A Socioeconomic Assessment of Projected Health Impacts From Outdoor Air Pollution Under Current European Climate Policy: A Modelling Study

Summary

Background

Outdoor air pollution poses a major threat to human health and is closely interlinked to climate change, as both stem largely from the same emission sources and pollutants, affecting both health and climate. Despite sustained mitigation efforts, over 239,000 premature deaths in Europe were attributed to PM_2.5 exposure in 2022, while climate policies are projected to yield substantial yet insufficient co-benefits by 2030, with uneven impacts across regions and socioeconomic groups.

Methods

We applied GCAM-Europe to model emission pathways under the National Energy and Climate Plans (NECPs) and Fit-for-55 (FF55) policy package. Premature mortality attributable to PM_2.5 was estimated at both 1 km² grid and NUTS3 levels using rfasst. Results were further analysed through socioeconomic lens, both between and within countries, using quantitative assessments and machine learning techniques.

Findings

Despite implementing the NECPs and FF55 packages, significant health risks persist by 2030, with average PM_2.5 levels exceeding the World Health Organisation guidelines in all but seven European countries. The Balkan region and northern Italy emerge as the most affected areas, with national averages reaching up to 12.4 μg m⁻³. Urban populations face the greatest health risks, with disproportionate burdens among low- and high-income groups and areas with average concentrations of elderly residents. Within-country analyses reveal distinct, country-specific vulnerability patterns.

Interpretation

Incorporating social and health dimensions into environmental policymaking is essential for designing strategies that jointly advance decarbonisation, reduce air pollution, and promote environmental justice. Attention to spatial resolution and geographic scale is critical, as coarse analyses risk obscuring vulnerable populations and regional disparities.

Funding

Funded as part of the DIAMOND Horizon Program.

Research in context.

Evidence before this study

We conducted an extensive literature search in PubMed, Scopus, and Google Scholar from inception up to September 15, 2025, using terms such as “outdoor air pollution”, “PM_2.5 premature deaths”, “socioeconomic inequality”, and “sociodemographic factors”. Most previous studies relied on single representative agents, in which entire populations are modelled as a homogeneous decision-maker with uniform preferences and constraints; presented results at broad regional scales, thereby determining outcomes across large geographic areas; and rarely examined the effects of realistic climate-policy portfolios, focussing instead on historical assessments or stylised global carbon-tax scenarios.

Added value of this study

This study introduces a novel approach to estimate PM_2.5 attributable premature mortality at a fine spatial scale (1 km × 1 km), linking it to key socioeconomic indicators: income per capita, settlement type, share of elderly population, and inequality index. By analysing both pan-European and country-specific data, we uncover region-dependent patterns and significant heterogeneity in outputs across grid cells as well as within national boundaries, offering a more nuanced understanding of pollution-related inequality.

Implications of all the available evidence

Our results show that urban areas and poor-income groups, particularly in Balkan countries, are disproportionately affected by PM_2.5 exposure. We also uncover within-country inequality dynamics and pinpoint areas at greatest health risk. These insights are critical for designing targeted social policies that can enhance the equity and effectiveness of climate mitigation efforts.

Introduction

Outdoor air pollution is a leading global environmental health threat, and Europe is no exception.¹ Among pollutants, fine particulate matter (PM_2.5) is the most harmful to human health, contributing not only to increased mortality but also to a wide range of illnesses and significant economic losses.^2,3 For instance, in 2022 alone, it contributed to an estimated 269,000 premature deaths across the European continent.⁴ Addressing this challenge requires effective policy interventions, including for climate change mitigation, which offers a valuable opportunity to deliver synergistic benefits. Because greenhouse gas (GHG) emissions and air pollutants largely stem from the same sources, most notably fossil fuel combustion, measures aimed at reducing emissions can simultaneously mitigate global warming and improve air quality.^5,6

Within the European Union (EU), the Fit-for-55 (FF55) policy package and the country-specific National Energy and Climate Plans (NECPs) represent a central policy framework, outlining Member States' strategies to meet the 2030 climate and energy targets set by the European Green Deal, the Climate and Energy Framework, and the bloc's Nationally Determined Contribution (NDC).⁷ Earlier research suggests that moving from countries' current policies toward the more ambitious NDCs could prevent nearly 200,000 deaths globally by 2030⁶ —underscoring, in the EU, the significant public health potential contribution of the climate policies to public health.

