	1.0656	0.9895	1.0082	0.8265	0.8438	0.8982	0.9785	1.0628	0.9971	0.9223	0.9496	0.8865	0.9419	1.0588	1.0928	1.0919	1.0178	1.0842	0.9842
Beijing	1.2609	1.3406	1.3905	1.3994	1.4080	1.5164	1.5195	1.6298	1.5324	1.5699	1.7853	1.7177	1.8208	1.7291	1.8597	1.8970	1.9442	1.9606	1.6268
Tianjin	1.0111	1.0080	1.0098	0.7751	0.7519	0.8291	1.0046	1.0565	1.0027	0.6978	0.7209	0.7217	0.7170	1.0013	1.0060	1.0079	0.7274	1.0019	0.8917
Hebei	1.0280	1.0126	1.0208	0.7712	0.7996	0.8758	1.0179	1.0150	1.0164	1.0099	1.0093	0.7830	1.0300	1.0499	1.0418	1.0377	1.0133	1.0210	0.9752
Shanxi	1.0196	1.0317	1.0246	0.6004	0.6356	0.6273	0.6498	1.0218	0.7833	0.7052	0.5975	0.5972	0.5021	0.5105	0.5227	0.4863	0.3939	0.4153	0.6736
Inner Mongolia	1.0085	0.5546	0.5951	0.5863	0.6238	0.6422	0.7006	0.5907	0.6508	0.6289	0.6350	0.6128	0.6395	1.0032	1.0338	1.0306	1.0100	1.0224	0.7538
East China	1.0195	0.9706	0.9754	0.9932	1.0136	1.0129	1.0619	1.1378	1.0118	0.9850	0.9707	0.9638	0.9529	1.0451	1.0496	0.9385	0.9223	1.0604	1.0047
Shanghai	1.1618	1.0781	1.0850	1.1379	1.1170	1.1075	1.1109	1.3595	1.0817	1.1148	0.8853	0.9107	0.9564	0.9535	0.9323	1.0176	1.0896	1.1096	1.0672
Jiangsu	1.1153	1.1529	1.1283	1.1695	1.1388	1.1359	1.1506	1.2141	1.1492	1.1520	1.0889	1.1862	1.1725	1.1880	1.1786	1.2075	1.1385	1.1203	1.1548
Zhejiang	1.0712	1.0642	1.0509	1.0074	1.0053	1.0098	1.0091	1.0728	1.0656	1.0563	1.1509	1.0626	1.0066	1.0317	1.0890	0.9739	1.0051	1.0381	1.0428
Anhui	0.7414	0.6938	0.5812	0.7517	0.7757	0.7926	1.0041	1.0317	0.8366	0.7152	0.7031	0.6144	0.5635	1.0004	1.0016	0.4536	0.3449	1.0096	0.7564
Shandong	1.0081	0.8639	1.0317	0.8997	1.0311	1.0187	1.0347	1.0108	0.9259	0.8869	1.0253	1.0450	1.0654	1.0518	1.0466	1.0396	1.0332	1.0244	1.0024
Northeast China	0.7293	0.7768	0.6949	0.7080	0.6410	0.5822	0.6040	0.8777	0.4387	0.5334	0.5146	0.5635	0.5187	0.4992	0.5622	0.4848	0.4937	0.5164	0.5966
Liaoning	0.8500	1.0133	0.8605	1.0094	0.7557	0.6755	0.6768	0.8759	0.6503	0.6027	0.5852	0.6758	0.6311	0.5707	0.6341	0.6021	0.5846	0.6068	0.7145
Jilin	0.6503	0.6322	0.5786	0.5480	0.4918	0.4838	0.5349	1.0060	0.0758	0.4742	0.4618	0.4962	0.4405	0.4633	0.5408	0.4292	0.4490	0.4639	0.5122
Heilongjiang	0.6876	0.6850	0.6454	0.5665	0.6754	0.5871	0.6004	0.7512	0.5901	0.5233	0.4970	0.5185	0.4844	0.4636	0.5116	0.4232	0.4475	0.4786	0.5631
Central China	0.8121	0.7466	0.7150	0.6948	0.7057	0.6704	0.6800	1.0117	0.6225	0.6694	0.5915	0.5771	0.5377	0.5133	0.4956	0.4563	0.4168	0.3906	0.6282
Jiangxi	0.6656	0.6734	0.5752	0.5665	0.5807	0.5394	0.5477	1.0110	0.5062	0.6492	0.4404	0.4274	0.4614	0.3865	0.3476	0.2682	0.1645	0.1268	0.4965
Henan	0.7586	0.7287	0.8406	0.7087	0.7398	0.7421	0.7992	1.0159	0.7409	0.7761	0.7337	0.6710	0.5607	0.5714	0.4972	0.4638	0.4343	0.3192	0.6723
