Dataset for "WHO_npjCleanAir_Data_Scripts" by Im et al. (2025)

npj | clean air Article https://doi.org/10.1038/s44407-025-00013-w Europe will struggle to meet the new WHO Air Quality Guidelines Check for updates Ulas Im1,ZhuyunYe 1,NinaSchuhen 2, Sourangsu Chowdhury2, Jesper H. Christensen1, Camilla Geels1, Risto Hänninen3,ØivindHodnebrog 2, Louis Marelle4, Mikhail Sofiev3, Jørgen Brandt1&KristinAunan 2 The World Health Organization (WHO) updated its Global Air Quality Guidelines in 2021 due to growing evidence on adverse health impacts of air pollution even at low concentrations. We used a suite of regional atmospheric chemistry models to simulate fine particulate matter (PM 2.5 ) and ozone (O 3 ) levels over Europe in 2015–2050 and assessed the compliance of European countries with the new guidelines under different emission scenarios. The results show that 65% of the EU countries will comply with the PM 2.5 target value (5 µg m−3) by 2050 under ambitious emission reductions (SSP12.6). Under less ambitious mitigation scenarios (SSP2-4.5 and SSP3-7.0), the compliance level is only 10%. In addition, none of the EU countries will comply with the O 3 target value (60 µg m−3), while interim values are achieved in most of the EU countries, partly under SSP2-4.5, and to a large extent under SSP1-2.6. These results highlight that reaching the new WHO limit values will be challenging for Europe, partly due to natural contribution to PM 2.5 reaching up to 50% in some regions. Our findings imply the necessity of more drastic emission reductions to meet the targets. Fine particulate matter (PM 2.5 )andozone(O 3 ) are associated with a range of negative health outcomes, including premature mortality, with large economic impacts on society1–3. The existing estimates of global premature mortality vary substantially, typically differing by a factor of about two. The World Health Organization (WHO) estimated 4.2 million premature deaths in 2016, globally1. In contrast, Im et al.3and Lelieveld et al.4estimated higher figures of 5.3 million and 8.8. million in 2015, respectively, related to exposure to PM 2.5 and O 3 . In Europe, specifically, the number of premature deaths has been estimated to be around 300 to 800 thousand2,4. Recently, Achebak et al.5estimated 115 thousand as the O 3 -related premature mortality in 2015–2017. According to the European Environment Agency (EEA), exposure to PM 2.5 was responsible for approximately 250 thousand premature deaths in 2023 in the EU countries, marking a reduction of 45% between 2005 and 20226,7. The differences in these assessments stem from different exposure-response functions used across studies, and differences in methodology of exposure calculations8. In particular, the upper estimate of Lelieveld et al.4is based on post-processed simulations of global atmospheric models with coarse resolutions and/or limited chemistry representations as well as accounting for all non-communicable diseases9. Due to a growing body of evidence on the health impacts of air pollution10 (https://www.sciencedirect.com/special-issue/10MTC4W8FXJ and references therein) even at lower concentrations than previously understood11–14, the World Health Organization (WHO) updated its 2005 Global Air Quality Guidelines in September 20211. The new air quality guidelines are ambitious and recommend limit values for six air pollutants: PM 10 ,PM 2.5 ,O 3 , nitrogen dioxide (NO 2 ), carbon monoxide (CO) and sulfur dioxide (SO 2 ). These guidelines are not legally binding but can potentially influence air quality policy across the globe in the nearto midterm future. Among these pollutants, annual mean concentration of PM 2.5 is recommended not to exceed 5 µg m−3, while the 99th percentile of daily mean PM 2.5 is recommended not to exceed 15 µg m-3. The guidelines for O 3 set the short-term exposure to (annual 99th percentile of daily maximum 8-hourly surface O 3 (DMAX8hrO 3 ) concentration) not to exceed 100 µg m−3, and the long-term exposure guideline recommends a peak season mean DMAX8hrO 3 of 60 µg m−3. Following these developments, the European Parliament published the new EU directive on air quality, which entered into force in December 2024. The new directive is an ambitious roadmap to the EU Zero Pollution Action Plan towards 2050, when air pollution is reduced to levels no longer considered harmful to health and natural ecosystems. The available literature on compliance to these new air quality guidelines are mainly limited to measurementsormodelingstudiesfocusingon the present-day, while we are not aware of any studies estimating if and when different regions or countries will comply in the future. According to EEA15,more than 90% of the European monitoringstations registered PM 2.5 and O 3 concentrations above the WHO annual guideline levels in 2022. Bowdalo et al.16 compared the new WHO limit values with ground station observations across Europe and found that the target levels are widely 1Aarhus University, Department of Environmental Science/iClimate, Roskilde, Denmark. 