Forest cover as a green buffer against urbanization pressures: evidence from the EU Open Access

The relationship between urbanization and economic development has long been studied. Urbanization is thought to be associated with increased prosperity and broad social transformation, supporting this process. However, it is uncertain whether this positive relationship will continue in the long term (Turok et al., 2023), particularly as the pace of global expansion puts pressure on environmental sustainability, making the process uncertain. This situation is particularly evident in industrialized environments such as the European Union, where population growth and migration-focused [push from the countryside and pull from the city (Heins, 2004)] dense settlement patterns and construction-focused growth intensify the pressure on ecosystems. The pressure on environmental sustainability manifests itself in increasing carbon dioxide (CO₂) emissions, deteriorating air quality, and widespread land use (Weber and Sciubba, 2019; Andrei, 2023). This process is particularly evident in Western Europe, where regional population growth, rising CO₂ emissions, and expanding urban land use are occurring simultaneously (Weber and Sciubba, 2019). The construction sector, which thrives in direct proportion to urban population growth, is central to this ongoing connection. It is incorrect to assess urban growth solely in terms of population increase. This process also means increased land area (at the expense of shrinking agricultural land), more material stock, increased structural volume, and more energy locked into the built environment. UNEP (2022) estimates that energy use in the sector in 2021 was associated with approximately ten gigatons of CO₂ equivalent emissions. Furthermore, the sector is the second-largest consumer of plastics. In 2019, 17% of plastic produced (77 million tons) was used by the building and construction sector. The transportation–construction–packaging sectors together account for 60% of total plastic demand (OECD, 2022). Given this scale of footprint, it is difficult to assume that environmental pressures will simply decrease as incomes rise; this is not merely an optimistic interpretation sometimes attributed to the Environmental Kuznets Curve.

This is precisely where the concept and importance of green infrastructure comes into play. From an urban forestry perspective, this scope is not limited to large forest blocks on the city periphery; it also includes peri-urban forests, green belts, city parks, and park forests, as well as tree systems embedded within the urban fabric (particularly street trees and tree-dominated green spaces). This approach is consistent with the framework that defines urban green infrastructure as a strategic network of natural and semi-natural areas designed to produce multiple ecosystem services (Hansen and Pauleit, 2014; European Commission, 2024; Kodym et al., 2025). This framework is particularly important in emphasizing the role of “stepping stone” elements such as street trees, private gardens, and green roofs/walls in ecological continuity (Badiu et al., 2019; Kodym et al., 2025). These tree-based green resources are central to the agenda of ecosystem services in cities and, therefore, their planning and management are directly related to air quality, ecological footprint, and climate outcomes. In this broader conceptual framework, however, the empirical focus of the present study is narrower. We do not measure green infrastructure in its full urban-planning sense; rather, we focus specifically on forest cover as a macro-level land-cover indicator that captures an important tree-based component of ecological buffering capacity. This raises the question: If urban stress factors intensify exposure and emissions, to what extent can forest-based natural cover absorb these negative effects? This is where the Green Buffer Hypothesis (GBH) comes in, suggesting that natural areas can reduce the damage associated with negative environmental impacts related to industrialization-driven growth and development, traffic, and urban heat and CO₂ emissions, as measured for ecosystems and human health (van den Berg et al., 2010). In fact, these mechanisms are not uncertain or unpredictable. Forests and natural vegetation affect air currents, promote the accumulation of particulate matter on leaf surfaces, and can absorb gases such as NO₂, SO₂, and O₃ through stomata (Jimenez et al., 2021). These processes can generally be considered urban forest ecosystem services, and their strength is often dependent on the size, type, and distribution of woody vegetation in urban and peri-urban areas. Forests play a key role in the climate system by sequestering carbon. While this role is important, its preventive effect and strength cannot be fully guaranteed. In fact, carbon losses and declines in ecosystem values have been well documented in cases of forest loss or degradation (IPCC, 2023). The conclusions that can be drawn from this depend on the answer to a simple empirical question: Does the extent of forest cover, used here as a macro-level indicator of tree-based natural cover rather than a full measure of green infrastructure, mitigate the environmental impacts associated with urbanization, and does this buffering role function similarly for air pollution and carbon emissions? The answer to this question also has the potential to clearly reveal the solutions.

In the literature, the effects of urban development and forest share on air quality are usually considered independently (Wu and Liu, 2023). This type of analysis ignores the interaction between the two elements in the urban landscape, which can hinder a full understanding of their impact on pollution dynamics (Venter et al., 2024). Panel data studies, especially at the macro- or regional-level, are insufficient to model an interaction term testing whether the pollution-enhancing effect of development is lower in regions with higher forest shares.

This study aims to econometrically test the interaction between urbanization and forest share using country-year panel data from EU member states. It thus aims to provide general empirical evidence on the mitigating effect of the GBH on pollution, thereby contributing to the development of integrated land-use and environmental policies that help the EU achieve its sustainability goals.

In summary, this study offers three key contributions: (1) it tests the Green Buffer Hypothesis, which has been examined only to a limited extent in the literature, both conceptually and empirically, at the level of European Union countries, and provides a new explanation for the conditional relationship between urbanization and environmental quality. (2) it more accurately represents spatial reality by using the built-up ratio (BU), derived from GHSL-based land-use statistics, instead of traditional population-based indicators to measure urbanization; (3) It offers an evidence-based policy perspective to co-design urban planning and green infrastructure policies by developing a multilevel panel analysis framework that jointly assesses forest cover, urbanization, and carbon and air pollution dynamics. At the same time, the policy implications should be interpreted in light of the study's measurement strategy: the empirical evidence speaks most directly to the role of forest cover as a tree-based natural buffer, rather than to every possible component of urban green infrastructure.

Over the past decade, a growing body of research has examined the interconnections between urbanization, green infrastructure, and air pollution, combining conceptual discussions with empirical modeling to clarify how cities can grow more sustainably. This section first examines the conceptual foundations underpinning the urbanization–green infrastructure–pollution relationship; it then addresses connections among urbanization, emissions, and carbon dynamics. Subsequent subsections discuss the effects of green infrastructure and forest cover on air quality at micro, meso, and macro scales, evaluate interaction models (BU × FOR), and assess measurement strategies for the relationship between urbanization and forest share.

