Future PD caused by several specific endpoints related to PM2.5 have been estimated using non-linear exposure–response functions, a methodology analogous to that previously implemented in . The health impact function was applied in each grid cell (Eq. ) to estimate the premature mortality: 1ΔM=y0×RR-1RR×Population. This equation is based on epidemiological relationships between air pollution concentration and mortality in each grid cell, where ΔM is the premature mortality due to a specific disease, y0 is the baseline mortality rate, RR is the risk ratio and Population refers to the exposed population (in this contribution, adults are considered). y0 varies according to the cause of mortality, age and European region (Fig. S1 in the Supplement) and is estimated by the WHO for each sex and year. Sex mixing values used in the present study account for both male and female dwellers during the year 2017 (the last available). The premature mortality and RR were estimated for each pathology and the different group ages included in this contribution: 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74, 75–79, +80 years and all ages.

Risk ratios were determined with the GEMM methodology developed by (Eq. ): 2RR=exp⁡θlog⁡[zα+1](1+exp⁡(-z-μν),where z=max⁡(0,PM2.5-2.4µgm-3), θ, α, μ and ν are variables obtained from for each pathology, and z refers to the mean annual concentration of PM2.5. The pathologies included in this work are lung cancer (LC), chronic obstructive pulmonary disease (COPD), lower respiratory infection (LRI), cerebrovascular disease (CEV), ischaemic heart disease (IHD) and non-accidental diseases (NCD + LRI). Other NCD is calculated by subtracting NCD + LRI and the rest of the categories. Uncertainty ranges are expressed as the 95 % confidence intervals (95 % CIs), adopted from .

Population data for Europe were taken from the gridded dataset. These data provide the population density by age and gender for the year 2010 (consistent with national censuses and population registers) with a resolution of 5 km2. Population data were interpolated to the working grid to make them consistent with the gridded air pollution data (Fig. S2, top, in the Supplement).

With respect to the future population, a projection for the year 2050 was estimated using information on population prospects from the United Nations (UN) Department of Economic and Social Affairs Population Dynamics . This included the changes in the total national numbers and the age distribution. The relative variation of the population from this dataset between 2010 and 2050 for each European country and age range was calculated in order to obtain the ratio of population for the future scenario (2050) in this study (Fig. S2, bottom). The population pyramids for the years 2010 and 2050 are presented in Fig. . This figure indicates a very slight projected decrease (-0.2 %) in the European population (807.5 M vs. 806.2 M dwellers in the present and future populations, respectively), especially over eastern Europe (-4.0 %). Conversely, the populations of western and central Europe increase in the UN 2050 projections (by +2.7 % and +5.7 %, respectively). In addition, the projected data include a higher population density over many urban areas and a clear ageing of European citizens. As an example, the population of over 80 year olds (80+) represents barely 4 % of the total European population today, while it is expected to increase to >9 % in the projected UN 2050 estimations.

Data on observed air pollution for conducting studies on the impacts of air pollution on human health are scarce. The network of stations for measuring air pollutants are generally insufficient for health purposes due to their spatial misalignment and low coverage . This limitation has led to the modelling of outputs to provide information about air pollution, especially when air quality projections are needed .

Here, air quality model data (PM2.5 dry aerosol mass) from the WRF-Chem model of the REPAIR initiative are inputted into Eq. () in order to estimate the annual PD associated with different endpoints in both current and future climate change scenarios. The parameterizations implemented in the WRF-Chem model are summarized in Table S1 of the Supplement. The domain covers Europe with a horizontal resolution of 0.11∘ under the Euro-CORDEX requirements . For future scenarios, climate forcing is derived from the RCP8.5 scenario, since RCP8.5 represents an upper limit to climate impacts . The reference periods span 1991–2010 for the present and 2031–2050 for future projections (PM2.5 concentrations are averaged for each 20-year period). The robustness of this simulation for representing PM2.5 is evaluated in the Supplement (Tables S2 and S3), where the model is compared with data from 108 stations belonging to the AirBase database of the European Environment Agency. The results are summarized in Fig. , and the numerical results for each station can be found in Table S3. Briefly, the low errors found (for example, an average mean bias of under 2 µgm-3 and a mean fractional bias of <9 %) guarantee phase accordance (timing) between and similar amplitudes for the simulated and observational series as well as the quantitative accuracy of the simulated climatologies, hence making us confident of the suitability of the modelling system for the purpose of this study.