2Center for International Climate Research (CICERO), Oslo, Norway. 3Finnish Meteorological Institute (FMI), Helsinki, Finland. 4Sorbonne Université, UVSQ, CNRS, LATMOS, Paris, France. e-mail: [email protected].dk npj Clean Air | (2025) 1:13 1 1234567890():,; 1234567890():,; exceeded, with up to 98% of stations exceeding the annual PM 2.5 guideline level of 5 µg m−3. Franke et al.17 showed that in Europe, PM 2.5 concentrations in 2016 exceed the annual WHO guideline levels more than a factor of 5andO 3 by up to a factor of 2, and that the mortality attributable to air pollution can be reduced by up to 70% depending on the emission scenario. The present study therefore aims to project the future PM 2.5 and O 3 levels over Europe and the individual EU countries until 2050 under different emission scenarios using a multi-model ensemble and assess if and when the new WHO guidelines will be achieved across Europe. Results and Discussion Compliance to WHO PM 2.5 interim and limit values The analysis has been performed for the bias-corrected simulated annual mean PM 2.5 surface concentrations (see Methods Section “Bias correction”) to alleviate the impact of models underestimating surface PM 2.5 concentrations by -4% to -57%, with a bias of -24% for the multi-model mean (MMM; Table 1). As the bias correction has been done using monthly observed PM 2.5 values (see Methods Section “Model evaluation”), we have not performed analyses on the 99th percentile of the daily mean PM 2.5 concentrations. Fig. 1shows the surface PM 2.5 concentrations over continental Europe as simulated by the MMM under the different scenarios, after bias-correction. Surface PM 2.5 concentrations decrease from 2015 to 2050 in all three scenarios, by 47%, 26%, and 13% under SSP1-2.6, SSP2-4.5, and SSP3-7.0, respectively. The largest decreases are projected to take place over central Europe, in particular under SSP1-2.6 (Fig. 2, upper panel). However, as seen in Fig. 1, SSP1-2.6 is the only scenario where the mean surface PM 2.5 concentration approaches the 5 µg m−3WHO target value, only after 2040, and only according to some models. Figure 3shows if and when the future surface PM 2.5 concentrations in the individual EU countries are projected to comply with the WHO guidelines under the different emission scenarios, based on the MMM. As seen in the figure, the majority of the EU countries are projected to not fulfill the WHO target value of 5 µg m−3for PM 2.5 . According to the MMM, 19 countries, mainly western and northern countries will comply with the WHO target level under the SSP1-2.6 scenario, and majority only after 2030 or 2040. Few countries in Scandinavia (Finland, Norway, and Sweden) also comply in the SSP2-4.5 and SSP3-7.0 scenarios, already during present-day. Compliance to the interim values (Fig. 3)iseasier.Thefirst interim value of 35 µg m−3is complied with in all EU countries and is therefore not showninFig.3. All EU countries would have already complied with the interim target 3 (15 µg m−3) if the developments during 2015–2024 would follow SSP1-2.6 or SSP2-4.5, and only two countries (Bulgaria and Cyprus) would not comply under SSP3-7.0. Similarly, all EU countries either already or will comply with interim target 4 (10 µg m−3) under SSP1-2.6, the majority before 2025, while Bulgaria and Cyprus will not comply in SSP24.5. Under SSP3-7.0, the majority of countries complies with the interim target 4, while 12 countries will fail reaching 10 µg m−3.Overall,resultsshow that compliance to WHO target and interim values for PM 2.5 will mainly be possible after 2040, and mostly in the western and northern Europe. Contribution of natural sources to PM 2.5 concentrations Two of the models from our ensemble, SILAM and WRF-Chem, have the natural emission of mineral dust aerosols calculated online (see Methods Section “Air Pollution Downscaling”). Current version of the DEHM model has natural mineral dust from the boundaries only and has no natural mineral dust emissions. Although all three models have online-calculated sea-salt aerosols, due to missing natural mineral dust emissions in the DEHM model, we estimated the contribution of natural aerosols (mineral dust and sea-salt) only with the SILAM and WRF-Chem models. Contribution of mineral dust and sea-salt to domain-mean PM 2.5 mass accounts for