Measuring urbanization is not limited to population growth or spatial expansion alone for understanding and improving urban planning. Measuring urbanization requires the combined use of criteria and methods that address the physical, social, economic, environmental, and spatial dimensions of the city within a broader framework (Zhou et al., 2023). This approach, based on the assumption that a single indicator (e.g. population urbanization rate) cannot fully reflect the “quality” of urbanization, necessitates the use of multi-indicator assessment frameworks and different weighting techniques (e.g. PCA, entropy) (Zhou et al., 2023). The importance of measurement increases particularly during periods of rapid urbanization. For example, according to Northam's (1979) S-curve approach, the rapid urbanization phase corresponds to the stage where the level of urbanization ranges between approximately 25% and 70% (Zhang et al., 2022). The acceleration of urbanization increases energy and natural resource consumption through the effects of its sub-elements, thereby intensifying environmental degradation, carbon emissions, and pressures on urban ecosystems (Wang et al., 2019).

At this point, green infrastructure provides a complementary framework for sustainable urban management as an approach that aims to integrate natural and semi-natural systems into urban areas in a planned manner, providing ecosystem services, supporting biodiversity, and encompassing the planning of natural and semi-natural areas (European Commission, 2024). Accordingly, urban green infrastructure defines not only a physical system but also an important interconnected ecological and social ecosystem. This ecosystem encompasses a range of services, including the retention of rainwater for reuse without contamination, the reduction and removal of pollutants, microclimate regulation, and carbon sequestration (Grabowski et al., 2022). It also encompasses natural habitat connectivity and recreational opportunities for urban communities (Herath and Bai, 2024). Furthermore, urban green infrastructure enhances ecosystem services in cities and increases communities' resilience to environmental stressors (Yao et al., 2025). As emphasized by Vliet and Hammond (2025), the strategic use of urban green infrastructure also involves its implementation in design and adaptive planning frameworks for cities. In fact, urban design and planning become fundamental elements of the task. In the present study, however, green infrastructure is treated as the broader conceptual context rather than as the directly observed empirical variable. Our empirical analysis focuses specifically on forest cover, which we use as a macro-level land-cover indicator of tree-based ecological buffering capacity, not as a complete measure of all green-infrastructure components.

A review of the literature reveals that since the mid-2010s, the green infrastructure literature has expanded rapidly (approximately 210,000 studies on Google Scholar and 8,850 studies in WOS as of December 20, 2025). One of the most important reasons for this is that it offers a language that links environmental quality to the daily urban experience (Shao et al., 2021). Green infrastructure is naturally compatible with the sustainable city agenda. This compatibility does not mean that implementation is simple. It is an important and challenging process. When green infrastructure projects move from principles to implementation, certain trade-offs immediately emerge. This is a costly process because preserving ecological functions, meeting social needs, and sustaining economic activity do not always point to the same design or location, and because priorities can conflict in multi-faceted processes (Hanna and Comín, 2021). For that reason, equity and participation should be read as conditions of performance rather than as moral footnotes. They influence which issues are considered preferable, which neighborhoods are renovated, and whether green investments are used to rectify or reproduce existing spatial inequalities (Grabowski et al., 2023). Thus, green infrastructure can improve livability, but it does so reliably only when it is planned and maintained as part of the city's primary development decisions, rather than being treated as a cosmetic layer added afterward.

At the same time, the evidence on forests under urban expansion warns against expecting a single, uniform trajectory. Kathmandu Valley shows a clear land-use exchange: between 2013 and 2021, residential land grew by nearly 90 km² while forest cover fell by roughly 24 km² (Bhomi et al., 2024). The pattern for the United States is less linear, with population expansion linked to a slight mitigation of deforestation, but, again, the topology of settlements was linked to higher deforestation rates from 2001 to 2006 (Clement et al., 2015). Long-term evidence for Türkiye's Trabzon region confirms a similar trend: it is linked to population expansion due to urbanization, and the rate of deforestation is 0.42% per annum (Keleş et al., 2008). In coastal China, the situation is further complicated: forests declined by about 103 km² per year near the coast, but increased inland by about 26 km² annually (Zhu et al., 2019). Taken together, these studies underscore a simple point in the pollution debate: whether forest cover can buffer urbanization-related emissions and exposure depends on where growth occurs and on whether it erodes the very land cover expected to provide that buffer. That is why the urbanization–pollution relationship is best interpreted in conjunction with land-cover change, and why forest cover is analytically more meaningful as a moderator than as a background condition.

Urbanization is often treated as a direct driver of CO₂ emissions and air pollution, but the relationship is rarely so straightforward. Two cities can grow at a similar pace and end up with very different emission trajectories. What usually explains the difference is not urban growth itself, but the composition of that growth: the energy system that feeds it, the industrial mix it attracts, and the capacity of planning and regulation to keep pace with construction, transport, and rising demand. Some studies do report a fairly tight coupling between urban expansion and emissions. In northern China, increases in urbanization and CO₂ emissions are interrelated, with evidence of both unidirectional and bidirectional causality (Siqin et al., 2022). In the CEMAC countries, growth through urbanization is associated with increased environmental pressure (Ngong et al., 2022). Neither of these instances would be particularly surprising in countries in which energy-intensive growth is combined with urbanization and industrialization, because increased growth and mobility can be quickly manifested as increased fuel consumption and emissions.

Elsewhere, the relationship differs. Indonesia, for instance, exhibits an inverted U-shaped pattern: emissions rise in the early phase of urbanization and then decline after an income threshold is reached (Ahmed et al., 2019). That drop is often interpreted as a sign that later-stage urban development can coincide with efficiency gains, technological upgrading, or changes in the sectoral structure. A less favorable variant of “declining emissions” has been reported in some low-income settings. When urban growth slows, emissions may fall simply because economic activity contracts or energy supply is constrained, not because a low-carbon transition has taken hold (McGee and York, 2018). This is why it is difficult to discuss a single urbanization–emissions curve. Where infrastructure, land-use control, and energy policy keep pace with expansion, emissions can plateau and, in some cases, begin to ease. When urban expansion outpaces infrastructure and regulatory capacity, environmental pressures tend to accumulate rather than disperse. In such settings, CO₂ rises alongside local pollutants and their precursors, including NOx and SO2 (Salahodjaev, 2014; Ma and Ogata, 2024). Differences in industrial structure and energy intensity help explain why similar rates of urbanization can yield markedly different outcomes (Zhao et al., 2025). This is also the core policy point: low-carbon urbanization is less about slowing urban growth in the abstract and more about the energy, land-use, and technological decisions that shape how cities grow.