However, while mitigation policies can bring major environmental and health benefits, they also risk exacerbating existing social inequalities,⁸ particularly in industrialised (e.g., coal-dependent) EU regions that are highly vulnerable to such policy transitions.⁹ Similarly, disparities in the exposure to air pollution based on income,^10,11 ethnicity,¹² education,¹³ age,¹⁴ and household settlement type¹⁵ influence associated health risks. Recognising these differences is essential to ensure that climate policies maximise co-benefits, minimise trade-offs, and ultimately secure stronger public support for climate action.¹⁶

To inform such policymaking, researchers use integrated assessment models (IAMs), sophisticated tools that simulate the interactions between climate, energy, and socioeconomic systems in a single computational platform.¹⁷ IAMs are widely applied to assess climate policies and their co-benefits,^18,19 and can be coupled with chemical transport models to estimate future air pollution and related health impacts.20, 21, 22 However, traditional air pollution studies often focus on broad regional scales, limiting the detail of their insights.^5,22,23 Even when spatial granularities are improved, health impacts are usually aggregated at coarser levels (e.g., national or supranational).^24,25 Moreover, climate policies are frequently not explicitly modelled²⁶ or modelled in overly simplistic ways, such as uniform global carbon taxes.^27,28 And, although IAMs have recently begun incorporating inequality by stratifying results by income deciles,^29,30 other key socioeconomic dimensions remain overlooked, limiting the models’ ability to deliver more targeted and inclusive policy insights.

In this context, we assess the health impacts related to PM_2.5 outdoor air pollution under the NECPs and wider FF55 policies across EU+ (defined in the Methods section), focussing on different key socioeconomic dimensions, including age, income, and urbanicity. Our analysis captures PM_2.5-related deaths and inequalities at a high-resolution 1 km² grid level, revealing disparities both between EU+ areas and within EU+ countries, and further extends the assessment to the Nomenclature of Territorial Units for Statistics–level 3 (NUTS3),³¹ the most widely used spatial aggregation level for specific diagnoses.³² Finally, we identify the most socioeconomically vulnerable populations and map out pollution hotspots, offering valuable insights to support targeted social policies that complement and enhance climate action.

Methods

Methodological framework

The workflow of this study has four sequential main steps: (1) estimate the emission pathways deriving from the POLICY55 scenario, which includes the NECPs and FF55 policy packages, modelled through the novel Global Change Analysis Model - Europe (GCAM-Europe)³³; (2) estimate the PM_2.5 concentrations and the associated premature deaths at the 1 km² grid level through the latest version of rfasst³⁴; (3) analyse the results in the EU+ countries from a broad socioeconomic perspective using quantitative statistical assessments and machine learning techniques; and (4) perform a EU+ national analysis at the NUTS3 level (Fig. 1). This workflow is best suited for scenario-based analysis across geographical and socioeconomic dimensions, rather than for providing quantitative forecasts. Considering data availability, the EU+ region is considered as the conjunction of the following countries: Albania, Austria, Belgium, Bosnia and Herzegovina, Bulgaria, Croatia, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Moldova, Montenegro, the Netherlands, North Macedonia, Norway, Poland, Portugal, Roumania, Serbia, Slovakia, Slovenia, Spain, Sweden, Switzerland, and the United Kingdom.

Fig. 1.

Methodology sketch. The background colours indicate the workflow step.

GCAM-Europe³³ is a geographically and sectorally expanded version of the Global Change Analysis Model (GCAM), a well-reputed multisector integrated assessment model.³⁵ GCAM includes technology-rich representations of the economy, energy sector, land use, and water systems linked to a climate model that can be used to explore climate change mitigation policies, including carbon taxes, carbon trading, regulations and accelerated deployment of energy technology. GCAM has been extensively applied in global and regional scenario analyses, as well as in national and international assessments, including recent Intergovernmental Panel on Climate Change (IPCC) reports³⁶ or the Scenario Model Intercomparison Project (ScenarioMIP)³⁷ as it is designed to analyse alternative “what-if” scenarios through the end of the century within a unified computational framework, providing insights into the potential consequences of uncertain futures.³⁵

GCAM-Europe³³ is a recently released expansion of GCAM that offers significantly higher regional and sectoral detail for Europe. While the core GCAM model divides the European continent into only five regions, GCAM-Europe disaggregates all EU Member States and several non-EU key countries into individual model regions. This finer level of detail allows for country-specific assessments of European policy packages and transformational strategies, as well as for the evaluation of their broader international implications, such as on international trade.³⁸ GCAM-Europe substitutes the default international datasets for all newly defined European countries with Europe-specific sources whenever possible. For countries not fully covered in these datasets, alternative sources are used, most often reverting to the default (international) data sources of the core GCAM version. Leveraging these improved datasets, GCAM-Europe further expands sectoral and technological detail, with particular emphasis on energy demand for buildings. Notable additions include new demand categories, such as hot water, cooking, and household appliances, and the explicit representation of low-carbon technologies like heat pumps. Beyond regional disaggregation and updated datasets, GCAM-Europe introduces several structural improvements, reflecting the European electricity grid structure and the EU internal market in terms of international trade flows. These refinements improve the realism of European dynamics and enable a better assessment of future policies such as the Carbon Border Adjustment Mechanism (CBAM) or the NECPs.

