County-Level Energy-Related Carbon Emissions and Sustainable Low-Carbon Transition in the Central-Southern Liaoning Urban Agglomeration: Spatiotemporal Evolution and Spatial Spillover Effects

Abstract

For old industrial urban agglomerations, low-carbon planning requires emission information at a finer spatial scale, but county-level energy statistics are often incomplete. This study focuses on the Central-Southern Liaoning Urban Agglomeration, a typical heavy-industrial region in Northeast China. County-level energy-related carbon emissions for 73 units from 2005 to 2024 are reconstructed by combining socioeconomic panel data with harmonized DMSP-OLS-like nighttime light data. On this basis, global and local spatial autocorrelation, Moran scatterplots, Markov and spatial Markov transition matrices, and a spatial STIRPAT-based Spatial Durbin Model are used to examine the spatial pattern, transition process, and driving factors of emissions. The results show that emissions continued to increase during the study period, although the growth rate became slower and no clear regional peak was observed. Moran’s I rose from 0.627 in 2005 to 0.675 in 2024, which means that county-level emissions became more spatially clustered. The traditional Markov matrix shows strong state persistence, with diagonal probabilities ranging from 0.8793 to 0.9852. The spatial Markov results further suggest that counties surrounded by high-emission neighbors face greater pressure to move upward. In the SDM results, the spatial autoregressive coefficient is significant at the 1% level, with rho = 0.537. GDPPC and POP show negative direct effects, SEC increases local emissions but has a negative indirect effect, and PE is positively related to local emissions. Spatially, high-emission counties are mainly distributed around Shenyang, Anshan, Liaoyang, Dalian, and other industrial cores, while eastern ecological counties remain at relatively low emission levels. These findings provide county-scale evidence for differentiated low-carbon governance in old industrial regions.

1. Introduction

Carbon peaking before 2030 and carbon neutrality before 2060 have moved carbon mitigation from a national pledge into a concrete regional sustainability task [1]. In old industrial regions, this task is inseparable from industrial restructuring, cleaner energy substitution, land-use coordination, and the reorganization of urban-agglomeration functions [2,3]. Urban agglomerations concentrate people, capital, infrastructure, and energy demand; they also transmit mitigation pressure and industrial spillovers across administrative boundaries. The 2018 policy document on regional coordinated development stresses the role of central cities and interregional linkages in urban-agglomeration governance [4]. An urban-agglomeration perspective is therefore needed to connect China’s dual-carbon targets with spatially explicit sustainable-development policies.

The Central-Southern Liaoning Urban Agglomeration is a representative growth pole in Northeast China’s old industrial base. Its urbanization and industrialization have been shaped by steel, equipment manufacturing, petrochemicals, energy extraction, and port logistics. These sectors have supported regional output, yet they have also produced persistent pressure on energy consumption, ecological quality, and carbon intensity [5,6]. A county-level analysis that accounts for spatial spillovers can reveal where carbon risks are locked in, where low-emission ecological functions remain stable, and how dual-carbon policies can be differentiated for a sustainable transition in heavy-industrial regions [7,8].

Existing research offers useful but fragmented evidence. Urban-agglomeration studies emphasize regional integration, industrial division of labor, and collaborative mitigation [9,10,11]. Nighttime-light studies show that harmonized DMSP-OLS and SNPP-VIIRS products can approximate socioeconomic activity and emissions when conventional statistics are incomplete [12,13,14]. County-scale studies reveal heterogeneity that is obscured in provincial or prefecture-level data [15,16], while spatial econometric, LISA, and Markov approaches clarify spillovers, club convergence, and transition paths [17,18,19,20,21]. Recent work further links collaborative governance, functional division, urban form, innovation, traffic, land-use carbon balance, and multiscale accounting with geographically differentiated mitigation [22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39]. Together, these studies suggest that sustainable low-carbon governance should treat counties as connected units embedded in regional networks, rather than as isolated administrative observations.

Despite these advances, old heavy-industrial urban agglomerations in Northeast China remain underrepresented in county-level carbon research. Many studies still rely on national, provincial, or prefecture-city scales [9,10,11,12,13,14,15,16]. Fewer reconstruct county-level energy-related emissions and then examine both neighborhood-conditioned transition probabilities and spatial spillover mechanisms within the same old industrial base. The county scale is important here because steel, petrochemical, equipment-manufacturing, power-generation, and resource-processing activities create compact emission hotspots that can be diluted in coarser administrative averages.

This study fills that gap by reconstructing energy-related emissions for 73 county-level units in the Central-Southern Liaoning Urban Agglomeration and assessing how local emission states interact with surrounding emission environments. The contribution is threefold. First, the study links harmonized nighttime-light data with city-level energy accounting to provide a fine-scale emission dataset for a data-scarce old industrial region. Second, it combines spatial Markov transition analysis with a spatial STIRPAT-style SDM, allowing emission persistence, neighborhood pressure, and socioeconomic drivers to be interpreted together [19,20,21,40,41]. Third, it translates the spatial evidence into sustainability-oriented policy implications for core urban districts, heavy-industrial corridors, resource-based counties, and ecological peripheral areas.

In the current statistical system, energy-consumption data are usually reported at provincial or city levels, and county-level time series are rarely complete. The empirical design therefore calibrates nighttime light intensity against city-level energy-related carbon emissions and reallocates the city totals to 73 county-level units for 2005–2024. The model uses the updated Harvard Dataverse DMSP-OLS-like nighttime light product, including its SNPP-VIIRS-based extension. A spatial STIRPAT-style log-linear SDM identifies direct and spillover effects of socioeconomic drivers, while traditional and spatial Markov matrices describe movements among emission states. The resulting evidence supports carbon-accounting tools and place-specific mitigation measures for the sustainable transformation of China’s old industrial bases.

2. Materials and Methods

2.1. Research Methodology

2.1.1. Spatial Autocorrelation Analysis

Spatial autocorrelation analysis was employed to examine whether county-level carbon emissions were randomly distributed across space or exhibited clustering patterns. The global Moran’s I statistic was first calculated to assess the overall spatial dependence of carbon emissions within the study area. Its formula is given as follows:

$$I=\left[n\sum _{i=1}^n\left(x_i-\overline{x}\right)\sum _{j=1}^n{W}_{\mathrm{ij}}\left(x_j-\overline{x}\right)\right]/\left[\sum _{i=1}^n\left(x_i-\overline{x}\right)\sum _{i=1}^n\sum _{j=1}^n{W}_{\mathrm{ij}}\right]$$

(1)

where I denotes the global Moran index; n represents the number of county-level units; x_i and x_j denote the carbon emissions of counties i and j, respectively; $\overline{x}$ is the mean carbon emission level; and w_ij represents the spatial weight between the two counties. In this study, a first-order queen contiguity matrix was constructed based on shared county boundaries and subsequently row-standardized before analysis.

Although the global statistic can reveal the presence of spatial dependence at the regional level, it cannot identify the specific locations where clustering occurs. Therefore, Local Moran’s I was further applied to investigate local spatial associations. The statistic is expressed as follows:

$$I_i={\displaystyle \frac{(x_i - \overline{x})}{S^2}}{\displaystyle \sum _{j=1}^n}{w}_{ij}(x_j - \overline{x})$$

(2)

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

(3)

where S² denotes the sample variance, while the remaining variables are defined as above. Based on the sign and statistical significance of Local Moran’s I, counties were classified into four categories: High–High, Low–Low, High–Low, and Low–High. These categories were then used to identify local emission hotspots, low-emission clusters, and spatial outliers.

