Do European Regions Converge in Carbon Intensity? The Role of Climate-Driven Energy Demand

1. Introduction

Decarbonizing economic activity has become a central objective of contemporary energy and climate policy, requiring economies to reduce greenhouse gas emissions while sustaining economic growth. Within this context, the European Union has established an ambitious policy framework through the European Green Deal and the European Climate Law, setting the path towards climate neutrality by 2050 [1,2]. Despite these policy commitments, the transition to a low-carbon economy has progressed unevenly across European regions. Differences in economic structure, industrial specialization, labor market conditions, energy demand and climatic characteristics have created diverse regional decarbonization pathways, leading to substantial disparities in environmental performance. Understanding how decarbonization evolves across regions is therefore essential for designing effective place-based policies that account for territorial heterogeneity rather than relying exclusively on national-level evidence.

Assessing regional decarbonization, however, requires an indicator that captures not only the level of emissions but also the environmental efficiency of economic activity. Carbon intensity, commonly measured as carbon dioxide emissions per unit of gross domestic product, provides such a measure by linking environmental pressure to economic output [3]. Unlike absolute emissions, which are strongly influenced by the size of an economy, carbon intensity reflects the amount of emissions generated to produce a given level of economic value, making it particularly suitable for evaluating decarbonization performance across regions with different economic structures and development levels [4]. By combining information on economic activity and environmental pressure, carbon intensity has become a central indicator in climate economics and decomposition analysis within the Kaya identity framework and is widely used to assess the drivers of emission reductions and the effectiveness of energy transition policies [5].

A growing body of literature has investigated carbon intensity dynamics, focusing primarily on convergence processes, structural transformation, technological progress, and the determinants of regional decarbonization. Several studies have examined whether regions or countries with initially higher carbon intensity gradually converge towards cleaner production patterns, reporting mixed evidence on absolute, conditional, and club convergence while consistently highlighting the importance of structural, technological, and institutional factors in shaping regional decarbonization outcomes [6,7,8]. More recently, researchers have shown that carbon intensity cannot be analyzed independently across regions, as technological diffusion, industrial linkages, and environmental policies generate significant spatial spillover effects that shape regional decarbonization trajectories [9]. Within the European context, recent evidence further reveals that carbon intensity exhibits persistent regional heterogeneity, with distinct convergence clubs reflecting differences in production specialization and integration into European value chains rather than a uniform transition towards lower-emission development paths [10].

Despite the progress made in analyzing regional decarbonization, the relevant evidence remains divided across largely separate strands of research. Recent studies have documented heterogeneous carbon-intensity trajectories and convergence clubs across European NUTS-2 regions but have primarily linked these patterns to productive specialization and participation in European value chains rather than to the climatic determinants of regional energy demand [10]. A related body of research has established that emissions in European regions are spatially interdependent and that socioeconomic and technological conditions generate effects extending beyond regional boundaries [11]. These studies, however, focus mainly on emission levels or sector-specific emissions rather than on the spatial mechanisms underlying carbon-intensity convergence. At the same time, heating and cooling requirements are widely used as indicators of climate-driven energy demand, and previous country-level evidence suggests that heating needs can significantly affect carbon intensity [12,13]. Their role within a regional convergence framework nevertheless remains insufficiently examined. Consequently, the literature provides limited evidence on whether European regions converge in carbon intensity once climatic energy requirements, spatial dependence, and structural heterogeneity are considered jointly.

This study addresses this gap by examining conditional carbon-intensity convergence across European NUTS-2 regions over the period 2000–2024. It makes three main contributions. First, it extends the predominantly national and sector-specific evidence by providing a long-run, harmonized assessment of convergence at the European regional level. Second, it explicitly incorporates climate-driven energy demand by combining heating and cooling degree days into a regional climate-load measure, thereby accounting for systematic differences in the energy required to maintain thermal comfort across European territories. Third, it integrates conventional β-convergence analysis with tests of spatial autocorrelation and spatial panel models, allowing convergence, regional heterogeneity, and interregional spillovers to be examined within a unified empirical framework. This integrated approach distinguishes the study from research that considers convergence, climatic energy demand, or spatial dependence separately and provides a stronger basis for designing place-based decarbonization policies.

The remainder of the paper is organized as follows. Section 2 reviews the relevant literature on carbon intensity convergence and regional decarbonization. Section 3 describes the study area, dataset, variables, and empirical methodology. Section 4 presents the empirical findings, including the results of the convergence analysis and spatial econometric models. Finally, Section 5 discusses the main findings and their policy implications, while Section 6 concludes the paper.

2. Literature Review

2.1. Carbon Intensity as an Indicator of Regional Decarbonization

Carbon intensity is widely used to assess whether economic growth is becoming progressively less dependent on greenhouse gas emissions. Its relevance extends beyond the measurement of environmental pressure, as it reflects the capacity of an economy to generate output while limiting the emissions associated with production. A decline in carbon intensity may therefore result from several distinct processes, including improvements in energy efficiency, changes in the energy mix, technological progress, and the reallocation of economic activity toward less carbon-intensive sectors. This multidimensional character makes carbon intensity particularly useful for evaluating regional decarbonization, where differences in production structures and energy systems can generate markedly different environmental outcomes even among regions displaying similar levels of income.

The literature identifies economic structure and technological progress as two of the principal mechanisms shaping carbon-intensity dynamics. Regions with a high concentration of manufacturing, extractive activities, heavy industry, or other energy-intensive sectors tend to generate more emissions per unit of economic output. Structural transformation toward services and knowledge-intensive activities can reduce this dependence, although the environmental effect of industrial restructuring is not necessarily linear and may vary across stages of economic development [14,15,16]. Technological progress can further lower carbon intensity by improving production efficiency, reducing the energy required per unit of output, and facilitating the adoption of cleaner production processes. Nevertheless, its effectiveness depends on the nature of technological change and the capacity of regional economies to absorb and diffuse innovation. Rebound effects may also offset part of the expected environmental gains when efficiency improvements reduce production costs and stimulate additional economic activity [14,15].

The composition of the energy system represents a second central determinant. Economies relying heavily on coal and other fossil fuels generally exhibit higher carbon intensity, whereas the expansion of renewable energy can lower the emissions associated with electricity generation and productive activity. Recent international evidence indicates that renewable energy deployment is generally associated with lower carbon intensity, although its effect varies across renewable technologies, income groups, institutional conditions, and stages of energy-system development [17,18]. Hydropower, solar, wind, and geothermal energy cannot therefore be assumed to produce identical decarbonization outcomes. Their effectiveness depends on existing generation infrastructure, grid integration, energy-storage capacity, and the extent to which renewable production replaces rather than supplements fossil-fuel generation. Green technological innovation reinforces this process by improving renewable-energy technologies and energy efficiency, but its effects likewise differ between developed and developing economies [18].