In our case, simulations do not assimilate observational data. Ground-based observations and satellite products are often used to improve modelling results for present-day simulations concerning particulate matter (e.g. ). However, these bias correction techniques that are widely used in climate impact modelling are limited when future scenarios are included in the simulations, since no observations can constrain future modelling results. Instead, we have decided to use the so-called “delta method” to present the results and future changes in air pollution, as recommended in . In the simple terms applied in this contribution, we assume that the results of the evaluation presented in the Supplement point to accurate results (small biases) for present-day PM2.5 simulations. In addition, the delta method assumes that model errors for the future time slice (2031–2050 in this contribution) will cancel out when compared with the simulation of the present climate (1991–2010, taken as the present reference time slice). This is related to bias correction methods. In particular, delta changes are insensitive to local shift bias correction methods. It is true that more complex bias correction techniques could have been applied (e.g. quantile mapping), but the bias-corrected and delta-change projections differ for those methods , leading to a new source of uncertainty. Therefore, this contribution uses the delta method (assuming the cancellation of present and future biases), as also implemented in other works related to air pollution impacts on health issues (e.g.  or the contributions of (among many others), which rely on these very simulations).

Further details about the methodology of the model simulations are included below. The GOCART aerosol module , the aerosol scheme used in this work, includes a bulk approach for black carbon (BC), organic carbon (OC) and sulfate, as well as a sectional scheme for mineral dust and sea salt using brittle fragmentation theory, a simple and cheap computational approach . In the present work, this scheme is coupled with the RACM-KPP (KPP: kinetics preprocessor; ). ISORROPIA is used for the thermodynamic partitioning of aerosols.

In order to isolate the possible effects of climate change in atmospheric pollutants on pathologies, constant anthropogenic emissions for all present and future simulations are assumed. Anthropogenic emissions for the year 2000 by country and sector with a spatial resolution of 0.1∘ were obtained from the ACCMIP database . This allows possible impacts to be anticipated if mitigation strategies for regulatory pollutants are not carried out, and characterizes the climatic penalty for air quality levels. ACCMIP compiled a global emission dataset with annual official or scientific inventories at the national or regional scale for CH4, NMVOC, CO, SO2, NOx, NH3, PM10, PM2.5, black carbon and organic carbon. Climate-dependent natural emission sources include desert dust, sea salt aerosols and biogenic volatile organic compounds (VOCs). The emissions were pre-processed according to .

As stated in , the estimation of the PM2.5 is carried out by the subroutine sum_pm_gocart in module_gocart_aerosols.F. This estimation considers dust and sea salt concentration in bins 1 (ranges 0.1–1.0 and 0.1–0.5 µm, respectively) and 2 (1.0–1.8 and 0.5–1.5 µm, respectively), black and organic carbon, and sulfate. GOCART does not include the treatment of secondary organic aerosols (SOA). The authors are aware of this limitation; however, the WRF-Chem version requires the use of the GOCART scheme if desert dust and sea salt aerosols are to be included . Nitrate aerosols are also not explicitly included in the simulations conducted here.

Last, it should be mentioned that the GOCART aerosol scheme in the WRF-Chem simulations presented here does not allow full coupling of aerosol–cloud interactions . For instance, convective wet scavenging and cloud chemistry are not available. However, the Morrison microphysics scheme acts as a double-moment scheme here. Hence, the configuration of the model used here allows double-moment microphysics with greater flexibility when representing size distributions and hence microphysical process rates . When the double-moment scheme is activated (as it is here), a prognostic droplet number concentration is estimated using gamma functions and mixing ratios of cloud ice, rain, snow, graupel and hail, cloud droplets, and water vapour . Finally, the interaction of cloud and solar radiation with the Morrison microphysics scheme is implemented in WRF-Chem. Therefore, the droplet number will affect both the droplet mean radius and the cloud optical depth calculated by the model, affecting the cloud and precipitation in the model.

Figure  shows the PM2.5 mean annual concentration over Europe for the present period (1991–2010) and the changes projected for the future scenario (2031–2050, RCP8.5) with emissions from ACCMIP and the mitigation scenario (2031–2050, REN80). As also stated in previous works , chemistry/climate simulations reveal the presence of some areas in Europe that exceed the annual PM2.5 limit value (25 µgm-3) established by the European Directive 2008/50/EC on ambient air quality and cleaner air for Europe. Currently, eastern Europe and some large cities and conurbations have the highest concentrations of fine particles, with some cities such as Paris (France), Krakow (Poland) and Moscow (Russia) presenting elevated levels that exceed 25 µgm-3.

The future changes in PM2.5 concentration attributable to GHG-induced climate change under the RCP8.5 scenario indicate a mean increase of 0.7 µgm-3, with higher increases over southern Europe. These results are in agreement with those of and , among others, who show an overall increase in surface PM2.5 concentration over most land regions. On the other hand, slight decreases are mainly projected over the Scandinavian countries and some Baltic areas (-0.2 µgm-3). The decrease in PM2.5 over this area under the mitigation scenario (REN80) can be > 0.5 µgm-3. Although air quality improves overall in Europe, areas such as Paris, Krakow and Moscow will keep exceeding the European Directive threshold.

