The global model used to provide climate change projections in this study, is the HadGEM2-ES Earth System Model (Collins et al., 2011). HadGEM2-ES is a coupled atmosphere-ocean model, with additional Earth system components such as dynamic vegetation, interactive chemistry and aerosols, and a terrestrial and ocean carbon cycle (Collins et al., 2011). HadGEM2-ES was used extensively to contribute model outputs from ensembles of historical and future simulations to Phase 5 of the Coupled Model Intercomparison Project (CMIP5; Taylor et al., 2012). Further details on the implementation of forcings for all CMIP5 simulations are provided in Jones et al. (2011). 3-D distributions of temperature, specific humidity, U and V wind components, surface pressure, soil temperature and moisture from HadGEM2-ES were used as initial and 6-hourly boundary conditions for nested regional numerical weather prediction simulations described below (Sect. 2.2).

The EMEP4UK model is based on the European Monitoring and Evaluation Programme Meteorological Synthesizing Centre-West (EMEP MSC-W) chemistry transport model, used by the UNECE Convention on Long-range Transboundary Air Pollution to assess trans-boundary air pollution in Europe. The EMEP4UK model version applied in this study to simulate regional atmospheric composition and air quality metrics is based on EMEP MSC-W rv4.6 (Simpson et al., 2012, 2015). It uses a one-way nested approach with two domains: an outer domain covering the majority of Europe which provides boundary conditions for a UK nested domain (Vieno et al., 2010, 2014, 2016). The EMEP4UK model meteorological driver is the WRF model version 3.6.1 (https://www2.mmm.ucar.edu/wrf/users/wrf_wps_v3_info.html, last access: 1 September 2016). In addition to using initial and 6-hourly lateral boundary conditions from HadGEM2-ES, WRF simulations are nudged every 6 h in 3-D, using temperature and U and V wind components from HadGEM2-ES.

The EMEP4UK and WRF models use the same grid definition, with a horizontal resolution of 50km×50km for the European domain and 5km×5km for the UK domain. An intermediate domain with 10km×10km horizontal resolution is also used for WRF. The models also share the same vertical grid that employs 21 vertical levels from the surface to 100 hPa, with the lowest vertical layer ∼50 m deep. Modelled air pollutant concentrations described here as surface concentrations have been adjusted to correspond to 3 m above the surface (Simpson et al., 2012).

EMEP4UK uses the CRI-v2-R5 gaseous chemical mechanism (Watson et al., 2008), which has 220 species and 609 reactions. Five classes of fine and coarse particles are represented in EMEP4UK. Gas–aerosol partitioning of secondary inorganic aerosol utilises the Model for an Aerosol Reacting System (MARS) equilibrium module (Simpson et al., 2012); secondary organic aerosol formation uses the volatility basis set approach (Bergström et al., 2012). PM_2.5 is the sum of fine ammonium (NH4-), sulphate (SO42-), fine nitrate (NO3-), fine elemental carbon (EC), fine organic matter (OM), fine sea salt (SS), fine mineral dust, and 27 % of the coarse nitrate aerosol (Vieno et al., 2016). The WRF-EMEP4UK nested model system used here has been thoroughly evaluated against measurements (Vieno et al., 2010; Ots et al., 2016, Lin et al., 2017), including provision of evidence on air quality to the UK government (e.g., AQEG, 2021).

Anthropogenic and greenhouse gas emissions are annually invariant to permit a clearer isolation of the climate change signal. Anthropogenic emissions of NOx, NH3, SO2, primary PM_2.5, primary coarse PM (PM_2.5–10), CO and non-methane VOCs for 2012 are derived from the EMEP Centre for Emission Inventories and Projections (CEIP, https://www.ceip.at, last access: 1 August 2018). The National Atmospheric Emission Inventory (NAEI, https://naei.energysecurity.gov.uk/data-archive, last access: 1 December 2016) is used for anthropogenic emissions for the UK for 2012 at 1 km resolution. Shipping emission estimates for the UK domain are derived from ENTEC (2010), projected to 2012. Annual total anthropogenic emissions derived from the inventories are resolved to hourly resolution using prescribed monthly, day-of-week and diurnal hourly emissions factors and distributed vertically (Simpson et al., 2012). Biomass burning emissions are not included here. The standard EMEP4UK model uses prescribed daily biomass burning emissions derived from satellite-based inventories. To ensure that interannual variability in the model experiments is due to climate alone, biomass burning emissions are turned off altogether. Biomass burning makes a relatively small contribution to total regional emissions over Europe (∼1 % for NO, 8 %–10 % for CO and PM_2.5), but estimates vary substantially (see e.g., Pan et al., 2020) and there is considerable uncertainty in how they will respond to climate change.