around 27% in SILAM (Fig. 4: top left panel) and around 37% in WRF-Chem (Fig. 4: top right panel) in the present day. The contributions in 2015 over the continental Europe are upwards of 20% in SILAM, largest around the coastal areas in the north and western regions due to sea-salt aerosols, and in the Mediterranean also due to mineral dust (Fig. 4,lowerleft panel). Same pattern is modeled by the WRF-Chem model (Fig. 4,lower right panel), however with the contributions up to 50%, particularly over Scandinavia and the northwestern coast. Differences are linked to the different dust and sea-salt parameterizations in the two models (see Methods Section “Air Pollution Downscaling”). These contributions are comparable with available literature, reporting up to 16% natural contribution in southern Europe18,19.Guetal. 20 attributed up to 20% of PM 2.5 -related premature mortality in Europe during 2005–2015 to natural sources. Im et al.21 estimated more than 30% of sea-salt contribution to surface PM 2.5 in the Table 1 | Mean bias (MB), root mean square error (RMSE), normalized mean bias (NMB), fractional gross error (FGE), Pearson correlation (r), and the regression coefficient (r2) for simulated daily maximum 8-hour moving average O 3 concentration (DMAX8hrO 3 ) and annual mean PM 2.5 concentrations against EBAS measurements in the 1980–2014 period for each model and their ensemble mean Model MB RMSE NMB FGE r r2 PM 2.5 (Annual mean) DEHM −5.17 6.70 −0.57 0.60 0.51 0.26 WRF-Chem −0.86 4.55 −0.04 0.30 0.41 0.17 SILAM −3.21 5.14 −0.32 0.38 0.53 0.28 EnsMean −2.64 4.88 −0.24 0.33 0.50 0.25 O 3 (DMAX8hrO 3 ) DEHM −0.30 25.98 0.03 0.27 0.52 0.27 WRF-Chem −3.11 27.36 −0.01 0.29 0.42 0.18 SILAM 9.31 31.15 0.13 0.32 0.40 0.16 EnsMean 2.18 26.08 0.06 0.27 0.48 0.23 Fig. 1 | Bias-corrected surface PM 2.5 concentrations as predicted by the ensemble mean in 2015–2050 over continental Europe under different emission scenarios (solid lines of the corresponding color). The black line shows the WHO guidelines target value of 5 µg m−3. Shaded ranges show the standard deviation between the individual models. https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 2 Nordic region. Both models predict that the relative contribution due to natural sources increases in the future, and the rate of increase depends on how significantly the anthropogenic PM 2.5 is mitigated; SSP1-2.6 leads to the largest increase in the natural contribution while SSP3-7.0 leads to the smallest increase. This large contribution, which can be spatially up to 50% in SILAM and 70% in WRF-Chem, shows the challenge in parts of Europe to reach the WHO target value of 5 µg m−3in the future. This contribution can grow even larger if other natural contributions, such as biogenic Fig. 2 | PM 2.5 (upper panel) and O 3 (lower panel) surface concentrations in 2050 with respect to 2015 under different emissions scenarios: SSP3-7.0 (left panel), SSP2-4.5 (middle panel), and SSP1-2.6 (right panel). Fig. 3 | WHO interim and target values for surface PM 2.5 and the compliance across European countries under SSP1-2.6 (right panel), SSP2-4.5 (middle panel), and SSP3-7.0 (left panel). https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 3 secondary organic aerosols, are singled out from secondary aerosols. Figure 4also shows that in 2015, surface PM 2.5 concentrations over the continental Europe, except for northern Scandinavia, are above 5 µg m−3. Regions with PM 2.5 concentrations above 5 µg m−3decrease in the future, largest in the SSP1-2.6 scenario (Fig. S4). The exceedances remain in all scenarios over regions, which are largely affected by natural PM 2.5 ;inparticularoverthecoastalareas,aswellasoversouthernEurope,whichis largely affected by dust transport. Compliance to WHO O 3 interim and limit values Peak season DMAX8hrO 3 averaged over the EU countries as simulated by the MMM are presented in Fig. 5, along with the WHO target value of 60 µg m−3. The bias correction is not applied over the simulated O 3 concentrations, as the normalized mean bias (NMB)isonlyabout6%forthe MMM [-1% to 13% across the models] (Table 1). Fig. 5(top) shows that the peak seasonal DMAX8hrO 3 projections change slightly within the simulation period, with interannual variability often exceeding the systematic trends. Under SSP1-2.6, the peak season concentrations decrease by 17 µg m−3(-20%) from 90 µg m−3in 2015-2025 to 72 µg m−3in 2041-2050, while under SSP2-4.5, the decrease is 4 µg m−3(-3%). On the other hand, under SSP3-7.0, there is a slight increase of 2 µg m−3(+2%). Under SSP12.6, where there is an overall decrease across the domain, O 3 concentrations increase slightly over the Benelux region and Po