How Coupled Carbon Flows Reshape Urban Carbon Neutrality: Spatial Patterns and Differentiated Pathways Across Chinese Cities

Abstract

Urban carbon neutrality is increasingly shaped by cross-regional interactions rather than a closed balance between local emissions and sequestration. From an open-system perspective, this study conceptualizes urban carbon neutrality as the outcome of interactions between embodied carbon transfer (ECT) and carbon sequestration service flows (CSSFs). Using panel data for 297 Chinese cities in 2012, 2017, and 2022, an integrated measurement framework is developed to examine spatiotemporal patterns, typological heterogeneity, and driving mechanisms. The results reveal significant disparities in emission responsibility and ecological support across city types. Ecological conservation-oriented cities act as major carbon sequestration providers, while industrial- and service-oriented cities face higher emission pressures and weaker local sequestration capacity. The joint effects of ECT and CSSF reshape urban carbon neutrality through responsibility reallocation and ecological support transfer, enhancing overall performance while intensifying inter-city differentiation. Spatial Durbin model results indicate that carbon neutrality is jointly influenced by socioeconomic development, energy structure, factor mobility, ecological conditions, and institutional regulation, with both local and spillover effects. These findings suggest that urban carbon neutrality is a relational process embedded in production–consumption linkages and ecosystem service networks, highlighting the need for differentiated governance pathways to support coordinated mitigation and ecological compensation.

1. Introduction

Global climate change has become a central environmental challenge constraining the sustainable development of human society. The Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) indicates that, without more effective mitigation measures, the global average temperature could rise by 2.5–3.0 °C by the end of this century, posing profound risks to ecosystem security and socioeconomic stability [1]. The Paris Agreement established the global goal of limiting warming to well below 2 °C and pursuing efforts to keep it within 1.5 °C [2], making net-zero transitions a key direction of global climate governance [3]. Against this backdrop, how to scientifically assess regional low-carbon transition performance and develop evaluation frameworks that reflect actual carbon responsibility structures and ecological support capacities remains a critical research challenge.

A growing body of literature has examined net-zero transitions from multiple perspectives, including level assessment, scenario simulation, driver identification, and mitigation pathway optimization [4,5,6,7]. However, most existing studies have focused on national or provincial scales. Although recent research has gradually extended the analysis to finer spatial units [8,9], systematic city-scale investigations remain relatively limited. Cities are core spatial carriers of energy consumption and industrial activities, contributing more than 70% of global carbon emissions [10], and are key operational units for implementing low-carbon transition policies [11]. Meanwhile, substantial differences in development stage, industrial structure, resource endowment, and ecological foundation lead to pronounced heterogeneity in mitigation responsibilities, sequestration capacities, and transition pathways across cities. Therefore, establishing a city-scale assessment framework and identifying differentiated mechanisms are essential for optimizing inter-city responsibility sharing and promoting coordinated regional mitigation governance.

The central issue in carbon neutrality assessment lies in defining accounting boundaries and measurement approaches [12]. In this regard, carbon footprint theory provides an important foundation for moving beyond a single production-based accounting boundary. Wiedmann and Minx defined the carbon footprint as the total amount of carbon dioxide emissions directly and indirectly caused by a given population, system, or activity, emphasizing the supply-chain emission responsibilities embedded in consumption activities [13]. Hertwich and Peters further revealed the spatial transfer of carbon emission responsibilities across countries and regions from the perspective of global trade and consumption [14]. In recent years, city-scale carbon footprint studies have continued to advance. Existing studies have not only focused on the construction of urban carbon emission inventories and refined carbon monitoring but have also emphasized the practical value of megacity carbon footprint accounting for low-carbon governance and carbon neutrality pathway identification [15,16]. Meanwhile, carbon emission network research has further revealed interregional carbon emission linkages and spatial differentiation, providing a new analytical perspective for identifying cross-regional carbon responsibility transfer [17]. The resulting research tradition of consumption-based carbon accounting, urban carbon footprint assessment, and carbon emission network analysis provides an important theoretical basis for identifying the redistribution of emission responsibilities across regions. The embodied carbon transfer (ECT) examined in this study follows this research tradition and further characterizes the separation between production- and consumption-based responsibilities and the cross-regional flows of carbon responsibility at the city scale.

Existing carbon neutrality assessments generally adopt administrative units as accounting boundaries and measure regional carbon neutrality using the difference or ratio between “territorial emissions and local carbon sinks” [18,19]. This closed-system accounting logic implicitly relies on two theoretical assumptions: first, that carbon emission responsibility is entirely attributed to the location where emissions physically occur; and second, that the carbon sequestration function of ecosystems operates solely within local boundaries. However, in an economic system characterized by deep integration into global value chains and highly specialized regional division of labor, significant spatial separation commonly exists between production and consumption locations [20,21]. Economically developed and consumption-intensive regions tend to externalize part of their emission responsibilities to production regions through trade [22,23], while resource-based or industrial-oriented regions bear higher production-related emission pressures [24,25]. This mechanism of embodied carbon transfer (ECT) reconstructs emission responsibilities across space [26,27]. Persisting with closed-boundary accounting under such conditions may systematically underestimate the true carbon responsibility of consumption-driven regions. In recent years, several studies have further incorporated ECT results into carbon neutrality evaluation frameworks, arguing that both production-based direct emissions and net embodied carbon inflows should be jointly considered to better reflect actual carbon neutrality performance [28]. However, at the city scale, the coupled relationship between ECT and carbon neutrality assessment remains insufficiently explored and warrants further investigation.

Meanwhile, the carbon sequestration function provided by ecosystems also exhibits cross-regional spillover characteristics. CSSF has emerged as an important analytical tool for revealing cross-regional supply–demand relationships in carbon sequestration services [29,30] and was initially applied to identify sequestration service supply areas and demand areas. Regions with higher forest coverage or stronger ecological conservation functions not only absorb local carbon emissions but also provide sequestration support to highly urbanized and industrialized areas through ecosystem service flows [31,32]. Carbon sequestration demand is predominantly concentrated in highly urbanized and industrialized cities [33], enabling these regions to achieve carbon neutrality on paper while relying heavily on ecological support from external areas [34]. Neglecting such cross-regional ecological support flows may likewise lead to biased assessments of regional carbon neutrality. In recent years, CSSF-based approaches have begun to extend into carbon neutrality assessment [35]. However, the existing body of work remains limited in scope and is primarily conducted at national or provincial scales, making it difficult to reveal inter-city mismatches in sequestration supply and demand and their spatial transmission mechanisms.

In terms of driving mechanisms, carbon neutrality levels are considered to be jointly influenced by multiple factors, including economic development, energy structure, industrial structure, ecosystem conditions, and institutional factors. Empirical evidence suggests that simple economic scale expansion often leads to increased energy demand and higher carbon emissions [36], whereas energy structure optimization, improvements in energy efficiency, and industrial upgrading play significant roles in promoting carbon neutrality [37,38]. Urbanization, technological progress, and factor mobility have been found to exert heterogeneous effects on carbon neutrality across different development stages [39,40,41,42]. Meanwhile, ecosystem carbon sequestration capacity provides essential natural support for carbon neutrality [43,44], while institutional factors such as environmental regulation influence mitigation behavior through both constraint and incentive mechanisms [45]. In terms of analytical approaches, existing studies on identifying the drivers of carbon neutrality have mainly employed regression analysis [46,47], structural decomposition [48,49], machine learning and threshold models [50], as well as spatial econometric methods [51]. Among these, the spatial Durbin model (SDM) can simultaneously capture both local effects and spatial spillover effects of driving factors [52], making it particularly suitable for analyzing the cross-regional transmission mechanisms of urban carbon neutrality under the influence of ECT and carbon CSSF. However, overall, existing studies have paid insufficient attention to the integrated relationships among carbon responsibility transfer, ecological support flows, urban typological differences, and spatial spillover linkages.

Focusing on the core scientific question of how cross-regional carbon flows reshape urban carbon neutrality patterns, this study develops an analytical framework of “measurement construction–spatiotemporal evolution–mechanism identification–governance implications” under an open-system perspective. Using panel data for 297 Chinese cities in 2012, 2017, and 2022, the study proceeds in four steps (Figure 1). At the measurement construction stage, the conventional closed accounting logic of “local emissions–local sequestration” is extended by incorporating ECT and CSSF from the emission (source) and sequestration (sink) perspectives, respectively. Based on this, an open-system framework of urban carbon neutrality is developed to capture the balance between responsibility-adjusted emissions and ecological support capacity. At the spatiotemporal evolution stage, city typologies and temporal comparisons are employed to identify how cross-regional carbon flows reshape the spatial patterns and typological differentiation of urban carbon neutrality. At the mechanism identification stage, SDM is applied to examine both direct and spatial spillover effects from multiple dimensions, including economic development, energy structure, factor mobility, ecological conditions, and institutional regulation, thereby revealing the cross-regional transmission mechanisms of responsibility reallocation and ecological support. At the governance implication stage, based on the responsibility-sharing and ecological support relationships revealed by ECT and CSSF, this study synthesizes the differences across city types in terms of emission constraints, sequestration capacity, and interregional linkages, and proposes differentiated pathways for urban carbon neutrality. These findings provide a basis for regionally coordinated mitigation, horizontal ecological compensation, and targeted policy design.

Compared with existing studies, the marginal contributions of this study are reflected in the three aspects. First, at the measurement construction level, this study integrates ECT and CSSF into urban carbon neutrality assessment, thereby extending the evaluation logic from the closed-system perspective of “local emissions–local sequestration” to the open-system perspective of “responsibility adjustment–ecological support”. Second, at the research-scale level, this study conducts a city-level assessment covering 297 cities in China and extends the temporal scope to 2022, thereby deepening the understanding of the spatiotemporal evolution of urban carbon neutrality patterns in China. Third, at the mechanism-identification level, this study combines urban functional typology with spatial econometric modelling to examine heterogeneity in carbon responsibility structures and ecological support capacities across different city types, providing more targeted evidence for regional carbon governance and differentiated policy design.