After estimating the emission pathways through GCAM-Europe by 2030, we run rfasst,³⁹ an R-based implementation of the TM5-FAst Scenario Screening Tool (TM5-FASST).⁴⁰ As a reduced-form atmospheric model, TM5-FASST utilises pre-calculated emission-concentration coefficients to efficiently estimate pollutant concentrations and associated health impacts. The most recent version³⁴ enhanced spatial resolution, enabling the downscaling of mean PM_2.5 concentrations and resulting population-weighted premature mortality to a fine spatial scale (1 km² grid cells). Subsequently, these high-resolution estimates can be re-aggregated, for example to the national or subnational level, flexible to different analytical scopes. In this study, we use the aggregation level of the NUTS3 territorial units.⁴¹ Premature mortality from PM_2.5 exposure is calculated using the Integrated Exposure Response functions developed by the Global Burden of Disease study⁴² (see Section S11 for further details). While this architecture is uniquely suited for analysing the interplay between health outcomes and socioeconomic data, we acknowledge that its reliance on a linearised framework comes with specific limitations, which are thoroughly described in Section S11.1.

Finally, to better characterise the socioeconomic dimensions of exposure and health impacts from PM_2.5, we analyse key indicators at two spatial scales. At the 1 km² grid-cell level, we examine settlement type (city, town/suburb, and rural), disposable income (expressed in 2015 EU27 PPP€/capita), and the share of elderly people (aged 65 and above). At the NUTS3 level, we extend this analysis to include the Gini index as a proxy for inequality, alongside the three previous indicators. The Gini index quantifies the deviation of an income distribution from perfect equality; higher values indicate greater inequality.⁴³ These variables exhibit negligible to weak correlations with one another, with pairwise correlation coefficients below 0.15 (Table S1), indicating that they capture distinct dimensions of vulnerability. This lack of strong collinearity supports the independent analysis of each socioeconomic factor independently, allowing for a more granular examination of how different aspects of social and demographic structure contribute to pollution exposure and its health consequences. See Section S4 for further details.

To identify country-level patterns of intra-national inequality in air pollution exposure and the associated health impacts, we employ hierarchical clustering, an unsupervised machine learning technique well-suited for exploratory analysis of structured data.⁴⁴ This technique entails clustering countries based on similarities in the distribution of PM_2.5 concentrations or related premature mortality across socioeconomic quintiles. Unlike partitioning methods, hierarchical clustering does not require the number of clusters to be specified in advance, allowing for a data-driven identification of relevant grouping. The resulting dendrograms reveal systematic cross-country variation in environmental inequality within the EU+ and are presented in Figures S25–S34. See Section S10 for further details.

Scenario

Thanks to the geographical detail of GCAM-Europe, we implement a scenario consistent with the NECPs⁴⁵ and the wider FF55⁴⁶ policies by 2030, hereafter referred to as POLICY55. This portfolio is the central planning and governance instrument coordinating and streamlining the EU's transition toward a climate-neutral economy at the national level. Member States' NECPs outline how each country will address the five dimensions of the Energy Union: decarbonisation, energy efficiency, energy security, internal energy market, and research, innovation, and competitiveness. Beyond policy alignment, NECPs also serve as investment roadmaps, helping to mobilise both public and private capital for the energy transition. On the other hand, the FF55 policy package combines EU level regulations with country-specific strategies, with the goal to reduce GHG emissions by 55% below 1990 levels by 2030.⁴⁷

The NECPs are implemented as country-specific targets for energy efficiency, renewable energy, and coal phase-out plans. These policies are further combined within a wider FF55 context, including EU-wide policies such as the Emissions Trading System (ETS 1 and 2) and vehicle emission standards defining average emissions per vehicle-kilometre of the EU-wide car and truck fleet.⁴⁸ Furthermore, the scenario reflects existing energy taxation for electricity, gas and transport fuels based on observed values up to 2024, assuming these to stay constant throughout 2030. Because the scenario solely focuses on fossil emission mitigation, proposed national contributions in the Land Use, Land Use Change, and Forestry (LULUCF) sector are taken as given and are not modelled. Besides, we assume technological advancements, such as end-of-pipe control measures, and implicit air-pollution-related regulations, exogenously decreasing the emission factors (emissions per unit of output), extensively documented in literature.⁴⁹ The scenario framework and detailed underlying assumptions are further described in Frilingou et al.⁵⁰

The combination of the planned national and EU-wide policies implemented in the scenario approximately reach the 55% GHG reduction target in 2030 relative to 1990, while country-specific GHG emissions are close to the targets specified in the NECPs. This methodology, based on the implementation of specific policies, provides a better approximation of reductions by sector and related impacts on non-GHG pollutants, than, for instance, a counter-factual economy-wide GHG target that guarantees achieving emission targets but with alternative sectoral contributions to mitigation. Figure S6 shows the impact of this scenario in terms of GHG and non-GHG emissions relative to a no-policy baseline, which has direct implications for PM_2.5 concentrations and its health impacts. The contribution of individual sectors varies by precursor, with some sectors playing a more dominant role in specific emissions (for details, see Section S3).