valley, which are the known hot spots (Fig. 2, lower panel). The largest increases are projected to take place over central Europe under SSP2-4.5, and over most of the study domain under SSP3-7.0. EU countries on average are not projected to comply with the WHO guideline target value of 60 µg m−3by 2050 under any of the emission scenarios. On the other hand, the first interim target of 100 µg m−3can be achieved after 2035. While the models suggest that achieving the peak seasonal target value for long-term exposure to O 3 will not be possible by 2050, the annual 99th percentile target value of 100 µg m−3(short-term exposure) might be achieved by some countries under the SSP1-2.6 scenario, according to some models (Fig. 5bottom). Compliance with the WHO guidelines in the individual EU countries are presented in Fig. 6.Asseeninthefigure, the ensemble-mean projects no compliance to the target long-term exposure value of 60 µg m−3among the EU countries under any of the emission scenarios, with only 9 countries projected to reach the target after 2040 under the minimum model simulations. On the other hand, the interim 1 value of 100 µg m−3is achieved in SSP2-4.5 in all countries. The interim 2 value of 70 µg m−3is achieved in 21 countries in SSP1-2.6, and in 15 western and northern European countries in SSP2-4.5. On the other hand, Finland and Sweden will be able to comply with the short-term exposure value of 100 µg m−3under SSP2-4.5, while Fig. 4 | Primary natural (mineral dust and sea-salt) contribution (%) to domainmean PM 2.5 mass over Europe in SILAM (upper left panel) and WRF-Chem (upper right panel) under different SSP scenarios (shaded areas represent the spatial variability). Lower panels show 2015 annual mean of the dust +sea-salt contribution (%) to PM 2.5 under SSP2-4.5 scenario for SILAM (left) and WRFChem (right). The dot-shaded regions are where PM 2.5 concentrations are above 5 µg m−3. https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 4 more than half of the countries will comply with this threshold under the SSP1-2.6 scenario. These results imply that the short term O 3 exposure (8hour) can be mitigated by emission reductions, particularly with ambitious reductions, whereas the long-term exposurewillbemorechallengingfor Europe. Particularly under the worst case scenario (SSP3-7.0), compliance to the new WHO target, and interim valuesare not possible in any European country. Similar to the compliance to PM 2.5 , meeting the O 3 target and interim values will be particularly challenging in southern European countries due to much higher O 3 levels because of higher photochemical activity, UV radiation, and precursor sources22. In conclusion, complying with the new WHO air quality guidelines for PM 2.5 and O 3 will be challenging for several European countries in the nearterm future until 2050. Our multi-model ensemble of state-of-the-art regional atmospheric chemistry-transport models, using several emission scenarios, and bias-correction for PM 2.5, showsthatitrequiresambitious emission mitigations (e.g. SSP1-2.6) to achieve the WHO target value. Under moderate mitigation scenarios (e.g. SSP2-4.5), only a few countries, and mainly only after the 2030 s, can comply with the PM 2.5 target value. This challenge is partly associated with the fact that the natural contribution from mineral dust and sea-salt can comprise up to 70% of the PM 2.5 ,which cannot be mitigated. It should be noted that while there are indications from epidemiological and experimental studiesthatdesertdustcanaggravate respiratory and potentially also cardiovascular health, its effects may differ in magnitude and mechanism from those of urban PM 2.5 23,24.Ouranalyses show that the O 3 target value is practically unreachable before 2050. Discussion With its new EU directive on air quality in 2024, the EU set an ambitious roadmap, within which air pollution is reduced to levels no longer considered harmful to human health. Even though our results show that Europe will not be able to comply with the new WHO guidelines for PM 2.5 and O 3 unless extremely ambitious emission reductions take place, it is still beneficial to reduce anthropogenic emissions to reduce air pollution and its negative health impacts. To show this,wehavecalculatedthepercentageof the population of each country that is exposed to PM 2.5 concentrations above the new WHO guideline in 2015 and 2050 (Fig. 7). On average,94% of the 2015 European population is exposed to PM 2.5 concentrations above 5µgm −3, while projections show that in 2050, under SSP1-2.6, exposure to above 5 µg m−3drops to 45% of the population. Under SSP2-4.5 and SSP37.0, it falls by 2-5%, to 89% and 