2.1.2. Spatial Econometric Models

County-level carbon emissions are shaped not only by local socioeconomic characteristics but also by the conditions of surrounding counties. Emissions in one county are therefore examined in relation to development patterns and emission levels in neighboring areas. To account for these spatial interactions, a Spatial Durbin Model (SDM) was employed in the econometric analysis. The SDM incorporates both the spatial lag of the dependent variable and the spatially lagged explanatory variables, allowing local influences and spatial spillover effects to be examined simultaneously. The model can be written as follows:

$$Y=\rho WY+X\beta +WX\delta +\epsilon$$

(4)

where Y is the column vector of the dependent variable, ρ is the spatial autoregressive coefficient of the dependent variable, W is the row-standardized queen contiguity matrix, X is the matrix of explanatory variables, WX represents the spatially lagged explanatory variables, δ is the corresponding coefficient vector, and ε is the random error term.

2.1.3. STIRPAT Model

The STIRPAT (Stochastic Impacts by Regression on Population, Affluence, and Technology) model extends the classical IPAT identity and has been widely used to examine how socioeconomic factors affect environmental pressure [38,39]. In the IPAT framework, environmental impact is related to population, affluence, and technology. Its traditional form is expressed as:

$$I=P\times A\times T$$

(5)

Dietz and Rosa extended the original model to the STIRPAT model, expressed as $\left(6\right)$:

$$C{E}_{\mathrm{it}}=\alpha \times P_{\mathrm{it}}^{\theta }\times A_{\mathrm{it}}^{\beta }\times T_{\mathrm{it}}^{\delta }\times {\mathcal{l}}_{\mathrm{it}}$$

(6)

In this study, the environmental factor considered is carbon emissions ($CE$), with $i$ denoting the county and $t$ the year. $P$, $A$, and $T$ retain the same meanings as in the original IPAT model. $\alpha$, $\theta$, $\beta$, and $\delta$ are parameters to be estimated, representing the influence of each variable, and ${\epsilon }_{it}$ is the stochastic error term.

Natural logarithms are taken for CE, POP, GDPPC, PE, and HS to make the regression coefficients more comparable, while SEC is retained as a share variable. The transformed STIRPAT equation is expressed as follows:

$$\ln C{E}_{it}=\ln \mathsf{\alpha }+\mathsf{\theta }\ln P_{it}+\mathsf{\beta }\ln A_{it}+\mathsf{\delta }\ln T_{it}+{\mathsf{\epsilon }}_{it}$$

(7)

The variables used in this paper are urban affluence (GDPPC), population size (POP), industrial structure (SEC), public expenditure (PE), and household savings (HS). Substituting these variables into the model gives:

$$l\ln C{E}_{it}=\ln \mathsf{\alpha }+{\mathsf{\beta }}_1\ln G DPP{C}_{it}+{\mathsf{\beta }}_2\ln P O{P}_{it}+{\mathsf{\beta }}_{3}SE{C}_{it}+{\mathsf{\beta }}_4\ln P E_{it}+{\mathsf{\beta }}_5\ln H S_{it}+{\mathsf{\epsilon }}_{it}$$

(8)

The log-linear STIRPAT specification is then combined with the spatially lagged dependent variable and spatially lagged explanatory variables to form a spatial STIRPAT-style SDM. This model is treated as a spatial econometric extension of the STIRPAT framework, not as a direct derivation from the original IPAT identity:

$$\begin{array}{cc}\mathrm{l}\mathrm{n} C{E}_{it}=& \rho W\mathrm{l}\mathrm{n} C{E}_{it}+\mathrm{l}\mathrm{n} \alpha +{\beta }_1\mathrm{l}\mathrm{n} GDPP{C}_{it}+{\beta }_2\mathrm{l}\mathrm{n} PO{P}_{it}+{\beta }_{3}SE{C}_{it}+{\beta }_4\mathrm{l}\mathrm{n} P{E}_{it}+{\beta }_5\mathrm{l}\mathrm{n} H{S}_{it}\\ & +{\delta }_{1}W\mathrm{l}\mathrm{n} GDPP{C}_{it}+{\delta }_{2}W\mathrm{l}\mathrm{n} PO{P}_{it}+{\delta }_{3}WSE{C}_{it}+{\delta }_{4}W\mathrm{l}\mathrm{n} P{E}_{it}+{\delta }_{5}W\mathrm{l}\mathrm{n} H{S}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}\end{array}$$

(9)

2.1.4. Markov Transition Probability Matrix

The Markov transition matrix was used to trace changes in county-level emission categories over time. Rather than focusing solely on increases or decreases in total emissions, this method identifies whether a county remained in its original emission category or moved to a higher or lower category during the study period.

• Traditional Markov Transition Probability Matrix

County-level energy-related carbon emissions were classified into four states according to the pooled-sample quartile thresholds of 0.25, 0.50, and 0.75. State 1 represents the lowest emission category, whereas State 4 represents the highest emission category. The same classification thresholds were applied throughout the study period to ensure comparability across years. The transition probability from state i in year t to state j in year t + 1 is calculated as follows:

$$P_{\mathrm{ij}}={\displaystyle \frac{z_{\mathrm{ij}}}{z_i}}$$

(10)

where $z_{ij}$ is the total number of counties that transition from type $i$ at time $t$ to type $j$ at time $t+1$ over the entire study period, and $z_i$ is the total number of counties in category $i$ during the study period.

2.

Spatial Markov Transition Probability Matrix

Although the traditional Markov matrix captures changes in a county’s own emission state, it does not account for the influence of neighboring areas. To incorporate spatial context, a spatial Markov transition matrix was constructed by introducing the emission characteristics of surrounding counties.

The spatial lag of carbon emissions was calculated as follows:

$$L_{it}={\displaystyle \sum _{j=1}^n}{w}_{ij}C{E}_{jt}$$

(11)

where $L_{it}$ denotes the spatial lag value of county i in year t; CEⱼₜ represents the energy-related carbon emissions of neighboring county j in year t; wᵢⱼ denotes the spatial weight between counties i and j; and n is the total number of county-level units. Before calculating the spatial lag values, the contiguity-based spatial weight matrix was row-standardized.

The resulting spatial lag values were then classified into four neighborhood categories using the pooled-sample quartile method. These categories represent low-, medium-low-, medium-high-, and high-emission neighborhood environments. A conditional transition matrix was subsequently estimated for each neighborhood category. The conditional transition probability is expressed as follows:

$$P_{ij\mid k}={\displaystyle \frac{n_{ij\mid k}}{n_{i\mid k}}}$$

(12)

where Pᵢⱼ_|ₖ denotes the probability that a county moves from state i to state j under neighborhood type k; nᵢⱼ_|ₖ represents the number of transitions from state i to state j within neighborhood type k; and nᵢ_|ₖ denotes the total number of observations initially belonging to state i under neighborhood type k. This framework makes it possible to compare the transition behavior of counties with the same initial emission state under different surrounding emission environments.

2.2. Study Area and Data Sources

2.2.1. Study Area

The Central-Southern Liaoning urban agglomeration is located in the central-southern part of Liaoning Province and is an important component of the Bohai Economic Rim. It includes ten prefecture-level cities: Shenyang, Dalian, Anshan, Dandong, Fushun, Benxi, Yingkou, Liaoyang, Tieling, and Panjin. The area covers about 96,000 km², approximately two-thirds of Liaoning’s land area. Figure 1 shows the spatial distribution of these cities (used ChatGPT (GPT-5.5) for image editing and refinement).

The Central-Southern Liaoning urban agglomeration has a long industrial history and a high level of urbanization, making it a major heavy-industry base in China. Shenyang and Dalian are the two main population and economic centers. Anshan, Benxi, and Liaoyang are closely related to steel and metallurgical production; Fushun, Panjin, and Yingkou are associated with petrochemicals, energy, port logistics, and resource processing; and equipment manufacturing remains important in Shenyang and surrounding cities. As a result, energy demand and carbon emissions are shaped not only by population and urbanization but also by heavy industry, power demand, transport corridors, and resource-based production systems.

Table 1. Socioeconomic status of the Central-Southern Liaoning Urban Agglomeration in Liaoning Province in 2024.