At this point, per capita emissions bring another dimension into focus. In the early stages of urbanization, per capita CO₂ emissions generally increase with the expansion of construction, transportation, and industrial activities, which is inherent to the nature of the process. However, over time, factors such as increased efficiency, cleaner technologies, or changes in the energy mix (increased environmental awareness, legal restrictions) come into play and have the potential to reverse this initial trend. For example, Jordan is a notable case where per capita emissions decreased and renewable energy use increased during the urbanization process (Bashayreh et al., 2024). Research articles focusing on “new type” urbanization in China also show a similar increase followed by a decline with the spread of cleaner technologies (Zhao et al., 2025). Research conducted in SAARC countries, however, highlights an alternative threshold scenario. According to this, the crises caused by the urbanization process are mild at first, but as urbanization intensifies beyond a certain point, they tend to increase (Anser et al., 2020). The common point among all these examples is that urbanization is neither positive nor negative in itself. Yes, the rate of urbanization inherently exacerbates negative impacts, but the pattern of growth significantly and substantially determines emission dynamics.

The impact of green infrastructure is felt through the effects of elements visible at the micro scale. It is most noticeable at the street level, particularly through roadside trees and vegetation adjacent to buildings. Due to their biological structure, the leaves and bark of trees and smaller plants can capture certain airborne particles, reducing local exposure (Jeanjean et al., 2016). Furthermore, all living organisms inhabiting this environment can have a positive impact on the process. However, this does not always mean greener, cleaner air. In a street canyon with poor air circulation, dense and poorly placed vegetation can further slow airflow and delay the dispersion of pollutants; in some cases, it has even been observed to increase concentrations near the ground. Therefore, the outcome depends more on plant positioning and the street's physical conditions than on their mere presence; density, layout, street geometry, and wind regime are decisive factors. Indeed, over very short distances, within a radius of approximately 15–60 m, the relationship between total green space change and pollution was found to be weak on average; it has been reported that vegetation cover can also lead to adverse effects if aerodynamic conditions are unfavorable (Venter et al., 2024).

At the mesoscale, which is a combination of micro- and macro-scale processes, vegetation retention and dispersion mechanisms are better balanced. An increase in the percentage of green areas at the neighborhood level is associated with a decrease in PM₁₀. Likewise, the negative correlation between tree cover density and pollutant concentration is clearer at the meso-scale (Douglas et al., 2019). Concentrations of CO, NO₂, SO₂, and PM₁₀ are lower in areas with high tree cover across the city, while traffic density, population density, and other factors increase pollutant concentrations. In addition, a strong relationship exists between the ratio of forest area to inhabited structures and PM₂.₅ levels, and pollutant concentrations at the mesoscale significantly decrease in cities with increased forest cover (Feng et al., 2017).

Macro-scale forest cover and large green areas, which have a direct impact on air pollution, are shaped by environmental policies, meteorological variables, and emission sources (Grabowski et al., 2022). Changes in green space (especially tree cover) in the US and Europe between 2010 and 2019 are associated with significant reductions in city-scale NO₂, PM₁₀, and PM₂.₅ concentrations of approximately 1–3% per year, indicating that the long-term effects are gradual and moderate. In addition, simulations predicted that increasing forest cover in cities to a certain level could bring annual average PM₂.₅ and PM₁₀ concentrations closer to national air quality standards (Zhou et al., 2019). Because the present study uses forest share rather than a full inventory of urban green-infrastructure elements, the macro-panel analysis should be read primarily as evidence on the buffering role of tree-based land cover at the national scale.

These findings indicate that the impact of green infrastructure is meaningful when evaluated in conjunction with local morphology and emission patterns, rather than as a single average effect.

Studies on the environmental impacts of urbanization generally consider human activities and natural infrastructure as separate axes of analysis. Human activity indicators (share of built-up area, population density, transportation infrastructure density, etc.) are generally considered environmental pressure variables, while land cover indicators (forest cover, green area ratio, vegetation index, etc.) are generally examined as indicators of environmental regulation or of carbon-sink capacity (Kumar et al., 2019; Muresan et al., 2022). However, interaction (moderation) analyses that examine how these two indicators condition each other, i.e. the environmental impacts of human activity in relation to natural cover, are very limited in the literature. Most existing studies treat urbanization and forest cover as independent determinants and largely ignore the effects of their interaction, described as “development and forest share,” on air quality (Jennings et al., 2021). Meta-analyses of the impact of green infrastructure on air quality or carbon emissions generally focus on single effects and do not quantitatively test how the buffering effects of green spaces vary with urbanization density (Bočkarjova and Kačalová, 2021). However, recent spatial analyses show that urban green areas can significantly reduce PM₂.₅ and NO₂ levels, but this effect weakens as the density of surrounding housing increases (Muresan et al., 2022). These findings suggest that the interaction between urban green infrastructure and land-use patterns can be decisive for environmental outcomes. This broader conditioning logic is also consistent with recent forest-related panel evidence showing that the effects of economic growth, ICT, and human capital vary across countries with high versus low forest density, underscoring that forest-related environmental mechanisms are often context-dependent (Soyyigit et al., 2025). In our empirical setting, this broader interaction logic is examined through forest share, which serves as the observed macro-level indicator of tree-based natural buffering capacity.

In this context, the study aims to fill this gap by conceptualizing the Green Buffer Hypothesis. According to this hypothesis, as the forest share increases, the polluting effect of construction weakens. Therefore, in regions with a high proportion of forest, the marginal impact of an increase in the share of built-up area on air pollution (PM₂.₅) and on carbon emissions is attenuated. This relationship is empirically tested using the interaction term “construction and forest share”, and the panel fixed-effects model controls for both unobserved heterogeneity at the country-year level and common shocks. With this approach, the study considers the environmental impacts of forest-based natural cover as an independent variable and a mechanism that conditions (moderates) the negative impacts of urbanization. Thus, the role of forest share in areas under pressure from urbanization is quantified, and the air quality and climate co-benefits of co-designing urban plans and forest-management policies are evaluated. In this respect, the study brings methodological and conceptual innovation to the literature on the forest cover–urbanization–pollution relationship. Therefore, explicitly testing the BU × FOR interaction term at the panel level is a critical step toward filling this notable gap in the literature.