92%, respectively. However, the change in the population fraction not exposed to concentrations above the target limits strongly depends on the country. In some countries such as Denmark, Estonia, Finland, Poland, Sweden, and Switzerland, large increases in the low-exposure population fraction are achieved under all SSPs, by more than 20%. In most of the other countries, such increases are by only a few percent under SSP2-4.5 and SSP3-7.0. These reductions in exposure will however still avoid premature mortality and morbidity, leading to reductions in associated costs for the society, and therefore are beneficial. Our results indicate that fulfilling WHO AQ recommendations requires mitigation strategies that introduce strong policies, fast renewable adoption, reduced consumption, widespread electrification in transport and industry, and global cooperation and coordination, leading to deep emissions cuts, following the SSP1-2.6 scenario. Continued reliance on fossil fuels and weaker carbon reduction policies than in SSP1-2.6, such as in SSP2-4.5, will lead to not fulfilling these recommendations. The above analysis took 2015 as the starting year for emission scenarios. One has to keep in mind that during the period of 2015–2024 the global emissions followed the SSP3-7.0 trend (Fig. 8), i.e., the obtained results should be considered as optimistic. One should also note that the natural contribution to PM 2.5 canbeuptomorethan50%insomecountries, which would require even more drastic emission reductions to comply with the guidelines. A strength of our approach is the use of multiple atmospheric chemistry models and bias correction to simulate surface concentrations as close to the observed concentrations as possible. On the other hand, it also highlights the differences across models in their physical and chemical Fig. 5 | Top panel: Peak season surface daily maximum 8 h O 3 (DMAX8hrO 3 ); bottom panel: annual 99th percentile of DMAX8hrO 3 concentrations as simulated by the ensemble mean in 2015–2050, averaged over EU member countries under different emission scenarios. The black lines show the WHO guidelines target values of 60 µg m−3 and 100 µg m−3, respectively. Shaded areas show the standard deviation between the individual models. https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 5 mechanisms and thus representation of atmospheric chemistry, aerosol processes, and natural emissions, as summarized in Methods Section “Air Pollution Downscaling”and Table S1. Therefore, our results advocate the use of a suite of models for more robust policy recommendations. Methods Climate Downscaling with WRF The WRF model version 4.125 has been used for downscaling global climate scenarios. In this study, the model domain has been set up with a 20 km horizontal resolution in a Polar stereographic projection centered on 60 N, and up to 10 hPa with 54 vertical layers. The study domain covers continental Europe at 25-72 N, 25W-45E. Meteorological initial and boundary conditions to WRF are provided by the NCAR CESM2 model, which has a spatial resolution of 0.9 × 1.25 degree. This resolution is not sufficient to resolve the highly variable air pollution concentrations within cities and countries. Boundary conditions of 3-D temperature, humidity, horizontal winds and geopotential height are updated every 6 h in WRF, while sea surface temperatures and sea ice are updated daily. In the present study, we have used future scenarios from the Coupled Model Intercomparison Project Phase 6 (CMIP626:), which feeds to the recent IPCC 6th Assessment Report (AR6). Three shared socioeconomic pathways (SSP) scenarios are used: SSP1-2.6, SSP2-4.5 and SSP37.0, to address different levels of mitigation and adaptation27.SSP1andSSP3 define various combinations of high or low socio-economic challenges to climate change adaptation and mitigation, while SSP2 describes medium challenges of both kinds and is intended to represent a future in which development trends are not extreme in any of the dimensions, but rather follow middle-of-the-road pathways27. The SSP1-2.6 scenario aims to Fig. 6 | WHO guidelines interim and target values (color coded) for long-term exposure (top panel) and short-term exposure (bottom panel) to O 3 concentrations along with the compliance across the EU member countries under SSP1-2.6 (right panel), SSP2-4.5 (middle panel), and SSP3-7.0 (left panel). Target (Min) is based on the minimum values from three models, and all the rest are based on the MMM. https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 6 achieve a 2100 radiative forcing level of 2.6 W m−2, keeping the temperature increase below 2 °C compared to the preindustrial