2. Materials and Methods

2.1. Data Sources and Preprocessing

This study uses 297 Chinese cities as the research sample. The datasets mainly include socioeconomic data, energy data, carbon-emission data, city-level MRIO data, spatial ecological data, population-density data, and MRIO validation and robustness data. To improve the transparency and reproducibility of the data-processing procedures,

Table 1. Summary of datasets, indicators, sources, and preprocessing procedures.

Data Category	Main Indicators Used	Data Source/Reference	Pre-Processing and Derived Indicators
Socioeconomic data	GDP, population, urbanization rate, industrial structure, sectoral value added, household consumption, fixed asset investment, and related city-level indicators	Provincial and prefecture-level statistical yearbooks, including China City Statistical Yearbook and local statistical yearbooks	City names, administrative units, and measurement units were harmonized across years. Missing or inconsistent values were checked using provincial and local yearbooks. These data were used to construct socioeconomic variables and the growth factors for the 2022 MRIO table.
Energy data	Sectoral energy consumption, energy balance indicators, energy intensity, and carbon intensity	China Energy Statistical Yearbook and local statistical yearbooks	Energy indicators were matched to the six-sector classification. Sectoral energy consumption shares were calculated and used to allocate total city-level carbon emissions to sectors.
Carbon emission data	Total city-level CO2 emissions and sectoral CO2 emissions	China Emission Accounts and Datasets (CEADs), University College London, London, United Kingdom. Available online: https://www.ceads.net.cn (accessed on 3 June 2026).	CEADs provides total city-level CO2 emissions. Sectoral emissions were estimated by proportionally allocating total city emissions according to sectoral energy consumption shares.
City-level MRIO data	Intermediate transactions, total output, final demand, and interregional input–output linkages	City-level MRIO tables from CEADs; the 2022 table was constructed in this study based on the 2017 table and 2017–2022 statistical data	Original MRIO tables were adjusted to match the final analytical units. Cities with missing emission data or incomplete sectoral statistics were excluded or aggregated as described in Section 2.1. The original 42 sectors were aggregated into six sectors to align with energy data. The 2022 MRIO table was constructed using growth-factor extrapolation, RAS bi-proportional balancing, and provincial consistency adjustment.
Spatial ecological data	Land use, net primary productivity (NPP), and carbon sequestration-related indicators	Resource and Environment Science and Data Center of the Chinese Academy of Sciences, Beijing, China. Available online: https://www.resdc.cn (accessed on 3 June 2026).	Raster datasets were projected, resampled, and harmonized to a 1 km × 1 km spatial resolution. Values were extracted and aggregated by city boundary. These data were used to estimate local carbon sequestration capacity.
Population density data	Gridded population density	WorldPop, University of Southampton, Southampton, United Kingdom. Available online: https://www.worldpop.org/ (accessed on 3 June 2026)	Population density data were harmonized to a 1 km × 1 km resolution and aggregated to the city level. These data were used to characterize demand-side spatial distribution and support the calculation of carbon sequestration service flows.
MRIO validation and robustness data	Matrix balance, non-negativity, provincial consistency, and sensitivity analysis results	Constructed 2022 city-level MRIO table and official provincial/national statistical totals	The constructed 2022 MRIO table was validated by checking mathematical balance and non-negativity, comparing aggregated city-level results with provincial and national totals, and conducting ±5% and ±10% perturbation tests on technical coefficients. Detailed results are provided in Appendix B.

summarizes the main data categories, indicators, sources, and preprocessing methods used in this study.

Socioeconomic indicators were obtained from provincial- and prefecture-level statistical yearbooks for the years 2013, 2018, and 2023. Energy-related statistics were derived from the China Energy Statistical Yearbook (2013, 2018, and 2023 editions). Spatial datasets, including land-use data and net primary productivity (NPP), were obtained from the Resource and Environment Science and Data Center of the Chinese Academy of Sciences (https://www.resdc.cn/). Population density data were sourced from the WorldPop dataset (https://www.worldpop.org/). All raster-based spatial datasets were harmonized to a spatial resolution of 1 km × 1 km.

2.1.1. Carbon Emission and MRIO Data

Both carbon emission data and MRIO tables were obtained from the China Emission Accounts and Datasets (CEADs, https://www.ceads.net.cn/). The CEADs database provides total city-level carbon emissions but does not report sector-specific emissions. To derive sectoral carbon emissions, this study disaggregated total city emissions by applying proportional allocation based on sectoral energy consumption shares derived from city-level energy balance tables reported in local statistical yearbooks. Sectoral carbon emissions were thereby estimated for each city.

The city-level MRIO tables used in this study cover 309 cities across China (excluding Hong Kong, Macao, and Taiwan). Cities in Tibet were excluded due to missing carbon emission data. For several prefecture-level cities in Xinjiang where sectoral statistics are incomplete, MRIO data were aggregated and treated at the provincial level. In addition, because city-level MRIO tables are not yet available for Yunnan Province, Yunnan was also treated at the provincial scale. After these adjustments, the final analytical dataset includes MRIO tables for 297 city- or province-level units.

2.1.2. Sector Aggregation

The original city-level MRIO tables include 42 economic sectors. Following sector aggregation approaches adopted in prior studies [53], the sectors were consolidated into six major groups: (1) agriculture, forestry, animal husbandry, and fishery; (2) industry; (3) construction; (4) wholesale, retail, accommodation, and catering; (5) transportation, storage, and postal services; and (6) other services. This sector classification scheme is fully aligned with the six-sector energy consumption structure reported in the energy balance tables, ensuring consistency between intersectoral input–output relationships and sectoral energy use patterns (see Appendix A).

2.1.3. Construction of the 2022 City-Level MRIO Tables

Due to the absence of officially released city-level input–output data for 2022, this study constructs a six-sector city-level MRIO table for 2022 using the 2017 city-level MRIO table as the base year, combined with sectoral statistical data for 2017–2022. The construction follows a procedure of growth-factor extrapolation, RAS bi-proportional balancing, provincial consistency adjustment, and validation with sensitivity analysis. Specifically, following previous studies that assume relative stability of input–output technical coefficients in the short term, the direct consumption coefficients across cities and sectors in 2022 are assumed to largely retain the structural characteristics of 2017 [54]. First, sector-specific growth factors are constructed using statistical indicators such as value added of six major sectors, population, household consumption, and fixed asset investment over 2017–2022. These growth factors are then applied to extrapolate the base-year intermediate transaction matrix and final demand vector, generating an initial 2022 city-level input–output matrix. Subsequently, the RAS bi-proportional adjustment method is employed to balance the matrix across rows and columns. The city-level results are then aggregated to the provincial level and calibrated against official 2022 provincial statistics on sectoral total output, value added, and final demand to ensure consistency between the constructed city-level MRIO table and macro-level statistical accounts.

Considering that the assumption of relatively stable direct consumption coefficients from 2017 to 2022 is a key premise for constructing the 2022 city-level MRIO table, and that the COVID-19 pandemic during 2020–2022 may have disrupted interregional trade linkages, industrial chain structures, and final demand, this study further evaluates its reliability from three perspectives. First, a mathematical validity test is conducted to verify the balance and non-negativity of the extrapolated matrix. Second, a macro-level consistency check is performed by comparing aggregated city-level results with provincial and national statistical totals. Third, a sensitivity analysis is implemented by introducing ±5% and ±10% perturbations to the technical coefficients to examine the responsiveness of ECT and urban carbon neutrality levels to parameter changes. It should be noted that this sensitivity analysis is mainly used to examine the stability of the results when technical coefficients fluctuate within a certain range, and cannot fully simulate complex shocks such as the restructuring of regional trade patterns, industrial chain disruptions, and changes in demand structures during the pandemic. Therefore, the 2022 city-level MRIO table and the related carbon-flow estimates should be interpreted as estimated results derived from the available data and constrained balancing procedures. The results indicate that the constructed 2022 city-level MRIO table exhibits strong consistency with macro-level statistics, and the main findings remain robust under different perturbation scenarios. Detailed results are provided in Appendix B.

2.2. Methods

2.2.1. ECT Estimation Based on MRIO

The MRIO model has become a mainstream analytical tool for quantifying ECT, as it systematically captures intersectoral input–output relationships embedded in cross-regional trade and the associated carbon emission transfers Following the city-level ECT accounting approaches proposed in prior studies [55,56], this study estimates ECT among 297 Chinese cities using panel data for 2012, 2017, and 2022. Specifically, sectors are first aggregated into six major groups, and inter-city intermediate input matrices and final demand matrices are constructed to form the MRIO tables. Second, the direct requirement coefficient matrix is derived, and the Leontief inverse matrix is obtained based on the input–output balance relationship. Third, a sectoral carbon emission coefficient matrix is developed using city-level sectoral emission data, which is then combined with the Leontief inverse framework to quantify carbon emissions embodied in inter-city trade of intermediate and final goods. Finally, the resulting ECT matrix is aggregated by region to calculate, for each city, embodied carbon inflows, embodied carbon outflows, and net embodied carbon transfers (see Appendix C).

2.2.2. CSSF Estimation

The cross-regional transmission of carbon sequestration services is jointly shaped by socioeconomic conditions and ecosystem characteristics. Its spatial direction generally reflects regulatory flows from sequestration supply areas to emission-demand areas [20], and the transmission effect exhibits a clear distance-decay pattern as spatial separation increases [8]. Following established approaches in the ecosystem service flow literature [57], this study develops a city-level CSSF estimation framework from a supply–demand perspective. First, carbon sequestration service supply is estimated for each city based on NPP and carbon sequestration conversion coefficients. Carbon sequestration demand is proxied by city-level energy-related carbon emissions. Second, a supply–demand matching relationship is established to identify sequestration service supply areas and demand areas. Third, a breakpoint–field strength model is introduced to estimate the service flow radius, flow intensity, and transfer volume of CSSF from supply areas to demand areas. Finally, the inflow, outflow, and net flow of carbon sequestration services were calculated for each city to characterize the intercity spatial flow pattern of carbon sequestration services (see Appendix D).