Carbon-intensity dynamics are also influenced by socioeconomic and institutional conditions. Labor-market performance, human capital, research capacity, environmental regulation, and governance affect the ability of firms and regions to adopt cleaner technologies and adjust their productive structures. Strong innovation systems and effective environmental institutions can accelerate technological diffusion and discourage carbon-intensive activities, whereas structurally weaker regions may face financial, technological, and institutional constraints that delay adjustment. These differences imply that apparently similar climate and energy policies may generate unequal outcomes across territories. Decarbonization should therefore not be understood as a uniform response to common policy objectives but as a territorially differentiated process conditioned by the economic and institutional capacity of each region.

The importance of territorial heterogeneity is increasingly apparent in subnational research. Evidence from European regions demonstrates that industrial decarbonization is shaped not only by local socioeconomic and technological characteristics but also by geographical proximity and interregional relationships [11]. Broader subnational evidence similarly identifies considerable disparities in the pace and composition of the transition toward decarbonized power systems, indicating that aggregate national trends may conceal structurally disadvantaged regions and divergent local transition pathways [19]. Climatic conditions introduce an additional source of heterogeneity because heating and cooling requirements influence energy demand and consequently the emissions associated with regional economic activity. Regions may therefore display different carbon-intensity trajectories even when their levels of development, industrial structures, or formal policy commitments appear comparable.

Taken together, the literature indicates that carbon intensity is shaped by the interaction of structural transformation, technological progress, energy-system composition, institutional capacity, and territorial conditions. Because these factors evolve unevenly across space and time, reductions in carbon intensity need not occur at the same rate or lead regions toward a common long-run position. This observation gives rise to a central empirical question: whether initially carbon-intensive economies and regions gradually catch up with better-performing ones or whether existing disparities persist. The literature addressing this question through absolute, conditional, stochastic, and club-convergence approaches is reviewed in the following subsection.

2.2. Carbon Intensity Convergence: From Absolute to Conditional and Club Convergence

The concept of convergence originates from neoclassical growth theory, according to which economies with lower initial levels of development are expected to grow faster than more advanced ones, gradually approaching a common long-run equilibrium [20]. This theoretical framework distinguishes between absolute β-convergence, where all economies converge towards the same steady state, and conditional β-convergence, which recognizes that differences in structural characteristics, technological capabilities, institutions, and policy environments lead economies to converge towards different long-run equilibria [21]. The latter approach has become the dominant framework in empirical research because it better reflects the substantial heterogeneity observed across countries and regions, particularly in environmental and energy-related studies.

The convergence framework was subsequently extended to environmental indicators, initially focusing on greenhouse gas and carbon dioxide emissions before being applied to broader measures of environmental efficiency such as energy intensity and carbon intensity. Early cross-country studies generally reported evidence of emissions convergence, although the magnitude and speed of adjustment differed considerably across samples and estimation methods [22,23]. As literature evolved, increasing attention shifted toward carbon intensity because it simultaneously captures environmental performance and economic productivity, providing a more comprehensive indicator of decarbonization than absolute emissions alone. Recent evidence further suggests that carbon-intensity dynamics are shaped by renewable-energy deployment, technological progress, industrial specialization, and institutional quality, indicating that convergence depends on structural conditions rather than representing an automatic process [17,24].

These findings have strengthened the empirical relevance of conditional convergence, which assumes that economies characterized by different structural conditions converge toward different steady-state equilibria. Differences in energy systems, industrial composition, technological innovation, governance quality, and environmental policies generate heterogeneous adjustment paths, causing convergence speeds to vary substantially across countries and regions. Consequently, empirical studies increasingly incorporate structural control variables to explain not only whether convergence exists but also the mechanisms through which it occurs. This perspective is particularly relevant in the context of carbon intensity, where improvements in environmental performance depend on both economic transformation and technological change rather than solely on reductions in emissions.

The growing recognition of structural heterogeneity has also stimulated the development of club convergence approaches. Instead of assuming that all economies eventually converge toward a common equilibrium, club-convergence theory argues that groups of countries or regions sharing similar structural characteristics may converge toward different steady states [25]. This framework has gained considerable attention in environmental economics because countries differ substantially in resource endowments, industrial structures, institutional quality, technological capabilities, and energy-system characteristics. Consequently, multiple convergence clubs frequently emerge, reflecting distinct transition pathways rather than a universal decarbonization process. Recent empirical applications increasingly confirm the existence of heterogeneous convergence regimes in both national and regional energy systems, highlighting that structural similarities often dominate geographical proximity when determining long-run environmental trajectories [10].

More recently, the literature has progressively shifted from national to regional analyses, acknowledging that country-level averages often conceal substantial spatial disparities. Regional studies reveal that territories operating under the same national regulatory framework may nevertheless experience markedly different decarbonization trajectories because of differences in industrial specialization, innovation capacity, labor-market conditions, infrastructure, and local governance [11]. Furthermore, increasing evidence suggests that geographical proximity facilitates technology diffusion, economic interactions, and policy spillovers, implying that regional convergence cannot be fully understood without explicitly considering spatial dependence. These developments have considerably expanded the scope of convergence analysis, moving beyond purely temporal dynamics toward geographically interconnected transition processes.

Collectively, the literature has evolved from asking whether carbon intensity converges to identifying under which conditions, through which structural mechanisms, and across which groups of economies or regions convergence is more likely to occur. This evolution demonstrates that convergence is not a universal outcome but rather the product of interactions among economic structure, technological progress, institutional quality, energy-system characteristics, and spatial interdependence. Consequently, recent research increasingly incorporates geographical interactions into convergence models, providing the theoretical foundation for the spatial perspective discussed in the following subsection.

2.3. Spatial Dependence and Regional Carbon Intensity Dynamics

Recent research has increasingly recognized that regional carbon-intensity dynamics cannot be adequately explained by analyzing regions as isolated economic units. Within integrated economic systems such as the European Union, neighboring regions are interconnected through trade, labor mobility, infrastructure networks, technological diffusion and common environmental policies. These interactions create geographical interdependencies whereby improvements in environmental performance achieved in one region may influence neighboring territories, generating spatial spillover effects that shape regional decarbonization trajectories [6,11].

The existence of spatial dependence has been consistently confirmed in recent empirical studies. Using a dynamic spatial panel approach, Huang et al. (2019) [6] demonstrate that carbon intensity exhibits significant spatial autocorrelation, implying that regional carbon-intensity levels are influenced not only by local characteristics but also by developments occurring in neighboring areas. Their findings further reveal that technological progress generates positive spillover effects, accelerating carbon-intensity reductions beyond regional administrative boundaries. These results suggest that ignoring geographical interactions may lead to an incomplete understanding of regional decarbonization processes.