On the other hand, the REN80 scenario indicates an overall improvement in air pollution related to PM2.5 in Europe, especially over eastern Europe. In that area, the effect of REN80 largely counteracts the climate penalty (reductions of >-1.0 µgm-3 with respect to present concentrations and average changes of >-1.4 µgm-3 with respect to the RCP8.5 scenario, reaching -2.5 µgm-3 over certain hotspots). The reason for this decrease over eastern Europe is the high ratio of fossil fuels in the energy mixes of countries in that area . Again, these results are in agreement with previous works found in the scientific literature. For instance, obtain a reduction of almost 0.9 µgm-3 for a future scenario with a reduction of 20 % in anthropogenic emissions from power and industrial sources.

It is also noticeable that northern Europe will benefit from both the climate penalty and the mitigation scenarios. This benefit cannot be observed over southern Europe, where the benefits of mitigation strategies might not compensate for the effect of climate change on PM2.5 levels (fine particles will increase in concentration in southern Europe as a consequence of decreasing precipitations and increased emissions from natural sources in this target area) (Fig. ).

Figure  depicts the annual excess PD associated with the present PM2.5 pollution (PRE-P2010). For the present period (1991–2010), 895 000 (95 % CI 725 000–1 056 000) annual PD are estimated over the European area. This estimation is in agreement with – although slightly higher than – that of (647 000 PD) and exceeds that of (546 000), who only include the contribution from primary PM2.5 and secondary inorganic aerosols. report 260 000 PD associated with PM2.5; the difference from the number of annual PD estimated in the present work is primarily due to the domain covered in the simulations presented here. The target domain in the present contribution includes the most populated areas of Russia and Turkey, which are not included in the numbers reported for Europe in the aforementioned study.

In this contribution, the highest incidence of excess PD is estimated for eastern Europe (467 000 annual PD, 95 % CI 378 200–551 100) (Table ), while the highest incidence of normalized PD per 100 000 adult inhabitants is found over central Europe (229.4 PD/100 000 inhabitants per year). The latter result is in good qualitative agreement with the estimation of , who identify the Benelux region as the European region with the highest excess mortality rates associated with air pollution.

When considering individual endpoints (Fig. ), the main cause of the excess PD associated with PM2.5 in the European domain is cardiovascular diseases (IHD + CEV), especially over eastern Europe. Moreover, CEV and IHD are the causes of PD that show the largest differences between target areas. The PD incidence over eastern Europe is caused mainly by IHD + CEV: together, they contribute >60 % of the premature mortality burden in Europe (Fig. ). This is because of (1) the higher population (51 % of the total European population contemplated in this contribution) of eastern Europe in comparison with central Europe (19 % of the European population) and western Europe (30 % of European inhabitants) and (2) the higher PM2.5 concentrations in eastern Europe (as previously shown in Fig. ). Similar results are seen for the total PD by all causes: the PD are proportional (with respect to their percentage) to the population held in the respective areas. In this sense, eastern Europe contributes more than 50 % of the mortality in Europe, while central and western Europe add around 20 % and 30 % of the PD, which are proportional to their populations.

It is noticeable that for COPD, LRI and LC, the three studied regions contribute similar percentages of the total PD over Europe (around 30 %), which is indicative that the COPD, LRI and LC ratios are much higher over western Europe (30 % of the European population) and central Europe (19 % of the total European population) than over eastern Europe (51 % of the population).

In this section, the role of the climate penalty is assessed by comparing two simulations differing only in the radiative forcing (1991–2010 vs. RCP8.5 2031–2050; i.e. the two model runs FUT-P2010 and PRE-2010). For this, the population was kept constant at 2010 levels. The results presented in Table  indicate that the climate penalty has a very limited impact on air-pollution-related future premature deaths over Europe. When comparing the total incidence of PD in the PRE-P2010 and FUT-P2010 cases, only a +0.2 % increase is projected in the future scenario. However, this increase is uneven across the different regions of Europe. While the total PD per 100 000 adults (Table ) increases by nearly 2 % in western Europe and by 0.2 % in eastern Europe, the mortality decreases by -0.8 % over central Europe. When considering a future RCP8.5 scenario, these variations in the distribution of PM2.5 in Europe are mainly due to changes in precipitation (Fig. ).