2.2.3. Measurement of Urban Carbon Neutrality Under an Open-System Framework

Based on the estimation of urban carbon emissions, embodied carbon transfer, and carbon sequestration service flows, this study builds on the carbon neutrality measurement approach at the provincial scale [58] and extends it to the city level to develop an open-system framework for measuring urban carbon neutrality. Unlike the conventional closed accounting framework of “local emissions–local sequestration,” the open-system perspective emphasizes that urban carbon neutrality is shaped not only by local carbon budgets but also by cross-regional carbon flows. This approach provides a more realistic representation of urban carbon neutrality embedded in production–consumption networks and ecosystem service networks. Within this framework, ECT reallocates emission responsibility through production–consumption linkages, reflecting responsibility reconfiguration, while CSSF adjust the capacity of carbon sequestration support through ecosystem service supply–demand relationships, reflecting ecological support reconfiguration. Although these two processes differ conceptually, they operate on the emission (source) and sequestration (sink) sides, respectively, and can therefore be integrated within a unified “source–sink matching” framework to assess urban carbon neutrality.

Accordingly, the urban carbon neutrality level (Carbon Neutrality Level, CNL) is defined as:

$$CNL={\displaystyle \frac{CS+CSSF}{CE+ECT}}$$

(1)

where CNL denotes the urban carbon neutrality level; CS represents local ecosystem carbon sequestration within the city; CSSF denotes carbon sequestration service inflows received from other regions; CE represents territorial carbon emissions; and ECT denotes net embodied carbon transfer (defined as positive for net inflows and negative for net outflows).

When ECT > 0, it indicates a net inflow of embodied carbon, implying increased emission responsibility and thus being unfavorable for carbon neutrality; when ECT < 0, it indicates a net outflow of embodied carbon, reducing emission responsibility and thus being conducive to carbon neutrality. When CSSF > 0, it indicates a net inflow of carbon sequestration services, enhancing ecological support and promoting carbon neutrality; conversely, when CSSF < 0, it indicates a net outflow of carbon sequestration services, weakening local ecological support and thereby hindering carbon neutrality. As CNL is a ratio-based indicator, it may be sensitive to small denominator values, namely CE + ECT. We therefore checked all city-year observations and found no zero-denominator cases. In addition, a Winsorization-based robustness check was conducted to reduce the potential influence of extreme values on the econometric results.

When CNL > 1 it indicates that, after accounting for cross-regional carbon flows, the carbon sequestration support is sufficient to offset responsibility-adjusted emissions, implying that the city achieves carbon neutrality. When CNL < 1, it suggests that the city remains in a state of carbon overload. To characterize heterogeneity in carbon neutrality performance, cities are classified into six levels according to CNL values: Level I (high carbon overload) for CNL < 0.2; Level II (moderate carbon overload) for 0.2 ≤ CNL < 0.5; Level III (mild carbon overload) for 0.5 ≤ CNL < 1.0; Level IV (primary carbon neutrality) for 1.0 ≤ CNL < 1.5; Level V (intermediate carbon neutrality) for 1.5 ≤ CNL < 2.0; and Level VI (high-quality carbon neutrality) for CNL ≥ 2.0.

To reveal the synergistic mechanisms through which ECT and CSSF jointly influence urban carbon neutrality, this study further classifies urban carbon neutrality states based on the measured carbon neutrality levels (

Table 2. Classification of urban carbon neutrality types and their interpretations.

Type	Condition	Interpretation
Internal-spillover carbon-neutral	CNL ≥ 1, ECT ≤ 0, CSSF ≥ 0	Carbon neutrality has been achieved. The city exhibits net outflow of carbon responsibility and net inflow of ecological support, reflecting an “responsibility spillover–ecological benefit” pattern.
External-spillover carbon neutrality	CNL ≥ 1, ECT > 0, CSSF ≥ 0	Carbon neutrality has been achieved, but relies on external carbon responsibility transfer and ecological support; carbon neutrality is sustained based on external inputs.
Internal-spillover carbon-overload	CNL < 1, ECT ≤ 0, CSSF < 0	Carbon neutrality has not been achieved. The city shows net outflow of carbon responsibility and net loss of ecological support, forming a “dual outflow” pattern and facing high carbon pressure.
External-spillover carbon overload	CNL < 1, ECT > 0, CSSF < 0	Carbon neutrality has not been achieved. The city both receives external carbon responsibility and loses ecological support, facing a dual constraint of “responsibility inflow–ecological outflow”.

).

2.2.4. City Development Typology

To identify differences in development patterns across cities in terms of industrial structure, functional roles, and ecological endowments, this study classifies 297 Chinese cities into four types based on their functional attributes: ECO, INO, SVO, and CDO cities (Figure 2, see Appendix E for the full list). This classification was not conducted ex post based on the CNL results calculated in this study. Instead, it was determined by comprehensively considering factors such as urban ecological function, resource-based city attributes, industrial structure, service-function agglomeration, and comprehensive development capacity, so as to avoid circular interpretation between city typology and carbon neutrality measurement results. The list of cities is provided in Appendix E. It should be noted that comprehensive development-oriented cities have certain composite and residual-category characteristics, and their internal heterogeneity may be relatively high. Therefore, this study mainly focuses on the overall differences among functional city types, rather than treating this classification as a strict clustering boundary. Detailed classification criteria are provided in

Table 3. Criteria for urban development type classification.

Type	Classification Criteria	Representative Cities
Ecological conservation–oriented (ECO)	Located in national key ecological function zones, ecological conservation zones, or important ecological source areas; characterized by high shares of ecological land (e.g., forests and grasslands) and strong ecological protection and ecosystem service provision functions.	Gannan Tibetan Autonomous Prefecture, Hanzhong, Shangluo, Ankang, Aba Tibetan and Qiang Autonomous Prefecture, Yanbian Korean Autonomous Prefecture
Industrial-oriented (INO)	Characterized by resource-based attributes and a dominant secondary industry; typically associated with strong industrial production capacity and higher carbon emissions.	Tangshan, Baotou, Panzhihua, Dongying, Ordos, Zhuzhou
Service-oriented (SVO)	Characterized by a high share of the tertiary sector, high levels of urbanization and population concentration; serving as national or regional centers for comprehensive services, technological innovation, finance and trade, or administrative management.	Beijing, Shanghai, Guangzhou, Shenzhen, Hangzhou, Nanjing
Comprehensive development–oriented (CDO)	Do not meet the defining criteria of the above three types; characterized by relatively balanced industrial structures and a mix of industrial, service, and integrated development functions.	Suzhou, Wenzhou, Quanzhou, Jiaxing, Xiangyang, Guilin

.

2.2.5. Model Specification for Carbon Neutrality Drivers

Considering that regional carbon neutrality levels may exhibit spatial dependence and spillover effects, this study applies spatial econometric methods to systematically analyze the driving factors of urban carbon neutrality performance. First, Global Moran’s I is employed to test the spatial autocorrelation of carbon neutrality levels. The results indicate a significant positive spatial autocorrelation at the city scale throughout the study period, thereby justifying the use of spatial econometric models. On this basis, SDM is constructed to examine both local effects and spatial spillover effects of explanatory variables on urban carbon neutrality levels. The general specification of the model is given as follows:

$${CN}_{it}=\rho {WCN}_{it}+X_{it}\beta +{WX}_{it}\theta +{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(2)

where ${CN}_{it}$ denotes the carbon neutrality level of city i in year t; W is the spatial weight matrix; $\rho$ is the spatial autoregressive coefficient, capturing the spatial dependence of carbon neutrality across cities; $X_{it}$ is the vector of explanatory variables; ${\mu }_i$ and ${\lambda }_t$ represent city fixed effects and time fixed effects, respectively; and ${\epsilon }_{it}$ is the random error term.

Prior to spatial econometric modeling, spatial dependence tests were conducted based on OLS regression residuals (

Table 4. Diagnostic and specification tests for spatial econometric models.

Test Category	Test Name	Statistic	d.f.	p-Value	Conclusion
LM tests	LM-lag	84.695	1	0	Spatial lag dependence exists
	LM-error	25.034	1	0	Spatial error dependence exists
Robust LM tests	Robust LM-lag	2.512	1	0.113	Not significant
	Robust LM-error	36.227	1	0	Spatial error dependence dominates
Hausman test	FE vs. RE	14.67	—	0.0021	Fixed effects preferred
LR tests	LR (SDM → SAR)	15.874	5	0.0072	Reject simplification to SAR
	LR (SDM → SEM)	16.38	5	0.0058	Reject simplification to SEM
Wald tests	Wald spatial lag	136.78	k	0	Reject simplification to SAR
	Wald spatial error	100.55	k	0	Reject simplification to SEM

Notes: (1) LM and Robust LM tests are used to detect spatial dependence in the residuals of the non-spatial panel model. (2) The Robust LM tests help distinguish whether spatial dependence is primarily driven by spatial lag or spatial error effects. (3) The Hausman test is employed to choose between fixed effects (FE) and random effects (RE) specifications. (4) Likelihood ratio (LR) and Wald tests are conducted to examine whether the SDM can be simplified to the Spatial Autoregressive Model (SAR) or the Spatial Error Model (SEM). (5) k denotes the number of spatially lagged explanatory variables included in the SDM. (6) Statistical significance is evaluated at the 10%, 5%, and 1% levels.

). Both the LM-lag and LM-error statistics are significant at the 1% level, indicating the presence of strong spatial dependence in urban carbon neutrality levels and supporting the use of spatial econometric models. Further Robust LM tests show that Robust LM-error is significant whereas Robust LM-lag is not, suggesting that spatial dependence primarily arises from spatial error transmission mechanisms. On this basis, the SDM is adopted as the general specification, and its possible simplifications are examined using likelihood ratio (LR) and Wald tests. The results indicate that the null hypotheses that the SDM can be reduced to a spatial autoregressive model (SAR) or a spatial error model (SEM) are both rejected at conventional significance levels, confirming the appropriateness of the SDM specification. Hausman test results support the fixed-effects specification. Accordingly, this study employs an SDM with both city fixed effects and time fixed effects to examine the relationships between urban carbon neutrality and the explanatory variables, as well as their spatial spillover characteristics. It should be noted that the estimations are based on observational panel data; therefore, the results are primarily used to identify statistical associations and spatial transmission patterns rather than to establish strict causal inference. The baseline specification adopts a Queen contiguity matrix, and alternative spatial weight matrices are further employed for robustness checks to assess the sensitivity of the estimation results to different spatial adjacency specifications.