Urbanization and housing growth, urban-scale energy consumption, traffic density, and improper land use are the main drivers of increased PM₂.₅ concentrations and CO₂ emissions (Luqman et al., 2023). In 91 cities, a positive relationship was found between urbanization and CO₂ emissions. Similarly, green infrastructure, especially urban forests and urban green spaces, can contribute positively to air quality by improving particulate matter retention, dry deposition, and air circulation. These mechanisms have been systematically examined (Wróblewska and Jeong, 2021). However, no study in the literature examines whether the urbanization rate and the share of forest cover interactively moderate pollutants using panel data. This study aims to fill this gap. Therefore, the following methodological section aims to make measurement-based and policy-relevant contributions by testing the GBH within a panel framework using the BU × FOR interaction term.

The variables used in the study are presented in Table 1.

Table 1 defines the variables used in the analysis and also helps clarify their basic distributional features. In the dependent variables, the mean ln(PM₂.₅) is 2.75 and the mean ln(CO₂ emissions per capita) is 1.94. The main explanatory variables, BUw (urbanization) and FORw (forest cover), are constructed as two-way within residuals after removing country and year fixed effects. For this reason, their sample means are approximately zero, and their remaining variation reflects within-country, over-time fluctuations net of common year shocks and time-invariant country characteristics. Per capita income (GDPpc) has a mean of about 9.87 on the log scale. The urban population ratio (URB) averages 70%, while the share of arable land (ARB) averages 26%. The panel includes 850 country-year observations and provides meaningful variation across both countries and time. Overall, these features indicate that the dataset is well suited to examining how air pollution and carbon emissions covary with economic and environmental conditions.

The built-up area variable (BU) is obtained from the GHSL Country Stats database and indicates how much of a country's land area is built-up. It is calculated as built-up land area (km²) divided by total land area (km²), multiplied by 100. In constructing this measure, built-up and land-area values are first taken from the GHSL country-level statistics and then aggregated to the country-year level. We prefer this measure because it reflects the physical land-use side of urbanization more directly than population-based indicators. In the empirical analysis, BUw and FORw capture the within-country changes over time in built-up area and forest share after controlling for country and year fixed effects.

Two basic models were established in the study. The established models and hypotheses are presented below.

1. Empirical Model (PM₂.₅)

Hypotheses and expected signs for Model 1 are defined as follows. (see Table 2)

1. Empirical Model (CO₂ Emissions Equation)

Hypotheses and expected signs for Model 2 are defined as follows. (see Table 3)

This section summarizes the methodological approach used to assess the validity of the GBH using a panel of EU countries for the period 1990–2023. The empirical approach is a multiple-assessment method that combines advanced panel-data methodologies and Fourier expansions to correct for cross-sectional dependence (CSD), heterogeneity, and differing orders of integration among the variables. We work at the country-year level because both outcomes are monitored and reported in a nationally comparable form across EU member states, and this is also the scale at which many EU policy instruments are designed and assessed. Using country-level PM₂.₅ exposure and per-capita CO₂ emissions therefore allows us to speak directly to EU-wide policy discussions, including Fit for 55 and the LULUCF Regulation (European Union, 2018, 2023; European Commission, 2021). At the same time, national forest share serves as a macro indicator of the urban and peri-urban woodland resource that anchors the wider green infrastructure network, which is central to an urban forestry and urban greening perspective.

In our study, we follow a carefully planned, balanced strategy to ensure that empirical results are based on sound diagnoses. Accordingly, we first test for cross-sectional dependence using “the LM, CDlm, CD, and LMadj statistics” (Breusch and Pagan, 1980; Pesaran, 2004; Pesaran et al., 2008). This step is particularly important in the EU context. This is because common shocks and spillover effects can spread across member countries and ignoring this can distort traditional panel inferences and lead to misinterpretations. In our study, we also assess slope homogeneity using the Pesaran and Yamagata (2008) tests to determine whether the underlying relationships differ across countries. These preliminary checks guide subsequent estimations and ensure that the reported coefficients and inferences are appropriate for the EU sample. Since the formulas for the relevant tests are comprehensively covered in the literature, they will not be repeated here (see El-Montasser et al., 2016; Kazak et al., 2025).

To determine the degree of integration, the Panel Fourier LM Unit Root Test Nazlioglu and Karul (2017), which accounts for structural breaks and nonlinear trends that may vary across countries and over time, was used.

The PLM and ZLM values for the Fourier panel unit root test using Equation (3) are calculated using Equations (4) and (5), respectively.

The unit root tests used in this study indicate that the variables are integrated in different orders. Therefore, to assess whether there is a long-term equilibrium relationship, we apply the Fourier panel cointegration test defined by Olayeni et al. (2020). This approach is implemented using the fractional frequency flexible Fourier form (FFFFF), which incorporates soft structural changes in the deterministic component. The relevant test in the study will be referred to as the Panel Cointegration Test Results (FFFFF). The test statistics are defined as follows (Equation 6):

We estimate both Model 1 (PM₂.₅) and Model 2 (CO₂) using high-dimensional fixed-effects (HDFE) regressions to accommodate the panel structure of the data (Correia, 2016). HDFE extends the standard fixed-effects framework by absorbing multiple sets of fixed effects simultaneously, such as country and year effects, doing so controls for unobserved, time-invariant cross-country differences as well as common shocks that affect all countries in a given year, thereby reducing omitted-variable bias.

Estimations were based on the within transformation. To ensure robust inference under potential serial correlation and cross-sectional dependence, we report alternative standard-error estimators across specifications: standard errors clustered by year in M1, two-way clustered by country and year in M2, and Driscoll–Kraay standard errors with lag 3 in M3 (Driscoll and Kraay, 1998).

This approach is widely used in empirical work in environmental, energy, and development economics, so its formal derivation is not repeated here (see Correia, 2016).