Indicators	Unit	CSL Urban Agglomeration	Liaoning Province	Share of CSL in Liaoning (%)
GDP	Yuan	2.84 × 1012	3.26 × 1012	87.2
Total Population	persons	3.04 × 107	4.07 × 107	74.6
Public Fiscal Revenue	10,000 yuan	2.33 × 107	2.75 × 107	84.6
Road Freight Volume	Tons	1.04 × 109	1.53 × 109	68.0
Total Imports and Exports	USD	10.57 × 1010	10.89 × 1010	97.1

summarizes the socioeconomic status of the region.

2.2.2. Data Sources

(1) Nighttime Light Data

Nighttime light (NTL) data provide an observable measure of the spatial intensity of human activities and are therefore useful for estimating carbon emissions in areas where small-scale energy statistics are incomplete. Existing NTL products mainly come from two satellite systems: the Operational Linescan System (OLS) of the Defense Meteorological Satellite Program (DMSP) and the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership satellite. In this study, we use the updated DMSP-OLS-like dataset from Harvard Dataverse (V7). This V7 product provides a 1992–2024 annual series by integrating historical DMSP-OLS observations with SNPP-VIIRS observations through inter-sensor harmonization; therefore, the 2020–2024 observations used in this study are derived from the V7 harmonized extension rather than from the original DMSP-OLS sensor. The product provides gridded data at a spatial resolution of 30 arc seconds. The DMSP-like DN values range from 0 to 63; values close to the upper bound were treated cautiously because saturation in dense urban cores can weaken the marginal response between light intensity and emissions.

Pixel values were aggregated to calculate the Total Digital Number (TDN) and Total Light Pixels (TLP) over time. The calculation is expressed as follows:

$$TDN=\sum D{N}_i\times C_i$$

(13)

where $D{N}_i$ denotes the DN value of a pixel, $C_i$ represents the number of pixels at location $i$, and TDN is the sum of all light pixel values.

(2) Carbon Emissions Accounting

The accounting boundary covers energy-related emissions from final energy consumption in the study area. Fossil-fuel consumption and purchased electricity are included, whereas industrial process emissions, agricultural emissions, forestry carbon fluxes, solid-waste emissions, and other non-energy greenhouse-gas sources are excluded because consistent county-level data are unavailable. The estimates should therefore be interpreted as energy-related carbon emissions, not as a complete regional greenhouse-gas inventory. Electricity is treated as a final-energy item using the coefficient in Table 2, and fuels used for power generation are not allocated again to counties as purchased-electricity emissions, reducing double-counting. The conversion and emission coefficients are listed in Table 2, with the original source acknowledged in the table note. The calculation is expressed as follows:

$$CE=\sum _{i=1}^{9}E{N}_i\times E{F}_i$$

(14)

where CE denotes total energy-related carbon emissions, EN_i is the amount of the i-th energy type after conversion into standard coal equivalent, EF_i is the carbon emission factor for that energy type, and i indexes the energy category. The coefficients used for the accounting procedure are reported in Table 2 and cited from Tian et al. [42]. Where China-specific factors were unavailable, IPCC default parameters were used only as supplementary references.

Table 2. Standard coal conversion coefficients and carbon emission factors used for energy-related carbon emission accounting [42].

Energy Type	Standard Coal Coefficient (tce per Physical Unit)	Carbon Emission Coefficient (t C/tce)
Coal	0.7143 tce/t	0.7559
Coke	0.9714 tce/t	0.8556
Crude oil	1.4286 tce/t	0.5860
Gasoline	1.4286 tce/t	0.6182
Kerosene	1.4714 tce/t	0.5538
Diesel oil	1.4714 tce/t	0.5743
Fuel oil	1.4574 tce/t	0.5918
Natural gas	1.33 × 10−3 tce/m3	0.4483
Electricity	0.3450 tce/103 kWh	0.2720

Note: Source: Adapted from Table A1 in Tian et al. [42]. The original table was published in Polish Journal of Environmental Studies under a Creative Commons Attribution-NonCommercial 4.0 International license.

Table 2. Standard coal conversion coefficients and carbon emission factors used for energy-related carbon emission accounting [42].

Energy Type	Standard Coal Coefficient (tce per Physical Unit)	Carbon Emission Coefficient (t C/tce)
Coal	0.7143 tce/t	0.7559
Coke	0.9714 tce/t	0.8556
Crude oil	1.4286 tce/t	0.5860
Gasoline	1.4286 tce/t	0.6182
Kerosene	1.4714 tce/t	0.5538
Diesel oil	1.4714 tce/t	0.5743
Fuel oil	1.4574 tce/t	0.5918
Natural gas	1.33 × 10−3 tce/m3	0.4483
Electricity	0.3450 tce/103 kWh	0.2720

Note: Source: Adapted from Table A1 in Tian et al. [42]. The original table was published in Polish Journal of Environmental Studies under a Creative Commons Attribution-NonCommercial 4.0 International license.

The nighttime light dataset was used to estimate energy-related carbon emissions in the Central-Southern Liaoning Urban Agglomeration from 2005 to 2024. Nighttime light intensity is treated as a proxy for the spatial distribution of human activity and energy demand, not as a direct observation of emissions. This distinction is important in a heavy-industrial region, where emissions from steel, petrochemical, power, equipment-manufacturing, and resource-processing activities do not necessarily increase linearly with observed light intensity. The calibration was therefore checked using city-level linear, zero-intercept, and log-linear specifications. The final proportional allocation was applied only after city-level energy-related emissions had been fixed, so the county-level estimates preserve the calculated city totals. The baseline relationship is expressed as follows:

$$y=kx$$

(15)

In Equation (15), TDN refers to the aggregated nighttime-light value within the study area, calculated by summing the NTL values of the county-level units. CE is the energy-based carbon emission estimate, and k is the fitted coefficient linking nighttime-light intensity with carbon emissions. Separate linear regressions were conducted for the ten prefecture-level cities. As shown in Figure 2, the R² values of the city-level linear models are between 0.7639 and 0.9435, while those of the log-linear models range from 0.7636 to 0.9348. These results suggest a generally close association between nighttime-light intensity and energy-related carbon emissions, although the strength of this relationship differs among cities.

After city-level categorical variables were introduced, the fixed-effects model achieved a high goodness of fit. The estimates show that nighttime light intensity remains strongly and positively related to energy-related carbon emissions after regional differences are controlled. Even so, the coefficients differ across cities, and some high-emission industrial facilities have weak or nonlinear light signals. For this reason, the county-level reconstruction is interpreted as an NTL-assisted allocation of city-level energy-related emissions rather than a direct measurement of all county-level greenhouse gas emissions.

Because the dependent variable is reconstructed with nighttime light information, mechanical correlation is a potential concern if the explanatory variables are themselves close proxies for nighttime light. Pairwise diagnostics were calculated using the raw county-level panel data to assess this risk. The correlation between ln(county NTL) and lnGDPPC is 0.402, and the correlation with lnHS is 0.280, indicating moderate associations, as summarized in

Table 3. Descriptive statistics of county-level panel variables, 2005–2024.

Variable	Unit	Mean	Std. Dev.	Min.	Max.
CE	10,000 tons	517.51	464.46	3.87	1519.93
POP	10,000 persons	41.92	24.30	6.49	115.00
GDPPC	yuan/person	57,564.47	45,109.51	3159.61	294,659.37
SEC	share	0.440	0.190	0.012	0.925
PE	10,000 yuan	230,610.07	292,281.97	8732.00	4822,656.00
HS	10,000 yuan	44,986.42	41,023.77	2077.08	287,861.44

Note: CE denotes energy-related carbon emissions; POP is year-end registered population; GDPPC is per capita GDP; SEC is the share of the secondary industry; PE is local government general public budget expenditure; HS is household savings. Statistics are calculated from 73 county-level units over 2005–2024 (N = 1460).