Hubei	0.8956	0.8878	0.7815	0.8415	0.8232	0.7728	0.7485	1.0126	0.6313	0.6267	0.5945	0.6092	0.6161	0.5744	0.6045	0.5792	0.5927	0.6314	0.7124
Hunan	0.9285	0.6965	0.6626	0.6624	0.6789	0.6271	0.6248	1.0072	0.6114	0.6257	0.5974	0.6007	0.5125	0.5209	0.5332	0.5138	0.4756	0.4851	0.6313
South China	0.9995	1.0011	0.9682	0.9398	0.8442	0.7359	0.7633	1.0882	0.7337	0.9733	0.8663	0.7476	0.7139	0.6908	0.7094	0.6883	0.6978	0.6725	0.8241
Fujian	1.0219	1.0178	1.0057	0.8214	1.0022	0.7451	0.8014	1.0407	0.7635	0.7368	0.6954	0.7426	0.7080	0.6500	0.6777	0.7116	0.7269	0.6722	0.8078
Guangdong	1.1855	1.1624	1.1892	1.1319	1.1287	1.1309	1.1014	1.1569	1.1523	1.1304	1.1591	1.1783	1.2213	1.1038	1.1487	1.1625	1.1797	1.2111	1.1575
Guangxi	0.7728	0.7953	0.6492	0.7818	0.7345	0.6538	0.6315	1.0199	0.5521	1.0189	0.6058	0.5993	0.4532	0.4473	0.4111	0.3482	0.3739	0.2844	0.6185
Hainan	1.0179	1.0290	1.0286	1.0240	0.5113	0.4137	0.5188	1.1354	0.4668	1.0072	1.0050	0.4703	0.4732	0.5622	0.5999	0.5308	0.5108	0.5224	0.7126
Southwest China	0.8087	0.6898	0.6521	0.6346	0.7403	0.7020	0.7033	0.9598	0.5981	0.6067	0.6398	0.7050	0.8753	0.6106	0.5591	0.5281	0.5065	0.5164	0.6687
Chongqing	0.5703	0.4749	0.5532	0.5038	0.5209	0.4544	0.8044	1.0156	0.4799	0.4578	0.4648	0.5244	0.5405	0.5394	0.4997	0.5071	0.5459	0.6293	0.5603
Sichuan	0.8168	0.7745	0.7588	0.7801	0.7678	0.7642	0.8049	1.0470	0.7676	0.6659	0.8303	0.9193	0.6038	0.7129	0.5823	0.4936	0.3869	0.3176	0.7108
Guizhou	1.0042	0.8425	0.7217	0.6490	1.0234	1.0153	0.6222	0.7575	0.4893	0.5894	0.5114	0.5260	0.6565	0.5001	0.4853	0.5138	0.5066	0.5330	0.6637
Yunnan	0.8434	0.6674	0.5747	0.6055	0.6491	0.5742	0.5817	1.0190	0.6557	0.7138	0.7527	0.8504	1.7006	0.6901	0.6691	0.5980	0.5868	0.5857	0.7399
Northwest China	0.8000	0.7112	0.6211	0.6335	0.5327	0.4807	0.5874	0.9355	0.7255	0.7990	0.5173	0.5432	0.5211	0.5866	0.5610	0.6000	0.5782	0.6874	0.6345
Shaanxi	0.5649	0.6385	0.5627	0.4938	0.5461	0.5326	0.6119	1.0349	1.0071	1.0047	0.5470	0.6313	0.5601	0.4942	0.5329	0.5297	0.5250	0.5468	0.6313
Gansu	1.0058	1.0046	0.6018	0.6321	0.4859	0.4351	0.5273	1.0138	0.4527	0.4050	0.4074	0.3886	0.3325	0.3480	0.3691	0.3291	0.3203	0.3422	0.5223
Qinghai	1.0171	0.5501	0.5409	0.6251	0.7731	0.7034	0.6761	1.1108	0.6895	1.0442	0.0001	0.5791	0.5838	0.5314	0.6032	0.5339	0.5314	0.5148	0.6449
Ningxia	1.0165	1.0230	1.0260	1.0347	0.4289	0.4096	0.7306	1.0493	1.0378	1.0202	1.0136	0.6428	0.6434	1.0020	0.6244	0.5956	0.4833	1.0157	0.8221
Xinjiang	0.3957	0.3399	0.3739	0.3815	0.4296	0.3229	0.3910	0.4688	0.4402	0.5210	0.6183	0.4742	0.4857	0.5571	0.6756	1.0117	1.0308	1.0174	0.5520
Total sample	0.9032	0.8479	0.8150	0.7822	0.7678	0.7379	0.7846	1.0184	0.7602	0.8043	0.7374	0.7259	0.7381	0.7403	0.7420	0.7099	0.6852	0.7342	0.7797