Although the panel homogeneity diagnostics are informative, they should not be taken to imply fully identical country-specific relationships across all EU members. The sample includes countries with different urbanization trajectories, forest-management traditions, and environmental-policy settings. Accordingly, the HDFE coefficients reported below are interpreted as average within-country effects for the pooled EU sample after controlling for country and year fixed effects. The empirical strategy is therefore intended to identify the direction and average magnitude of the Green Buffer relationship at the macro panel level, while allowing its exact strength to vary across national contexts.

Finally, the Panel Fourier Toda–Yamamoto (PFTY) test, developed by Yilanci and Gorus (2020) as a panel extension of the Fourier Toda–Yamamoto test, was employed to examine causal relationships among CO₂, PM₂.₅, and the study's independent variables. This test was used because it is suitable for variables with different integration orders (I(0) or I(1)), can incorporate soft structural breaks and nonlinearities using Fourier functions, and provides more reliable results for small-to medium-sized panels. In this test, the null hypothesis (H0) indicates the absence of causality. The fact that the relationships revealed by the HDFE/FE regression model were also analyzed as bivariate causal relationships using a Fourier expansion added depth to the study. For this purpose, the following bivariate panel VAR model was estimated.

The Panel Fourier Toda-Yamamoto (PFTY) causality test is based on the generalized vector autoregressive (VAR) model for panel data. The model is represented by the following equations (Equations 7 and 8) for each cross-sectional unit (i = 1, 2, 3, …, N) and each time period (t = 1, 2, 3, …, T). The optimal lag order, denoted by “$ki$” in the model, is estimated using a model selection criterion such as the Akaike Information Criterion (AIC) or the Schwarz Bayesian Criterion (SBC). “$dmax$” indicates the maximum integration order of the variables in the model. Fourier functions “$sin(2πtfiT)$ and $cos(2πtfiT)$” are used to include soft structural shifts and nonlinear trends in the analysis; here, “$π≈227$” represents the time trend term, “T” represents the sample size, and “$fi$” represents the frequency. To test the fundamental hypothesis of the absence of causality, Equations (7) and (8) are estimated separately for each variable in the model. The Fisher test statistic for PFTY causality tests can be derived using the following formula, which is based on the sum of squared residuals of the unrestricted and restricted models obtained from estimating these equations:

In this study, the PFTY approach is employed to explore the direction of Granger-causality within the Green Buffer framework across EU countries while allowing for smooth structural shifts over time. We examine whether built-up density (BU), forest cover (FOR), and their interaction (BUFOR) help predict changes in PM₂.₅ exposure and CO₂ emissions, and whether reverse causal links are also present. To align with the baseline specifications, we additionally report causality patterns involving income (GDP and GDP²), urbanization (URB), and arable land (ARB). The causality findings are summarized separately for PM₂.₅ and CO₂.

As detailed in the Methodology section, cross-sectional dependence (CSD) and the homogeneity of slope coefficients were tested in the first stage of our empirical analysis to determine the basic econometric properties of the variables in the panel dataset. In this context, the LM, CDlm, CD, and LMadj tests were applied to determine the degree of dependence among the variables. Furthermore, the two tests were used to examine whether the slope coefficients were consistent across countries in the panel and thus to assess the homogeneity assumption. The findings of these preliminary analyses are summarized in Table 4 and form the basis for determining the most appropriate econometric methods for panel data estimation.

In both models, the dependent variables are PM₂.₅ and CO₂ and the independent variables are BU, FOR, BUFOR, GDP, GDP², ARB, and URB; all variables have p-values less than 0.01. Therefore, cross-sectional dependence was observed in both models.

The probability values of the Delta and Delta-adjusted results for both dependent variables (PM₂.₅ and CO₂) exceeded 0.1. Therefore, it was concluded that the slope coefficients were homogeneous, as shown in Table 5.

The panel Fourier unit root test, applied to 9 variables (2 dependent and 7 independent), indicated that FOR and BUFOR were stationary in levels, while the remaining variables had unit roots in levels, as shown in Table 6. Therefore, causal relationships among the variables will be investigated using the panel Fourier Toda–Yamamoto causality test, which can detect causal relationships even when variables are integrated of different orders.

Based on the GLS and PP tests applied to the dependent variables PM₂.₅ and CO₂ in the two established models, the cointegration test detected a cointegrating relationship between the dependent and independent variables at the 1% level, as shown in Table 7. The long-term coefficient estimates for these models are as follows.

In the models reported in Table 8, the dependent variable is the population-weighted annual average PM₂.₅ concentration at the country level, and all coefficients are interpreted as semi-elasticities. For regressors expressed in percentages (e.g. BU, FOR, URB, ARB), a one-unit change corresponds to a one-percentage-point change. In this context, BUw and FORw denote the residual (two-way within) components of the urbanization rate and forest share, respectively obtained after removing country and year fixed effects (i.e. the variation in BU and FOR that remains net of time-invariant country heterogeneity and common time shocks). The BUw × FORw interaction term captures how the marginal association between urbanization and PM₂.₅ varies with forest cover, conditional on the controls and fixed effects. Among the control variables, URB represents the ratio of the urban population to the total population (%), ARB represents the ratio of arable land to total land area (%), and GDPpc represents per capita income. All models include country and year fixed effects, while standard errors are estimated using year clustering in M1, two-way (country and year) clustering in M2, and Driscoll–Kraay standard errors (lag = 3) in M3.

When the results are examined, the FORw coefficient is negative and highly significant (p < 0.01) in M1 and M3; a one-percentage-point increase in forest share (relative to the level implied by country and year fixed effects) is associated with an approximately 1.1% decrease in PM₂.₅ in the same year. The BUw × FORw interaction is negative in all models and statistically significant in M1 (−0.029, p < 0.05); this supports the GBH by indicating that the pollution-increasing effect of urbanization weakens (and may even reverse at sufficiently high levels of forest cover) as forest share increases. When considered alone, the coefficient of BUw is small and statistically uncertain; however, its effect is conditional on the level of FORw through the interaction term. Under the interactive specification, the marginal effect of urbanization is (⁠$∂ln(PM2.5)∂BU=βBUw+βBUw×FORw·FORw$⁠). Using the M1 coefficients, when FORw = +5, a one-percentage-point increase in BU reduces PM₂.₅ by approximately 16% (−0.162); when FORw = −5, it increases PM₂.₅ by approximately 13% (+0.128). The threshold at which the sign of the effect changes is (⁠$FORw≈−βBUw/βBUw×FORw≃−0.59$⁠) percentage points, indicating that when forest share falls slightly below its fixed-effects-implied level, the pollution-increasing effect of increased urbanization becomes dominant. Overall, the sign structure is consistent across models, implying that forest cover plays a measurable buffering role against air pollution in the context of short-to medium-term domestic fluctuations.