3. Results

3.1. Spatiotemporal Evolution of Carbon Emissions and Carbon Sequestration

3.1.1. Spatiotemporal Evolution of Carbon Emissions

From the perspective of spatial pattern evolution, the three-period carbon emission distribution (Figure 3) shows that high-emission cities were consistently concentrated in the Beijing–Tianjin–Hebei region, the Yangtze River Delta, and the Chengdu–Chongqing region during 2012–2022, forming stable high-emission core clusters. Among them, Shanghai, Chongqing, and Tianjin remained in the highest emission tier (>150 Mt) throughout the study period. In contrast, the southwestern mountainous areas, the northwestern plateau region, and the northeastern forest zone persistently exhibited low emission levels, revealing a clear spatial gradient characterized by a decline from east to west and from plains to mountainous areas. Marked differences are observed across city development types in both emission levels and growth rates. SVO cities consistently recorded the highest average carbon emissions, increasing from 74.01 Mt to 86.83 Mt, with a ten-year growth rate of 17.31%. INO cities ranked second, with average emissions rising from 36.28 Mt to 44.03 Mt, corresponding to a growth rate of 21.37%. CDO cities exhibited a pattern of medium scale but rapid growth, with average emissions increasing from 26.96 Mt to 35.48 Mt, representing the fastest growth rate among all four types (31.59%). By contrast, ECO cities maintained the lowest emission levels, with average emissions increasing only from 13.79 Mt to 15.91 Mt and a comparatively modest growth rate of 15.39%, indicating a stable low-emission, low-growth profile. Overall, urban carbon emissions across China display a clear stepwise hierarchy of SVO–INO–CDO–ECO, highlighting the strong influence of urban functional orientation and industrial structure on emission trajectories.

3.1.2. Spatiotemporal Evolution of Carbon Sequestration Capacity

From the perspective of spatial pattern evolution, the three-period distribution of carbon sequestration capacity (Figure 4) indicates a generally stable national pattern from 2012 to 2022, characterized by high-value cores in the southwest and northeast, low-value zones across the eastern and central plains, and a gradual nationwide increase in sequestration capacity. High-sequestration areas are concentrated in Aba, Ganzi, and Liangshan in Sichuan Province, Chongqing, Baise in Guangxi, Ganzhou in Jiangxi, Heyuan in Guangdong, Xilingol, Hulunbuir, and Chifeng in Inner Mongolia, as well as Heihe and the Greater Khingan forest region in Heilongjiang, where sequestration volumes consistently remain in the highest tier (>150 Mt). In contrast, densely urbanized areas such as the North China Plain, the Middle–Lower Yangtze Plain, and the Pearl River Delta exhibit relatively low sequestration capacity, mostly within the 0–20 Mt range. Clear and stable stratification is also observed across city development types. ECO cities consistently maintain substantially higher carbon sequestration levels than other types. CDO cities and SVO cities fall within the intermediate range, while INO cities show the lowest sequestration capacity. Specifically, the average sequestration volume of ECO cities increases steadily from 102.65 Mt to 111.43 Mt, corresponding to a growth rate of 8.55%, which is notably higher than that of other types. CDO cities show a modest increase from 36.76 Mt to 38.60 Mt (4.98%). SVO cities rise from 39.58 Mt to 42.44 Mt (7.22%). INO cities increase from 30.25 Mt to 32.58 Mt, with a growth rate of 7.72%.

3.2. Spatiotemporal Evolution of ECT Patterns

3.2.1. Spatial Characteristics of the ECT Scale Across City Types

During 2012–2022, inter-city ECT exhibits a structurally stable yet continuously expanding pattern at the national scale (Figure 5). CDO cities consistently function as the core hubs of embodied carbon circulation, with the internal ECT volume within this group increasing from 3774.80 Mt in 2012 to 5166.35 Mt in 2022. INO cities serve as the primary embodied carbon exporters. Their ECT outflows to CDO and SVO cities increase from 751.36 Mt and 206.64 Mt to 869.43 Mt and 235.62 Mt, respectively. ECO cities, by contrast, function as typical embodied carbon inflow areas, with ECT inflows from CDO and INO cities rising from 314.08 Mt and 244.76 Mt to 455.31 Mt and 293.05 Mt, respectively. SVO cities, characterized by high population density and intensive consumption activities, consistently exhibit net embodied carbon inflows. Their ECT inflows from CDO and INO cities grow from 506.40 Mt and 206.64 Mt to 590.07 Mt and 235.62 Mt, respectively. Overall, the national ECT network increasingly forms a clear transfer pathway from INO and CDO cities toward SVO and ECO cities, revealing a well-defined inter-city carbon flow structure.

3.2.2. Spatial Characteristics of Net ECT Across City Types

During 2012–2022, the net ECT pattern across city types remains highly stable and continues to intensify structurally (Figure 6a). SVO cities consistently constitute the most prominent net embodied carbon inflow group nationwide, with total net inflows increasing from 1101.81 Mt in 2012 to 1391.98 Mt in 2022, highlighting the strong dependence of high-population, high-consumption areas on external production systems. INO cities stably function as major net embodied carbon exporters, with net outflows of −564.60 Mt, −538.59 Mt, and −645.63 Mt across the three periods, indicating that energy-intensive industrial clusters continuously bear national embodied carbon supply responsibilities. CDO cities also represent major net outflow areas, with total net outflows expanding from −446.07 Mt to −666.78 Mt, reflecting their key role in national production and processing networks and associated emission burdens. ECO cities exhibit relatively small but persistent net outflows, with net transfers changing from −91.13 Mt to −79.57 Mt, suggesting that under strengthened ecological protection policies, these areas show a modest increase in reliance on external production supply. Overall, the national ECT structure maintains a stable configuration characterized by SVO cities as the principal consumption-side receivers, INO cities and CDO cities as the primary supply-side exporters, and ECO cities as a secondary supporting supply group, with this spatial responsibility division becoming progressively more pronounced over the past decade.

3.3. Spatiotemporal Patterns and Flows of CSSF

3.3.1. Spatial Patterns of Carbon Sequestration Service Supply–Demand Relationships

To reveal spatial differences in urban carbon sequestration service supply–demand relationships, this study uses net CSSF outflow as the core indicator and classifies cities into five categories: strong net supply, moderate net supply, near-balance, moderate net demand, and strong net demand (Figure 7). The results indicate pronounced spatial differentiation in CSSF supply–demand structures across Chinese cities. In terms of spatial distribution, the number of strong and moderate net supply cities remains relatively stable at around 120, with a persistent spatial pattern concentrated in the southwest, northeast, and parts of central–western China where ecological endowment is strong and forest and natural ecosystem coverage is high. These areas exhibit high forest and grassland coverage and strong ecosystem sequestration capacity, and consistently serve as major CSSF supply regions at the national scale. By contrast, the number of strong and moderate net demand cities increases from 128 to 138 and is highly clustered in the eastern coastal and central regions with dense population and economic activity, showing clear spatial agglomeration. These areas are characterized by high urbanization levels, intensive energy consumption, and high carbon emission intensity, with a steadily increasing dependence on external CSSF support. From a temporal perspective, CSSF supply capacity in major net outflow regions continues to strengthen, while sequestration deficits in major net inflow regions further expand. The east–west contrast becomes more pronounced, and the interregional imbalance in sequestration service supply and demand shows a trend of structural deepening.

3.3.2. Spatial Flow Characteristics of CSSF Across City Types

From the perspective of spatial flow structure, the three-period CSSF results (Figure 8) indicate that a highly stable and directionally structured inter-city sequestration service flow network has formed nationwide. CDO cities consistently act as the largest CSSF suppliers, with internal sequestration circulation increasing from 3194.51 Mt in 2012 to 3304.49 Mt in 2022, while CSSF outflows to other city types continue to expand. ECO cities, as major sequestration service supply areas, show steadily increasing CSSF outflows to CDO cities and SVO cities, rising from 1010.03 Mt and 129.18 Mt in 2012 to 1050.86 Mt and 144.95 Mt in 2022, respectively. By contrast, INO cities exhibit relatively limited CSSF supply capacity. Their CSSF outflows to CDO cities remain broadly stable at around 1080 Mt, while flows to SVO cities increase moderately from 137.22 Mt to 146.93 Mt. SVO cities function primarily as the receiving side of CSSF. CSSF inflows from CDO cities and ECO cities increase from 529.39 Mt and 75.44 Mt in 2012 to 543.23 Mt and 76.87 Mt in 2022, respectively. Overall, the national CSSF network exhibits a stable spatial flow configuration characterized by CDO cities and ECO cities as the principal supply sources and SVO cities as the primary receiving end.

3.3.3. Spatial Characteristics of Net CSSF Across City Types

During 2012–2022, the net CSSF pattern across city types remains structurally stable while showing clear differentiation (Figure 6b). ECO cities consistently function as net sequestration service suppliers, with net CSSF outflows expanding from −3998.97 Mt in 2012 to −4298.35 Mt in 2022, indicating a continuous increase in net supply capacity. SVO cities remain the principal net recipients of CSSF, with net inflows of 1380.33 Mt, 1331.42 Mt, and 1373.98 Mt across the three periods, consistently maintaining a high level of net inflow. INO cities also exhibit pronounced net inflow characteristics, with net CSSF increasing from 2312.54 Mt in 2012 to 2386.00 Mt in 2022. CDO cities occupy a relatively balanced intermediate position between supply and demand, with net CSSF rising from 306.10 Mt to 538.37 Mt over the study period. Overall, the national CSSF supply–demand structure remains broadly stable: ECO cities continue to expand their net supply role, whereas SVO cities and INO cities persist as major net demand centers, indicating a deepening spatial mismatch between ecological carbon sequestration supply and demand across regions.