Biogenic emissions of isoprene and monoterpene (α-pinene) in EMEP4UK are calculated interactively using surface temperature and insolation (Guenther et al., 1995; Simpson et al., 2012) and therefore respond to the changes in climate as simulated in these nudged WRF simulations. CO2 inhibition of isoprene emission (Arneth et al., 2008) is not included in this EMEP4UK model version. Emissions of NOx from soils, which are temperature dependent, and of wind-driven sea salt are also calculated interactively (Simpson et al., 2012). Other natural emissions (lightning, DMS, volcanic) are fixed and hence invariant between the present-day and future coupled model simulations. Lightning NOx emissions are prescribed here. Interactive schemes used in other studies have shown strong responses to climate change globally, but relatively small responses over Europe (Finney et al., 2018), hence the impact of this simplification is likely small. The impacts of climate change on DMS are uncertain (Thornhill et al., 2021; Zhao et al., 2024; Joge et al., 2025). The import of Saharan dust is treated using a monthly climatology of fine and course dust concentrations. Since the response of dust to climate change is uncertain (Sect. 1), Saharan dust is also treated as invariant between present-day and future.

Longer lived gaseous species are provided as boundary and initial conditions for 2012 and these are used for all EMEP4UK simulation years. Atmospheric CO2 concentrations, which influence sulphate production and dry deposition of sulphur dioxide in this version of EMEP are fixed at 392 ppmv. Mixing ratios of methane are specified across the whole model domain at 1780 ppb for every year. O3 boundary conditions at the edge of the European domain are based on climatological ozone-sonde data, modified monthly against climatological clean-air surface observations (Simpson et al., 2012). Hence, the same present-day O3 boundary conditions for the outer European domain are used for each year. Boundary conditions for gas-phase (CO, PAN, NOx, SOx, HNO3, H2O2, VOCs) and inorganic aerosol species are also prescribed climatologies based on measurements (Simpson et al., 2012). This excludes the influence of climate-driven concentration changes outside the European domain on European air quality but allows us to isolate the signature of climate change over Europe alone.

For O3 and NO2 model evaluation (performed using the R openair package), hourly observational data over Europe for 2012 from “rural” sites were obtained from EMEP (https://www.eea.europa.eu/data-and-maps/data/airbase-the-european-air-quality-database-8, last access: 1 April 2026), with 365 and 265 sites respectively meeting the 75 % hourly data capture criteria. Slightly fewer sites met additional data capture criteria within the peak season required for calculating MDA8 O3 (332). Observations from urban and suburban background sites were also included in the comparisons for PM_2.5, giving a total of 170 sites, due to the small number of rural PM_2.5 monitoring sites available in 2012 (38). Model-observation comparisons are performed by extracting values from the closest EMEP4UK grid cell to each monitoring location.

The Atmospheric Dispersion Modelling System (ADMS) Urban model version 3.4.6 is used in this study. ADMS-Urban is a quasi-Gaussian model that simulates the dispersion of emissions based on meteorological stability, which is influenced by urban land use and building morphology (Carruthers et al., 1994; Stocker et al., 2012; Hood et al., 2018). The model uses meteorological profiles of wind speed and direction, among other parameters, to define atmospheric conditions. Emissions from industrial, domestic and road traffic sources are included, either explicitly with detailed time-varying profiles e.g., for major road and industrial sources, or as 1 km grid-averaged emissions. A street canyon module modifies the dispersion of emissions from all roads in the modelling domain with adjacent buildings, while an urban canopy module calculates modified wind speed and turbulence flow profiles to represent larger-scale urban conditions. ADMS-Urban uses a semi-empirical NOx photolytic chemistry module (Venkatram et al., 1994), which accounts for fast, near-road oxidation of NO by O3 to form NO2 (Smith et al., 2017) and a simplified sulphate chemistry scheme for conversion of SO2 to PM_2.5. The NOX chemistry scheme performance has been compared with the detailed Master Chemical Mechanism over London by Hood et al. (2018). This highlighted reasonable agreement (within 20 %–40 %) except during summer air pollution episodes (Malkin et al., 2016). Further details of the ADMS-Urban model set-up can be found in Hood et al. (2018).