The importance of spatial interactions is further reinforced by the convergence literature. Analyzing global energy-intensity dynamics, Balado-Naves et al. (2023) [26] show that convergence is conditional upon geographical proximity and spatial spillovers. Technological diffusion, renewable-energy deployment and policy transmission mechanisms significantly increase convergence rates, while neighboring economies tend to follow similar adjustment paths. Although their analysis focuses on energy intensity, the underlying mechanisms provide a strong conceptual foundation for understanding regional carbon-intensity dynamics, where technological and institutional spillovers likewise influence the pace of environmental transition.

Evidence from European regions provides particularly strong support for the spatial nature of decarbonization. Examining 238 NUTS-2 regions, Vagnini et al. (2025) [27] identify significant endogenous spatial interactions in industrial carbon emissions, demonstrating that neighboring regions tend to exhibit similar emission trajectories over time. Their results further indicate that higher educational attainment and regional investments in research and development generate both direct and indirect reductions in emissions, highlighting the importance of knowledge diffusion and innovation spillovers in promoting regional decarbonization. Complementary spatio-temporal analyses also reveal persistent geographical clusters of industrial emissions that frequently extend across national borders, suggesting that economic integration and coordinated European climate policies contribute to common regional transition pathways rather than isolated local adjustments.

Beyond spatial spillovers, recent regional research increasingly emphasizes the role of path dependence in shaping carbon dynamics. Bourdin and Perrot (2025) [28] introduce the concept of the regional carbon trap, arguing that some European regions remain locked into carbon-intensive development trajectories because of persistent industrial specialization, weak economic diversification and institutional constraints. Their analysis identifies distinct regional transition pathways across EU NUTS-2 regions, demonstrating that historical development patterns and regional structural characteristics substantially influence future decarbonization prospects. These findings reinforce the view that neighboring regions may not only interact spatially but may also display persistent similarities in their transition dynamics due to shared economic structures and institutional environments.

Taken together, the recent literature suggests that regional carbon-intensity dynamics emerge from the interaction between local structural characteristics and geographical interdependencies. Technology diffusion, innovation networks, regional knowledge spillovers and coordinated climate policies enable environmental improvements to propagate across space, while historical development paths influence regions' capacity to benefit from these interactions. Therefore, analyzing carbon-intensity convergence without considering the spatial dependence may underestimate both the speed and the mechanisms of regional decarbonization. These considerations provide a strong theoretical justification for adopting a spatial econometric framework to investigate carbon-intensity convergence across European NUTS-2 regions.

2.4. Climate-Driven Energy Demand and Regional Carbon Intensity

Climate conditions constitute an important source of regional heterogeneity in energy demand and, consequently, in carbon-intensity dynamics. Outdoor temperature directly affects the energy required to maintain comfortable indoor conditions, producing higher heating needs during cold periods and higher cooling needs during warm periods. Heating degree days (HDD) and cooling degree days (CDD) are therefore widely used as temperature-based indicators of climate-driven energy requirements. HDD measure the accumulated deviation of outdoor temperatures below a heating threshold, whereas CDD capture deviations above a cooling threshold. Although these indicators do not represent observed energy consumption directly, they provide comparable measures of the climatic pressure placed on regional energy systems [29,30].

The relationship between climate conditions and energy use is nevertheless more complex than a simple response to average temperature. Actual consumption also depends on building characteristics, insulation, heating and cooling technologies, energy prices, household behaviour, income and the penetration of air-conditioning systems. Consequently, similar degree-day values may generate different energy and emissions outcomes across regions. Deroubaix et al. (2021) [29] distinguish the climatic component of heating and cooling demand from its socioeconomic and technological determinants, while Mutschler et al. (2021) [31] demonstrate that future cooling requirements depend jointly on climate change, population development and the adoption of cooling equipment. Degree-day indicators should therefore be interpreted as measures of climate-related energy pressure rather than as direct substitutes for final energy use or greenhouse gas emissions.

Recent evidence points to a substantial rebalancing of heating and cooling requirements under climate change. Rising temperatures generally reduce HDD and the associated need for space heating, while increasing CDD and electricity demand for cooling. However, these opposing effects are neither geographically uniform nor necessarily equivalent in their energy and emissions consequences. Using high-resolution climate projections for Europe, Larsen et al. (2020) [32] identify declining heating requirements alongside increasing cooling needs, with substantial variation across countries, climatic zones and future scenarios. At the global level, Deroubaix et al. (2021) [29] similarly find that the increase in climate-driven cooling demand is generally more pronounced than the reduction in heating demand, although the magnitude of the cooling response remains subject to considerable climate-model uncertainty.

This geographical asymmetry is particularly relevant for European regional decarbonization. Southern and Mediterranean regions are increasingly exposed to prolonged cooling seasons and rising peak electricity demand, whereas northern regions remain more strongly characterized by heating requirements. At the same time, the carbon implications of these shifts depend on the energy technologies used to satisfy the additional demand. Reduced heating requirements may lower fossil-fuel consumption where buildings depend on natural gas or oil, while growing cooling demand may increase electricity consumption and peak-load pressures. The resulting effect on carbon intensity will therefore be smaller in regions with low-carbon electricity generation, efficient buildings and modern cooling technologies, but stronger where marginal electricity demand continues to be supplied by carbon-intensive sources. Hu et al. (2024) [33] show that temperature–electricity demand relationships differ substantially across European countries and evolve with electrification, building insulation and the adoption of active and passive cooling technologies.

Climate-driven demand may affect regional carbon intensity through both the level and timing of energy consumption. Changes in HDD and CDD influence annual energy requirements, while increasingly concentrated heating and cooling episodes can create seasonal peaks that place additional pressure on electricity generation, transmission networks and reserve capacity. Filahi et al. (2024) [34] show that climate change alters not only the magnitude but also the temporal distribution of European heating and cooling needs. Such temporal fragmentation is relevant for emissions because peak demand may require the activation of less efficient or more carbon-intensive generation capacity, even in energy systems with rapidly expanding renewable production. Climate exposure may therefore influence carbon intensity through a combination of total demand, demand seasonality and the carbon composition of the marginal energy supply.

These mechanisms justify the inclusion of a combined climate-load indicator in regional carbon-intensity models. By integrating HDD and CDD, the Climate Load Requirement captures the total annual temperature-related pressure for space conditioning and permits comparison across regions characterized by different combinations of heating and cooling needs. In the context of European NUTS-2 regions, this measure is particularly useful because it incorporates climatic heterogeneity that is not captured by conventional socioeconomic controls. Its estimated effect should nevertheless be interpreted as the average association between aggregate thermal pressure and carbon intensity, recognizing that heating and cooling may differ in their energy carriers, conversion efficiencies and marginal emissions.