Figure. 2(a) shows the distribution of the TPCEE during the study period. The interval of 0–0.2.2 has the least number of occurrences, only four times, accounting for 0.78%; the TPCEE greater than 1 had the highest frequency of (187 times), accounting for 36.38%. During the study period, the frequency of TPCEE greater than 0.6 was 317 times, accounting for 61.67%. This indicated that, although the TPCEE showed a degrading trend, it was still at a relatively favorable level.

Fig. 2.

Overall carbon emission efficiency of China’s thermal power generation sector

Provincial perspectives

Beijing, Guangdong, Jiangsu, Shanghai, and Zhejiang were the top-five provinces (cities) with average TPCEE greater than 1, with value of 1.6075, 1.1498, 1.1482, 1.0647, and 1.0410, respectively (Table 4). These regions have a high economic level and a strong industrial base, which makes it easy to improve and implement power generation technologies, such as cogeneration. Jiangxi Province had the lowest TPCEE (0.4912), and was the only province with an average TPCEE below 0.5. The province’s energy transition has been slow, with its share of renewable power generation predicted to be 25.31% in 2022, which is well below the national average (31.6%). During the study period, a total of 10 provinces, accounting for 33.33%, realized positive growth in the TPCEE, including Shandong, Hebei, and Chongqing, which are more distributed in the central and eastern China belt.

We selected representative provinces to visualize the time trends of their TPCEE (Fig. 2(b)). Beijing had the highest TPCEE, Jiangxi had the lowest TPCEE, Hebei had the highest rate of TPCEE improvement, and Henan had the highest rate of TPCEE decrease. Furthermore, Beijing consistently maintained a high level with a stable upward trend. Specifically, it increased steadily from 1.26 in 2005 to 1.96 in 2022, representing an increase of approximately 55.49%, without any significant decline, demonstrating the exceptional proficiency of thermal power enterprises in enhancing energy efficiency. Jiangxi remained within the lowest national range, with an efficiency of only 0.13 in 2022, which is the lowest among the four provinces. Beyond its overall low TPCEE, the province also exhibited significant fluctuations, such as during 2011–2013. However, in most years, it remained below 0.5, lacking sustained improvement. Henan showed a pronounced downward trend overall. From a high of 1.19 in 2005, it steadily declined to just 0.32 in 2022, a steep drop of 73.07%, accelerating particularly after 2016. Meanwhile, Hebei’s TPCEE demonstrated a pronounced upward trajectory, increasing from 0.65 in 2005 to 1.02 in 2022—an increase of approximately 57%. The efficiency values frequently exceeded 1 after 2011. While Hebei’s TPCEE exhibited notable fluctuations during the study period, including declines in 2008 and 2016, potentially linked to phased environmental production restrictions and policy adjustments, it also exhibited a strong overall recovery capacity.

Regional perspectives

From a regional perspective, the TPCEE was highest in East China (1.0047), followed by North China (0.9842), South China (0.8241), Southwest China (0.6687), Northwest China (0.6345), Central China (0.6282) and Northeast China (0.5966). In Fig. 3(a), it can be observed that the density curves in North China and Southwest China appear to be trailing on the right side, with anomalously high values, while the left side of Northwest and Northeast China is clearly trailing, with anomalously low values appearing near 0.0. The peaks in East and North China were at the 1.0, and the TPCEE in this region was high. The peaks (main peaks) in Central, Southwest, Northwest, and Northeast China were between 0.5 and 0.75, and there was major room for efficiency improvement. North China, South China and Northwest China showed multiple peaks, and the efficiency distribution was non-uniformity. North China had the longest box and a high degree of dispersion between groups, indicating large differences in the TPCEE within the region, and the need to balance the green development of energy within the region.

Fig. 3.

Carbon emission efficiency of the thermal power generation sector in each region

Figure. 3(b) shows the time evolution of the TPCEE in each region. During the study period, the TPCEE of all the regions showed significant volatility. The efficiency values across all regions generally peaked in 2012, with East China reaching 1.1378 and Central China at 1.0117. Southwest, Northwest, and South China also achieved exceptionally high values within their respective ranges. Following 2012, efficiency declined significantly in most regions, particularly in Northeast China, where it dropped to 0.4387 in 2013. Central China’s efficiency dropped to 0.3906 in 2022, representing a 51.90% decline from 2005 and continuing its downward trajectory. This region exhibited the widest TPCEE range with an extreme value difference of 0.6211, indicating the highest TPCEE instability. This volatility primarily stemmed from efficiency fluctuations in Jiangxi and Henan provinces. The North China region consistently led in TPCEE, maintaining high annual averages, reaching 1.0842 in 2022 with an overall upward trend. East China’s TPCEE remained stable at elevated levels, rebounding to 1.0604 in 2022. South China gradually declined from 0.995 in 2005 to 0.6725 in 2022. Southwest China experienced significant fluctuations, rising to 0.8753 in 2017 before falling back to 0.5164 in 2022. Northwest China recorded 0.6874 in 2022, representing a 14.08% decrease from 2005. Northeast China’s TPCEE remained persistently low, reaching 0.516 in 2022, showing an overall gradual decline.

Spatial analysis of carbon emission efficiency in the thermal power sector

Figure. 4 shows the spatial patterns of the TPCEE in 2005, 2011, 2017, and 2022. Overall, a high TPCEE was more likely to appear in North China and coastal provinces; the distribution of low TPCEE values was decentralized, located in the Northwest, Northeast, Central, Southwest, and South China zones. The spatial distribution could be broadly summarized as follows. Eastern provinces generally exhibited high and contiguous TPCEE values. Most provinces in the central, western, and northeastern regions showed relatively low TPCEE values, forming extensive low-value zones. Within the northwest region, the TPCEE levels varied significantly among provinces, showing alternating patterns of high and low values with spatial discontinuity.

Fig. 4.