Emissions for all sources other than road traffic are from the London Atmospheric Emissions Inventory 2010 (GLA, 2013), projected from the LAEI base year 2010 to the modelled year 2012. Time-varying profiles are applied. Road traffic emissions were calculated using activity data from the LAEI with adjustments to NOX, NO2 and PM_2.5 emissions factors to improve consistency with real-world emissions measurements as described in Hood et al. (2018). The ADMS-Urban outputs for this study are for specified receptor locations corresponding to 56 reference air quality monitoring sites from the Automatic Urban and Rural Network (AURN) and from the London Air Quality Network (LAQN) (Hood et al., 2018).

The coupled WRF-EMEP4UK-ADMS-Urban regional to urban model system is used to simulate continental and urban street-scale air pollution and relevant metrics. Consistent anthropogenic emissions for the year 2012 are used in WRF-EMEP4UK and ADMS-Urban. Hourly meteorological and chemical boundary concentrations from WRF-EMEP4UK grid-cells are used as input to ADMS-Urban through one-way coupling. This ensures that the long-range transport and chemical environment is adequately represented in terms of physical and chemical processes at all relevant time and spatial scales, from regional to street scale. Model evaluation (performed using the R openair package), used hourly observational data for London for the years 1996–2005 and 2012 from the LAQN for “background” and “near-road” sites that met the requirement of at least 70 % data capture of hourly data during the relevant year. For 2012, 20, 11 and 42 sites for O3, PM_2.5 and NO2 met this requirement (see Hood et al., 2018). For the 1996 to 2005 period, considerably fewer data were available: for O3 (NO2) the number of sites increased from 7 (9) in 1996 to 16 (40) in 2005. For PM_2.5 for this period only “roadside” data were available (1 site from 1998 and 2 sites from 2004), hence the measurements do not reflect ambient conditions well; also, observations exhibit considerable year-to-year variation (between 28.9 and 56.0 µgm-3), which may reflect equipment error. Model-observation comparisons are made at directly corresponding locations due to the high spatial resolution of ADMS-Urban model output. The coupled system and the standard ADMS-Urban model configuration for 2012 is extensively evaluated in Hood et al. (2018).

Present-day and future experiments were performed using the regional and urban coupled models to quantify changes in atmospheric composition due to climate change under RCP 8.5 at the regional scale over Europe and the UK and at urban scale over London. Air quality metrics were calculated to evaluate the implications for attaining World Health Organisation (WHO) air quality guidelines and interim targets (WHO, 2021).

Coupled model simulations were performed for two 10-year time periods: present-day (1996–2005) and future (2090–2099) following RCP8.5 using HadGEM2-ES climate outputs based on historical and RCP8.5 (Lamarque et al., 2010; Meinshausen et al., 2011) future simulations from CMIP5. The difference between these two periods results in a large global mean near-surface temperature change of 4.7 °C (Met Office Hadley Centre, 2012). Anthropogenic emissions representative of the year 2012 (Sect. 2.3) are used for all 10-year present-day and future simulations in order to isolate the impacts of climate change alone on atmospheric composition. The model experiment set-up, shown in Table 1, enables quantification of the coupled EMEP4UK and ADMS-Urban responses to climate change due to the combined effects of changes in atmospheric chemistry and physics processes, climate-sensitive emissions, deposition and transport (Sect. 1).

Annual mean near-surface temperatures increase by more than 4 °C across Europe and up to 8 °C in northern Scandinavia and Alpine regions under RCP 8.5 in the 2090s (2090–2099) compared to 2000s (1996–2005; Fig. 1d). HadGEM2-ES suggests little change in temperature over the North Atlantic, a common feature of other CMIP models, and this has been attributed to a reduction in the meridional overturning in this Atlantic Ocean region (e.g., Park and Yeh, 2024).