levels. The SSP2-4.5 describes a “middle of the road”socio-economic family with a 4.5 W m−2 radiative forcing level by 2100. The SSP3-7.0 scenario is a medium-high reference scenario with the highest CO 2 , methane and air pollution precursor emissions among the three scenarios used in the present study. In the present study, we use three state-of-the-art atmospheric chemistry and transport models to simulate the surface PM 2.5 and O 3 concentrations in the period of 2015 to 2050 over Europe under different emission scenarios to assess the compliance of European countries to the new WHO target and interim values. The emission scenarios are referred to as taking “the green road”(SSP1-2.6), “the middle of the road”(SSP2-4.5) and “arockyroad”(SSP3-7.0). We also discuss the implications on the population exposure to these pollutants. Air Pollution Downscaling We have used three state-of-the-art chemistry transport models: the Danish Eulerian Hemispheric Model (DEHM), the System for Integrated Fig. 7 | Percentage of population exposed to surface PM 2.5 levels over the WHO target value of 5 µg m−3in 2015, and in 2050 under different SSPs. Fig. 8 | Reported and projected (SSPs40:) global CO 2 emissions. Reported CO 2 emissions are taken from the Global Carbon Budget (https://globalcarbonbudget. org/)underhttps://ourworldindata.org/co2-emissions. SSP are taken from https://ourworldindata.org/ explorers/ipcc-scenarios?Metric=Greenhouse+gas +emissions&Rate=Total&Region=Global&country= SSP3+-+Baseline~SSP1+-+2.6~SSP2+-+4.5. https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 7 modeLling of Atmospheric coMposition (SILAM), and the Weather Research and Forecasting model with chemistry (WRF-Chem). DEHM and SILAM are members of the Copernicus Atmospheric Monitoring Service (CAMS) regional forecasting ensemble (https://atmosphere.copernicus.eu/ regional-services), providing regional forecasts and analysis of air pollution overEurope.FireemissionsinallmodelsareproducedbytheFireForecasting Model, an extension of the IS4FIRES system28.Thephysical,chemical, and aerosol mechanisms used in the models, along with their treatment of natural aerosols and precursor gases are provided in Table S1. The DEHM model was originally developed mainly to study the transport of SO 2 and sulfate (SO 4 )totheArctic 29 and has been extended to different applications during the last decades. It has been documented extensively30 and evaluated in several intercomparison studies2,21,31–33.The DEHM version used in the present study has a 20 km × 20 km spatial resolution over Europe, extending up to 100 hPa through 29 vertical levels, with the first layer height of approximately 20 m. The gas-phase chemistry module includes 71 chemical species, 9 primary particles, including natural particles such as sea-salt and 122 chemical reactions (Brandt et al.30). The current version of the DEHM model does not include wind-blown or re-suspended dust emissions (Table S1). SILAM (http://silam.fmi.fi34,) version 5.8 is used both for the global boundary runs in 2 degree resolution and for the regional European runs with 0.4 × 0.3 degree resolution (approximately 30 km). The global setup extends up to 3.7 hPa with 26 vertical layers. For regional runs the lowest 19 layers from the global setup are used, the top being at 106 hPa. Boundaries for these regional runs are taken from the global runs, updated every 3 h. The chemistry and aerosol mechanisms used in the SILAM model is provided in Table S1. WRF-Chem35 version 3.9.1.1 has been set up for a European domain at 35 km x 35 km horizontal resolution and 50 vertical layers, ranging from the surface and up to 50 hPa. The chemistry is online with the meteorology, and both direct and indirect (i.e., activation as cloud droplets) aerosol feedback is accounted for. Simulations are run as multiple 15-month time slices whereof the first 3 months (Oct-Dec) of each run are considered spin-up. Meteorological boundary conditions from CESM2 were updated every 6 h and spectral nudging towards the global data was applied in the free troposphere (i.e., above the planetary boundary layer) for temperature, horizontal winds, and geopotential height. Chemical boundary conditions are from the global SILAM simulations. The global SILAM runs that are used for boundary conditions for European regional runs have been conducted at a spatial resolution of 2 × 2 degrees and with a temporal resolution (averaging time window) of 3 h. For the ERA5 meteorology, they cover a time window of 1980–2019. For the CESM meteorology the historical part covers the years from mid-1940’s to 2014. However, the project interest is mainly in the time window of 1980–2014, which