Taken together, recent research indicates that climate-driven energy demand is not merely an external background condition but a structural determinant of regional decarbonization pathways. Regional differences in heating and cooling requirements influence energy consumption, system peaks and the technologies used to satisfy demand, thereby affecting both current carbon intensity and the capacity of regions to converge toward lower-carbon equilibria. Incorporating the Climate Load Requirement into the convergence framework therefore enables the analysis to distinguish climate-related energy pressure from the effects of industrial structure, labor-market conditions and spatial interactions.

2.5. Literature Synthesis and Research Gap

The existing literature has substantially advanced the understanding of carbon-intensity dynamics by demonstrating that decarbonization is influenced by a broad set of economic, technological, institutional and geographical factors. Recent studies consistently identify carbon intensity as a comprehensive indicator of sustainable development, while providing growing evidence that convergence processes are characterized by considerable heterogeneity across countries and regions. At the same time, the literature increasingly recognizes that regional carbon-intensity dynamics are shaped by spatial spillovers and that climate-related energy demand represents an additional source of regional differentiation.

Despite these important advances, several research gaps remain. First, although numerous studies investigate carbon-intensity convergence, relatively few simultaneously consider the joint effects of regional structural characteristics, spatial dependence and climate-driven energy demand within a unified analytical framework. Existing contributions generally examine these mechanisms separately, thereby providing only a partial explanation of regional decarbonization dynamics. Second, recent evidence suggests that carbon-intensity trajectories are far from homogeneous. Rather than converging toward a single equilibrium, countries and regions frequently exhibit distinct adjustment patterns reflecting differences in technological development, institutional quality and climate-policy effectiveness [35].

These limitations become even more evident in the European context. Although the European Union operates under a common climate-policy framework and increasingly integrated energy markets, its NUTS-2 regions differ substantially in terms of industrial structure, innovation capacity, labor-market conditions and climate exposure. Recent regional evidence further indicates that decarbonization follows multiple development pathways, with some regions remaining locked into persistent carbon-intensive trajectories because of structural and institutional constraints, while others successfully transition toward low-carbon development [28]. These findings imply that regional carbon-intensity dynamics cannot be adequately understood without simultaneously accounting for regional heterogeneity, spatial interactions and differences in climate-related energy requirements.

Motivated by these observations, the present study develops an integrated framework for analyzing carbon-intensity convergence across European NUTS-2 regions. Specifically, it combines conditional convergence analysis with spatial econometric modelling while explicitly incorporating Climate Load Requirement (CLR) as an indicator of climate-driven energy demand. By jointly considering structural characteristics, spatial spillovers and climatic energy pressure within a common empirical framework, the study provides a more comprehensive explanation of regional carbon-intensity convergence and extends the emerging literature on regional decarbonization in Europe.

3. Materials and Methods

3.1. Data

The empirical analysis is conducted at the NUTS-2 level and includes European regions for the period 2000–2024. NUTS-2 regions represent the principal territorial units for the implementation and evaluation of European regional and cohesion policies, making them an appropriate spatial scale for investigating regional disparities in energy transition and decarbonization. The initial dataset consists of 200 regions observed over 25 years (5,000 observations).

Regional Greenhouse Gas (GHG) emissions are used as the main environmental indicator. The data are obtained from the EDGAR NUTS-2 emissions dataset, developed by the European Commission's Joint Research Centre (JRC) [36]. Economic performance is measured by regional Gross Domestic Product (GDP) per capita [37]. The analysis uses the regional unemployment rate [38] and the share of industrial employment [39] to characterize regional socioeconomic conditions. Climatic conditions are considered based on Heating Degree Days (HDD) and Cooling Degree Days (CDD) [40]. To capture overall climate-driven energy requirements, the Climate Load Requirement (CLR) is constructed as the sum of HDD and CDD.

After constructing the dependent variable and the required lagged regressors, the final estimation sample consists of 189 regions and 4,287 observations. Eleven regions are excluded because of incomplete information, primarily due to missing greenhouse gas emissions (Swiss regions and Iceland), unavailable industrial structure data (Melilla and Åland), or missing climatic indicators (North Macedonia). The resulting panel is unbalanced but preserves broad geographical coverage across Europe.

3.2. Carbon Intensity and Convergence Variables

Carbon intensity has become one of the most widely used indicators for assessing progress toward decarbonization because it combines environmental and economic performance into a single measure. Unlike absolute greenhouse gas emissions, carbon intensity reflects the emissions generated relative to economic output, providing an indicator of the carbon efficiency of regional production systems [17]. Consequently, reductions in carbon intensity indicate improvements in the ability of an economy to generate value while producing fewer emissions, making the indicator particularly relevant for evaluating regional energy transition pathways.

For each region i and year t, carbon intensity is defined as

$${CI}_{it}={\displaystyle \frac{{GHG}_{it}}{{GDP}_{it}}},$$

(1)

where ${GHG}_{it}$ denotes greenhouse gas emissions and ${GDP}_{it}$ regional gross domestic product per capita. To facilitate econometric estimation and the interpretation of convergence dynamics, the analysis employs the logarithmic transformation

$$log{CI}_{it}=log{GHG}_{it}-log{GDP}_{it},$$

(2)

The annual logarithmic change in carbon intensity is calculated as

$${growth}_{it}=\Delta log{CI}_{it}=log{CI}_{it}-log{CI}_{it-1}.$$

(3)

This measure captures the year-to-year change in regional carbon intensity, with negative values indicating reductions and positive values indicating increases.

Differences in labor market conditions, industrial structure and climate-driven energy requirements are considered. The unemployment rate reflects the regional labor market conditions, while the share of industrial employment captures the differences in production structure. Climate-driven energy requirements are represented by CLR. Higher CLR values indicate greater combined heating and cooling requirements.

Several additional variables were also considered, including renewable energy penetration, research and development personnel, tertiary education attainment and population density. They were not included in the final model because of limited data coverage and/or the absence of statistically significant effects.

Table 1 summarizes the variables employed in the analysis, their definitions, expected effects and corresponding data sources.

Table 2 reports the descriptive statistics of the variables included in the analysis. The variables exhibit substantial variation across European regions and over time, supporting the use of panel data techniques to investigate regional carbon intensity convergence. Differences in the number of observations reflect data availability and the construction of lagged variables used in the econometric analysis.

3.3. Conditional β-Convergence Model

To examine whether European regions converge in carbon intensity, the analysis adopts the conditional β-convergence framework [20] in a panel-data setting [21]. This approach is appropriate for regional energy systems, since regions may converge towards different long-run equilibrium levels determined by their structural characteristics.