Spatial evolution of carbon emission efficiency in the thermal power sector across provinces

In 2005, Beijing, Jiangsu, Shanghai and Guangxi had high TPCEE values, whereas Xinjiang, Shaanxi, and Chongqing lagged behind. The high-efficiency areas formed a semi-ring from Qinghai to Hainan. In 2011, the pattern of the TPCEE changed to “low in the north and south; high in the middle,” and the spatial distribution was more clustered than that in 2005. The high-efficiency zones were located in Beijing-Tianjin, the Yangtze River Delta, and Guangdong Province. The northwest region showed a particularly significant shift, With the TPCEE of Qinghai, Gansu, and Ningxia dropping from high to low levels, whereas that of Shaanxi improved slightly. In 2017, Beijing and Yunnan formed a “northeast-southwest” high-efficiency zone, whereas low-efficiency zones were clustered in the northeastern, northwestern, and central regions, including Heilongjiang, Jilin, Xinjiang, Gansu, Chongqing, and Hunan. In 2022, the spatial pattern will change to an alternating “low-high” distribution from the south to the north. The high-TPCEE region extended from Xinjiang eastward to Inner Mongolia, and then downward to Guangdong Province, around the northern and eastern parts of China. The lowest TPCEE were mostly located in the outer region, concentrated in Gansu, Sichuan, Henan and other provinces.

Characteristics of temporal and spatial variation of carbon emission efficiency in the thermal power generation industry

Spatial relevance

Using Arcgis10.8, this study verified the spatial correlation of the research object, and the results are presented in Table 5. All provinces had spatial correlation characteristics, except in 2008, 2012, 2014 and 2016. Since 2018, Moran’s I has been greater than 0 and significant at the 1% level, the Z-value exceeded 1.96, and the spatial aggregation effect was significant. This indicates that the effectiveness of China’s regional coordinated emission reduction policy is remarkable, the diffusion of green technology and environmental protection regulation have increased, and spatial spillover effects played a key role in environmental governance. Of note, Moran’s I for the 2020–2022 period showed a slight decline compared to those in 2018 and 2019. This reflects a weakening of the spatial clustering effect in China’s TPCEE, with fluctuations in the synchrony of coordinated emission reductions across regions. The impact of the pandemic was likely the primary cause of this change.

Table 5.

Global moran’s I, Z-value and P-value

Year	Moran’s I	Z-value	P-value	Year	Moran’s I	Z-value	P-value
2005	0.09734	1.7445	0.0811	2014	0.0691	1.3819	0.1670
2006	0.1480	2.4010	0.0163	2015	0.1423	2.4331	0.0150
2007	0.2294	3.4588	0.0005	2016	0.0524	1.2134	0.2250
2008	0.0530	1.1589	0.2465	2017	0.0891	1.6974	0.0896
2009	0.1052	1.8531	0.0639	2018	0.3466	5.0971	0.0000
2010	0.1830	2.9146	0.0036	2019	0.3685	5.5677	0.0000
2011	0.3488	5.1461	0.0000	2020	0.3101	4.7339	0.0000
2012	0.0142	0.6771	0.4983	2021	0.1948	3.1677	0.0015
2013	0.2826	4.2361	0.0000	2022	0.2646	4.0650	0.0000

To further explore the spatial correlations, this study conducted a cluster analysis based on the TPCEE results. A high-high (H-H) cluster indicated that both the province and its surrounding provinces exhibited relatively high TPCEE values. A low-low (L-L) cluster indicated that both the province and its neighbors had relatively low efficiency scores. High-low (H-L) and low-high (L-H) clusters represented anomalous situations in which a province and its neighbors exhibited inconsistent states, specifically high efficiency in a province with low efficiency in surrounding areas, and low efficiency in a province with high efficiency in surrounding areas, respectively. As shown in Fig. 5, the 2005 national TPCEE data exhibited H-H clustering only in Hebei and Tianjin, marking the period with the fewest cluster types. By 2011, H-H clusters were concentrated in North and East China, whereas L-L clusters emerged in Qinghai, Northwest China. In 2017, H-H clusters persisted only in Hebei, whereas L-L clusters persisted in Qinghai. In 2022, H-H clusters primarily occurred in North China and Shandong, whereas L-L clusters were concentrated in Southwest China and Hunan. Overall, both H-H and L-L clusters exhibited an initial increase, followed by contraction, and then renewed growth. H-H clusters were predominantly distributed across North and East China, including Hebei, Tianjin, and Shandong, whereas L-L clusters were primarily found in Northwest and Southwest China, such as Qinghai and Guizhou. H-L clusters were more common in Guangxi and Sichuan, while L-H clusters were more readily identified in northern provinces such as Shanxi and Liaoning.

Fig. 5.

The clustering characteristics of carbon emission efficiency in China’s thermal power generation industry

Geographic description of LISA time trajectories

With reference to relevant studies [38], we produced Fig. 6. The relative length of LISA reveals the local spatial dependence and the stability of the spatial structure (Fig. 6(a)). A relative length >1 is greater than the national average, and the local spatial structure is more dynamic. There were 13, 16, and 11 provinces with relative lengths >1 in the three phases of 2006–2011, 2011–2017, and 2017–2022, accounting for 43.33%, 53.33%, and 36.67%, respectively. In terms of spatial distribution, the TPCEE in the Chongqing, Hubei, Guangdong, and Jilin regions remained relatively active.

Fig. 6.