Present day annual-mean surface O3 mixing ratios averaged over the period 1996–2005 (that use anthropogenic emissions for 2012) show a North–South gradient, with values greater than 30 ppbv in southern parts of Europe (higher over the Mediterranean and the Alps) and lower values ∼20 ppbv in northern Europe (Fig. 1b). These annual average modelled O3 concentrations are compared to observations for rural sites for the year 2012, assuming that the dominant influence on air pollutant concentrations arises from underlying anthropogenic emissions. The observed annual-average O3 values are generally well captured by the model although slightly underestimated, with small magnitudes of mean bias (-1.8 ppbv), normalised mean bias (-6 %) and RMSE=6 (Fig. 1). Some of the largest underestimates for individual sites occur when monitors are located at high elevation (>1500 m; Fig. A1) e.g., in the Alps/Balkans. The spatial correlation coefficient is moderate r=0.56 (Fig. 1b); this may partly reflect differences in meteorological impacts on concentrations that arise through a comparison of observations from the year 2012 against the 1996–2005 modelled period average. The annual mean O3 response to climate change shows a strong regional contrast, as noted in previous studies for summer O3 (e.g. Colette et al., 2015; Schnell et al., 2016; Sect. 1), with statistically significant increases of up to 20 % over southern and central Europe (centred on the Alps) and decreases of up to 15 % over northern Europe (Fig. 1e).

Present-day annual-mean PM_2.5 concentrations are typically between 10–20 µgm-3 but show hotspot locations e.g., northern Italy ∼50 µgm-3 (Fig. 1c). Annual average PM_2.5 concentrations are overestimated by the model in 2012, with a mean bias of 3 µgm-3 (NMB=21 %) and an RMSE=7, and a moderate spatial correlation (r=0.63). Annual mean PM_2.5 concentrations are elevated over most of Europe (up to 30 %; typically between 5–10 µgm-3) under RCP8.5 (Fig. 1f); these values are similar in sign but higher than reported from global models following RCP8.5 and SSP3-7.0 pathways in previous studies (Silva et al., 2017; Turnock et al., 2022; Gomez et al., 2023).

To understand the drivers of these changes, winter and summer mean changes in O3, PM_2.5 and key precursor species are shown in Fig. 2. The response of surface O3 to climate change exhibits a strong contrast between winter and summer. In winter, surface O3 decreases significantly over a substantial part of continental Europe (2–8 ppbv; Fig. 2a), whilst in summer substantive O3 increases are evident across nearly all of continental Europe (5–10 ppbv; Fig. 2f); with small decreases (that are statistically significant at the 95 % confidence interval) over Nordic regions and the UK, consistent with the findings of Colette et al. (2015). In this warmer climate, increased water vapor concentrations reduce background O3 levels, explaining reductions across the Atlantic in both seasons, and may largely drive the O3 decreases across northern Europe in winter. The response of surface PM_2.5 to climate change is far more muted in winter compared to summer which shows significant increases of up to 15 µgm-3 in parts of Southern Europe, the Mediterranean and Northern Africa (Fig. 2b and g).

Whilst anthropogenic emissions remain unaltered, biogenic isoprene emissions respond to elevated temperatures increasing from 10.4 to 22.5 mgm-2 (116 %) associated with a 4.8 °C increase over the European domain (35–70° N and 20° W–40° E), corresponding to 24 % increase per 1 °C (Table A1). This doubling of annual-mean isoprene emissions (Fig. A1c), results in isoprene mixing ratios significantly elevated by up to 4 ppb in summer in parts of southern Europe, with much smaller changes in winter (Fig. 2c and h). These elevated isoprene levels are the main driver of higher O3 and PM_2.5 in summer across continental Europe. Andersson and Engardt (2010) found increases in isoprene emissions of 83 % over Europe in the 21st Century under the SRES A2 scenario, smaller than the changes reported here, but for a smaller temperature increase. Langner et al. (2012) showed that under the SRESA1B scenario isoprene emissions over Europe increased by 21 %–26 % in four out of five regional model simulations over the first four decades of the 21st Century associated with a 1.27 °C temperature increase. The sensitivity of the isoprene response to climate change over Europe was comprehensively examined using the MEGAN-MOHYCAN model by Bauwens et al. (2018). Under RCP8.5 they suggest isoprene emission increases of 83 % over the 21st Century for an average increase in temperature of 4 °C (21 % per 1 °C). Larger increases were found when CO2 fertilisation effects were included, but there were substantial decreases when considering CO2 inhibition. Overall, the changes in isoprene emissions reported in our study are large but consistent with other studies that do not include CO2 inhibition or fertilisation effects. Large uncertainties remain in the interplay between these complex effects on isoprene emissions (Do et al., 2025). Consistent with changes in isoprene, α-Pinene concentrations also strongly increase with temperature in both winter and summer throughout Europe (except in northern Scandinavia)(not shown).