The estimation results presented in Table 9 detail the roles played by income, urbanization, land use, and forest cover in per capita CO₂ emissions. The findings indicate that the coefficient on the per capita income variable is positive and highly significant across all model specifications. This result confirms that carbon emissions increase with economic activity. Similarly, the urban population ratio (URB) and agricultural land share (ARB) are generally positive and significant. These findings indicate that both urban concentration and agricultural activities are associated with increased emissions.

The individual effects of urbanization (BUw) and forest area (FORw), which are the model's main focus, are not statistically significant. However, the interaction term between these two variables (BUw × FORw) is negative and statistically significant at the 5% level in models M1 and M3. This finding indicates that the effect of urbanization on emissions diminishes and may even reverse direction as forest area increases. Indeed, the marginal effects analysis reveals that urbanization increases CO₂ emissions when the forest share is below average, but decreases them when it is at or above average. The implied threshold lies roughly one percentage point below the country-specific average forest share.

To clarify the economic significance of this interaction, the marginal effect of built-up expansion on environmental pressure can be written as ∂Y/∂BUw = β₁ + β₃FORw. Setting this marginal effect equal to zero yields the implied threshold value of FORw, namely -β₁/β₃. In the CO₂ estimates, this threshold is negative, which indicates that the marginal environmental effect of built-up expansion is already attenuated at approximately average within-country forest conditions and becomes even weaker as forest cover rises further. Because the interaction is estimated using within-transformed variables, this threshold should not be interpreted as a single universal absolute forest-share requirement for all EU countries. Rather, it indicates that, relative to a country's own average forest conditions, higher forest availability strengthens the buffering role of natural land cover against urbanization pressures. Therefore, urbanization increases carbon intensity in countries with relatively low forest cover, but can reduce emissions pressure where forest availability is stronger.

In terms of explanatory power, the R² values in the M1 and M2 models are high (≈0.90), indicating that they explain a significant portion of the variance in the dependent variable. Although R² is lower in the M3 model with Driscoll–Kraay standard errors, the interaction term retains its sign and significance, confirming the methodological robustness of the findings. The analyses of CO₂ emissions support the validity of the GBH in the carbon context.

These findings have clear policy implications. When urbanization proceeds without protecting forest resources, carbon emissions tend to increase. By contrast, planned urban growth accompanied by the protection or expansion of forest areas is more likely to be associated with lower carbon intensity. This shows that urbanization strategies should be designed to go beyond physical expansion alone. Forest cover and green infrastructure should therefore be treated as core components of urban planning for long-term carbon management and sustainable growth.

Table 10 evaluates whether the EKC model holds for per capita CO₂ emissions and PM₂.₅ exposure. Empirical evidence clearly supports the EKC model for CO₂. The income term (GDPpc_c) is positive, while the quadratic term (GDPpc_c²) is negative and highly significant, consistent with the expected EKC sign model for an inverted U-shaped relationship. These results indicate that emissions increase at low income levels and then tend to decline as countries cross a certain income threshold. According to the coefficients we estimated in the model, this turning point occurs at approximately upper-middle income levels for our sample.

The PM₂.₅ results, however, do not confirm the EKC hypothesis. In this model, the coefficients on income and income squared are not statistically significant. This indicates that per capita income growth does not imply a systematic inverse-U-shaped relationship with air pollution levels. In other words, economic growth does not produce a clear downward trend in PM₂.₅ exposure; air quality may be more sensitive to factors other than income. This divergence between CO₂ and PM₂.₅ is substantively important. While CO₂ is a cumulative and more globally mediated pollutant that is closely linked to the energy system, technological change, and long-run structural transformation, PM₂.₅ exposure is more strongly shaped by local emission sources, urban form, traffic intensity, heating patterns, and spatial planning conditions. For that reason, income growth alone may be insufficient to generate an EKC-type adjustment for PM₂.₅ within the EU sample, even when it does so for CO₂.

Other control variables are significant for both dependent variables. In the CO₂ model, the urbanization (URB) and agricultural area (ARB) variables exhibit positive and statistically significant coefficients, indicating that urbanization and land-use dynamics are associated with higher carbon emissions. In the PM₂.₅ model, urbanization is positive and significant, while the agricultural area share is negative and weakly significant. Furthermore, in both models, the interaction between building density (BUw) and forest share (FORw) is negative and significant; this confirms that forest cover mitigates the negative effects of building density on both carbon emissions and particulate matter concentrations. At the same time, these coefficients should be interpreted as pooled average within-country effects for the EU sample rather than as evidence of a perfectly uniform buffering mechanism across all member states. Given cross-country differences in urbanization trajectories, forest-management structures, and environmental-policy settings, the magnitude of the Green Buffer effect may remain context-dependent even where the overall panel evidence supports the hypothesis.

Overall, while the CO₂ results reveal the classic EKC pattern between income and emissions, the PM₂.₅ results do not support this hypothesis. These findings emphasize that greenhouse gas emissions and local air pollution indicators respond differently to economic growth and therefore require different policy instruments. More specifically, CO₂ mitigation is more tightly linked to energy transition, technological upgrading, and carbon-management strategies, whereas PM₂.₅ reduction depends more directly on local land-use regulation, transport management, and spatially targeted environmental planning. Although the pooled estimates support the Green Buffer Hypothesis, they should not be read as implying fully identical country-specific relationships throughout the EU. This point is also consistent with the country-level causal evidence reported below, where the direction and intensity of the relationships vary across member states. Taken together, the results suggest that the buffering role of forest cover is present at the macro panel level, but its exact strength is likely to differ across national contexts.

In the final stage of the empirical analysis, causal relationships between variables at different levels of stationarity were examined using the panel Fourier Toda–Yamamoto test. First, causality was examined for PM₂.₅, and the results are presented in Table 11.

Finally, the causality of CO₂ was examined; the results of the causality test are presented in Table 12.