Geometric characterization of the LISA time route

A larger value of curvature indicates that the TPCEE is more affected by the spatial role of the neighborhood, and the spatial dependence has greater volatility (Fig. 6(b)). The average curvatures of the three stages were 7.5136, 6.6481, and 5.5809, respectively. In 2006–2011, there are 10 provinces had curvature values that were higher than the average value (33.33%). In 2011–2017, there were 11 provinces with curvature values exceeding the average value (36.67%). In 2017–2022, there are 12 provinces had curvature values exceeding the average value (40.00%). The regions with high curvature (> 9) in the TPCEE LISA time paths were concentrated in Beijing, Chongqing, Inner Mongolia, Jiangxi, Shanxi, and Zhejiang, reflecting the high volatility of spatial dependence in these regions. Regions with lower curvature were mainly concentrated in Gansu, Liaoning and Shaanxi, reflecting the relatively stable spatial dependence variation in these regions.

The direction of movement revealed a synergistic change of in the TPCEE among the provinces, and the results are shown in Fig. 6(c). From 2006 to 2011, there were 18 provinces with synergistic growth, among which 10 (33.33%) showed positive synergistic growth and eight (26.67%) showed negative synergistic growth. Positive synergy growth was mainly observed in the Beijing-Tianjin-Wing and East China regions, whereas adverse synergy development was primarily observed in the northern and southern ends of China. From 2011 to 2017, there were still 18 provinces with synergy growth, of which eight (26.67%) showed positive synergy growth and 10 (33.33%) showed negative synergy growth. At this time, the Central and East China regions were dominated by negative synergies, whereas positive synergistic growth occurred in Northeast China, Inner Mongolia, Qinghai, and Guizhou. From 2017 to 2022, 10 provinces (33.33%) had positive synergies, and 7 provinces (23.33%) had negative synergies. Positive synergistic growth occurred mainly in Northeast, North China, and Northwest China, whereas negative synergistic growth was concentrated in South China. Overall, the number of provinces experiencing synergistic growth in the three phases was slightly more than 50%, with a stronger degree of spatial integration, greater percentage of positive synergies, and a greater positive spatial integration impact.

LISA time-shift matrix

The mutual transfer status of local spatial correlation types can be further analyzed using a spatiotemporal leap matrix. As can be seen from Table 6, the spatial correlation pattern of the TPCEE from 2006 to 2022 was relatively stable, with fewer leaps between the different types. Moreover, the spatiotemporal cohesion probabilities were all above 70%, showing a certain degree of transfer inertia as a whole. Among the four types of spatiotemporal leaps, Type I had the largest number of provinces with a higher degree of SC, indicating that the local spatial structure of the TPCEE in each province was characterized by evident path dependence or spatial locking throughout the study period.

Table 6.

Spatiotemporal leaps and probability matrices for different stages of CEE

Period of time	t/t + 1	HH	LH	LL	HL	Type	Quantity	Percentage	SC
2006–2011	HH	0.2611	0.0167	0.0056	0.0056	I	137	0.7611	0.7667
	LH	0.0278	0.0889	0.0333	0.0111	II	25	0.1389
	LL	0.0000	0.0333	0.3111	0.0333	III	14	0.0778
	HL	0.0056	0.0056	0.0611	0.1000	IV	4	0.0222
2011–2017	HH	0.2167	0.0556	0.0000	0.0111	I	129	0.7167	0.7167
	LH	0.0278	0.1111	0.0333	0.0000	II	34	0.1889
	LL	0.0000	0.0278	0.3222	0.0556	III	16	0.0889
	HL	0.0167	0.0056	0.0500	0.0667	IV	1	0.0056
2017–2022	HH	0.2400	0.0133	0.0000	0.0000	I	126	0.8400	0.8400
	LH	0.0200	0.1600	0.0333	0.0067	II	11	0.0733
	LL	0.0000	0.0467	0.3667	0.0267	III	12	0.0800
	HL	0.0000	0.0000	0.0133	0.0733	IV	1	0.0067

Analyzing the influencing factors of carbon emission efficiency in the thermal power generation industry

In this study, the multicollinearity among the influencing factors was tested and the results are presented in Table 7. The regression results showed that the VIF values < 5, indicating that the influencing factors did not exhibit multicollinearity and further analysis should be conducted.

Table 7.

Covariance diagnostics of influencing factors

Variable	VIF
IS	1.96
GM	1.38
LEI	2.10
SSTS	2.51
EI	2.30
LED	3.90
ECS	2.70
LGI	2.44

This study employed a Tobit model analysis to assess influencing factors at both the national and regional levels (see Table 8). The regression results from the Tobit model revealed that the determinants of China’s TPCEE exhibited distinct national commonalities alongside regional heterogeneity.

Table 8.