. By contrast, correlations with lnPOP (−0.034), SEC (−0.017), and lnPE (0.002) are weak. These diagnostics do not remove endogeneity concerns, but they indicate that the explanatory variables are not merely transformations of nighttime light. The SDM results are therefore interpreted as conditional associations rather than strict causal effects.

Following related studies, the intercept was fixed at zero to obtain a transparent proportional allocation coefficient. The final fitted equation is expressed as Equation (16):

$$SUMDN=0.04x$$

(16)

where $\mathrm{SUMDN}$ denotes the total nighttime light value for each county-level unit, and ${\mathrm{CO}}_2$ represents the estimated carbon emissions for the sample cities.

(3) Socio-economic Statistics Data

The study area was delineated using China’s administrative boundary data for the Central-Southern Liaoning urban agglomeration, and county-level units were identified from National Bureau of Statistics classifications. Energy activity data for 2005–2024 were collected mainly from the Liaoning Statistical Yearbook and the China Energy Statistical Yearbook. Socioeconomic variables were compiled from statistical yearbooks and official reports. The standard coal conversion and carbon-emission parameters used in Table 2 follow the coefficient set reported by Tian et al. [42], with China-specific greenhouse-gas factor databases and statistical guidelines used for consistency checks. The 2006 IPCC Guidelines for National Greenhouse Gas Inventories were consulted only where localized factors were not available.

Dependent variable: Energy-related carbon emissions (CE) of the Central-Southern Liaoning Urban Agglomeration. Because county-level energy-consumption data are limited, harmonized DMSP-OLS-like nighttime light data were used to establish a fitting relationship with city-level energy-related carbon emissions and to estimate county-level energy-related carbon emissions from 2005 to 2024.

Independent variables: GDPPC captures economic development and is measured as per capita GDP; POP represents year-end registered population; SEC denotes the proportion of secondary industry in GDP; PE is local general public budget expenditure; and HS is the year-end balance of household savings for urban and rural residents. Because HS is a stock of savings rather than a flow of income or consumption, it is used only as a proxy for household wealth and financial capacity [43].

3. Results

3.1. Spatiotemporal Characteristics of Energy-Related Carbon Emissions

County-level energy-related carbon emissions in the Central-Southern Liaoning Urban Agglomeration were visualized for 2005, 2011, 2017, and 2024 using the natural breaks (Jenks) classification method in ArcMap (version 10.8; Esri, Redlands, CA, USA). Emissions were divided into five categories: low, relatively low, medium, relatively high, and high emission regions, as shown in Figure 3.

Over 2005–2024, energy-related carbon emissions in the Central-Southern Liaoning urban agglomeration increased in stages. The trajectory changed from rapid expansion in the early period to slower growth in the later period. In 2005, regional emissions were still relatively low, with the county-level average at the scale of several hundred thousand tons. The spatial pattern was strongly polarized around Shenyang, and the increase in emissions during this stage was closely related to the revitalization of the old industrial base in Northeast China.

Around 2011, carbon emissions rose more rapidly as industrialization advanced. Total emissions were roughly 20–30% higher than in 2005. Emission intensity in core cities increased markedly during this period, reflecting the concentration of population and the expansion of secondary industry. By 2017, total emissions were still increasing, but the growth momentum had begun to weaken. Differences within the urban agglomeration also became clearer: central cities continued to maintain high emission levels because of stronger agglomeration effects, while many peripheral counties showed slower growth because of more limited development dynamics.

By 2024, total energy-related carbon emissions were still rising, but the growth rate had slowed substantially. The results therefore point to decelerating growth rather than a confirmed emission peak. At the prefecture-city scale, the reconstructed 2024 emissions were highest in Shenyang (64.27 million tons), followed by Yingkou (49.78 million tons), Dandong (49.46 million tons), Liaoyang (49.28 million tons), Anshan (48.47 million tons), Dalian (48.28 million tons), Tieling (47.87 million tons), Benxi (47.56 million tons), Panjin (46.77 million tons), and Fushun (45.81 million tons). These city-level totals provide a scale reference for interpreting the spatial distribution in Figure 3.

The temporal pattern is consistent with the socioeconomic structure of the region, the model estimates, and findings from existing studies on industrial agglomeration and county-level carbon emissions [7,8,10,12,13]. The Central-Southern Liaoning Urban Agglomeration is an old industrial base with concentrations of steel, petrochemical, machinery, and other energy-intensive industries. Table 1 shows that the urban agglomeration accounts for a large share of Liaoning Province’s GDP, population, fiscal revenue, freight volume, and trade activity. The descriptive statistics in Table 3 also show substantial cross-county differences in energy-related emissions, economic development, fiscal expenditure, and household savings, which supports the analysis of spatially differentiated emission patterns.

Spatially, carbon emissions shifted from a relatively scattered pattern toward stronger clustering and then to a more stable configuration. In 2005, the distribution was still fragmented and emission levels were generally low. High-emission areas were mainly found in a few core locations, such as the central districts of Shenyang and traditional industrial areas such as Tiedong District in Anshan. Eastern cities, including Dandong and Fushun, were mostly composed of low-emission counties, with weak spatial connections between them. At this stage, the overall pattern was shaped more by the individual development conditions of administrative units than by strong interregional linkages, although a basic west-high/east-low gradient had already appeared.

By around 2011, the emission pattern showed clearer spatial connections. Some previously concentrated high-emission areas expanded outward, especially in the Huanggu and Hunnan districts of Shenyang and the central districts of Dalian. The Shenyang–Liaoyang–Anshan corridor developed into a continuous high-emission band. Eastern counties such as Huanren maintained lower emission levels than the western and central industrial areas. This redistribution is consistent with the stronger spatial autocorrelation reported in

Table 4. Global Moran’s I indices of county-level carbon emissions in the Central-Southern Liaoning Urban Agglomeration, 2005–2024.

Year	Moran’s I	Z	p-Value
2005	0.627	7.802	0.000
2006	0.627	7.800	0.000
2007	0.626	7.789	0.000
2008	0.621	7.729	0.000
2009	0.630	7.830	0.000
2010	0.623	7.744	0.000
2011	0.624	7.753	0.000
2012	0.624	7.753	0.000
2013	0.620	7.691	0.000
2014	0.642	7.964	0.000
2015	0.653	8.086	0.000
2016	0.647	8.019	0.000
2017	0.661	8.197	0.000
2018	0.671	8.328	0.000
2019	0.675	8.385	0.000
2020	0.655	8.133	0.000
2021	0.661	8.216	0.000
2022	0.663	8.245	0.000
2023	0.662	8.247	0.000
2024	0.675	8.385	0.000

and the expansion of High-High clusters in the local spatial autocorrelation results.

After 2017, the spatial pattern became more closely aligned with the main development corridors. The high-emission pattern along the Shenyang-Dalian axis gradually formed a belt-like structure, and contiguous high-emission areas expanded. This corridor-like clustering is consistent with prior evidence that industrial corridors, port logistics, and urban agglomeration effects can reinforce spatial dependence in carbon emissions [10,12,18,20]. The local spatial autocorrelation analysis shows that eastern counties such as Kuandian and Huanren maintained lower emission levels than the western industrial core. High-emission and low-emission areas therefore became more clustered, suggesting stronger spillover effects from core cities and industrial corridors.

In 2024, the spatial pattern was relatively stable. High-emission zones were concentrated around major urban and industrial hubs, especially Shenyang, Dalian, and Anshan, forming a larger high-emission cluster. Peripheral counties such as Qingyuan, Xinbin, Huanren, Kuandian, and parts of northern Tieling remained in lower-emission or more stable categories. The regional distribution can therefore be understood as a combination of relatively stable central industrial zones and peripheral ecological or agricultural zones with slower changes. The gap between the western and eastern parts of the region continued to widen, and spatial dependence strengthened.