Ethics statement

Ethical approvals for this study were not required, as we only used open-access, publicly available data that contained no identifiable personal information.

Role of the funding sources

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results

PM_2.5 and health impacts in an NECP-compliant EU+

Despite the decarbonisation strategies associated with the POLICY55 scenario resulting in high GHG emission reductions, high pollution levels and associated health damages are still estimated in 2030. Notably, only seven countries in EU+ are expected to have average PM_2.5 concentrations below the threshold outlined in the World Health Organisation (WHO) guidelines (5 μg m⁻³)⁵¹ (Figure S1a), with Italy, Switzerland, and the Netherlands bearing the highest pollution burdens. In terms of mortalities, 153,288 premature deaths are still projected to occur in EU+ in 2030 due to PM_2.5 outdoor air pollution under the POLICY55 scenario. The most affected countries are Bulgaria, Serbia, and North Macedonia (Figure S1b), where the widespread use of solid fuels (e.g., coal and wood) for household heating, industrial processes, and power generation, combined with less ambitious climate targets, drives higher pollution levels and, consequently, greater health impacts.⁵² The highly industrialised and densely populated northern Italy also shows considerable projected mortality linked to air pollution, albeit slightly lower than in the Balkan countries. The difference lies in the distinct meteorological and topography conditions in the region, which lead to pollutant accumulation but involve different sources and compositions than those found in the Balkan area.⁵³ Additionally, population density differs substantially between the two regions, directly influencing the population-weighted health estimates. Furthermore, northern Italy represents only a portion of the country's territory, whereas the pollution burden in smaller Balkan nations is more uniformly distributed (Figure S10), contributing to a higher national premature mortality average.

Zooming into the grid-level detail, PM_2.5 concentrations peak in the Po Valley, in northern Italy, highlighting it as one of the most polluted areas (Fig. 2a). In terms of health impacts, the major EU+ cities show the highest numbers of PM_2.5-related premature deaths (Fig. 2b), with these densely populated areas experiencing concentrations well above the WHO air quality thresholds (Figure S1a). Consistent with previous research,^27,54 the Balkan region bears the highest health burden, with estimated premature deaths reaching levels up to seven times those in other EU+ regions, such as the Iberian Peninsula or France (Fig. 2c).

Fig. 2.

Short-term (2030) PM_2.5 effects in EU+ under POLICY55 scenario at the 1 km² grid cell level: a) PM_2.5 concentration, b) premature deaths in absolute numbers (log scale), and c) premature mortality rate by grid cell, (i.e., premature deaths normalised by grid cell population) per 1M inhabitants.

In the following subsections, we examine these impacts through a socioeconomic lens, both at the EU+ level, i.e., treating the region as a single aggregate, and at the individual country level. The assessment is conducted using data at two spatial resolutions: a fine 1 km² grid and the NUTS3 territorial units. This dual perspective allows us to identify not only broad regional patterns but also local disparities, highlighting the populations and areas most at risk.

PM_2.5 exposure and health impact disparities in EU+ wide

To assess the spatial disparities in PM_2.5 exposure and its health consequences at the EU+ wide level—i.e., considering EU+ as a single block—we first categorise each 1 km² grid cell by settlement type: urban, town/suburb, and rural (further details available in Section S9). As expected, urban areas showcase the highest PM_2.5 concentrations, reflecting the existence of multiple emission sources such as traffic, industry, and dense heating systems. In contrast, rural and suburban areas generally experience lower ambient pollution levels (Fig. 3a). When examining premature mortality in absolute terms, the largest (urban) cities account for the highest death tolls (Fig. 2b and Figure S7a), driven by both higher pollution and denser population. A strikingly different pattern emerges when premature mortality is normalised by population at the grid cell level, as rural areas appear to exhibit higher per capita mortality burdens (Fig. 3b). This counterintuitive result is a consequence of the different population counts per grid cell, that directly affect the normalised population-weighted premature deaths estimates, and the manifestation of Simpson's paradox⁵⁵: while urban areas often show higher mortality within individual countries (Fig. 4b), the global average is skewed because when analysing at EU+ level as a single node, rural grid cells vastly outnumber urban ones (Figure S23). As a result, the overall rural average is elevated, even though this does not necessarily reflect within-country patterns (analysed in the next section) across EU+.

Fig. 3.