Tobit model results

Variable	National	Northeast China	North China	East China	South China	Central China	Northwest China	Southwest China
IS	0.5261** (0.014)	0.1465 (0.684)	1.2600*** (0.000)	1.2807*** (0.001)	0.7804 (0.143)	1.3057** (0.024)	0.0596 (0.948)		−0.4593 (0.638)
GM	0.1894* (0.034)	0.6538** (0.021)	0.1502 (0.662)	0.3618** (0.012)	0.5762*** (0.002)	−0.1155 (0.663)	0.1538 (0.402)		−0.2508*** (0.005)
LEI	0.0527 (0.989)	35.4351* (0.062)	8.9242 (0.101)	8.8639*** (0.000)	6.2390 (0.203)	20.0011 (0.440)	−36.1166 (0.154)		−2.8635 (0.877)
SSTS	−0.3385 (0.861)	6.8723 (0.197)	4.2285* (0.071)	−2.3837 (0.363)	23.9182*** (0.000)	−4.0852 (0.576)	0.0837 (0.987)		4.9136 (0.469)
EI	0.1505*** (0.000)	0.1076 (0.715)	0.3161*** (0.000)	0.0145 (0.891)	−0.3851 (0.228)	0.0754 (0.614)	0.0456 (0.365)		0.0741 (0.210)
LED	0.0000 (0.233)	0.0000 (0.808)	0.0000*** (0.000)	0.0000** (0.043)	0.0000 (0.226)	0.0000 (0.688)	0.0000 (0.298)		0.0000 (0.238)
ECS	− 0.0529 (0.825)	0.0121 (0.983)	2.1605*** (0.002)	−0.6861 (0.172)	−0.1569 (0.851)	−0.4308 (0.642)	−0.5544 (0.298)		1.6161** (0.016)
LGI	0.0000*** (0.001)	− 0.0000 (0.731)	−0.0000 (0.420)	0.0000** (0.047)	−0.0002*** (0.001)	−0.0001 (0.149)	0.0000*** (0.005)		−0.0002** (0.036)

* p < 0.1, ** p < 0.05, *** p < 0.01

At the national level, IS, GM, EI, and LGI exerted positive effects on the TPCEE at the 1% or 5% significance level. This indicated that they had stable promotional effects on the TPCEE. Other variables, such as ECS, did not show consistent significance at the national level, suggesting that their impacts may have been masked by regional heterogeneity.

The regression results across regions revealed significant variations in the influencing factors and their statistical significance. IS exhibited a significant positive impact in North, East, and Central China, indicating that these regions had effectively enhanced their TPCEE through industrial upgrading and agglomeration, as well as the diffusion of clean technologies. In Northeast, East, and South China, the positive effect of GM primarily stemmed from economies of scale and technological spillovers. However, in Southwest China, seasonal fluctuations in hydropower generation reduced the efficiency of thermal power peak regulations. Coupled with an imperfect clean-energy consumption mechanism, this led to a negative impact on GM. Regarding LEI, the positive driving effect in Northeast and East China was mainly attributed to technological spillover effects from the introduction of technological equipment.

Under the influence of SSTS, the TPCEE will further increase in South and North China, indicating that technological investments in these regions are precisely focused on low-carbon technologies for thermal power generation. Regarding EI, only North China exhibited a significant positive impact, suggesting that this region achieved economies of scale and technological synergy through energy efficiency upgrades. The effects in other regions were insignificant, potentially because of the limited room for efficiency improvement and delayed progress in upgrades.

LED technology will drive TPI emission reductions and efficiency gains in North and East China, where high-quality economic development synergizes with industrial upgrades. Economic growth is boosting the share of low-energy-consumption industries, such as high-end manufacturing and the digital economy, thereby reducing the absolute demand for high-energy-consuming thermal power. Contrastingly, other regions presented distinct challenges: Northeast China, undergoing economic transformation, directs more capital toward traditional industrial recovery, while underinvesting in low-carbon thermal power technologies; Southwest and Northwest China, with weaker economic foundations, see growth dominated by high-energy-consuming industries that increase thermal power demand; and South China, where economic growth is centered on services with limited ties to the thermal power sector, fails to generate direct drivers.

From the perspective of ECS, North China and Southwest China exhibited significant positive effects, indicating that increased per-capita electricity consumption in these regions has prompted thermal power enterprises to optimize production processes, adopt advanced energy-saving technologies, enhance energy utilization efficiency, and drive improvements in the TPCEE. In East and Northwest China, the TPCEE was enhanced owing to government intervention, reflecting the notable effectiveness of carbon reduction policies. Conversely, South and Southwest China face issues where policies are disconnected from regional realities.

Notably, the direction and significance of individual variables varied considerably across different regions. For instance, LGI exhibited a significant positive effect in East and Northwest China, yet a significant negative effect was observed in Southwest China. This indicated that the driving mechanisms of the TPCEE demonstrated strong regional dependence, necessitating tailored enhancement strategies that accounted for regional characteristics.

Discussion

Carbon emission efficiency of the thermal power generation industry and the theme of sustainable development

Fossil-fuel energy, which is inefficient and harmful, contributes to a significant emission of greenhouse gases and pollutants such as PM2.5, contributing greatly to the energy used by the TPI [40]. Carbon emission reduction in this sector is critical for advancing the “double carbon” goal. Therefore, considering environmental deterioration and global warming, carbon-related research and the optimization of the energy transition in the TPI have become a major research topics [41, 42]. Furthermore, many countries have recognized the importance of the TPI in human life, as well as the destruction of the natural environment, and have attempted to achieve coordination between the two from an industry perspective. For example, Germany has emphasized sustainability in the power industry and proposed eliminating the use coal in power generation by 2038 [43].