3.4. Spatiotemporal Evolution of Carbon Neutrality Levels

3.4.1. Temporal Dynamics of Carbon Neutrality Grades

From the perspective of national spatial distribution, the three-period carbon neutrality level (CNL) results (Figure 9) show a clear stage-based evolution pattern across Chinese cities. In terms of CNL grades, 115 cities were classified as Level VI (high-quality carbon neutrality) in 2012, accounting for approximately 39% of all cities, suggesting a relatively strong overall CNL foundation at the national scale. However, about 27% of cities remained in carbon overload status (Level III and below). By 2017, the number of Level VI cities increased markedly to 151, representing more than 50% of the total, indicating a rapid nationwide improvement in CNL performance. At the same time, the share of carbon-overload cities rose to 35%, suggesting that emission pressures intensified in part of the urban system and that structural differentiation began to emerge. By 2022, the number of Level VI cities declined to 132 (44%), while the proportion of carbon-overload cities further increased to 38%. Nevertheless, cities classified within Levels IV–VI still accounted for 62% of all cities, indicating that the national CNL pattern shifted from mid-period rapid improvement toward late-period structural divergence.

From the perspective of city types, CNL differs substantially across functional categories. ECO cities consistently maintained the highest CNL performance, with most reaching Level VI by 2022, indicating that their carbon balance was primarily supported by local ecosystem sequestration capacity. CDO cities generally sustained relatively high CNL values, with Level VI cities remaining dominant overall, although carbon overload emerged in a subset of cities, leading to increasing internal heterogeneity. By contrast, SVO cities were predominantly in carbon overload status, mostly concentrated in Levels I–III, reflecting a typical pattern of limited sequestration support combined with strong consumption-driven spillover effects. INO cities, after incorporating ECT and CSSF into the accounting framework, showed comparatively high CNL grades overall, with more than 50% of cities reaching Level VI in all three periods. This pattern suggests that, as major embodied carbon exporters, some INO cities bear substantial production-side emissions while part of the consumption-driven responsibility is redistributed through cross-regional trade linkages. Overall, the inclusion of cross-regional carbon flows is associated with notable differences in the spatial pattern and type-specific interpretation of urban CNL under the open-system framework.

To examine the potential influence of extreme values in the CNL indicator, this study further conducted descriptive statistics for CNL values in 2012, 2017, and 2022 (

Table 5. Distributional characteristics of CNL in 2012, 2017, and 2022.

Year	Mean	Median	Min	Max	Std. Dev.	Skewness	P99	Number of Outlier Cities
2012	2.7599	1.5189	0.3598	30.5919	3.4562	3.6665	15.7813	3
2017	3.9183	2.0519	−10.0667	104.2136	8.9172	7.3815	28.5793	3
2022	3.5121	1.7136	−12.5106	94.1844	8.6639	7.6043	23.1902	3

). The results show that the mean CNL values in all three periods are higher than the corresponding medians, and the skewness values are all positive, indicating that the CNL distribution is clearly right-skewed and that a small number of high-value cities exert an upward influence on the overall distribution. Specifically, the skewness values of CNL in 2012, 2017, and 2022 are 3.6665, 7.3815, and 7.6043, respectively, suggesting that inter-city disparities in carbon neutrality levels became more pronounced in the latter two periods. Using the 99th percentile of each year as the threshold for identifying extreme-value cities, three such cities are identified in each period: Alxa League, Jiayuguan, and Qianjiang in 2012; Daxing’anling, Gannan, and Garzê in 2017; and Baise, Daxing’anling, and Garzê in 2022. Overall, the number of extreme-value cities is limited and has little influence on the classification of carbon neutrality levels for most cities or on the overall temporal evolution pattern.

3.4.2. Temporal Shifts in Carbon Neutrality Types

From a national perspective (Figure 10), the composition of urban carbon neutrality types in China shows significant structural divergence between 2012 and 2022. In terms of type distribution, internal-spillover carbon-neutral cities, which achieve carbon balance primarily through local emissions and local sequestration, decline markedly in number from 106 to 50, indicating that reliance solely on local ecological carrying capacity is becoming less common. By contrast, external-spillover carbon-neutral cities remain dominant and increase from 112 to 133. This suggests that these cities maintain carbon neutrality by reallocating emission responsibilities through cross-regional ECT while relying on local sequestration capacity, and they constitute the principal support for the overall improvement in national carbon neutrality performance. The number of internal-spillover carbon-overload cities remains relatively stable at around 60 throughout the study period, indicating that in these cities carbon overload is mainly driven by local emissions exceeding local sequestration capacity. Meanwhile, external-spillover carbon-overload cities increase substantially from 17 to 49, showing that under conditions of industrial chain restructuring and expanding consumption, a growing number of cities shift into carbon overload status due to outward emission responsibility transfer combined with insufficient sequestration capacity.

Clear differences are observed in carbon neutrality types across city development categories. ECO cities are dominated by internal-spillover carbon-neutral types, primarily achieving carbon balance through local ecosystem sequestration and territorial emission–sequestration equilibrium. INO cities exhibit a mixed pattern of external-spillover carbon neutrality and external-spillover carbon overload, indicating that they function both as major emission concentration areas and as key regions for outward shifting of emission responsibility through interregional linkages. CDO cities are mainly characterized by external-spillover carbon neutrality, reflecting their role in sharing emission responsibilities through cross-regional trade networks; however, a subset of CDO cities remains in internal-spillover carbon-overload status where local emissions exceed local sequestration capacity. SVO cities are predominantly classified as external-spillover carbon-overload types, indicating concentrated consumption activities combined with limited sequestration capacity, with emission responsibilities largely transferred outward.

3.4.3. Sources of Urban Carbon Neutrality

To enhance interpretability, this study decomposes urban carbon neutrality into four components: local carbon emissions, ECT, local carbon sequestration, and CSSF. Based on this decomposition, pathways to carbon neutrality for different city types are examined from the perspectives of carbon-source constraints and carbon-sink support. Specifically, local carbon emissions and ECT constitute the carbon-source side, while local carbon sequestration and CSSF represent the carbon-sink side. The relative contributions of these components to the two sides are quantified using a share-based approach (see Appendix G). It should be noted that, because ECT and CSSF may take negative values (i.e., net outflows) in some cities, the calculated shares may exceed 1 or be less than 0. This does not indicate computational errors but rather reflects the directional contributions of different components within the carbon-source and carbon-sink structures. When ECT is negative, it indicates the outward transfer of emission responsibility, thereby increasing the relative share of local carbon emissions on the carbon-source side. Similarly, when CSSF is negative, it indicates that the city is a net exporter of ecological services, which further strengthens the dominant role of local carbon sequestration on the carbon-sink side.

From the carbon-sink perspective (Figure 11a), ECO cities are primarily supported by local carbon sequestration, while CSSF remain consistently negative, indicating that these cities act as net exporters of ecological services and function as key “ecological supply centers.” For INO cities, the carbon-sink structure evolves from being dominated by local sequestration in the early stage to a combination of local and external support, with the share of CSSF increasing from 0.02 to 0.34, suggesting a growing reliance on cross-regional ecological support. SVO cities exhibit a continuous increase in the share of local sequestration (from 0.48 to 0.90), accompanied by a gradual decline in external dependence, reflecting an improvement in local ecological carrying capacity. CDO cities shift from net exporters of carbon sequestration services (negative CSSF) in the early stage to a more balanced structure (CSSF approaching zero), indicating a transition from an “outflow-dominated” to a “balanced” pattern.

From the carbon-source perspective (Figure 11b), INO cities consistently exhibit significant net embodied carbon outflows (ECT < 0), resulting in the share of local emissions remaining above 1 (approximately 1.75–1.82). This suggests that their relatively high carbon neutrality levels are associated with the redistribution of carbon responsibility driven by the separation of production and consumption, rather than solely by reductions in local emissions. In contrast, SVO cities are characterized by net embodied carbon inflows (ECT > 0), with emission responsibilities largely driven by consumption demand, reflecting a “consumption-driven” emission pattern. ECO and CDO cities both exhibit moderate net embodied carbon outflows (ECT < 0), though with smaller magnitudes, indicating that while they assume part of the production-related responsibility, their carbon balance is still primarily regulated by local emission–sequestration dynamics.

3.5. Analysis of Carbon Neutrality Drivers

3.5.1. Construction of the Indicator System for Influencing Factors

To identify the influencing factors of carbon neutrality and their spatial correlation characteristics, this study constructs an indicator system consisting of 12 variables across five dimensions: economic development, energy structure, factor flows, ecosystem, and institutional regulation (

Table 6. Indicator system for carbon neutrality drivers.

Dimension	Indicator	Variable	Description
Economic development	Economic development level	pgdp	GDP per capita (CNY), reflecting the regional development stage
	Urbanization level	urban	Share of urban population (%), indicating population agglomeration
	Spatial development intensity	built_den	Built-up area density (%), reflecting urban expansion intensity
Energy structure	Industrialization level	ind2	Share of secondary industry value added in GDP (%)
	Energy intensity	energy_int	Energy consumption per unit of GDP
	Carbon emission intensity	carbon_int	Carbon emissions per unit of GDP
Factor mobility	External openness	open	Total imports and exports as a share of GDP (%)
	Logistics development level	logistics	Freight volume as a share of GDP (%)
Ecosystem	Carbon sequestration capacity	forest	Forest coverage rate (%)
	Natural ecological conditions	precip	Mean annual precipitation (mm)
Institutional regulation	Environmental regulation intensity	env_reg	Environmental governance investment as a share of GDP (%)
	Technological innovation input	rd_int	R&D expenditure as a share of GDP (%)

Notes: Variables are constructed at the city level for 2012, 2017, and 2022. Continuous variables are logarithmically transformed where appropriate and tested for multicollinearity before SDM estimation. All spatial regressions include city and time fixed effects.

).