Short-term (2030) PM_2.5 effects in EU+ under POLICY55 scenario at the 1 km² grid cell level, by (a–b) settlement type, (c–d) income level, and (e–f) share of elderly population. First column shows PM_2.5 concentration exposure and second column shows PM_2.5 premature deaths rate. Top panels show probability density functions; bottom panels display jittered points and box plots. Colours indicate settlement type, income, or elderly share quintile. Box plots represent the median, and the 66% and 90% distribution ranges. Each subfigure is based on a random sample of 10.000 grid cells.

Fig. 4.

Short term (2030) PM_2.5a) mean concentration and b) related premature mortality rate by 1 km² grid cell in the EU+ countries under the POLICY55 scenario by i) urban type, and quintile of ii) income per capita and iii) share of elderly people.

To explore the influence of income on environmental health outcomes, we disaggregate all grid cells into five income quintiles, from Q1 (lowest) to Q5 (highest), whereby each quintile represents 20% of the total population ordered by income level. The distribution of PM_2.5 concentrations (Fig. 3c) reveals that the wealthiest quintile (Q5), largely concentrated in urban hot spots and the centre of Europe (Figure S8), experiences the highest pollution exposure. Interestingly, both the lowest (Q1–Q2) and highest (Q5) income groups face the largest premature mortality burdens (Fig. 3d). The poorest populations are mainly located in Eastern Europe (Figure S8), where mortality impacts are particularly severe (Fig. 2c), while high concentrations of Q5 populations in polluted urban centres also result in elevated mortality estimates.

Finally, to analyse the demographic composition across grid cells, we classify them into five quintiles based on the concentration of elderly population, ranging from Q1 (lowest share of elderly residents) to Q5 (highest). The relationship between the share of elderly population and both PM_2.5 exposure and the corresponding premature mortality exhibits an inverse U-shaped pattern (Fig. 3e and Fig. 3f). At the EU+ level, grid cells with intermediate levels of elderly population (Q2–Q4) experience the highest pollution and mortality burden. This traces to the geographies of livelihoods, with lower presence of urban grid cells for Q1 and Q5, and higher presence of urban grid cells for Q2–Q4 (Figure S3b).

While these results offer valuable insights into broad EU+ wide patterns, they may obscure important country-specific dynamics. As shown in Figures S8 and S9, the distribution of income and elderly population across the EU+ region varies considerably by country. These disparities highlight the importance of conducting a detailed country-level analysis. Therefore, we deepen our investigation by examining the socioeconomic characteristics within each country individually.

PM_2.5 exposure and health impact disparities within EU+ countries

To better understand country-specific patterns, we analyse the distribution of PM_2.5 concentration and related premature deaths over the grid cells by settlement type, income, and elderly population quintiles within each country (Figures S12, S13, and S23). There is a consistent pattern from a settlement type perspective regarding PM_2.5 concentration: within countries, urban areas tend to concentrate the highest levels of PM_2.5, while rural areas are generally less affected (Fig. 4a). However, some countries deviate from this trend: Belgium, Bulgaria, Croatia, Denmark, Estonia, Germany, Latvia, and Serbia. In Denmark, Estonia, and Latvia, overall PM_2.5 concentrations are low, and a significant share of grid cells fall into rural or town/suburb categories. Conversely, Belgium, Bulgaria, Croatia, Germany, and Serbia exhibit uniformly high PM_2.5 levels across all settlement types, indicating a widespread pollution burden that is not spatially differentiated (Figure S10).

To better understand heterogeneity in exposure by income per capita and the share of elderly population, we apply a clustering machine learning algorithm (for details, see Section S10). This unsupervised approach identifies distinct groups of countries with similar multidimensional profiles—including sociodemographic, environmental, energy, and economic factors—, enabling a more nuanced interpretation of structural differences across regions.

Regarding the relationship between air pollution and income per capita, countries can be grouped based on how PM_2.5 exposure varies across income quintiles (Fig. 4a and Figure S25). The first group includes Albania, Croatia, Czechia, Estonia, Germany, Hungary, and Portugal, where an inverse relationship is observed between income and PM_2.5 exposure: the poorest quintile (Q1) is the most exposed, while the wealthiest quintile (Q5) is the least exposed. This pattern can be attributed to several interacting factors. First, the spatial distribution of industrial facilities and coal power plants often coincides with economically deprived regions, thereby increasing exposure among lower-income groups. For example, in Czechia, the Silesia industrial area exhibits higher pollution levels (Figure S10). Second, the role of residential emissions, which are typically higher among lower-income households because of greater reliance on solid fuels for heating and cooking, further reinforce these disparities.⁵⁶ This is evident in Czechia, Estonia, and Germany, which rank among the highest residential emitters in Europe.⁵⁷ Third, agricultural practices, including fertiliser use and livestock management, elevate NH₃ concentrations primarily in rural areas, which are often inhabited by lower-income populations⁵⁸ (Figure S3a); such as in the eastern part of Germany (Figure S12 and S23). Finally, transboundary air pollution substantially shapes exposure patterns in regions adjacent to Eastern Europe, where PM_2.5 concentrations are particularly elevated. This effect is evident in countries such as Albania and Croatia, where inland regions, closer to the Balkan region, experience higher exposure compared to the wealthier populations concentrated along the Adriatic coast (Figures S10 and S12). The second group of countries comprises Italy, Latvia, Poland, Slovakia, and Spain. They exhibit an N-shaped relationship between income and PM_2.5 exposure, indicating that Q1 and Q3 are the least exposed quintiles, while Q2, Q4, and Q5 suffer the highest burdens. The third group, comprising Austria, Belgium, Denmark, Finland, France, Greece, Lithuania, Moldova, the Netherlands, Serbia, and the United Kingdom, consists primarily of advanced economies and exhibits a direct relationship between income and PM_2.5 exposure, whereby higher-income populations tend to experience larger concentrations.