The results of the Fourier TY causality test generally support the fundamental assumptions of the “Green Buffer Hypothesis.” According to the PM₂.₅ model results (Table 11), causal relationships, either bidirectional or unidirectional, between PM₂.₅ and built-up areas (BU) and forest areas (FOR) have been identified in many countries. This finding indicates that increased pollution during urbanization is related to the environmental buffering capacity of forests. Furthermore, in some countries, significant causal relationships between GDP and URB, on the one hand, and PM₂.₅, on the other, confirm that economic growth and urbanization affect air pollution.

A similar pattern was observed in the results of the CO₂ model presented in Table 12. Empirical findings indicate significant relationships, particularly in the BU→CO₂ and FOR→CO₂ directions. This reveals that emissions related to urbanization are partially offset by forest cover. These relationships, observed in many European countries, are important in that they show that green spaces help reduce urban carbon emissions. The findings obtained from the Fourier-TY causality test in our study support the “Green Buffer Hypothesis” regarding PM₂.₅ and CO₂ and confirm that forest areas reduce environmental pressures. This finding makes an important contribution to the literature in terms of supporting the “Green Buffer Hypothesis.”

Recent work has made it increasingly clear that “urbanization” is not a single-dimensional process and that its environmental footprint depends on how we measure urban expansion and how we model land cover as part of that process (Wang et al., 2019; Zhang et al., 2022; Zhou et al., 2023). A parallel literature argues that green infrastructure should be treated as a functional component of the urban system rather than a decorative add-on (Grabowski et al., 2022, 2023; European Commission, 2024; Herath and Bai, 2024; Yao et al., 2025). Yet, much of the empirical evidence still tests “direct effects” separately: urbanization on pollution, or forests on pollution, without putting the conditioning relationship at the center of inference (Kumar et al., 2019; Jennings et al., 2021; Bočkarjova and Kačalová, 2021; Muresan et al., 2022).

This is where the value-added of this study sits. Instead of asking whether built-up expansion or forest cover matters on its own, the empirical strategy tests whether forest cover changes the slope of built-up expansion's environmental impact. That emphasis aligns directly with the GBH logic and is operationalized through the core interaction term in both equations, which is also explicitly reflected in the core hypotheses H1c and H2c.

The PM₂.₅ results speak most clearly to the “buffer” mechanism. The negative association between forest share and PM₂.₅ supports H1b and is consistent with the view that vegetation can contribute to particulate regulation through capture and deposition processes. The broader literature also stresses that such effects are scale-sensitive, with micro-scale street-level conditions and aerodynamic constraints shaping outcomes (Jeanjean et al., 2016; Venter et al., 2024), while at meso and macro scales the balance between dispersion, land cover, and emission sources becomes decisive (Feng et al., 2017; Zhou et al., 2019; Douglas et al., 2019). In that sense, the fixed-effects EU panel evidence offered here complements the scale-differentiated findings by testing the buffer mechanism in a macro-comparative setting.

More importantly, the interaction term BU × FOR is negative and statistically meaningful, which directly supports the core hypothesis H1c: as forest cover increases, the pollution-increasing effect of built-up expansion weakens and can turn downward at higher forest shares. This conditionality also explains why H1a should be read carefully. A positive “average” built-up effect on PM₂.₅ is a reasonable expectation in many settings, but the present results indicate that the relationship is not stable across forest conditions. Built-up expansion looks pollution-increasing when forest cover is relatively low, yet it becomes much less harmful, and in the marginal-effects interpretation can even become pollution-reducing, when forest cover is relatively high. That is precisely the theoretical point of the Green Buffer framing: the same urban expansion can produce different air-quality outcomes depending on the surrounding green matrix.

The remaining controls fit a plausible story but are secondary to the paper's main claim. Where income is expected to reduce PM₂.₅ through structural change and regulation (H1d), the evidence does not support a strong EKC-type pattern for PM₂.₅ in this dataset. This is not an empirical failure; it is an informative distinction between local air-quality dynamics and carbon dynamics, and it anticipates the difference observed in the CO₂ equation.

The CO₂ results sharpen the contribution in a slightly different way. The individual coefficients on built-up expansion and forest cover do not carry the main explanatory weight on their own, whereas the interaction term does. This pattern supports H2c more directly than H2a or H2b in isolation: what matters for CO₂ is not simply “more built-up raises emissions” or “more forests lower emissions,” but the combined land-cover configuration in which urban expansion occurs.

This result is consistent with the ongoing discussion in the current global context concerning the possibility of the existence of non-linear links between urbanization and emissions (Siqin et al., 2022; Ngong et al., 2022). Additionally, this result is supported by the existing body of data concerning EKC, which confirms that emissions has a tendency to rise during the initial stages of development but eventually decouple during the latter stages with the help of technology (Ahmed et al., 2019; Ma and Ogata, 2024; Salahodjaev, 2014; McGee and York, 2018). Here, the income variables give an EKC-capturing relationship for CO₂ emissions consistent with the implication in H2d, correctly interpreted with dynamic reading, with income increasing emissions at the beginning but later reduced by the non-linearity part of the relationship. Such results showing how income, energy intensity, and policies together determine CO₂ emissions paths across multiple countries analyzed with panel data are consistent with this implication (Anser et al., 2020; Bashayreh et al., 2024; Zhao et al., 2025).

Crucially, the buffer interaction offers a land-use complement to the EKC narrative. Even when income-driven dynamics are present, forest cover still conditions whether built-up expansion is associated with higher or lower carbon pressure. That is the paper's core message for the carbon context and the most direct way the study advances the literature it reviews.

A useful way to read both equations is through thresholds rather than average signs. The results imply that the environmental slope of built-up expansion depends on whether forest share is below or above a country's typical level. For PM₂.₅, the marginal-effects interpretation indicates a sign change around a level slightly below the national average forest share. For CO₂, the sign change occurs around a similarly intuitive reference point, close to the country average and described in the results as roughly one percentage point below that benchmark. The substantive implication is that the buffer is not an abstract idea; it behaves like a state-dependent mechanism. Where forest cover is already under pressure, built-up expansion tends to be environmentally costlier. Where forest cover is relatively strong, the same expansion comes with a weaker penalty and may even coincide with lower measured pollution pressure.