The low carbon economy concept considers the need to use fewer energy inputs to achieve increased economic production and to reduce the production of unwanted products, such as CO₂. This concept provides us with new ideas for measuring the TPCEE, facilitating the construction and application of a more comprehensive path for assessing the TPCEE. China’s TPCEE exhibited a declining trend during the study period, underscoring the need to implement structural reforms in energy supply, innovate electricity market mechanisms, and accelerate low-carbon transformation in the TPI [42, 44]. The TPCEE in East and North China (including Shanghai, Zhejiang, Jiangsu, and Beijing) consistently appeared to show high efficiency, consistent with the results of related studies [3, 13]. This confirms the validity of the results of this study. Regions with high LED, rapid green finance development, solid industrial foundations, and comprehensive industrial systems provide conditions conducive to the iterative upgrading of power generation technologies. For instance, in Shanghai, a pilot region for carbon emissions trading, local thermal power enterprises have improved their efficiencies via carbon quotas [45]. In addition, Beijing, as China’s capital, drives structural energy emission reductions and has achieved significant policy exemplification [18]. Regulations such as the Beijing Atmospheric Pollution Prevention and Control Ordinance strictly regulate air pollutant emissions. Conversely, regions such as Central China and Northwest China (e.g., Jiangxi and Henan) exhibit lower CEE, which is attributed to the poor technical operating components of equipment. This has resulted in increased CO₂ emissions, and these regions are economically underdeveloped and have low total industrial outputs [15]. Additionally, the low utilization of new energy sources is a major factor hindering improvements in the TPCEE, particularly in Jiangxi’s energy structure. Although the TPCEE in South and Southwest China has remained at a medium-to-high level, there is still room for improvement. This requires deepening the coordinated optimization of hydropower and thermal power to achieve efficiency enhancement and steady-state maintenance [46].

Synergistic carbon reduction strategies

China, with a large territory and notable regional variations, and thermal power plants in various regions are affected by a combination of factors, such as resource endowment and energy demand, occupying a unique functional position and performing activities focused on regional adaptability. According to Fang et al., power plants in southern China serve primarily provide electricity, while those in northern China are predominantly used to supply heating [15]. However, the spatial autocorrelation test and ESTDA showed that the Chinese TPCEE had a spatial clustering tendency and strong spatiotemporal cohesion. Although the intensity of this agglomeration characteristic has weakened to some extent since 2020 owing to the impact of the pandemic, its inherent agglomeration properties have not been fundamentally altered. This finding aligns with the conclusions of Guo et al. regarding the spatiotemporal distribution of China’s CO₂ emissions post-pandemic [47]. This suggests that a regional synergistic emission reduction strategy is essential for ensuring that the TPI realizes low-carbon transformation. In the future, a multifaceted or dual periodic communication system will be necessary for enhancing interregional economical and technical transfers and to direct the flow of capital, innovation, skills, and other factors from East and North China to the central, western, and northeastern regions [13].

The Tobit model accurately identified each region’s green shift factors and provided new ideas and perspectives for building a cross-regional green transition synergy and cooperation mechanism. For example, North and South China will benefit from increased scientific and technological innovation. They will effectively improve their TPCEE through measures such as increasing investment in low-carbon technologies and cleaner production processes, as well as establishing an innovation platform for cooperation among industry, academia, and research [48]. The carbon trading market and carbon offset policies should be actively promoted in East and North China, and an inter-regional joint system should be constructed to facilitate the industry-scale effect [2, 49]. Through multi-dimensional initiatives, such as regional synergistic cooperation, precise policy guidance, and the optimal allocation of factors, we will gradually move towards a new stage of low-carbon evolution and ecological sustainability of the TPI, achieving an ideal balance between economic growth and environmental safeguarding.

Applicability analysis

This study represents remarkable progress in the construction of indicators for method selection. We expanded the evaluation dimensions of the TPCEE, and analyzed its drivers more comprehensively; moreover, we did not utilize the traditional mode of spatiotemporal data analysis and, rather, adopted ESTDA.

In terms of methodology, the ESTDA-based analysis highlighted the time-dependence of the evolution of the TPCEE, uncovering the spatiotemporal coupling law of the TPCEE, and enabling a constant realization of the localized spatial dependency of the TPCEE from a “transient scenario” to a “dynamic scenario of spatial-temporal interaction” [30]. This allowed for a thorough exploration of the spatiotemporal characteristics of the samples. The Tobit model has the unique advantage of being able to handle truncated or censored data [1]. In addition, this study categorized the research objects based on regional differences and analyzed the influencing factors, thereby overcoming the limitations of the traditional model that makes it difficult to thoroughly analyze the heterogeneity of the factors influencing the TPCEE.

Research limitations

This study was conducted at the provincial level. Future research may extend to finer levels, such as prefecture-level cities and enterprises, to more precisely analyze the variations in the TPCEE and its influencing factors.

Summary

Findings

This study analyzed the spatiotemporal changes in China’s TPCEE, along with the influencing factors, and improved the theoretical system of the TPCEE to provide a reference for future research. Based on the findings of this study, the government can develop regionally specific energy-saving and emission-reduction plans and explore ways to enhance the TPCEE. Below are the findings:

(1) The TPCEE generally showed a fluctuating downward trend over the period 2005–2022, but a significant increase was observed in 2012 owing to strict energy-saving and emission reduction policies. High-CEE areas were primarily located in North China and coastal provinces, whereas low-CEE regions were scattered across Northwest, Northeast, Central, and Southwest China.