Dry deposition of O3 is also altered significantly by climate change. Across almost all of Europe, future wintertime deposition velocities that are up to 0.1 cms-1 larger contribute to lower surface O3 mixing ratios in winter, whilst deposition velocities up to 0.1 cms-1 (∼15 %) smaller support higher surface O3 in summer (Fig. 3a and d). Climate change simulations under the SRES A2 scenario found reductions in O3 deposition velocities in summer of up to 40 % over southern Europe between 1961–1990 and 2071–2100, leading to increases in summer mean O3 of up to 6 ppbv (Andersson and Engardt, 2010). Vieno et al. (2010) also noted severely restricted O3 dry deposition during the 2003 heatwave in the UK.

Changes in oxidants were also examined. Hydroxyl radical (OH) concentrations generally increase slightly in winter but decrease more prominently in summer over most of Europe and the Mediterranean by up to 0.75×106 moleculescm-3, with small OH increases for the UK and Benelux regions (Fig. 2e and j). These changes are statistically significant at the 95 % confidence interval. Most oceanic regions show summer increases reflecting greater OH production associated with higher humidity and O3 destruction. HO2 and RO2 radical species also change slightly in winter, but exhibit an opposing response to that of OH, that is similar or greater in magnitude. There are many formation and loss processes affecting oxidant levels that may be influenced by climate change. In summer, the primary influence on OH changes across continental Europe is likely to be higher abundances of biogenic VOCs, whose oxidation acts as a OH sink that enhances HO2 and RO2 levels. In winter, the small uniform OH increases may be a consequence of greater OH formation as a result of higher O3 destruction due to humidities.

Annual-mean soil NO emissions increase under this large climate signal (Fig. A1f) across almost all of Europe yielding a 64 % increase in the future compared to present-day or a 13 % increase per 1 °C (Table A1), potentially influencing land summer NOx concentrations (Fig. 2i). Previous experimental studies have suggested a 100 % (or doubling) of NO emissions for each 10 °C rise, although sub-ranges of temperatures showed differing levels of linearity (Laville et al., 2009); with the parametrisation used in this study based on such results (Simpson et al., 2012). Few studies have reported soil NO emissions changes over Europe. A 9 % increase in NO emissions averaged over Europe was simulated under a regional warming of 1.8 °C on average across Europe in the 2030s compared to the 1990s (a 5.2 % increase per 1 °C) by Kesik et al. (2006). The lower sensitivity to climate change compared to this study, is likely due to their inclusion of soil moisture effects on NO emissions. A recent experimental field study found that dryer soils in a warmer climate could reduce NO emissions (Huang et al., 2025), suggesting there is also uncertainty in the impact of this climate-sensitive emission process. Unlike isoprene, this natural soil emission source is minor compared to current anthropogenic NOx emissions as noted by Simpson et al. (2012). In response to climate change, surface NOx mixing ratios show distinct patterns of change over Europe in winter that are most prominent over major source regions. NOx increases strongly over the UK, Benelux region and northern France (∼4 ppbv) but generally decrease elsewhere, most notably over northern Italy (4 ppbv; Fig. 2d and i), although these changes are not found to be statistically significant. In contrast, in summer there are small but significant NOx increases over Europe, with large increases apparent in a few emission regions, and notably over the Mediterranean coincident with OH reductions. The most prominent changes in NOx mixing ratios over land are more localised than the O3 and PM_2.5 responses reflecting the shorter atmospheric lifetime of NOx.