When examining the relationship between air pollution and the share of elderly population across grid cells, we identify again three distinct country-level patterns (Fig. 4a and Figure S26). First, countries such as France, Norway, and Spain exhibit an inverse relationship between PM_2.5 exposure and the share of elderly residents. These countries might be characterised by relatively low overall PM_2.5 concentrations, limited industrial emissions, and a more rural distribution of elderly populations. As a result, areas with higher shares of elderly individuals tend to experience lower levels of air pollution. In contrast, Luxembourg and Switzerland display the opposite trend: despite their small size, both countries show a clear pattern in which grid cells with the highest shares of elderly residents are also the most exposed to PM_2.5. This suggests a spatial overlap between pollution hot spots and elderly population centres, potentially posing elevated health risks. Finally, a nonlinear U-shaped pattern emerges in countries such as Czechia, Denmark, and Italy, where grid cells with an average share of elderly individuals tend to face the highest PM_2.5 exposure—thereby mirroring the EU+ wide results.

Turning to PM_2.5-related premature deaths, the within-country patterns shift significantly. While there is a strong correlation between PM_2.5 concentration and premature mortality, the relationship is not strictly linear, as other demographic and socioeconomic factors modulate the health burden of air pollution (Fig. 4b, Figure S4 and S11). The most pronounced burden of PM_2.5-related deaths is concentrated in urban grid cells, mimicking the PM_2.5 concentration exposure pattern. The only two exceptions are Denmark and Latvia. These countries, which also diverged from the broader PM_2.5 concentration trends, show the highest health impacts in rural areas, followed by towns/suburbs and cities. Nevertheless, the differences between settlement types are relatively small in magnitude.

When analysing income-based disparities, a direct relationship emerges in many countries: the wealthiest quintile suffers the highest burden of PM_2.5-related deaths. This pattern is clearly observed in Austria, Belgium, Bulgaria, Denmark, Finland, France, Greece, Italy, the Netherlands, Slovakia, Spain, and the United Kingdom. In contrast, countries such as Albania, Czechia, and Germany exhibit an inverse pattern, where the poorest quintile suffers the most. A distinctive inverse U-shaped relationship is visible in Czechia, Hungary, or Lithuania, where middle-income quintiles (Q2–Q4) bear the largest health impacts. Lastly, countries like Latvia and Roumania reflect the inverse N-shape pattern, where exposure and health burdens first decrease, then rise with income, and go down again in the wealthiest quintile (Fig. 4b and Figure S27).

Regarding age structure, the most common pattern reflects a direct relationship: grid cells with a higher share of elderly residents experience larger PM_2.5-related mortality. This trend is evident in countries like Austria, Belgium, Bulgaria, Germany, Greece, Hungary, Lithuania, Luxembourg, Poland, Portugal, Spain, and Switzerland. Conversely, Finland, France, and Norway exhibit an inverse pattern, where areas with more elderly residents show lower mortality impacts. Finally, an inverse U-shaped curve appears in Czechia, Denmark, and Italy, where the highest mortality burden is concentrated in grid cells with a moderate share of elderly individuals, as also pointed out by the PM_2.5 concentration exposure (Fig. 4b and Figure S28).

PM_2.5 exposure and health impact disparities within EU+ countries at the NUTS3 level

The NUTS3 level represents the most granular tier of the hierarchical geographical classification system used within the EU.³¹ This territorial aggregation, commonly used to inform policy planning, corresponds to subnational administrative regions with varying degrees of decisional and budgetary authority. Moreover, the richer data availability at this level enables the exploration of relationships between PM_2.5 and key socioeconomic dimensions, such as economic inequality, measured by the Gini index. However, aggregating gridded results at this scale may introduce some aggregation bias, potentially masking dynamics that are more visible at finer resolutions (Figure S14).