This threshold framing also helps reconcile mixed findings in the literature. Case-based work shows that land-use trajectories under urban pressure do not follow a single uniform path (Bhomi et al., 2024; Clement et al., 2015; Keleş et al., 2008; Zhu et al., 2019). The present results suggest that part of this variation can be understood as a conditioning effect: not only do places urbanize differently, but the environmental consequence of that urbanization is filtered through existing land cover.

A recurring theme in the literature review is that studies often treat gray development and green infrastructure as separate axes, even though real-world urban systems combine them (Jennings et al., 2021; Kumar et al., 2019; Muresan et al., 2022). By placing BU × FOR at the core of the empirical specification, the paper tests a relationship that is frequently discussed conceptually but less frequently estimated directly in macro panel form.

This also links to measurement. Several strands emphasize that urbanization should not be reduced to population shares alone and that built-up land change is a distinct process with its own environmental channels (Zhou et al., 2023; Zhang et al., 2022; Wang et al., 2019). Using built-up area to represent the spatial footprint of urban expansion, while also controlling for urban population share, allows the analysis to separate “where the city spreads” from “how concentrated the population is.” That distinction matters for interpretation: a pollution effect attributed to urbanization in one study may actually be a land-conversion effect, while in another it may be a density and activity-concentration effect. The present approach makes that separation more explicit and therefore improves interpretability in a way that connects tightly to the reviewed debate.

The analysis is designed for EU-wide comparability and policy relevance, but its country-level structure inevitably comes with limits. First, it cannot observe within-country spatial heterogeneity, such as whether forests are contiguous or fragmented, and whether they are positioned near emission sources, issues that the scale-based green infrastructure literature flags as important for air-quality outcomes (Jeanjean et al., 2016; Venter et al., 2024; Douglas et al., 2019). A natural extension would thus be a subnational approach that maintains comparability while acknowledging land cover configuration in a more direct manner.

Secondly, forest share is a measure of quantity but does not in any way account for variations in the quality of the forests which could also contribute to variations in land use in case studies (Bhomi et al., 2024; Clement et al., 2015; Zhu et al., 2019). Future work could integrate indicators that distinguish forest area from forest quality to test whether the buffering mechanism strengthens when forest condition is higher.

Even with these limitations, the main point of the study remains clear and, importantly, it is legible to the international discourse described in the literature review: forest cover is not only a direct environmental asset. It also changes how built-up expansion translates into air-quality and carbon outcomes. That is the study's central contribution and the most direct response to the editor's concern about clarifying what the paper adds to the state of the art.

This paper tests the Green Buffer Hypothesis (GBH) using annual data from 25 European Union member countries between 1990 and 2023. In this context, the study analyzes whether urban sprawl and forest cover move in tandem with PM₂.₅ exposure and CO₂ emissions, two key indicators of environmental stress. The main result is consistent across all specifications. Accordingly, forest cover measurably mitigates environmental pressures associated with urban expansion. Urbanization is generally associated with worse environmental outcomes, but the magnitude of this effect varies depending on land cover. In both the PM₂.₅ and CO₂ specifications, the BU × FOR term is negative and statistically significant. This result clearly shows that as forest share increases, the environmental damage associated with urban expansion decreases. Forest cover is also evident in the data as a direct determinant of local air quality: countries with a higher forest share tend to have lower average PM₂.₅ exposure. This result supports the idea that forests provide tangible air quality benefits in addition to their broader ecological roles.

Another finding presented in the study is that income models differ according to pollutants. The results for CO₂ emissions support an Environmental Kuznets Curve-type relationship. According to this, emissions increase at low income levels and then tend to decline once countries reach higher income levels. This finding is consistent with the roles of technology, energy transitions, and regulatory capacity. The same EKC model, however, was not observed for PM₂.₅ exposure. This difference is not coincidental but informative. While carbon outcomes appear more sensitive to income-related technological and policy changes, PM₂.₅ exposure is more closely linked to spatial form, land-use preferences, and the direct environmental conditions created by urban development.

These findings point to a practical policy agenda: urbanization policy and forest policy should not be treated as separate areas. When construction and land conversion intensify without concurrent protection of forest resources, CO₂ pressures are likely to increase. When forest protection or expansion accompanies urban growth, CO₂ pressures are more likely to be contained. For that reason, minimum standards for tree-based green infrastructure, including urban and peri-urban forests, should be written into formal planning instruments, such as statutory land-use plans, zoning and development-control rules, and local urban greening or urban forestry standards. When defined this way, they become binding criteria that shape permitting decisions and project design rather than remaining aspirational targets. A second implication concerns project-level delivery. At the project level, permitting conditions can be used to mandate mitigation or compensation, for instance reforestation, ecological restoration, or dedicated financing for urban greening, with defined performance indicators and compliance checks. This is particularly relevant in areas where urban expansion is rapid and pressure on nearby forests is greatest. Thirdly, forest-based carbon management deserves to play a more robust role in the implementation of EU climate policy. Strengthening forest stocks not only expands carbon sequestration capacity but also supports resilience and co-benefits for local environmental quality. In this regard, aligning urban growth strategies with the EU's Fit for 55 package and the LULUCF framework demonstrates that the process carries a deeper meaning than being viewed merely as a climate accounting exercise. It is also a way to integrate green-focused and supported environmental and spatial planning, land management, and mitigation policy into a single, coherent approach. Finally, policy instruments must be appropriate to the current environmental indicator. CO₂ reduction is largely dependent on energy systems, technology adoption, development processes, and regulatory instruments across the economy. In contrast, reducing PM₂.₅ exposure is more dependent on spatial planning, local ecosystem restoration, and the deliberate expansion and protection of urban and peri-urban forests. In a broader sense, sustainability cannot be achieved through technology alone. It also depends on how natural and artificial systems are designed together. This design synergy and compatibility is particularly important in urban planning. In this sense, demonstrating that forests have an effect that alleviates environmental pressures associated with urbanization contributes significantly to both policy debates and empirical literature, and clearly supports the idea that nature-based solutions should be placed at the center of Europe's urban growth strategies. Future studies could be useful for testing the same mechanisms in other country groups and regions with different institutional structures, forest dynamics, and urbanization trends. This will help determine how far policy effects spread and which elements need to be adapted to local planning and governance contexts.