Differences in economic structure, labor markets and climate conditions may influence the regional decarbonization dynamics. These factors are incorporated into the conditional convergence model

$$\Delta log{CI}_{it}=\alpha +\beta log{CI}_{it-1}+{\gamma }_1{Unemp}_{it-1}+{\gamma }_2{Industry}_{it-1}+{\gamma }_3{CLR}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it},$$

(4)

where $\Delta log{CI}_{it}$ denotes the annual logarithmic growth of carbon intensity for region $i$ in year $t$, $log{CI}_{it-1}$ is the lagged level of carbon intensity, ${Unemp}_{it}$ is the lagged regional unemployment rate, ${Industry}_{it-1}$ represents the lagged share of industrial employment, and ${CLR}_{it}$ denotes the contemporaneous Climate Load Requirement. The terms ${\mu }_i$ and ${\lambda }_t$ capture unobserved regional and time-specific effects, respectively, and ${\epsilon }_{it}$ is the idiosyncratic error term.

The lagged level of carbon intensity ($log{CI}_{it-1}$) is the main convergence variable. A negative and statistically significant estimate of $\beta$ indicates conditional β-convergence, implying that regions with higher initial carbon intensity reduce their carbon intensity at a faster rate than regions with lower initial levels. Socioeconomic variables enter the model with a one-year lag to reduce potential simultaneity and to allow regional structural conditions to affect subsequent changes in carbon intensity. In contrast, CLR is included contemporaneously because heating and cooling requirements directly influence energy demand and associated emissions within the same period.

Based on the estimated convergence coefficient, the annual speed of convergence is computed as

$$\lambda =-\mathit{log}\left(1+\beta \right).$$

(5)

The corresponding half-life of convergence is obtained as

$$HL={\displaystyle \frac{log2}{\lambda }},$$

(6)

representing the number of years required to eliminate one-half of the initial disparity in carbon intensity.

The conditional convergence model is computed using a two-way fixed-effects (TWFE) panel estimator. Regional fixed effects control for unobserved time-invariant characteristics, such as geographical conditions, institutional quality, historical development patterns and technological progress, whereas time fixed effects account for common shocks affecting all regions simultaneously, including changes in European climate and energy policies, macroeconomic conditions and technological progress. Therefore, the estimated coefficients are identified from within-region variation over time.

The choice of the fixed-effects specification is supported by the Hausman specification test [41], which rejects the random-effects alternative. Model diagnostics indicate serial correlation, whereas the Breusch–Pagan test does not reject homoskedasticity at the 5% level. To obtain valid statistical inference, coefficient estimates are reported with Driscoll-Kraay standard errors [42]. This covariance estimator is robust to heteroskedasticity, serial correlation and general forms of cross-sectional dependence. Region-clustered standard errors are additionally considered as a robustness check.

3.4. Heterogeneity Analysis

To investigate whether convergence differs according to the initial level of carbon intensity, the sample is divided into three equally sized groups (terciles) based on the initial distribution of regional carbon intensity. Separate conditional β-convergence models are estimated for low-, medium- and high-carbon-intensity regions using the same econometric specification. This analysis examines whether convergence differs across regions with different initial levels of carbon intensity and whether highly carbon-intensive regions converge at the same pace as lower-carbon regions.

3.5. Spatial Analysis

Regional carbon-intensity dynamics may exhibit spatial dependence because neighboring regions are connected through energy infrastructure, production networks, technological diffusion and common climatic and policy environments. Consequently, the baseline convergence framework is extended to explicitly account for spatial interactions. The spatial analysis has three stages: construction of the spatial weights matrix, assessment of spatial autocorrelation, and estimation of alternative spatial panel specifications.

3.5.1. Spatial Weights Matrix and Global Spatial Autocorrelation

Neighboring NUTS-2 regions are identified using a first-order queen contiguity matrix $W=\left[w_{ij}\right]$ as it captures direct geographical interactions between adjacent NUTS-2 regions and is widely used in regional spatial econometric analyses. Two regions are defined as neighbours when they share either a common boundary or a common vertex:

$$w_{ij}=\left\{\begin{array}{cc}1,& \mathrm{if}\mathrm{regions}i\mathrm{and}j\mathrm{are}\mathrm{contiguous},\\ 0,& \mathrm{otherwise}.\end{array}\right.$$

(7)

The diagonal elements are set equal to zero ($w_{ii}=0$), and the resulting matrix is row-standardized so that the spatial weights associated with each region sum to one. Figure 1 illustrates the resulting neighborhood structure used in the spatial analysis.

Global spatial autocorrelation is examined using Moran's $I$ statistic [43]. For each year $t$, Moran's $I$ is calculated as

$$I_t={\displaystyle \frac{N_t}{S_{0t}}}{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^{N_t}}{\displaystyle {\sum }_{j=1}^{N_t}}{w}_{ij,t}\left(z_{it}-{\overline{z}}_t\right)\left(z_{jt}-{\overline{z}}_t\right)}{{\displaystyle {\sum }_{i=1}^{N_t}}(z_{it}-{\overline{z}}_t{)}^2}},$$

(8)

where $N_t$ denotes the number of regions observed in year $t$, $z_{it}$ is the variable of interest, ${\overline{z}}_t$ is its cross-sectional mean, and

$$S_{0t}={\sum }_i{\sum }_j{w}_{ij,t}.$$

(9)

Moran's $I$ is calculated separately for the logarithm of carbon intensity, $\mathrm{l}\mathrm{o}\mathrm{g}\left(C{I}_{it}\right)$, and its annual growth rate, $\mathsf{\Delta }\mathrm{l}\mathrm{o}\mathrm{g}\left(C{I}_{it}\right)$. The first measures spatial clustering in carbon-intensity levels, while the second examines spatial dependence in changes in carbon intensity. Positive values indicate that neighboring regions tend to exhibit similar carbon-intensity levels or growth patterns.

Lagrange Multiplier (LM) tests are used to identify the form of spatial dependence in the non-spatial fixed-effects model [44]. The analysis considers four tests: the LM-lag and LM-error tests and their robust versions. The LM-lag test examines spatial dependence in the dependent variable, while the LM-error test examines spatial dependence in the error term. The robust versions account for the possible presence of the alternative form of spatial dependence and are particularly useful when both standard LM tests reject their null hypotheses. The test results are used to determine whether spatial lag or spatial error specification is more appropriate.

3.5.2. Spatial Panel Models

To account explicitly for the detected spatial dependence, three alternative spatial panel specifications are estimated. All models retain the regional and time fixed effects of the baseline convergence specification and use the same row-standardized queen contiguity matrix [45].