Specifically, in the economic development dimension, per capita GDP, urbanization rate, and built-up area density are selected to capture the relationship between the stage of economic development, spatial agglomeration characteristics, and carbon neutrality levels. In the energy structure dimension, the share of the secondary industry, energy intensity, and carbon emission intensity are used to reflect the relationship between industrial structure, energy use patterns, and urban carbon neutrality. In the factor flow dimension, the level of openness and logistics development are introduced to characterize the spatial linkages of capital, technology, and material flows. In the ecosystem dimension, forest coverage rate and annual precipitation are used to measure ecosystem carbon sequestration capacity and natural endowment conditions. In the institutional regulation dimension, environmental regulation intensity and technological innovation investment are employed to reflect the statistical relationships between policy constraints, innovation activities, and carbon neutrality.

3.5.2. Spatial Effects of Multidimensional Drivers on Carbon Neutrality

Based on the decomposition of direct and indirect effects derived from the SDM (

Table 7. Direct and indirect effects of carbon neutrality drivers across periods.

Variable	2012 Direct	2012 Indirect	2017 Direct	2017 Indirect	2022 Direct	2022 Indirect
ln_pgdp	0.019 (0.462)	−0.021 (0.703)	−0.062 (0.365)	0.069 (0.675)	−0.170 * (0.033)	0.059 (0.685)
urban	0.222 (0.283)	0.591 (0.159)	−1.359 *** (0.000)	−0.713 (0.357)	−1.761 *** (0.000)	0.176 (0.842)
ln_built_den	−0.120 (0.283)	−0.078 (0.628)	0.415 *** (0.006)	−0.337 (0.293)	−0.036 (0.830)	−0.636 ** (0.045)
ind2	1.821 *** (0.000)	−0.686 (0.167)	−0.261 (0.153)	−0.713 ** (0.016)	−0.775 * (0.061)	−1.986 ** (0.016)
ln_energy_int	−4.299 *** (0.000)	−0.043 (0.981)	−2.117 (0.237)	4.168 (0.381)	0.826 (0.687)	10.779 ** (0.037)
ln_carbon_int	3.663 *** (0.000)	0.147 (0.910)	1.932 (0.108)	−2.650 (0.398)	−0.378 (0.775)	−6.683 ** (0.038)
ln_open	−0.200 (0.343)	−0.367 (0.342)	−0.258 (0.396)	−0.474 (0.567)	−0.470 (0.153)	−0.394 (0.600)
ln_logistics	−0.157 ** (0.019)	0.145 (0.224)	−0.187 ** (0.026)	−0.153 (0.461)	−0.155 (0.122)	−0.332 (0.167)
ln_forest	−0.671 *** (0.006)	0.123 (0.707)	2.313 *** (0.000)	0.228 (0.696)	2.096 *** (0.000)	0.555 (0.311)
ln_precip	0.093 (0.365)	−0.027 (0.746)	−0.021 (0.879)	0.129 (0.532)	−0.082 (0.587)	0.197 (0.295)
env_reg	−0.807 (0.335)	−2.570 * (0.095)	−0.630 (0.235)	−1.311 (0.260)	0.554 (0.369)	−0.769 (0.624)
ln_rd_int	−7.052 (0.217)	−15.801 (0.158)	−1.593 (0.352)	−1.053 (0.762)	−3.536 (0.327)	−11.278 (0.214)

Notes: (1) Reported values are direct and indirect effects from the spatial Durbin model (SDM). (2) Parentheses report p-values. (3) *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. (4) Results are based on cross-sectional SDMs e mated by ML with the force option for 2012, 2017, and 2022.

), and consistently supported by robustness tests, urban carbon neutrality levels are jointly influenced by local factors and spatial spillover effects from neighboring cities. To systematically characterize the transmission mechanisms of multidimensional drivers, the results are further interpreted across five dimensions—economic development dimension, energy structure dimension, factor flow dimension, ecosystem dimension, and institutional regulation dimension—over different periods.

Economic development dimension. Overall, the relationships between economic development variables and CNL are not consistently stable, whereas urbanization rate and built-up area density show more pronounced associations. Specifically, per capita GDP is not statistically significant in any of the three periods, suggesting that differences in economic development level alone are not sufficient to explain variations in CNL. Urbanization rate shows significantly negative direct associations in 2017 and 2022, with the magnitude of the association increasing over time. This indicates that higher population concentration tends to be accompanied by greater residential energy demand and more intensive construction activities, which are associated with lower CNL. This pattern is more evident in SVO and CDO cities. Built-up area density shows a significantly positive direct association only in 2017, suggesting that moderate development intensity may be linked to improved land-use efficiency and lower carbon emission pressure, particularly in CDO cities. By 2022, however, it shows a significantly negative indirect association, indicating that higher development density in one city may be spatially linked to carbon pressure in neighboring cities through cross-regional construction and land-development activities.

Energy structure dimension. Overall, energy structure variables are significantly associated with urban CNL and show notable spatial spillover associations. Specifically, the share of the secondary industry shows a significantly positive direct association in 2012, but this relationship becomes significantly negative in 2022. Its indirect associations remain significantly negative in both 2017 and 2022, indicating a clear stage-dependent pattern. In the early period, industrialization is positively associated with CNL, whereas in later periods, as energy-intensive and emission-intensive characteristics become more prominent, the association shifts to negative. Through inter-city industrial linkages, a higher share of secondary industry is also spatially associated with lower CNL in neighboring cities, a pattern that is particularly evident in INO cities. Both energy intensity and carbon emission intensity show significantly negative direct associations in 2012, indicating that higher energy use and emission intensity are generally related to lower CNL. Over time, these direct associations weaken. By 2022, their relationships with CNL are more evident through spatial spillover associations, suggesting that high-energy- and high-emission cities are linked to increasing carbon pressure in surrounding areas.

Factor flow dimension. Overall, factor flow variables are mainly associated with urban CNL through local direct relationships, while cross-regional spillover associations remain relatively limited. Specifically, logistics development shows significantly negative direct associations in both 2012 and 2017, suggesting that the expansion of transport and logistics activities is typically accompanied by increased intra-urban energy consumption and carbon emissions and is therefore negatively associated with CNL. This pattern is more pronounced in INO and CDO cities. By contrast, openness does not show significant direct or indirect associations in any of the three periods, indicating that the expansion of import and export activities has not formed a stable statistical relationship with urban CNL.

Ecosystem dimension. Overall, ecosystem carbon sequestration capacity shows a relatively stable positive association with urban CNL. Specifically, forest coverage rate shows a significantly positive direct association, indicating that stronger ecosystem carbon sequestration capacity is generally related to higher CNL. This relationship is more pronounced in ECO cities, where forest resource endowments are stronger and carbon sink supply characteristics are more evident at the regional scale. In contrast, SVO and INO cities show greater dependence on external ecological carbon sinks. Annual precipitation does not show significant direct or indirect associations in any of the three periods.

Institutional regulation dimension. Overall, institutional regulation variables do not show consistently stable associations with urban CNL, while their spatial spillover associations are relatively more evident in the early period. Specifically, environmental regulation intensity does not show significant direct associations in any of the three periods, but it shows a significantly negative indirect association in 2012. This may reflect regional differences in the structure and effectiveness of environmental governance investment, and may also be related to carbon emission patterns in neighboring cities, especially in INO cities. Technological innovation investment shows no significant direct or indirect associations in any period, indicating that its relationship with urban CNL has not yet become statistically stable.

3.5.3. Robustness Tests

To examine the robustness of the spatial econometric results and to avoid potential influences arising from model specification, variable selection, and the skewed distribution of the dependent variable, this study conducted robustness tests from three aspects: the specification of spatial weight matrices, the selection of explanatory variables, and the treatment of extreme CNL values. Regarding the spatial weight matrices, since the identification of spatial dependence may be sensitive to the definition of neighborhood structures, the Spatial Durbin Model (SDM) was re-estimated using a rook contiguity matrix, a k-nearest-neighbor matrix (k = 6), and a distance-threshold-based spatial weight matrix (300 km), in addition to the baseline queen contiguity matrix. Regarding explanatory variables, considering that energy intensity and carbon emission intensity are conceptually close to the dependent variable, namely the carbon neutrality level (CNL), and may introduce potential mechanical correlation, alternative models excluding these variables were further estimated. Regarding extreme CNL values, considering the pronounced right-skewed distribution of CNL, the dependent variable was Winsorized at the 1st and 99th percentiles, and the SDM was re-estimated using the Winsorized CNL to reduce the potential influence of extreme observations.

The results show that, under different spatial weight matrix specifications and variable-selection conditions, the estimated results of the core explanatory variables remain generally consistent in terms of sign direction and association patterns, and the spatial spillover associations also show similar characteristics. This indicates that the conclusions of the baseline model are relatively robust. Although the significance levels of some variables fluctuate across different spatial weight matrices, these differences mainly arise from the different ways in which the spatial weight matrices characterize spatial association structures and ranges, and do not alter the main conclusions. Furthermore, after excluding energy intensity and carbon emission intensity from the baseline model, the SDM was re-estimated and compared using the queen, rook, k-nearest-neighbor, and distance-based matrices. The results show that, after excluding these conceptually related variables, the estimated results of the core explanatory variables remain generally stable across different spatial weight matrices, suggesting that the findings are not driven by potential conceptual overlap or mechanical correlations among variables. Moreover, the Winsorization-based robustness check shows that the maximum value of CNL decreased from 104.21 to 24.57, and the skewness decreased from 8.62 to 2.82 after Winsorization, indicating that the influence of extreme observations was substantially reduced. The re-estimated SDM results based on the Winsorized CNL remain broadly consistent with the baseline estimates in terms of the main coefficient directions and spatial spillover associations, suggesting that the main findings are not driven by a small number of extreme CNL values. Overall, the analysis of the factors associated with urban CNL and their spatial spillover associations shows a relatively high degree of reliability and robustness. The detailed results are reported in

Table A5. Robustness test results of the SDM under alternative spatial weight matrices.