(2) The spatial pattern of the TPCEE was relatively stable, and the SC was above 70%, showing evident path dependence. In the future, interregional cooperation and communication should be strengthened to break the spatial locking effect, enhance the spatial spillover effect through the diffusion of green technology and environmental protection regulation linkages, and promote interregional synergistic emission reduction.

(3) At the national level, the IS, GM, EI, and LGI contributed to the overall improvements in the TPCEE. From a regional perspective, the impact of these factors on the TPCEE exhibited significant regional heterogeneity. This finding underscores the need for a differentiated policy design, requiring tailored emission reduction strategies that address the specific characteristics of different regions.

Policy recommendation

Nationwide efforts should be made to strengthen policy coordination and technical cooperation across regions, dismantle regional policy barriers, and promote the diffusion and sharing of green technologies. This can be achieved by enhancing data governance and building performance accountability systems. A unified national TPCEE monitoring platform should established to dynamically monitor efficiency changes across provinces and issue early warnings for regions experiencing three consecutive years of declining efficiency. CEE metrics should be integrated into the ecological civilization assessment systems of thermal power enterprises, and financial support and policy incentives should be provided to provinces and companies demonstrating significant efficiency improvements. Meanwhile, differentiated energy conservation and emission reduction policies should be implemented, and precise reduction strategies tailored to regional characteristics should be developed to drive balanced improvements in the TPCEE nationwide.

For provinces facing efficiency crises, such as Henan and Jiangxi, it is crucial to implement specialized provincial-level transformation plans. Henan should be included in the national key regulatory scope, with strict controls on new capacity additions in traditional high-energy-consuming industries such as those using nonferrous metals and building materials, while mandating energy-saving and carbon-reduction retrofits for existing capacity. Provincial green industry funds should be established to provide targeted support for the construction of high-efficiency ultra-supercritical units, cogeneration retrofits, and circular economy projects. Jiangxi should prioritize deploying distributed photovoltaic and decentralized wind power projects, explore piloting small-scale nuclear power units, and utilize interprovincial green electricity trading mechanisms to bridge clean energy supply gaps.

For regions in Northwest and Northeast China, where the TPCEE fluctuates at low levels and technical equipment is outdated, equipment upgrades and system flexibility retrofits are core breakthrough points. Northwest China should fully leverage its renewable energy endowment advantages, intensify renewable energy development, and increase the penetration rate of new energy sources such as wind and solar power, while optimizing thermal power load factors and reducing the operating hours of inefficient units. In addition, policy incentives should be strengthened to attract more technological innovation and capital investment, driving green transformation within the region. As a legacy industrial base, Northeast China faces considerable equipment aging and technological lock-in effects. Efforts should be made to intensify technical upgrades of outdated equipment, advance industrial restructuring, and reduce the share of high-energy-consumption industries. Policies should foster technological innovation and optimize resource allocation to enhance regional TPCEE. In addition, strengthened cooperation with North China and East China, given their successful experiences, would promote coordinated development across the broader area.

The TPCEE levels in North and East China remained relatively high. Efforts should be made to further consolidate their technological advantages, promote ultralow-emission retrofitting in high-energy-consumption industries, and enhance the dissemination and application of clean-energy development technologies. Emission reduction and efficiency gains should be achieved through the implementation and management of smart power plant systems and digital twin technologies. Additionally, pilot carbon emission trading programs should be advanced, carbon market mechanisms refined, and incentives provided to encourage enterprises to improve their CEE.

Southwest China boasts abundant hydropower resources; however, the coordination between hydropower and thermal power remains inadequate. Regional energy structure adjustments must be made to enhance the hydropower utilization efficiency and reduce the peak-shaving pressure on thermal power plants. South China should use incentives to encourage enterprises to adopt clean energy technologies and energy-saving emission-reduction measures, through carbon emissions trading and green finance policies. Cooperation between Hong Kong and Macao should be strengthened to introduce advanced technologies and management expertise, thereby enhancing the region’s TPCEE. Central China should bolster policy coordination and technological collaboration within the region to synchronize technological upgrades with demand growth. By optimizing the energy mix, the share of high-energy-consumption industries can be reduced while increasing the proportion of clean energy utilized. Concurrently, environmental oversight within regions must be strengthened to ensure effective policy implementation and drive improvements in regional TPCEE.

Author contributions

Yin Yan: Conceptualization, Writing - Original Draft, Writing - Review & Editing. Dalai Ma: Software, Formal analysis, Funding acquisition, Methodology, Resources. Chao Hu: Data Curation, Validation. Fengtai Zhang: Visualization, Project administration. Pengli Deng: Investigation. Kaihua Li: Supervision.

Funding

This work was financially supported by Chongqing Education Commission’s Humanities and Social Sciences project “Measurement and Evaluation of Chongqing’s High-quality Development” (No. 23SKGH248), and was supported by the Science and Technology Research Program of Chongqing Municipal Education Commission (Grant No. KJZD-K202501111).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participtae

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.