Using the same methodology applied at the grid cell level, we examine country-specific patterns at the NUTS3 level regarding the PM_2.5 concentration and related premature deaths (Figures S20, S21, S22, and S24). As previously noted, urban subnational regions generally exhibit the highest PM_2.5 concentrations. However, Czechia and the Netherlands are an exception (Fig. 5a), where few NUTS3 regions are classified as urban (Figure S24). When examining PM_2.5-related premature mortality, some rural NUTS3 regions, despite having lower concentrations, surpass urban areas in estimated deaths (Fig. 5b). This result is largely driven by normalisation effects in sparsely populated rural regions, which can inflate per-capita mortality estimates.

Fig. 5.

Short term (2030) PM_2.5a) mean concentration and b) related premature mortality by NUTS3 region in the EU+ countries under the POLICY55 scenario by i) urban type, and quintile of ii) income per capita, iii) share of elderly people, and iv) Gini index.

Income-related patterns are partially preserved at the NUTS3 level (Figure S29). Countries such as Croatia, Finland, France, Germany, and Italy retain the grid cell relationship between income and PM_2.5 exposure. In contrast, others, such as Austria or the Netherlands, show a shift in pattern due to aggregation effects (Fig. 5a). Interestingly, a greater number of countries display an N-shaped relationship at the NUTS3 level compared to the grid level. When analysing premature deaths by income quintile, we observe that the lowest income group (Q1) often ranks among the most affected. This is evident due to the frequent linear and inverse U-shaped patterns found in countries such as Croatia, Germany, Greece, Portugal, and Serbia (Fig. 5b and Figure S32). Nevertheless, the analysis is constrained in countries with limited regional units, such as Luxembourg, where the number of NUTS3 regions is insufficient to define five income quintiles.

While at the EU+ level the median PM_2.5 exposure across quintiles of the elderly population share appears relatively uniform (Figure S14), disaggregation by country at the NUTS3 level reveals a clearer socioeconomic gradient—consistent with, though often amplified compared to, the pattern observed at the grid level (Fig. 5a and Figure S30). For example, France and Spain continue to show an inverse relationship between PM_2.5 concentration exposure and elderly population percentage, while Switzerland maintains a direct relationship. When focussing on premature deaths, a direct linear relationship with elderly population share is observed across most countries, again more prominently than at the grid level. Examples include Austria, Belgium, Bulgaria, Germany, Greece, Hungary, Poland, Portugal, and Spain. Only France and Norway show an inverse trend, and the inverse U-shaped pattern previously observed in Czechia remains rare at this level of aggregation (Fig. 5b and Figure S33).

The analysis of the Gini index at the NUTS3 level does not reveal consistent within-country patterns in relation to PM_2.5 exposure or premature deaths. As shown in the hierarchical clustering and median distribution figures (Fig. 5, Figure S31 and S34), there is substantial heterogeneity in the distributions, with no robust evidence of direct, inverse, or non-linear associations. This heterogeneity suggests that additional factors beyond income inequality may drive exposure disparities, such as spatial distribution of pollution sources, demographic structure or urban forms. Moreover, limited data availability for the Gini index hampers the analysis in several countries.

Discussion

This study evaluated projected PM_2.5 concentrations and associated premature deaths in the EU+ region by 2030 under the implementation of the POLICY55 scenario—which integrates both the NECPs and the FF55 policy package—incorporating key socioeconomic indicators. Despite expected emissions reductions in all sectors, our results estimate 153,288 premature deaths attributable to PM_2.5 in the short term, consistent with previous projections.^6,24 These findings reaffirm that current climate policies alone, despite their ambition, may not be sufficient to fully mitigate health risks from air pollution, emphasising the urgent need for supplementing climate action with targeted health and social policies.

Spatially, the Po Valley in northern Italy and the Balkan countries emerge as the most affected areas. However, vulnerability is strongly mediated by socioeconomic factors rather than evenly distributed. At the EU+ level, rural areas exhibit lower average PM_2.5 concentrations but sometimes show disproportionately high mortality rates due to sparse populations, the population-weighted normalisation estimates, and the Simpson's paradox. Income further modulates exposure: the poorest quintile, concentrated in Eastern Europe, and the wealthiest quintile, concentrated in urban hot spots, both experience the heaviest health burdens. From a demographic perspective, areas with either very high or very low shares of elderly residents tend to show lower PM_2.5 exposure overall.

Within countries, settlement type and income are the primary drivers of disparities, with strong dependency on specific country characteristics. Air pollution burdens concentrate in urban areas, while income-related patterns vary depending on national energy mixes, industrial activity and its geographical distribution, agricultural and residential emissions, and transboundary effects. The share of elderly residents also emerges as a sensitive factor, with impacts strongly dependent on the livelihoods’ geographies.