Variables	Model 1: Queen Weights			Model 2: Rook Weights			Model 3: Distance Weights			Model 4: KNN-6 Weights
	SDM	Direct Effect	Indirect Effect	SDM	Direct Effect	Indirect Effect	SDM	Direct Effect	Indirect Effect	SDM	Direct Effect	Indirect Effect
pgdp	−0.0000 *** (0.0000)	−0.0000 * (0.0000)	0.0001 *** (0.0000)	−0.0000 * (0.0000)	−0.0000 (0.0000)	−0.0000 (0.0039)	−0.0000 ** (0.0000)	−0.0000 (0.0000)	0.0002 *** (0.0000)	−0.0000 ** (0.0000)	−0.0000 (0.0000)	0.0002 *** (0.0000)
open	−5.0524 *** (1.7560)	−4.8744 *** (1.7694)	2.5597 (5.0112)	−5.9431 *** (1.7712)	−5.6612 ** (2.2367)	79.7948 (410.0471)	−7.0511 *** (1.7402)	−6.5270 *** (1.7781)	10.1062 * (5.6687)	−7.0511 *** (1.7402)	−6.5270 *** (1.7781)	10.1062 * (5.6687)
precip	0.0017 (0.0021)	0.0016 (0.0019)	−0.0018 (0.0032)	0.0035 ** (0.0017)	0.0031 (0.0036)	−0.1329 (0.9268)	0.0056 *** (0.0020)	0.0052 *** (0.0018)	−0.0087 ** (0.0034)	0.0056 *** (0.0020)	0.0052*** (0.0018)	−0.0087 ** (0.0034)
env_reg	1.443 2 (4.2485)	3.1415 (3.9940)	27.9462 *** (8.7491)	5.5281 (4.3269)	6.6011 (7.4740)	364.0029 (1720.921)	7.8920 * (4.3627)	8.7137 ** (4.1009)	19.5723 * (10.1214)	7.8920 * (4.3627)	8.7137 ** (4.1009)	19.5723 * (10.1214)
ρ	0.4918 *** (0.0432)	0.4918 *** (0.0432)	0.4918 *** (0.0432)	0.7414 *** (0.1371)	0.7414 *** (0.1371)	0.7414 *** (0.1371)	0.5218 *** (0.0504)	0.5218 *** (0.0504)	0.5218 *** (0.0504)	0.5218 *** (0.0504)	0.5218 *** (0.0504)	0.5218 *** (0.0504)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	858	858	858	858	858	858	858	858	858	858	858	858

Notes: The table reports estimation results of the Spatial Durbin Model (SDM), including total (SDM), direct, and indirect effects. Standard errors are reported in parentheses. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. The Queen and Rook matrices are constructed based on spatial contiguity. The KNN matrix (k = 6) is based on k-nearest neighbors. The Distance matrix is defined using a spatial distance threshold of 300 km. All models include the same set of control variables, as well as city and year fixed effects. ρ denotes the spatial autoregressive coefficient.

,

Table A6. Robustness of the SDM after excluding conceptually overlapping variables.

Variables	Queen	Distance	KNN	Rook
pgdp	−0.0000 *** (0.0000)	−0.0000 * (0.0000)	−0.0000 ** (0.0000)	−0.0000 *** (0.0000)
open	−5.0984 *** (1.7522)	−6.0794 *** (1.7562)	−7.1695 *** (1.7304)	−5.0984 *** (1.7522)
precip	0.0016 (0.0021)	0.0034 ** (0.0017)	0.0055 *** (0.0020)	0.0016 (0.0021)
env_reg	1.1274 (4.2039)	5.1257 (4.2795)	7.4135 * (4.3209)	1.1274 (4.2039)
ρ	0.4909 *** (0.0432)	0.7424 *** (0.1363)	0.5227 *** (0.0503)	0.4909 *** (0.0432)
Controls	Yes	Yes	Yes	Yes
City fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
Observations	858	858	858	858

Notes: All models are estimated using the Spatial Durbin Model (SDM) with fixed effects after excluding energy intensity and carbon intensity. Robust standard errors are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

and

Table A7. Robustness of the SDM using Winsorized CNL as the dependent variable.

Variables	Main	Wx	Direct Effect	Indirect Effect	Total Effect
pgdp	−0.000006	0.000037 ***	−0.000001	0.000070 ***	0.000069 **
	−0.000006	(0.000012)	(0.000007)	(0.000025)	(0.000029)
urban	0.2055	−9.5474 ***	−1.2322	−19.2850 ***	−20.5171 ***
	−2.0302	−3.3272	−1.9294	−6.6901	−7.0868
built_den	0.3906	0.0775	0.464	0.5101	0.9741
	−0.345	−0.5859	−0.3318	−1.1084	−1.1948
ind2	1.3367	−3.1970 **	0.962	−4.9899 **	−4.0279 *
	−0.9944	−1.4454	−0.8938	−2.2496	−2.1198
open	0.1828	1.7315	0.4527	3.5654	4.0181
	−1.0592	−1.9081	−0.9998	−3.357	−3.5015
logistics	−0.0957 ***	0.0452	−0.0948 ***	−0.016	−0.1109
	−0.0362	−0.0594	−0.0358	−0.1094	−0.1181
rd_int	−6.2606	27.7569 **	−2.4995	53.4025 **	50.9029 *
	−8.0865	−13.8918	−8.1742	−25.0217	−26.2749
forest	7.7006	−22.5578	4.4319	−40.5099	−36.078
	−16.2701	−35.9921	−16.4828	−75.8412	−82.8181
precip	0.0008	−0.0009	0.0008	−0.0011	−0.0002
	−0.0012	−0.0015	−0.0011	−0.002	−0.0017
env_reg	1.6364	5.4975	2.6657	13.5770 *	16.2426 **
	−2.5508	−3.9559	−2.5139	−7.2799	−7.7407
rho	0.5545 ***
	−0.0368
sigma2_e	5.8282 ***
	−0.2887
City fixed effects	Yes
Year fixed effects	Yes
Observations	858

Notes: The dependent variable is the CNL Winsorized at the 1st and 99th percentiles. The model is estimated using the Spatial Durbin Model (SDM) with city and year fixed effects. Robust standard errors are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

.

3.6. Differentiated Urban Carbon Neutrality Governance Pathways from a Coupled-Flow Perspective

3.6.1. Responsibility Sharing and Ecological Compensation Under Coupled Flows

Existing studies suggest that cities at different development stages and with different functional types exhibit substantial heterogeneity in carbon neutrality processes, making it difficult to advance governance through a uniform pathway [59,60]. Building on this, the present study shows that, when both ECT and CSSF are considered, urban carbon neutrality is no longer solely a matter of local carbon balance, but rather the outcome of the joint effects of cross-regional redistribution of emission responsibilities and the spatial transfer of ecological support.

Accordingly, an open-system governance framework is required, in which carbon neutrality governance shifts from territorially bounded approaches toward a combination of responsibility sharing, ecological compensation, and differentiated strategies. Specifically, ECT captures the redistribution of carbon emission responsibilities between production and consumption locations, thereby providing a quantitative basis for cross-regional responsibility-sharing mechanisms. For cities with net embodied carbon outflows, it is important to continue assuming production-side mitigation responsibilities while establishing cooperative mechanisms with major consumption regions, including joint mitigation, cost-sharing, and responsibility accounting arrangements based on ECT results. In contrast, cities with net embodied carbon inflows should strengthen consumption-side responsibility constraints by promoting green procurement, low-carbon consumption, and supply chain management, thereby aligning mitigation responsibilities with their consumption demand.

Compared with ECT, CSSF reflects not only ecological support relationships but also quantifiable compensation linkages between supply and beneficiary regions. It thus provides an important basis for horizontal ecological compensation across regions. For areas with net outflows of carbon sequestration services, compensation mechanisms should be established based on the scale, direction, and intensity of CSSF, following the principle of “beneficiary pays.” Instruments such as interregional transfer payments, ecological compensation funds, and carbon sink benefit-sharing can help convert external ecological benefits into sustained support for ecological supply regions, thereby ensuring continued ecological investment and carbon sink stability. Conversely, regions with net inflows of carbon sequestration services should assume corresponding compensation responsibilities, supporting key supply areas through ecological co-investment, ecological product procurement, and carbon trading mechanisms in proportion to the benefits received. Taken together, CSSF provides a quantitative basis for determining compensation standards and implementation pathways. Overall, ECT underpins responsibility-sharing mechanisms between production and consumption regions, while CSSF supports horizontal ecological compensation between supply and beneficiary areas. Their integration constitutes the institutional foundation for urban carbon neutrality governance from an open-system perspective.

3.6.2. Differentiated Governance Pathways Across City Types

Building on the above analysis, differentiated governance pathways are required for different city types. INO cities are typically located at the net outflow end of embodied carbon and act as key nodes in the transmission of cross-regional emission pressures. Their governance priorities should move beyond territorially bounded mitigation toward a dual approach combining local structural optimization and cross-regional responsibility sharing. On the one hand, efforts should focus on reducing production-side emissions through industrial upgrading, improvements in energy efficiency, and technological retrofitting of energy-intensive sectors. On the other hand, these cities should establish cooperative mechanisms with major consumption regions, including carbon responsibility sharing, joint mitigation, and interregional transfer payment arrangements. SVO cities are generally positioned at the net inflow end of embodied carbon, with carbon constraints largely reflected on the demand side. Governance efforts should therefore prioritize consumption-based mitigation, including the promotion of green consumption, low-carbon transitions in buildings and transport, and supply chain carbon footprint management. Such measures can help align responsibility sharing among consumption centers, production regions, and ecological supply areas.

CDO cities perform multiple functions, including production, consumption, and factor coordination, and are more sensitive to changes in the coupled-flow structure. These cities should play a key role as regional coordination hubs. On the one hand, they need to balance local emission reduction with ecological restoration; on the other hand, leveraging their transport, industrial, and technological linkages, they can promote cross-regional joint governance, facilitate the diffusion of low-carbon technologies, and support the development of collaborative mitigation platforms, thereby enhancing overall regional carbon neutrality efficiency. ECO cities serve as major suppliers of carbon sequestration services. Their governance priorities should extend beyond ecological protection toward an integrated framework of “ecological conservation–value realization–benefit compensation.” This includes enhancing carbon sink stability, improving ecosystem quality, and establishing cross-regional ecological compensation mechanisms based on CSSF results. Policy instruments such as ecological product value realization and carbon sink benefit-sharing can help ensure the long-term sustainability of ecological supply functions.