The decrease in rainfall in southern Europe (e.g. ) and the projected increase in rainfall in northern and central Europe for this scenario (e.g. ) affect the wet scavenging of particles (neglecting the influence of modified photochemistry on the levels of particles). In this context, wet scavenging is the fundamental process for removing particles from the atmosphere (e.g. ).

It should be highlighted that the cause of premature mortality that is most sensitive to changes in PM2.5 is CEV + IHD (Fig. ), as also found by previous studies . The spatial patterns of change in the latter endpoint condition the patterns of total variation in PD by all causes. IHD mortality rates increase by nearly 3 % from 424 000 (95 % CI 356 100–487 600) to 425 000 (95 % CI 357 000–488 800) in the FUT-P2010 case. On the other hand, COPD, LC, and other NCD barely change, since these causes are not as sensitive to the PM2.5 concentration as IHD and LRI are at low PM2.5 concentrations (Fig. ), as also discussed in .

The previous section analysed the contribution of climate change alone (RCP8.5 scenario with the population kept constant) to the PD over Europe. Here, the impact of projected changes in the population according to UN2050 data is added to the climate penalty. This scenario is denoted FUT-P2050. As shown in Fig. , the highest mortality incidence in FUT-P2050 is related to IHD, which has an widely spread spatial pattern in the target domain. For FUT-P2050, the projected PD incidence increases to 1 540 000 (95 % CI 1 247 000–1 818 000), with eastern Europe being the most affected region, contributing almost half of the excess mortality rate over Europe (Table ).

The premature mortality burden associated with exposure to ambient PM2.5 in Europe is expected to increase by 72 % by the year 2050 in FUT-P2050 when compared with the PRE-P2010 simulation: from 894 000 (95 % CI 723 900–1 055 400) PD per year in 1990–2010 to 1 540 000 (95 % CI 1 247 000–1 818 000) PD per year in 2031–2050 (Table ). The leading cause of PM2.5-related mortality is IHD in both the present and future cases: 424 000 (95 % CI 356 200–487 800) PD per year at present, which increases by 74 % to 736 000 (95 % CI 618 200–846 400) PD per year in the FUT-P2050 case. IHD is followed by other NCD; this leads to 265 600 (95 % CI 140 800–329 300) PD per year in the PRE-P2010 case, which increases by 73 %, to 458 400 (95 % CI 243 000–568 400) PD per year in the FUT-P2050 case.

When assessing the relative contribution of each endpoint to the total PD burden in both PRE-P2010 and FUT-P2050, Fig.  indicates that future climate change and the change in the population by the year 2050 will barely change the relative percentage of each mortality cause by region, so the discussion is analogous to that presented in Sect. . The only exception is for LRI, which experiences several important changes from PRE-P2010 to FUT-P2050 (increasing from a 28 % contribution in western Europe for the present case to 36 % over the same area in the FUT-P2050 case, and from 35 % in PRE-P2010 over eastern Europe to 46 % in FUT-P2050 over the same domain).

This case takes into account the effect of climate change (RCP8.5, 2031–2050) together with the emissions scenario where it is assumed that 80 % of the energy production over Europe is provided by renewable sources (REN80). For the analysis of this impact, a scenario where the population remains constant at 2010 levels (REN80-P2010) and a second scenario where the population dynamics are taken into account (REN80-P2050) are analysed below.

Figure  shows that in REN80-P2010, the PD incidence will be lower than in the REN80-P2050 scenario that considers changes in the population, as previously discussed for FUT-P2010 and FUT-P2050. Large central European cities are the areas in which the PD incidence will increase most with a changing population (increases in the density and age of the population, as previously shown in Fig. ). A summary of the excess mortality incidence in all scenarios is presented in Table .

The incidence of all mortality causes studied decreases by 4 % in the future mitigation scenarios REN80-P2010 and REN80-P2050 when compared to the corresponding FUT-P2010 and FUT-P2050 simulations, respectively. This decrease is especially apparent in the central and eastern regions regarding the total PD (Table ). This is explained by the fact that an important part of the anthropogenic emissions of PM2.5 over Europe is associated with power generation . To be precise, this contribution is 15.1 % here. Table  indicates that CEV and IHD contribute the largest relative reductions to the reduction in the incidence ratio of PD in the REN80-P2050 scenario with respect to FUT-P2050 (-6 % changes in annual PD), while COPD and LRI show -5 % changes (Fig. )) and the change in LC is -4 % for the renewable scenario. Other NCD decreases by -2 % in the REN80 simulations. Other NCD and CEV are the endpoints that experience the largest improvements in the mitigation scenario, with decreases in the annual excess PD of 30 000 (95 % CI 15 900–37 200) and 10 000 (95 % CI 5300–12 400), respectively.