3.2. Analysis of Factors Influencing County-Level Carbon Emissions in the Central-Southern Liaoning Urban Agglomeration

3.2.1. Global Spatial Autocorrelation

The global Moran’s I results reveal a persistent spatial dependence in county-level carbon emissions during 2005–2024. As reported in Table 4, all Moran’s I values are positive and remain above 0.6, while the corresponding p-values are lower than 0.01. This confirms that counties with similar emission levels tended to be spatially clustered rather than randomly distributed.

In terms of temporal change, Moran’s I rose from 0.627 in 2005 to 0.675 in 2024. The increase is not large in annual terms, but it still points to a gradual strengthening of spatial association among county-level energy-related carbon emissions over the study period.

3.2.2. Local Spatial Autocorrelation

Local spatial autocorrelation was further examined to identify where the regional clustering pattern was mainly located. Moran scatterplots were produced in Stata (version 17; StataCorp LLC, College Station, TX, USA) for 2005, 2011, 2017, and 2024, as shown in Figure 4. In these plots, the horizontal axis represents standardized county-level carbon emissions, while the vertical axis represents the spatial lag of emissions. The four quadrants correspond to High-High, Low-Low, Low-High, and High-Low spatial association types.

The scatterplots are broadly consistent with the global Moran’s I results. County-level emissions were not randomly distributed, but showed a clear tendency to cluster with nearby counties. Moran’s I increased from 0.627 in 2005 to 0.675 in 2024, indicating that the spatial linkage among counties became stronger over time. This does not mean that all counties changed in the same direction, but it does show that the emission pattern became more dependent on neighboring areas.

High-emission clusters were mainly located in the western and central industrial belt. In 2005, they were concentrated around the central districts of Shenyang and the traditional industrial areas of Anshan. By 2011, the high-emission pattern had begun to extend outward from Shenyang, and the central districts and coastal counties of Dalian became another important pole. From 2017 to 2024, several industrial counties along the Shenyang–Anshan–Liaoyang–Yingkou axis became more closely connected, forming a more continuous high-emission corridor. This pattern reflects the influence of heavy industry, transport corridors, port logistics, and urban agglomeration effects.

Low-emission clusters were more stable over time. They were mainly found in northern agricultural counties, such as Changtu and Xifeng, and in eastern mountainous counties, such as Qingyuan, Xinbin, Huanren, and Kuandian. These areas have weaker heavy-industrial foundations and stronger ecological constraints, so their emission levels remained relatively low compared with the central and western industrial counties.

The outlier counties also deserve attention. High-Low outliers, such as Xinglongtai and Dawa in Panjin, represent local high-emission units located near lower-emission surroundings, which are closely related to petroleum extraction and petrochemical activity. Low-High outliers, including Xinmin, Liaozhong, Tai’an, Haicheng, and Liaoyang County in some years, indicate counties with relatively low emissions but strong exposure to nearby high-emission areas. These counties show spillover pressure from surrounding industrial centers and should be considered in cross-county mitigation coordination.

To identify statistically significant local clusters and spatial outliers, the Cluster and Outlier Analysis tool in ArcGIS Pro 3.7 was used to generate LISA cluster maps based on Anselin Local Moran’s I. The pseudo-significance threshold was set at p < 0.05. This method classifies counties into High-High, Low-Low, High-Low, and Low-High types according to the statistical significance of local spatial association. The LISA cluster maps provide a more direct representation of local emission hotspots, cold spots, and spatial outliers at the county scale, as shown in Figure 5.

From the perspective of overall spatial evolution, the LISA clustering trajectories from 2005 to 2024 confirm a strong core-periphery polarization pattern. High-emission clusters are concentrated in the western and central industrial belt, especially around Shenyang, Anshan, Liaoyang, Yingkou, Panjin, and Dalian, whereas low-emission clusters are mainly distributed in eastern ecological counties and northern agricultural counties. This spatial structure is consistent with the global Moran’s I results and the visual distribution shown in Figure 3.

The high-emission clusters show both persistence and expansion. Core urban districts in Shenyang and Dalian maintain high local clustering because of their population concentration, industrial agglomeration, transport functions, and energy demand. Counties along the Shenyang–Anshan–Liaoyang–Yingkou corridor further strengthen this pattern by connecting several traditional steel, petrochemical, and equipment-manufacturing bases into a continuous high-emission corridor.

The low-emission clusters also display strong stability. Qingyuan, Xinbin, Huanren, and Kuandian remain typical L-L counties because they are located in mountainous ecological-function zones with higher forest coverage and weaker heavy-industrial activity. Xiuyan County shows some transition from an L-L cluster to a statistically insignificant category, indicating that local industrialization and land-use change weaken the stability of low-emission clusters in some peripheral counties.

The L-H and H-L outliers identify transitional spaces between core industrial areas and peripheral low-emission zones. Counties such as Xinmin, Liaozhong, Tai’an, Haicheng, and Liaoyang County are frequently affected by nearby high-emission cores, whereas some resource-based and petrochemical districts in Panjin show isolated high-emission characteristics. These outliers are important for regional mitigation because they identify areas where carbon leakage, industrial transfer, and spillover pressure require priority attention.

The local spatial autocorrelation results show that the region’s emission pattern is not only a simple west–high/east–low gradient, but also a differentiated system of stable hotspots, stable cold spots, and transitional outliers. This evidence provides the spatial basis for differentiated policy recommendations across core urban districts, heavy-industrial corridors, resource-based counties, and ecological peripheral counties.

3.3. Spatial Spillover Effects and Heterogeneity of Driving Factors

3.3.1. Spatial Spillover Effects

The spatial correlation analysis shows significant spatial clustering in county-level carbon emissions across the Central-Southern Liaoning urban agglomeration. Conventional non-spatial linear regression models are therefore not sufficient to capture the effects of explanatory variables when spatial dependence is present. Spatial econometric models are used to assess this dependence and the associated spillover effects.

Table 5. Diagnostic tests for spatial panel models.

Testing Indicator	Statistic	p Value
LM-LAG	28.245	0.000
R-LM-LAG	9.963	0.002
LM-ERROR	1035.159	0.000
R-LM-ERROR	1016.877	0.000

reports the spatial diagnostic tests and their significance levels.

The Hausman test was first performed in Stata to choose between the fixed-effects and random-effects specifications. The test rejects the random-effects assumption at the 1% level, so the fixed-effects framework is preferred for the panel estimation.

The subsequent fixed-effects tests also support this choice. The statistics and p-values for the spatial fixed-effects, time fixed-effects, and two-way fixed-effects models are (28.80, 0.000), (46.96, 0.000), and (19.55, 0.000), respectively. Since both the spatial and temporal dimensions are significant, and the two-way specification is also strongly significant, the two-way fixed-effects model is used in the following estimation.

Spatial dependence was then examined using the Lagrange Multiplier (LM) tests. The test results reject the hypotheses of no spatial lag effect and no spatial error effect at the 1% level, which means that a conventional non-spatial panel model would be inadequate for these data. A spatial econometric specification is therefore required.

Finally, the likelihood ratio (LR) and Wald tests were used to examine whether the Spatial Durbin Model could be reduced to a simpler spatial lag or spatial error model. The results do not support such simplification, indicating that the SDM specification should be retained. The corresponding estimation results are reported in

Table 6. Estimation results of the Spatial Durbin Model (SDM).

Variables	Main	Variables	Spatial lag (W·X)
GDPPC	−0.499 ***	W × GDPPC	−0.121
	(0.0482)		(0.0988)
POP	−0.0771 ***	W × POP	0.0562 *
	(0.0168)		(0.0316)
SEC	0.248 ***	W × SEC	−0.685 ***
	(0.0623)		(0.118)
PE	0.132 ***	W × PE	−0.0341
	(0.0193)		(0.0328)
HS	0.0147	W × HS	−0.0281
	(0.0142)		(0.0261)
ρ	0.537 ***	σ2	0.0276 ***
	(0.0275)		(0.00105)
R2	0.6218
LR-SLM	20.19 ***	Wald-SLM	41.06 ***
LR-SEM	20.55 ***	Wald-SEM	23.13 ***

Notes: * and *** denote significance at the 10% and 1% levels, respectively.