will be used for the regional runs. All the three future scenario runs (SSP1-2.6, SSP2-4.5, SSP3-7.0, 4 TB per scenario was stored for long term usage) cover the time window of 2015–2099. The global runs have been conducted using the original (with a resolution of 0.5 degree) CMIP6 emissions for anthropogenic emissions36. They are complemented with NOx emissions from lightnings and MEGAN-MACC/CAMSv3.1 biogenic emissions37, together with halogen emissions based on the GEIA emission inventory and available data from the literature for yearly changes of different CFCcompounds. Additionally, the emissions of N 2 O from soils, together with anthropogenic N 2 O emissions from the EDGAR 5.0 inventory is included in simulations (https://edgar.jrc.ec.europa.eu/38,). For future SSP scenarios these anthropogenic N 2 O emissions are scaled with the different SSP scenario total emissions. For anthropogenic PM coarse emissions, which are not available in the CMIP6 emissions inventories, we used the EDGAR v4.3.2 emission inventory for PM 10 and PM 2.5 to make the coarse PM. The future development of PM coarse emission is estimated sector-wise from the development of CO and NOx emissions for different SSPs. Emissions The standard anthropogenic emissions are based on CMIP6 emissions at 0.5 × 0.5° resolution. Historical emission inventory is CEDS inventory36 and the future scenario emissions of Shared Socioeconomic Pathways (SSPs27:) are by Gidden et al.39 Fig. 8shows the global CO 2 emissions in the different SSP scenarios40 we have used in the present study, along with the reported global CO 2 emissions by the Global Carbon Budget (https:// globalcarbonbudget.org/). These emissions are directly used in SILAM global simulations. In regional (European scale) runs these emissions are remapped into 0.1 × 0.1 degree grid cells by using the CAMS-GLOB-ANT emission inventory version 2.1 for year 2015 as a proxy inside each 0.5×0.5 degree grid cell. If no emissions exist in CAMS inventory, a unity profile is used inside the original CMIP6 grid cell. Aviation emissions are not regridded into finer grids. Fig. 9shows the NOx emissions from the transportation sector in 2015 and 2050 in the original SSP2-4.5 scenario and its downscaled version used in this study. Model evaluation All models that participated in the study have been extensively evaluated in numerous research and operational projects2,28,41–44.Inaddition to the existing evaluations, a dedicated effort has been made to quantify the uncertainties of the setup used in this study. Surface O 3 and PM 2.5 measurements are taken from the EBAS database (https://ebas. nilu.no/)fortheperiod1980–2014. Model simulations are interpolated to observational stations using the nearest-grid interpolation. Only hourly measurements are used for O 3 evaluation and both hourly and daily measurements are used for PM 2.5 evaluations. The DMAX8hrO 3 concentrations are calculated as the daily maximum of 8-hours moving average of hourly O 3 observations and only keep the data that have at least 6 h in each 8 h window. Annual mean PM 2.5 concentrations are calculated from daily mean concentrations, and during each aggregation process, we use only measurements with 75% data availability for the time window. The evaluations metrics are calculated using these aggregated concentrations from all stations. As a result, DMAX8hrO 3 concentrations are overestimated by up to 13%, with rvalues around 0.4–0.5 (Table 1). The time series of observed and modeled surface annual PM 2.5 and DMAX8hrO 3 are presented in Figs. S2 and S3, respectively. The multi-model ensemble-mean overestimates DMAX8hrO 3 concentrations by 6%, with rvalues around 0.5. The models were less successful for surface PM 2.5 concentrations. All models underestimated annual mean surface PM 2.5 by up to 57%, with the DEHM model having the largest underestimation. The Pearson correlation of annual mean PM 2.5 is also around 0.4–0.5. These differences among models are due to several reasons including the different chemical mechanisms, which leads to differences in simulated O 3 concentrations, different aerosol mechanisms leading to differences in size distribution and secondary aerosol formation, native model spatial and vertical resolutions that lead to differences in transport as well as dry and wet removal in the atmosphere, and inclusion and simulation of natural aerosols, as described in the Methods Section “Air Pollution Downscaling”and presented in Table S1. For example, the largest biases of the DEHM model with respect to PM 2.5 levels compared to other models are partly because the DEHM model does not include mineral dust, which can be quite important in