The Spatial Autoregressive Model (SAR) introduces a spatial lag of the dependent variable:

$$y_{it}=\rho \sum _j{w}_{ij}{y}_{jt}+\beta \log \left(C{I}_{i,t-1}\right)+{\gamma }_{1}Unem{p}_{i,t-1}+{\gamma }_{2}Industr{y}_{i,t-1}+{\gamma }_{3}CL{R}_{it}+{\mu }_i+{\tau }_t+{\epsilon }_{it},$$

(10)

where $y_{it}=\mathsf{\Delta }\mathrm{l}\mathrm{o}\mathrm{g}\left(C{I}_{it}\right)$, and $\rho$measures endogenous spatial dependence in regional carbon-intensity dynamics. A statistically significant $\rho$indicates that changes in carbon intensity in one region are associated with contemporaneous changes in neighbouring regions.

The Spatial Error Model (SEM) retains the baseline convergence equation,

$$y_{it}=\beta \mathrm{l}\mathrm{o}\mathrm{g}\left(C{I}_{i,t-1}\right)+{\gamma }_{1}Unem{p}_{i,t-1}+{\gamma }_{2}Industr{y}_{i,t-1}+{\gamma }_{3}CL{R}_{it}+{\mu }_i+{\tau }_t+u_{it},$$

(11)

but allows the disturbance term to follow a spatial autoregressive process:

$$u_{it}=\varphi {\sum }_j{w}_{ij}{u}_{jt}+{\epsilon }_{it},$$

(12)

where $\varphi$ captures spatial dependence generated by omitted or unobserved factors shared among neighbouring regions.

Finally, the Spatial Durbin Model (SDM) extends the SAR specification by additionally incorporating spatially lagged explanatory variables:

$$y_{it}=\rho W{y}_{it}+X_{it}\beta +W{X}_{it}\theta +{\mu }_i+{\tau }_t+{\epsilon }_{it},$$

(13)

where $X_{it}=\left[\mathrm{l}\mathrm{o}\mathrm{g},,\left(C{I}_{i,t-1}\right),Unem{p}_{i,t-1},Industr{y}_{i,t-1},CL{R}_{it}\right].$The vector $\theta$ captures the association between neighbouring regions' characteristics and local carbon-intensity dynamics. The SDM permits both endogenous spatial interaction through $Wy$ and spatial dependence associated with the explanatory variables.

The LM tests are used to identify the most appropriate form of spatial dependence. The SAR, SEM and SDM models are then estimated to examine whether accounting for spatial dependence changes the baseline conditional β-convergence results.

4. Results

Our empirical findings are presented in this section. First, the estimation sample and the main characteristics of the variables are briefly examined. Next, the estimated conditional β-convergence model is presented using a two-way fixed-effects specification with Driscoll-Kraay standard errors. The analysis then investigates heterogeneous convergence across regions with different initial carbon intensity levels and finally examines the role of spatial dependence using global spatial autocorrelation measures and spatial panel econometric models.

Pairwise correlations among the explanatory variables are moderate, with the highest (absolute) correlation equal to 0.434. The estimated Variance Inflation Factors (VIFs) are all below 2, indicating no evidence of problematic multicollinearity (Table 3).

4.1. Conditional β-Convergence

Table 4 presents the estimates of the conditional $\beta$-convergence model. The Hausman test supports the use of the two-way fixed-effects specification ($X^2\left(4\right)=22.525$, $p<$0.001). The diagnostic tests indicate significant serial correlation ($X^2\left(8\right)=59.411$, $p<0.001$), while the tests for cross-sectional dependence provide mixed evidence. The Pesaran CD test is not statistically significant ($p=0.325$), whereas the Breusch–Pagan LM ($X^2\left(17766\right)=27688$, $p<0.001$) and scaled LM (z = 52.637, $p<0.001$) tests reject the null hypothesis of cross-sectional independence. The Breusch–Pagan test does not reject the null of homoskedasticity at the 5% significance level (BP = 9.103, df = 4, $p=0.059$). Given the presence of serial correlation and the potential for dependence across regions, statistical inference is based on Driscoll–Kraay standard errors.

The estimated coefficient of the lagged level of carbon intensity is negative and highly statistically significant $\beta =-0.1225$ ($p<0.001$), providing strong evidence of conditional $\beta$-convergence across European NUTS-2 regions. This finding indicates that regions with higher initial carbon intensity experience faster reductions in carbon intensity than regions with lower initial levels, after controlling for regional structural characteristics and common time effects.

The estimated coefficient implies an annual convergence speed of 13.1% and a half-life of approximately 5.3 years. This means that half of the initial differences in regional carbon intensity are expected to disappear within about five years.

Among the control variables, unemployment exerts a positive and statistically significant effect on carbon intensity growth, suggesting that weaker labor market conditions are associated with slower progress towards decarbonization. Likewise, regions with a larger industrial employment share exhibit significantly higher carbon intensity growth, reflecting the continued importance of industrial activity in shaping regional emissions. Climate Load Requirement (CLR) also enters the model with a positive and highly significant coefficient, indicating that higher heating and cooling requirements increase energy demand and consequently slow improvements in carbon intensity.

The overall results suggest that, although European regions are converging towards lower carbon intensity levels, the speed of convergence remains strongly influenced by regional economic structure and climate-related energy demand. These results underline the importance of considering both socioeconomic and climatic conditions when designing regional decarbonization strategies.

4.2. Heterogeneity Results

To examine whether convergence differs across regions with different levels of carbon intensity, the sample is divided into three equally sized groups according to their initial carbon intensity (low, medium and high terciles). The conditional $\beta$-convergence model is then estimated separately for each group using the same two-way fixed-effects specification.

Table 5 reveals substantial heterogeneity in regional convergence dynamics. The coefficient of lagged carbon intensity remains negative and statistically significant in all three terciles, confirming the existence of conditional $\beta$-convergence irrespective of the initial carbon intensity level. However, the estimated convergence speed decreases systematically from low- to high-carbon-intensity regions.

Low-carbon-intensity regions exhibit the fastest adjustment, with an estimated convergence speed of 17.0% per year and a half-life of approximately 4.1 years. Medium-carbon-intensity regions converge at an intermediate rate (13.4%; half-life 5.2 years), while regions with the highest initial carbon intensity display the slowest convergence, with an annual convergence speed of 10.8% and a half-life exceeding six years.

The estimated effects of the control variables also differ across terciles. The control variables are statistically significant only for the middle group. For this group, unemployment, industrial structure and CLR are all significantly associated with carbon-intensity growth, whereas none of the controls is statistically significant in the low- and high-carbon-intensity groups. The results indicate that regions with relatively low initial carbon intensity continue to improve their carbon efficiency more rapidly, whereas highly carbon-intensive regions require substantially longer periods. This finding points to heterogeneous regional energy transition pathways and suggests that regions starting from higher carbon intensity may face stronger structural constraints in achieving decarbonization.