.

As shown in Table 6, the spatial autoregressive coefficient ρ is 0.537 and statistically significant at the 1% level. This confirms the presence of spatial spillover effects in carbon emissions, meaning that local emissions are related to emissions in neighboring regions.

Before estimating the SDM, multicollinearity was checked using variance inflation factors for the five model regressors. The VIF values were 2.324 for GDPPC, 2.308 for POP, 1.120 for SEC, 3.237 for PE, and 1.465 for HS, all below the conventional threshold of 10. These diagnostics suggest that severe multicollinearity is unlikely to drive the baseline SDM results.

To examine the robustness of the GDPPC specification, a quadratic term of lnGDPPC and its spatial lag are introduced into the SDM-style two-way fixed-effects model. The results are reported in

Table 7. Quadratic GDPPC robustness check using an SDM-style two-way fixed-effects specification.

Variable	Coefficient	SE
W × lnCE	0.905 ***	0.0384
lnGDPPC	−0.0628	0.1688
(lnGDPPC)2	−0.000012	0.000803
W × lnGDPPC	0.143	0.291
W × (lnGDPPC)2	−0.00519	0.0138
SEC	0.255 ***	0.0647
W × SEC	−0.674 ***	0.119
N/R2	1460/0.986

Notes: Standard errors are reported in the SE column. *** denote significance at the 1% level. The model includes county fixed effects and year fixed effects. Baseline controls and spatially lagged explanatory variables are included according to the SDM-style specification. The quadratic GDPPC term and its spatial lag are introduced to examine the robustness of the GDPPC specification.

. The coefficient of W × lnCE remains significantly positive, indicating that the spatial dependence of county-level carbon emissions is stable after the quadratic GDPPC terms are included. The coefficient of SEC remains significantly positive, while W × SEC remains significantly negative, suggesting that the main industrial-structure result is also robust in this alternative specification. However, neither (lnGDPPC)² nor W × (lnGDPPC)² is statistically significant. The implied local turning point calculated from lnGDPPC and (lnGDPPC)² is not economically meaningful, and the implied turning point for the spatially lagged GDPPC term lies outside the observed range of lnGDPPC in the sample. Therefore, the quadratic specification does not provide statistical support for a formal EKC turning point. GDPPC is interpreted as a conditional association after controlling for industrial structure, population, fiscal expenditure, household savings, spatial dependence, and fixed effects.

To assess whether the SEC result is affected by its association with other socioeconomic variables, pairwise correlations between the SEC and the main regressors are reported in

Table 8. Pairwise correlations between SEC and socioeconomic variables.

Variable Pair	Pairwise Correlation
SEC and lnGDPPC	0.188
SEC and lnPOP	−0.176
SEC and lnPE	−0.086
SEC and lnHS	−0.035

Notes: Correlations are calculated using the county-level panel data. SEC denotes the share of secondary industry in GDP. The VIF of SEC in the baseline model is 1.120, suggesting that severe multicollinearity is unlikely.

. The correlations between SEC and lnGDPPC, lnPOP, lnPE, and lnHS are relatively low. In addition, the VIF of SEC in the baseline model is 1.120. These diagnostics suggest that the estimated SEC effect is unlikely to be driven by severe multicollinearity with the income, population, fiscal expenditure, or household savings variables. The result should still be interpreted as a conditional association, but the positive local coefficient of SEC and the negative spatially lagged SEC coefficient are not merely artifacts of simple pairwise correlation with GDPPC or POP.

Within the SDM setting, the estimated coefficients cannot be interpreted only as local marginal effects, because a change in one county can also affect nearby counties through the spatial weight matrix. For this reason, the effects of each explanatory variable were separated into three parts: the direct effect, which reflects the response of local carbon emissions; the indirect effect, which captures the response transmitted to neighboring counties; and the total effect, which combines the two.

Following the partial-derivative approach commonly used in spatial econometric models, this study reports the direct, indirect, and total effects of each explanatory variable. These results are used to evaluate both the local influence and the spatial spillover influence of the driving factors. The decomposition results are presented in

Table 9. Direct, indirect, and total effects of the Spatial Durbin Model.

Variables	Direct Effect	SE	Indirect Effect	SE	Total Effect	SE
GDPPC	−0.563 ***	0.0559	−0.777 ***	0.195	−1.340 ***	0.232
POP	−0.0764 ***	0.0153	0.0295	0.0589	−0.0469	0.0661
SEC	0.161 **	0.0725	−1.134 ***	0.236	−0.973 ***	0.273
PE	0.140 ***	0.0223	0.0737	0.0594	0.214 ***	0.0705
HS	0.00878	0.0166	−0.0423	0.0536	−0.0335	0.0611

Notes: **, and *** denote significance at the 5%, and 1% levels, respectively.

.

The coefficient of economic growth (GDPPC) is significantly negative, indicating a conditional negative association between per capita GDP and energy-related carbon emissions after spatial dependence and other socioeconomic factors are controlled. However, the quadratic robustness check in Table 7 does not provide statistical support for a formal EKC turning point. GDPPC is therefore interpreted as a conditional association rather than as evidence of a complete EKC pattern. The negative GDPPC coefficient is consistent with industrial upgrading and service-oriented development in more developed districts, but it should not be read as proof that the study area has reached an EKC turning point.

The direct effect of population size (POP) is significantly negative, whereas the indirect effect is positive but statistically insignificant. After spatial dependence and other socioeconomic factors are controlled, counties with larger registered populations tend to show lower local energy-related emissions in the current model. This finding is consistent with existing evidence on agglomeration efficiency and service-oriented functions in core urban districts [7,8,13,20], but it should not be interpreted as evidence that population growth itself automatically reduces emissions.

The secondary industry share (SEC) shows a more complex pattern. Its direct effect is positive, which is consistent with the expectation that counties with greater dependence on industrial production face stronger local emission pressure. The robustness diagnostics in Table 8 show that the correlations between SEC and lnGDPPC, lnPOP, lnPE, and lnHS are relatively low, and the VIF of SEC is 1.120. These diagnostics suggest that the estimated SEC effect is unlikely to be driven by severe multicollinearity with the income, population, fiscal expenditure, or household savings variables. The indirect effect of SEC remains negative and larger in absolute value, suggesting an association between the spatial organization of industrial activity and emission-reduction effects in neighboring counties through shared infrastructure, industrial specialization, and energy-efficiency spillovers. Because the magnitude of the indirect effect is relatively large, this result should be interpreted cautiously. It does not mean that expanding heavy industrial concentration is inherently beneficial. Rather, existing industrial corridors need coordinated low-carbon upgrading, cleaner production, energy-efficiency improvement, and carbon-management technologies.

Fiscal expenditure (PE) has significant positive direct and total effects, with coefficients of 0.140 and 0.214, respectively, while its indirect effect is not statistically significant. This indicates that fiscal expenditure is positively associated with local energy-related carbon emissions after spatial dependence is controlled. In an old industrial urban agglomeration, this association is linked to public expenditure, infrastructure construction, industrial support, and energy demand. Optimizing the structure of fiscal expenditure is therefore important for reducing the carbon intensity of regional development.

In contrast, household savings (HS) are not statistically significant in terms of direct, indirect, or total effects. Because HS measures savings stock rather than direct household consumption or disposable income, the insignificant coefficient should be interpreted as evidence that this particular wealth proxy has limited explanatory power for county-level energy-related carbon emissions in the current model.