southern Europe. The spatial and temporal breakdown of errors in thesemodelshavebeenanalyzedinpreviouswork 32. In addition, models do not capture the temporal evolution of the observed concentrations because the CESM2 climate model has been run without nudging, and therefore creates its own meteorology, which can largely deviate from the true meteorology. Therefore, it is not expected to obtain good statistics using CESM2-driven models for short time scales. Bias correction In the present study, we bias-corrected surface PM 2.5 concentrations due to large underestimations and large uncertainties in the aerosol https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 8 chemistry caused by physical (e.g. size distribution) and chemical (e.g. chemical composition) properties in the different models. On the other hand, surface O 3 concentrations were used as calculated by the models without any correction, as the biases here are considerably smaller across all models (Table 1) and bias correction did not result in substantial improvement. For PM 2.5 , a gridded satellite-derived dataset45 was used as observational reference for the correction, meaning we try to minimize any model biases relative to this dataset. It consists of global monthly mean concentrations from 1998 to 2022 at a horizontal resolution of 0.1 ˚x0.1˚,and combines measurements from multiple satellites (specifically MODIS, VIIRS, MISR, and SeaWiFS) with the GEOS-Chem chemical transport model and a Convolutional Neural Network for calibration to surface observations. Due to a lack of availability of high-quality gridded PM 2.5 observations at daily resolution, we decided to estimate bias correction parameters on monthly scales and then apply them to the respective daily projections. The 1998–2014 period was selected for parameter estimation (coinciding with the end of historical model simulations), and these parameters were then applied to simulations from the periods 1980–1997 and 2015–2050 (all emission scenarios). The remaining 2015–2022 reference data was used for out-of-sample evaluation of the bias corrected projections (see Fig. 10). Due to its versatility and ability to preserve long-term model trends, we chose the quantile delta mapping bias correction method (QDM46;), which removes biases by adjusting multiple quantiles and thereby addresses the entire distribution as compared to just and and/or variance. We here use the version implemented in the R software package MBC47.Inafirst step, quantiles from the empirical distribution function of future projections are detrended with respect to the quantiles of the historical/estimation period, which yields the model trend Δas the ratio between the future and historical quantiles. Next, the detrended quantiles are bias-corrected by mapping them onto the respective quantiles from the reference dataset (regular quantile mapping). Finally, we obtain bias-corrected future projections by re-applying the model trend Δto the bias-corrected quantiles from the previous step through multiplication. This method was applied to all three models on a grid-point-wise basis and using n equidistant quantiles, where n is the total amount of time points in the future period. The QDM approach explicitly preserves long-term trends (whereas they can deteriorate significantly when using regular quantile mapping) and presents a suitable option for biascorrection atmospheric variables lacking specialized techniques, in particular when models can be expected to represent future trends in a realistic manner. In Fig. 10,weshowmeanbias(MB) and root-mean-square error (RMSE) for historical and future model projections, based on comparison against the EBAS ground-based observations. The evaluations are conducted for monthly mean PM 2.5 values at all measurement stations. We only keep data with 75% availability during the process and use the nearest-grid interpolation for model simulations. In most instances, model skill has improved with bias correction, in particular for the DEHM model, which moved from a considerable negative bias to a slightly positive one. Of course, any bias correction approach relies heavily on the quality and accuracy of the reference data set, which can be a particular challenge for air pollution variables. In addition, Fig. S4 shows the time series of monthly mean observed PM 2.5 along with simulated concentrations before and after bias correction. Fig. 9 | NOx emissions from the transport sector in 2015 (upper panel) and 2050 (lower panel) as in CMIP6 emissions (SSP2-4.5: left panel) and downscaled in the present study (right panel). https://doi.org/10.1038/s44407-025-00013-w Article npj Clean Air | (2025) 1:13 9