$\beta$${\gamma }_1$${\gamma }_2$${\gamma }_3$

4.3. Spatial Dependence

Spatial dependence is first examined using Moran's I statistics. Persistent positive spatial autocorrelation is found in regional carbon intensity throughout the study period. The average Moran's I equals 0.475, while all annual estimates are statistically significant, indicating strong geographical clustering of regions with similar carbon intensity levels. Positive spatial autocorrelation is also observed for the annual growth of carbon intensity, with an average Moran's I of 0.301 and statistically significant values in 22 of the 24 study years. Although weaker than that observed for carbon intensity levels, these results suggest that neighboring regions also exhibit similar decarbonization dynamics, providing strong motivation for spatial econometric modelling (Table 6).

Table 7 reports the estimates of the spatial panel models. The robust LM diagnostics indicate that the SEM specification is the preferred spatial model, while SAR and SDM models are also estimated to examine alternative forms of spatial dependence.

In all spatial models, the coefficient of lagged carbon intensity remains negative and statistically significant, confirming conditional β-convergence after accounting for spatial dependence. Moreover, the estimated convergence coefficients remain close to those obtained from the baseline two-way fixed-effects model, indicating that the principal findings are robust to alternative spatial specifications.

The estimated spatial autoregressive and spatial error parameters are positive and highly significant, confirming spatial dependence among neighbouring European regions. In the SDM model, only the spatially lagged initial carbon intensity is statistically significant ($\theta =0.0289$, $p=0.021$). The spatial lags of unemployment, industrial structure and CLR are not statistically significant at the 5% level.

The spatial analysis confirms that regional decarbonization exhibits clear geographical interdependence. Nevertheless, accounting for spatial dependence does not alter the main conclusion of the paper, namely that European regions display significant conditional convergence in carbon intensity.

4.4. Robustness

The robustness of the results is assessed using alternative inference methods and model specifications. First, the two-way fixed-effects model is estimated using region-clustered standard errors instead of Driscoll–Kraay standard errors. The convergence coefficient remains negative and highly significant under both methods ($\mathit{\beta }=-0.1225$, $\mathit{p}<0.001$). Unemployment remains significant with Driscoll–Kraay ($\mathit{p}=0.016$) and region-clustered ($\mathit{p}<0.001$) standard errors. The industrial employment share is also significant under both approaches ($\mathit{p}<0.001$ and $\mathit{p}=0.029$, respectively), as is CLR ($\mathit{p}<0.001$ in both cases). The similarity of the estimated coefficients and inference indicates that the main findings are not sensitive to the choice of covariance estimator despite the presence of serial correlation and potential cross-sectional dependence. Further, the spatial analysis provides complementary evidence that the convergence result remains stable after accounting for different forms of spatial dependence.

5. Discussion & Policy Implications

The present study examined whether European NUTS-2 regions converge in carbon intensity while accounting for structural characteristics, climate-driven energy demand and spatial interactions. The findings provide strong evidence of conditional β-convergence, indicating that regions with initially higher carbon intensity experience faster reductions, leading to a gradual narrowing of regional disparities. However, convergence is not automatic, as the pace of decarbonization depends on socioeconomic conditions, industrial structure and climate-related energy requirements. These findings are consistent with the growing convergence literature, which emphasizes that regional development paths and structural characteristics shape long-run environmental performance and highlight the importance of place-based approaches to the European energy transition [6,20,21].

A second important finding concerns the pronounced heterogeneity in regional convergence dynamics. Although conditional β-convergence is observed across all groups of regions, the speed of adjustment declines systematically with initial carbon intensity. Highly carbon-intensive regions require considerably longer to converge than regions with lower initial carbon intensity. This pattern suggests that the regions facing the greatest decarbonization challenge also encounter the strongest structural barriers to transition [11,35]. Differences in industrial specialization, technological capabilities and institutional capacity are likely to increase adjustment costs, slowing the adoption of cleaner technologies and delaying improvements in carbon efficiency. These findings indicate that convergence is a gradual process whose speed remains conditioned by regional structural characteristics [11,27].

The slower adjustment observed among highly carbon-intensive regions also supports the concept of regional carbon traps discussed in recent literature [28]. Long-standing specialization in carbon-intensive activities, together with inherited industrial infrastructure and technological capabilities, may reinforce existing production patterns and make structural transformation more difficult. Such path dependence implies that historical development trajectories continue to influence current environmental performance despite increasingly ambitious European climate policies [10,28]. Rather than indicating policy failure, these persistent differences highlight that regions begin the energy transition from very different initial conditions. Consequently, the findings suggest that EU Cohesion Policy and the Just Transition Fund should acknowledge regional diversity by adapting support to local transition needs rather than assuming identical adjustment capacities across European territories.

The analysis further demonstrates that climate-driven energy demand represents an important determinant of regional carbon-intensity convergence. The positive relationship between the Climate Load Requirement (CLR) and carbon-intensity growth indicates that regions facing greater combined heating and cooling needs experience slower improvements in carbon efficiency [13,29,30]. Climatic conditions therefore represent more than an exogenous environmental constraint; they directly influence the energy required to sustain economic activity and consequently affect the pace of regional decarbonization. Although technological progress and the expansion of renewable energy have reduced the carbon content of European energy systems, greater thermal requirements continue to increase energy demand and place additional pressure on energy infrastructure, slowing the pace of regional decarbonization [29,30,33].

Beyond its statistical significance, the inclusion of CLR extends the literature on regional decarbonization by integrating climate-related energy demand into the convergence framework. Previous studies have separately highlighted the importance of climatic conditions, structural characteristics and technological change, whereas the present findings suggest that these mechanisms interact in shaping long-run regional trajectories [13,29,32]. Regions with similar industrial structures may therefore experience different decarbonization pathways because they face different thermal demands and rely on different technologies to meet those demands. Incorporating climate-driven energy requirements thus provides a more comprehensive explanation of regional carbon-intensity dynamics than models relying solely on conventional socioeconomic determinants.

Finally, these findings carry an important policy implication. Climate conditions should not be viewed as exogenous background factors but as structural drivers of regional energy transition [13,29]. The findings suggest that EU climate and energy policies aimed at reducing carbon intensity may also need to address climate-related energy demand alongside renewable-energy deployment. Possible measures include building renovation, energy-efficient heating and cooling systems, and improvements in energy infrastructure. Such measures are likely to be particularly beneficial in regions where climatic conditions create persistently higher energy requirements, allowing climate adaptation and mitigation policies to reinforce one another rather than operate independently.