The model results point to pronounced spatial spillover effects in the Central-Southern Liaoning urban agglomeration. Economic development and population agglomeration show conditional negative associations with emissions, the secondary industry share increases local emissions but has a negative spatial spillover effect, and fiscal expenditure has a positive local and total effect. These results are consistent with the High-High and Low-Low club convergence patterns identified in the local spatial autocorrelation analysis and with related spatial econometric evidence on county-level and urban-agglomeration carbon emissions [7,8,10,12,13,20].

For policy design, the results suggest three priorities: improving the quality of economic growth in core cities such as Shenyang and Dalian, promoting low-carbon upgrading in established heavy-industrial corridors, and restructuring fiscal expenditure to reduce carbon-intensive investment. The findings do not support a simple expansion of heavy industrial agglomeration. They instead point to coordinated industrial transformation and cleaner production within existing corridors.

3.3.2. Spatial Heterogeneity of Driving Factors

The SDM effect decomposition can be read together with the local spatial clustering pattern to understand the spatial heterogeneity of the driving mechanisms. Counties in the Central-Southern Liaoning Urban Agglomeration differ substantially in economic function and industrial base. Core urban districts in Shenyang and Dalian have stronger service functions and better conditions for industrial upgrading, whereas counties along the Shenyang–Anshan–Liaoyang–Yingkou corridor are more closely connected with steel, equipment manufacturing, petrochemical production, and other energy-intensive activities. Eastern mountainous counties, including parts of Benxi, Fushun, and Dandong, are more constrained by ecological functions and therefore show relatively low emission intensity.

The effect of economic development is more evident in core urban districts. The negative direct and spillover effects of GDPPC indicate that economic growth in the study area is not simply accompanied by higher emissions after spatial dependence is controlled. This pattern is consistent with the gradual upgrading of industrial structure and the expansion of service-oriented activities in more developed urban districts, but it should not be interpreted as a full EKC test. By contrast, in traditional industrial counties, the share of the secondary industry remains a more direct source of emission growth. The significant positive direct effect of SEC shows that counties with a higher dependence on industrial production still face stronger pressure from energy-related carbon emissions.

Fiscal expenditure also has spatially differentiated carbon effects. The positive direct and total effects of PE indicate that fiscal expenditure is closely associated with local emission growth, especially where public investment is linked to infrastructure construction, industrial support, and urban expansion. This interpretation is consistent with existing evidence that local fiscal expenditure and infrastructure-oriented investment can affect carbon emissions through development scale and investment structure [40]. The insignificant indirect effect suggests that this influence is mainly local rather than strongly diffused to neighboring counties. HS is not statistically significant, implying that county-level energy-related carbon emissions in the study area are shaped more by production-side factors than by the household savings proxy used in this model.

These heterogeneous effects call for policies tailored to county type. Core urban districts need to improve the quality of economic growth and strengthen technology spillovers. Heavy-industrial corridor counties need industrial restructuring, energy-efficiency improvement, cleaner production, and CCUS deployment. Resource-based and petrochemical counties should accelerate the low-carbon transformation of dominant industries, while ecological peripheral counties should maintain their low-emission development path and avoid receiving transferred high-emission industries. These recommendations are based on the SDM effect decomposition, LISA clustering, and Markov transition results.

3.4. Spatial Transition Probability Analysis of Energy-Related Carbon Emissions

To further examine the dynamic changes in county-level carbon emissions, both non-spatial and spatial Markov approaches were applied. Based on the quartile classification, county emissions were divided into four states: State 1 for low emissions, State 2 for relatively low emissions, State 3 for relatively high emissions, and State 4 for high emissions. A shift from a lower state to a higher state is regarded as an upward transition, whereas movement in the opposite direction is regarded as a downward transition.

Table 10. Traditional Markov transition probability matrix of county-level energy-related carbon emissions, 2005–2024.

t/t + 1	1	2	3	4
1	0.9157	0.0843	0	0
2	0.0402	0.8793	0.0805	0
3	0	0.0173	0.9191	0.0636
4	0	0	0.0148	0.9852

reports the estimated transition probabilities for county-level energy-related carbon emissions in the Central-Southern Liaoning Urban Agglomeration during 2005–2024.

The traditional Markov results reveal a clear tendency for counties to stay within their existing emission categories. The probabilities of remaining in States 1 to 4 are 0.9157, 0.8793, 0.9191, and 0.9852, respectively. Among the four groups, State 4 shows the strongest stability, implying that counties already in the high-emission category are unlikely to move downward in the short term.

Most observed changes occur between adjacent states. The probabilities of upward movement from State 1 to State 2, from State 2 to State 3, and from State 3 to State 4 are 0.0843, 0.0805, and 0.0636, respectively. The corresponding downward probabilities are 0.0402, 0.0173, and 0.0148. No leapfrog transition across non-adjacent states is observed. This indicates that the evolution of county-level energy-related carbon emissions is gradual, with evident category stability and lock-in characteristics.

Spatial dependence was then incorporated into the Markov framework. The spatial lag value of each county’s energy-related carbon emissions was calculated using the row-standardized queen contiguity matrix and divided into four neighborhood types according to pooled-sample quartile thresholds. Category 1 represents a low-emission neighborhood environment, Category 2 a medium-low-emission neighborhood environment, Category 3 a medium-high-emission neighborhood environment, and Category 4 a high-emission neighborhood environment. For each type of surrounding emission environment, a separate conditional transition matrix was estimated. The row sums were also examined to ensure that the probabilities were properly normalized; the remaining minor differences are only due to rounding.

Table 11. Spatial Markov transition probability matrix of county-level energy-related carbon emissions under different neighborhood conditions, 2005–2024.

Category	t\t + 1	1	2	3	4
1	1	0.9562	0.0438	0	0
	2	0.058	0.9275	0.0145	0
	3	0	0	0.9615	0.0385
	4	0	0	0	1
2	1	0.8242	0.1758	0	0
	2	0.0464	0.8742	0.0795	0
	3	0	0.0349	0.9302	0.0349
	4	0	0	0	1
3	1	0.7857	0.2143	0	0
	2	0.0252	0.8655	0.1092	0
	3	0	0.0152	0.9167	0.0682
	4	0	0	0.0253	0.9747
4	1	—	—	—	—
	2	0	0.7778	0.2222	0
	3	0	0.0098	0.902	0.0882
	4	0	0	0.0133	0.9867

Note: Category 1, Category 2, Category 3, and Category 4 represent low-, medium-low-, medium-high-, and high-emission neighborhood environments, respectively. The symbol “—” indicates that no transition observation is available for the corresponding state under that neighborhood condition.

presents the spatial Markov results for the period 2005–2024.

The spatial Markov transition probability matrix shows that neighborhood emission environments condition the dynamic transition of county-level energy-related carbon emissions. Compared with the traditional Markov matrix, transition probabilities vary across spatial lag categories, which means that local emission states do not evolve independently of surrounding counties.

Under Category 1, which represents a low-emission surrounding environment, the transition pattern is highly stable. Counties initially in State 1 have a 0.9562 probability of staying in that state, and only a 0.0438 probability of moving to State 2. The stability is also evident for States 2 and 3, with persistence probabilities of 0.9275 and 0.9615, respectively. For State 3, the probability of shifting upward to State 4 is only 0.0385, while counties already in State 4 remain unchanged, with a probability of 1.0000. These results imply that a low-emission neighborhood tends to preserve the existing emission structure and reduces the likelihood of large state changes.

Under Category 2, namely the medium-low-emission neighborhood, the transition pattern begins to show a stronger upward tendency. For counties in State 1, the probability of remaining at the original level decreases to 0.8242, whereas the probability of moving to State 2 rises to 0.1758. Counties in State 2 still mainly remain stable, with a probability of 0.8742, but the probability of moving to State 3 reaches 0.0795. For State 3, the probability of staying unchanged is 0.9302, and the probabilities of moving downward to State 2 and upward to State 4 are both 0.0349. Compared with Category 1, this indicates that once the surrounding environment shifts from low to medium-low emissions, low-emission counties show a higher probability of moving toward a higher emission state.