4. Discussion

4.1. Differences in Urban Carbon Neutrality Patterns from Closed-System and Open-System Perspectives

The carbon neutrality distribution pattern constructed solely from territorial carbon emissions and local sequestration capacity (Figure 12) shows that under the closed-system accounting framework, high-carbon-overload cities are mainly concentrated in highly industrialized regions, whereas high-level carbon-neutral cities are primarily located in ecological function zones. After incorporating ECT and CSSF into the accounting framework, the urban carbon neutrality grading system undergoes a systematic reconfiguration (see Figure 9).

From the perspective of grade structure, under the territorial accounting boundary, the number of Level VI high-quality carbon-neutral cities remains relatively stable at approximately 110–114 across the three periods, accounting for about 37–38% of all cities. Under the carbon-flow-based accounting boundary, however, the share of Level VI cities increases to 51% in 2017 and 44% in 2022. Meanwhile, the proportion of Level I high carbon-overload cities rises by about six percentage points compared with the territorial accounting results, whereas the share of Level II–III cities declines from around 38% under the territorial boundary to 22–26% under the carbon-flow boundary. Overall, the inclusion of cross-regional carbon flows is associated with a grade structure characterized by expansion at both the high-neutrality and high-overload ends and contraction in the middle tiers. This suggests that open-system accounting does not simply raise or lower urban carbon neutrality classifications, but provides a more differentiated assessment by highlighting the roles of cross-regional carbon responsibility transfer and carbon sequestration contribution flows.

From the perspective of city types, changes in CNL grades differ markedly across functional city categories under the closed-system and open-system accounting frameworks. Under the local accounting framework, all ECO cities are classified as Level VI. After incorporating carbon flows, although some grade redistribution occurs in 2012, ECO cities remain dominated by Level VI in 2017 and 2022, indicating that they still have strong carbon sequestration supply advantages within the open-system framework. INO cities are mainly concentrated in Levels I–III under the local accounting framework, whereas the number of high-grade cities increases under the carbon-flow accounting framework. This suggests that, after incorporating cross-regional production–consumption responsibility adjustment, the CNL classification of some industrial cities changes. CDO cities show the most pronounced grade changes. Under the carbon-flow accounting framework, the number of Level VI cities remains relatively high, while the number of Level I cities also increases, showing a coexistence of high-grade and high-load categories. This indicates that CDO cities are relatively sensitive to the inclusion of ECT and CSSF. SVO cities are mainly concentrated in Levels I–III across all three periods, and the share of low-grade cities further increases under the carbon-flow accounting framework, suggesting that consumption-driven embodied carbon inflows may strengthen their carbon neutrality constraints.

Overall, the comparison between closed-system and open-system accounting frameworks indicates that a single local carbon balance approach is insufficient to fully reflect the actual position of cities in cross-regional production division, consumption responsibility, and ecosystem service provision. Incorporating ECT and CSSF into urban carbon neutrality assessment helps extend the evaluation perspective from “local carbon balance” to “coordination between cross-regional carbon responsibility and carbon sequestration contribution”, thereby enabling a more comprehensive identification of the functional roles of different cities within the regional carbon circulation system.

4.2. Spatial Transmission Pathways of ECT and CSSF in Urban Carbon Neutrality

The effect decomposition results of the SDM indicate that different driving factors influence urban carbon neutrality through two distinct carbon-flow channels—ECT and CSSF—and exhibit markedly different spatial transmission patterns. This dual-channel mechanism provides a process-based explanation for the observed reconfiguration of carbon neutrality grade structures under different accounting boundaries.

On the one hand, variables related to industrial structure, energy use, and factor mobility exhibit significant indirect (spillover) effects in multiple periods, indicating that their influences are primarily transmitted across regions through the ECT channel. Industrialization level, energy intensity, and carbon emission intensity tend to redistribute emission responsibilities from production locations to external demand regions through industrial division of labor and interregional trade networks. Urbanization and logistics development further intensify dependence on external resource and product inflows, thereby enlarging embodied carbon inflows. These findings suggest that ECT constitutes the core channel for cross-regional reallocation of carbon emission responsibilities, which is consistent with the observed downward grade shifts in INO cities and CDO cities under the carbon-flow-based accounting framework. By contrast, ecosystem-related variables mainly display stable positive local effects, while their spatial spillover effects are relatively limited. This indicates that ecosystem sequestration capacity contributes to carbon neutrality primarily by strengthening local carbon sink supply rather than through strong cross-regional spillover. The “flow” reflected by CSSF represents a reconfiguration of sequestration service supply–demand relationships at the regional scale rather than a physical diffusion of ecological attributes; therefore, its effects do not necessarily manifest as statistically significant spatial spillovers. This mechanism also helps explain why ECO cities maintain relatively high grades under both accounting boundaries while showing a moderate grade adjustment under the carbon-flow-based framework.

Further analysis by city development type reveals clear functional differentiation in carbon-flow mechanisms. INO cities play a more prominent role in the outward transmission of emission responsibility through the ECT pathway. CDO cities exhibit strong dual-channel coupling characteristics, simultaneously showing substantial engagement in both ECT redistribution and CSSF support linkages. ECO cities are primarily characterized by their CSSF supply function and serve as the main providers of cross-regional sequestration support. SVO cities, in contrast, are more strongly influenced by consumption-driven embodied carbon inflows. Overall, ECT and CSSF correspond to two distinct but complementary mechanism channels—cross-regional diffusion of emission responsibility and cross-regional support of carbon sink capacity, respectively—which together constitute a dual-path mechanism underlying the reconfiguration of urban carbon neutrality patterns.

4.3. Limitations

Despite constructing a coupled analytical framework integrating ECT and CSSF at the city scale, this study is subject to several limitations. First, due to data availability constraints, carbon flow estimations for cities in Xinjiang, Qinghai, Yunnan, and Hainan are based on provincial-level data, which may reduce the granularity of city-level analysis. In addition, the 2022 MRIO table is not officially published but constructed based on available data, which may introduce estimation uncertainties. In particular, the COVID-19 pandemic during 2020–2022 may have disrupted interregional trade linkages, industrial chain structures, and final demand. Existing sensitivity analyses may not fully capture such structural shocks; therefore, the 2022 results should be interpreted with caution.

Second, this study employs an SDM to identify the determinants of urban carbon neutrality and their spatial spillover effects. However, as the analysis is based on observational data, potential endogeneity issues cannot be fully ruled out. Therefore, the results are more appropriately interpreted as associations rather than strict causal relationships. In addition, the city typology is based on ex ante attributes such as ecological function, resource-based characteristics, industrial structure, and service functions, which may involve some subjectivity. This is particularly relevant for comprehensive development-oriented cities, which have composite and residual-category features. Future research could further validate the typology using methods such as k-means clustering or latent class analysis.

Third, as a ratio-based indicator, CNL may generate high values when responsibility-adjusted carbon emissions are relatively small. This study has reported the distributional characteristics of CNL and identified extreme-value cities in Section 3. The results show that the number of extreme-value observations is limited and has little influence on the overall pattern or on the classification of most cities.

Finally, this study mainly analyzes urban carbon neutrality levels and their spatial associations based on historical cross-sectional data, without further conducting future scenario simulation or policy intervention assessment. Future research could combine dynamic spatial weight matrices, multilayer network models, scenario simulation, and uncertainty analysis to further assess changes in urban carbon neutrality levels under different policy combinations and industrial transition pathways, thereby improving the robustness and policy applicability of the open-system assessment framework.

5. Conclusions

This study incorporates ECT and CSSF into an urban carbon neutrality assessment framework, characterizes the spatiotemporal evolution of carbon neutrality across 297 Chinese cities from 2012 to 2022, and identifies key driving factors using an SDM. The main conclusions are as follows:

(1) Clear and persistent differences are observed across city development types in both carbon emissions and sequestration capacity. SVO cities and INO cities exhibit relatively high emission levels but comparatively weak sequestration capacity. CDO cities show the fastest emission growth with limited improvement in sequestration, whereas ECO cities form the national carbon sink core, characterized by low emissions and strong sequestration capacity.

(2) The national carbon-flow system evolves into a stable bidirectional network structure. ECT is primarily transmitted from INO cities and CDO cities toward other city types, while CSSF is mainly supplied by ECO cities and CDO cities to support other regions. Together, these two flows shape a spatial carbon neutrality pattern characterized by simultaneous emission spillover and sequestration compensation.

(3) From 2012 to 2022, overall urban carbon neutrality levels shift from predominantly low–medium carbon-overload grades toward medium–high carbon-neutral grades, followed by increasing structural divergence in the later stage. ECO cities achieve and consistently maintain high-quality carbon neutrality. CDO cities improve overall but show widening internal disparities. SVO cities remain dominated by medium–low carbon-overload grades. INO cities display relatively high overall levels while simultaneously exhibiting both improvement and divergence.

(4) Urban carbon neutrality outcomes result from the joint effects of internal drivers—such as urbanization, industrial structure, and energy efficiency—and cross-regional carbon-flow mechanisms represented by ECT and CSSF. ECT intensifies the spatial reallocation of emission responsibility, whereas CSSF provides cross-regional sequestration support through ecological compensation mechanisms. Together, these two channels reshape the realization pathways of urban carbon neutrality.

(5) The coupled-flow framework provides a new analytical basis and policy guidance for differentiated urban carbon neutrality governance. INO cities should prioritize industrial upgrading and cross-regional responsibility sharing; SVO cities should strengthen consumption-side carbon constraints and promote green demand; CDO cities should function as regional coordination hubs; and ECO cities should focus on enhancing carbon sink stability and improving ecological compensation mechanisms. Together, these pathways support the formation of a categorized and regionally coordinated open-system governance model for urban carbon neutrality.