The importance of regional heterogeneity becomes even more evident when the spatial dimension of decarbonization is considered. The results reveal persistent positive spatial dependence, indicating that neighboring regions tend to follow similar carbon-intensity trajectories over time [26,27]. This finding supports the view that regional decarbonization cannot be understood by treating regions as isolated economic units. Within the European Union, interconnected energy systems, production networks, technological diffusion and coordinated climate policies create multiple channels through which environmental improvements extend beyond administrative boundaries. The spatial econometric analysis further demonstrates that incorporating spatial dependence does not alter the evidence of conditional convergence but provides a more realistic representation of the mechanisms underlying regional adjustment. The superior performance of the Spatial Error Model indicates that carbon-intensity dynamics are influenced not only by observable socioeconomic characteristics but also by unobserved spatially correlated factors, such as institutional quality, historical industrial development, energy infrastructure and regional policy implementation. Ignoring these common regional influences may therefore lead to an incomplete understanding of the convergence process and underestimate the role of spatial interactions in shaping environmental performance.

These findings further suggest that decarbonization should be viewed as a geographically interconnected process rather than a collection of independent regional transitions [26,27]. Technological innovation, renewable-energy deployment and improvements in environmental performance are unlikely to remain confined within administrative borders. Instead, they diffuse through networks of economic and institutional interaction, allowing neighboring regions to benefit from common knowledge and shared policy environments. At the same time, persistent spatial dependence also implies that structural weaknesses may extend across neighboring territories, reinforcing regional disparities when groups of regions face similar economic constraints. This dual role of spatial spillovers highlights the need for coordinated regional policies capable of amplifying positive externalities while limiting the persistence of carbon-intensive development patterns.

From a policy perspective, the results suggest that although European regions are gradually converging towards lower carbon intensity, a uniform policy framework is unlikely to produce equally effective outcomes across all territories. The findings suggest that regions characterized by carbon-intensive production structures or greater climate-driven energy demand are likely to benefit from more targeted interventions addressing their specific structural challenges, including measures supporting industrial decarbonization and improvements in energy efficiency [11,27,28]. These findings reinforce the importance of place-based climate policies that acknowledge regional structural differences and allocate resources according to local transition needs rather than relying exclusively on harmonized policy instruments [1,2].

The significance of Climate Load Requirement suggests that regional climate exposure should be explicitly considered when designing mitigation strategies [13,29,30]. This may encourage closer integration between adaptation and mitigation policies. While European climate policy has traditionally emphasized emissions reduction and renewable-energy deployment, the significance of Climate Load Requirement demonstrates that regional exposure to heating and cooling needs also influences the pace of decarbonization. Examples of measures that may contribute to addressing climate-related energy demand include building renovation, thermal insulation, efficient heating and cooling technologies, district energy systems and demand-side management, which are also consistent with the objectives of the REPowerEU Plan to accelerate energy efficiency improvements and reduce dependence on fossil fuels [2]. These findings also suggest that closer integration between climate adaptation and mitigation policies may become increasingly relevant as climate change continues to reshape regional energy requirements across Europe [29,31,33].

More broadly, this study contributes to the regional decarbonization literature by combining three dimensions that have rarely been examined within a single empirical framework: conditional convergence, climate-driven energy demand and spatial dependence [6,11,26]. While previous studies have typically focused on one of these mechanisms in isolation, the present analysis demonstrates that regional convergence is jointly shaped by structural conditions, climatic factors and geographical interdependence. Considering these dimensions simultaneously provides a more comprehensive explanation of why European regions continue to follow different decarbonization pathways despite operating under a common climate-policy framework.

Overall, the findings indicate that the transition towards lower carbon intensity is progressing across European regions, but its pace remains strongly conditioned by regional characteristics and spatial interactions. Decarbonization should therefore be understood as a geographically differentiated process shaped by local economic structures, climate-related energy demand and interregional spillovers rather than by common policy objectives alone. Recognizing these multiple dimensions provides a stronger foundation for designing more effective regional climate policies and supports the growing emphasis on place-based approaches within the European Green Deal and Cohesion Policy [1,2]. By demonstrating that both climate-driven energy demand and spatial dependence influence long-run convergence, this study offers new evidence on the mechanisms shaping Europe's transition towards a more sustainable and low-carbon regional economy.

6. Conclusions

This study investigated whether European NUTS-2 regions are converging in carbon intensity while accounting for regional structural characteristics, climate-driven energy demand and spatial dependence over the period 2000–2024. By combining conditional β-convergence analysis with spatial econometric techniques, the study provides new evidence on the mechanisms shaping regional decarbonization within the European Union.

The empirical findings reveal that European regions are gradually converging towards lower levels of carbon intensity, although the pace of convergence remains conditioned by regional characteristics, industrial specialization and climate-related energy demand. High-carbon regions continue to adjust more slowly, indicating that structural constraints remain important determinants of regional decarbonization. The analysis also demonstrates that climate-driven energy demand and spatial dependence play a central role in shaping regional decarbonization dynamics, highlighting the importance of considering both climatic conditions and geographical interactions within convergence analyses. These findings support the need for place-based climate policies that combine common European objectives with targeted interventions reflecting regional structural and climatic conditions [1,2].

Beyond its policy relevance, this study contributes to the literature by integrating three dimensions that have rarely been examined simultaneously: conditional convergence, climate-driven energy demand and spatial dependence [6,11,26]. Considering these mechanisms within a unified empirical framework provides a more comprehensive understanding of why regional decarbonization follows heterogeneous pathways despite the existence of common European climate policies. The analysis therefore extends the convergence literature by demonstrating that long-run environmental adjustment is shaped not only by economic fundamentals but also by climatic conditions and spatial interactions.

Although the study provides robust evidence on regional convergence dynamics, several limitations should be acknowledged. The analysis focuses exclusively on European NUTS-2 regions and relies on a specific set of structural and climatic variables, while other determinants of regional decarbonization, such as technological innovation, institutional quality, environmental regulation or energy prices, may also influence convergence dynamics. Furthermore, the spatial relationships are modeled using a single spatial weighting structure, whereas alternative specifications could provide additional insights into the mechanisms of regional interaction.

Future research could extend the analysis by incorporating additional indicators related to innovation, renewable-energy deployment, digitalization and institutional performance, as well as by exploring the role of alternative spatial weighting matrices and nonlinear convergence dynamics. Comparative analyses across different world regions or between alternative territorial classifications could also provide a broader perspective on the factors driving regional carbon-intensity convergence under different economic and climatic conditions.

Our study demonstrates that achieving climate neutrality across Europe requires policies that combine common strategic objectives with sufficient flexibility to accommodate regional diversity [1,2]. By integrating convergence analysis, climate-driven energy demand and spatial econometric methods, new evidence is provided to support more effective and territorially sensitive decarbonization strategies.