Under Category 3, which corresponds to a medium-high-emission surrounding environment, upward movement becomes more apparent. The probability of shifting from State 1 to State 2 reaches 0.2143, while the probability of moving from State 2 to State 3 is 0.1092. For State 3, the upward probability to State 4 rises to 0.0682, which is higher than that observed under Categories 1 and 2. Counties already in State 4 remain largely unchanged, with a persistence probability of 0.9747. These results indicate that a medium-high-emission neighborhood not only increases the likelihood of upward movement among lower- and middle-level counties, but also helps maintain the stability of high-emission counties.

Under Category 4, which represents a high-emission surrounding environment, there are no transition observations for counties initially located in State 1. For counties in State 2, the probability of shifting to State 3 reaches 0.2222, showing stronger upward pressure than in the other neighborhood types. States 3 and 4 also remain highly stable, with probabilities of 0.9020 and 0.9867, respectively. This pattern suggests that high-emission surroundings tend to lock counties into medium-high and high-emission states, while also increasing the possibility that medium-low counties move upward.

The spatial Markov results indicate that county-level energy-related carbon emission transitions are strongly conditioned by neighborhood environments. Low-emission neighborhoods tend to stabilize low-emission counties, whereas higher-emission neighborhoods increase the probability of upward transitions and reinforce high-emission lock-in. Emission reduction policies should therefore consider not only individual high-emission counties, but also their surrounding emission environments and spatial linkages with neighboring counties.

4. Discussion and Conclusions

4.1. Main Findings

This study reconstructs energy-related carbon emissions for the Central-Southern Liaoning Urban Agglomeration from 2005 to 2024 by combining harmonized nighttime light observations with energy-based carbon accounting. On this basis, the temporal changes and county-level spatial patterns of emissions are examined. The main findings can be summarized as follows.

(1) Energy-related carbon emissions showed an overall upward trend during the study period, but their growth momentum weakened gradually over time. Peripheral and ecological counties, including Qingyuan, Xinbin, Huanren, Kuandian, and parts of northern Tieling, show lower or more stable emission trajectories, while large growth poles such as Shenyang and Dalian have not yet shown clear turning points. The results therefore support a conclusion of slowing growth rather than a confirmed regional emission peak.

(2) Spatially, carbon emissions show a strong core-periphery structure and a west-to-east gradient. High-emission areas are concentrated in western and central industrial counties around Shenyang, Anshan, Liaoyang, Yingkou, Panjin, and Dalian. Low-emission areas are mainly located in eastern ecological counties such as Qingyuan, Xinbin, Huanren, and Kuandian. Autocorrelation analysis reveals significant spatial dependence at both global and local levels, with Moran’s I increasing from 0.627 in 2005 to 0.675 in 2024. Both high-emission and low-emission counties show club-like convergence and spatial path dependence.

(3) The explanatory variables show different spillover patterns. GDPPC shows a negative conditional association with local emissions in the baseline spatial model. However, the quadratic robustness check does not provide statistical support for a formal EKC turning point. Therefore, the income-related result should be interpreted cautiously as a conditional association rather than as evidence that the study area has already reached an EKC turning point. The industrial-structure result remains relatively stable. SEC is positively associated with local emissions, while its spatially lagged term is negative. Pairwise correlations and the VIF diagnostic suggest that this result is unlikely to be driven by severe multicollinearity with GDPPC, POP, PE, or HS. Population size has a significantly negative direct effect but no significant spatial spillover effect, and local fiscal expenditure has a positive effect but no significant spillover effect.

(4) The Markov transition probability matrix reveals temporal path dependence and spatial conditional dependence in county-level energy-related carbon emissions. In the traditional Markov matrix, the diagonal transition probabilities are 0.9157, 0.8793, 0.9191, and 0.9852 for states 1 to 4, respectively, showing that counties tend to remain in their original emission states. The high-emission state has the strongest persistence, with a probability of 0.9852 of remaining in state 4. Downward transitions are relatively limited, with probabilities of 0.0402, 0.0173, and 0.0148 for transitions from state 2 to state 1, from state 3 to state 2, and from state 4 to state 3, respectively. The spatial Markov results further show that neighborhood emission environments condition local transition dynamics. Low-emission neighborhoods help stabilize low-emission counties, whereas medium-high and high-emission neighborhoods increase upward transition probabilities and reinforce the persistence of medium-high and high-emission states. County-level emission dynamics are therefore shaped by both internal path dependence and external spatial spillover effects.

4.2. Policy Implications

(1) The dual-core economy should strengthen its low-carbon transformation capacity, while the EKC-related interpretation should remain cautious. GDPPC and POP have negative direct associations with local energy-related emissions after spatial dependence and other variables are controlled, but these coefficients do not prove an Environmental Kuznets Curve or imply that population growth automatically reduces emissions. Shenyang and Dalian should focus on technological innovation, digital energy management, producer-service development, and green finance rather than relying on growth alone to deliver emission reductions.

(2) Heavy-industrial corridors need coordinated low-carbon upgrading rather than simple expansion of industrial concentration. The Shenyang–Anshan–Liaoyang–Yingkou corridor should prioritize steel and equipment-manufacturing upgrading, energy-efficiency retrofits, electrification of industrial processes where feasible, and shared carbon-management infrastructure. The Dalian coastal area should combine port logistics, petrochemical restructuring, clean energy substitution, and stricter control of coastal industrial emissions. These measures need to be linked with cleaner production, energy-structure transformation, circular industrial chains, and CCUS deployment where technically and economically feasible.

(3) Cross-jurisdictional governance is needed to address the spatial lock-in effects of carbon emissions. Eastern ecological barrier counties such as Kuandian, Huanren, Qingyuan, and Xinbin should maintain ecological conservation functions, develop low-carbon tourism and forest-resource industries, and avoid accepting high-emission projects transferred from other areas. Northern agricultural counties should focus on low-carbon agriculture, biomass resource utilization, rural energy substitution, and transport-efficiency improvement. Differentiated strategies for industrial corridors, coastal industrial areas, ecological barrier zones, and agricultural counties can reduce the risk that emission-control pressure is shifted across county boundaries.

4.3. Limitations and Future Research

Several limitations should be noted. The analysis covers energy-related carbon emissions and excludes industrial process, agricultural, forestry, and solid-waste emissions because consistent county-level data are unavailable. Nighttime-light-based reconstruction is subject to saturation, city-specific light-emission relationships, and weak light signals from some heavy-industrial facilities; the estimates should therefore be read as NTL-assisted allocations rather than direct observations. Pairwise diagnostics indicate that explanatory variables are not simple transformations of nighttime light, but mechanical correlation and endogeneity cannot be fully eliminated. The negative GDPPC coefficient should be interpreted as a conditional association rather than as proof of a formal EKC relationship. Future research should examine nonlinear income effects with quadratic, threshold, or panel spatial models once longer and more consistent county-level datasets become available.

4.4. Conclusions

County-level energy-related carbon emissions in the Central-Southern Liaoning Urban Agglomeration show slowing growth, strong clustering, and clear neighborhood-conditioned transition dynamics. GDPPC shows a negative conditional association with local emissions in the baseline spatial model. However, the quadratic robustness check does not provide statistical support for a formal EKC turning point, so the income-related result should be interpreted cautiously rather than as evidence that the study area has already reached an EKC turning point. The industrial-structure result remains relatively stable: SEC is positively associated with local emissions, while its spatially lagged term is negative. Pairwise correlations and the VIF diagnostic suggest that this result is unlikely to be driven by severe multicollinearity with GDPPC, POP, PE, or HS. These results support spatially differentiated mitigation strategies that combine technology spillovers from core cities, low-carbon upgrading of heavy-industrial corridors, protection of low-emission ecological counties, and cross-jurisdictional governance for sustainable regional development.