Finally, spatial aggregation can critically shape the observed patterns and introduce notable biases into inequality gradients. While merging heterogeneous areas into broad regional aggregates is useful for high-level policy design, it can obscure meaningful intra-regional differences—particularly for variables like settlement type or elderly population share, which vary markedly across small areas. Moreover, analysing health burdens purely at the aggregate EU+ level can mask local distributional inequities by over-representing settlement types, while income per capita tends to show more consistent patterns when compared across spatial scales. Consequently, to correctly analyse the regional and socioeconomic inequality gradients, it would be necessary to disaggregate the analysis in to national and sub-national levels.

This study has several limitations. At the modelling level, we implement cascade modelling rather than economically integrating air pollution modelling within our integrated assessment model framework⁵⁹; also, emission drivers are not disaggregated by sector nor at the subnational level, as GCAM-Europe operates at a country-level basis. Hence, we do not capture if emission reducing impacts from the POLICY55 scenario are geographically asymmetric within a country. We also rely on static downscaling and pre-calculated response coefficients, which preserve the historical spatial distribution of PM_2.5 rather than allowing it to evolve dynamically over time (Figure S5). Although such dynamic changes are expected to be more pronounced over longer time horizons than the one considered here (up to 2030), this further limits our ability to capture geographically asymmetric intra-country changes. Accordingly, this framework is optimised for analysing relative changes in concentrations and health impacts, rather than generating high-precision absolute exposure metrics for specific coordinates—a characteristic that directly motivates our focus on comparative geographical and socioeconomic disparities. Beyond modelling constraints, localised data gaps restrict certain dimensions of the analysis. For example, in countries such as Albania and Serbia, missing grid-level population data prevented income estimation at the NUTS3 scale, excluding them from parts of the socioeconomic assessment (as detailed in Section S12). These gaps are particularly relevant in the Balkan area, a region already identified as highly exposed to PM_2.5.⁶⁰ Finally, factors beyond socio-economic variables, such as biodiversity, diet, and behavioural trends, also influence the outcomes but are beyond the scope of this study.

Despite these caveats, the study provides timely and policy-relevant insights. A key contribution lies in the within-country analysis, that is so far rare in this field. By disaggregating data at the national level, we uncover regional dynamics often masked in EU-wide assessments. While the drivers of inequality vary across countries, recurring patterns point to common vulnerabilities. These consistencies not only strengthen the robustness of our findings but also suggest opportunities for coordinated EU+ action tailored to national contexts. Integrating social and health dimensions into climate and air quality policies is essential to design strategies that simultaneously advance decarbonisation, reduce pollution, and promote environmental justice. Equally important is the choice of spatial resolution: policies built on overly aggregated data risk overlooking vulnerable groups, whereas high-resolution analyses enable inclusive, effective, and evidence-based responses to the health impacts of air pollution in a changing socioeconomic and environmental landscape.

Contributors

Conceptualisation: CR, JS, DJ. Data curation: CR. Formal analysis: CR, JS, MA. Funding acquisition: KK, AN, DJ. Investigation: CR. Methodology: CR, JS, MA. Project administration: KK, AN, DJ. Resources: JS, KK, AN, DJ. Software: CR. Supervision: JS, DJ. Validation: CR, JS, MA, KK, AN, CB, DJ. Visualisation: CR. Writing—original draft: CR. Writing—review & editing: CR, JS, MA, KK, AN, CB, DJ.

Data sharing statement

The GCAM-Europe 7.2.0 version and the POLICY55 scenario used in this study are publicly available via Zenodo https://zenodo.org/records/15655568 and the corresponding GitHub repository https://github.com/bc3LC-GCAMEurope/gcam-core. The rfasst 2.1 package version can be accessed through Zenodo https://doi.org/10.5281/zenodo.17867248 and its GitHub repository https://github.com/bc3LC/rfasst. The code to reproduce the analysis and the figures is available at https://github.com/bc3LC/Rodes-Bachs_etal_AP_health_ineq_EU, while the data to reproduce the analysis can be accessed at https://zenodo.org/records/17913230.

The high resolution 1 km² grid data was retrieved from the following sources: settlement type⁶¹ https://human-settlement.emergency.copernicus.eu/download.php?ds=smod, disposable income⁶² https://entrepot.recherche.data.gouv.fr/dataset.xhtml?persistentId=doi:10.57745/QRBINB, and European population⁶³ https://ec.europa.eu/eurostat/web/gisco/geodata/population-distribution/geostat.

The NUTS3 data was retrieved from the following sources: settlement type⁶⁴ https://ec.europa.eu/eurostat/web/rural-development/methodology, harmonised gini index data⁶⁵ https://entrepot.recherche.data.gouv.fr/dataset.xhtml?persistentId=doi:10.57745/TTIOKI, and population data was retrieved from rfasst population data projections. The income per capita at the NUTS3 level was computed based on the gridded data. The Gini and income per capita aggregation procedures are detailed in Section S12.

Declaration of interests

The authors declare no competing interests.