In winter, the areas of largest change are coincident with strong NOx emission source locations. Winter NOx increases in northern European source regions are consistent with the findings of previous studies that highlight the reduced role of peroxyacetylnitrate (PAN) with warmer temperatures. Using a global model, Doherty et al. (2013) found a widespread increase in annual mean surface O3 of up to 1 ppbv over Europe under a global warming signal of 3 °C, as a result of higher NOx concentrations over source regions due to greater PAN decomposition. However, winter NOx decreases in southern Europe suggests photochemistry or mixing effects may be more important. In summer, reduced formation of PAN and OH decreases may increase NOx lifetimes and promote higher NOx over sources regions in Europe and the Mediterranean.

Higher latitude increases in surface NOx suggest titration of O3 by NO in winter, that leads to reduced O3 mixing ratios, as photochemistry is less active to replenish O3 in northern Europe. (Fig. 2a and e). In southern Europe, reduced surface NOx in winter is coincident with lower O3 mixing ratios. To investigate the O3 chemical environment the changes in NOx/VOC concentration ratios are depicted in Fig. A2. There is a strong contrast between northern and southern Europe in winter with the highest ratios, indicating most VOC-limited conditions, over Benelux and the UK. In summer marine regions with heavy shipping in the North Sea, English Channel/North Atlantic and the northern Mediterranean display the highest NOx/VOC ratios. NOx/VOC ratios increase in winter in northern Europe due to higher NOx suggesting climate change would lead a more VOC-limited regime in future here, and decrease elsewhere suggesting more widespread NOx-limited regimes (Fig. A2c). Conversely, in summer NOx/VOC ratios decrease over continental Europe because of higher biogenic VOC levels due to climate change, implying a more NOx-limited regime, but increase over the Mediterranean due to higher NOx (Fig. A2f). The different seasonal and latitudinal responses in the NOx/VOC ratio highlight the challenge for designing future O3 mitigation strategies across Europe when considering the impacts of climate change. Overall, it is clear that climate-driven changes in isoprene emissions, O3 dry deposition and chemistry have impacts on oxidant levels and surface O3 right across Europe, as reported in previous studies.

Climate change may also influence local mixing and transport patterns. Mean mixing heights over the 10-year present-day and future periods show distinct patterns of change over Europe. In winter, mixing heights decrease significantly over western Europe and ocean regions and increase over central, eastern and northern Europe by up to ∼100 m (Fig. 3b). In summer a strong land-ocean contrast is evident, with significant increases over continental Europe (up to 150 m) and decreases over oceans (Fig. 3e). In winter, over central Europe, the spatial pattern of surface O3 and NOx decreases resembles that of mixing height increases, suggesting that changes in mixing height may influence their responses to climate change. In summer, O3 and PM_2.5 responses do not appear to be impacted by changes in boundary layer mixing height. Precipitation in winter exhibits a similar east-west contrast, with increases in central and northern Europe and decreases in western Europe that are statistically significant (∼10 mm and up to 30 mm; Fig. 3c). Summer precipitation changes are mixed across Europe with increases in eastern areas and larger reductions over mountainous regions (Alps/Pyrenees) (Fig. 3f). These precipitation changes may also influence PM_2.5 concentrations e.g., increases over the Alps in summer (Fig. 2g). Several previous global model studies have suggested that reduced large-scale precipitation over northern hemisphere midlatitudes land regions under climate change especially in summer may lead to increases in surface PM_2.5 due to less wet deposition (Allen et al., 2016; Banks et al., 2022).

In the absence of anthropogenic emission changes, climate change impacts PM_2.5 through changes in natural aerosols and/or aerosol precursor emissions, oxidising capacity and secondary aerosol formation pathways, and/or aerosol sink processes. In this section we assess how the composition of annual and seasonal mean PM_2.5 may change in the future, and the key driving processes. The present-day (1996–2005) spatial distribution of annual-mean PM_2.5 components in Fig. 4 highlights that the inorganic contribution to PM_2.5 is dominated by sulphate from energy generation in south east Europe and shipping in the Mediterranean, as well as nitrate (and to a lesser extent ammonium) in the Alps/Po valley region. Primary black and organic carbon emission hotspots are responsible for a strong urban/anthropogenic fingerprint and a large natural Saharan dust component is also evident in overall PM_2.5 levels, particularly over south-west Europe. A substantial widespread secondary organic source of PM_2.5 is evident almost everywhere. In winter, the abundances of sulphate, nitrate and primary PM_2.5 are larger over land, whilst in summer sulphate is higher over the Mediterranean and SOA levels are substantially higher (exceeding 10 µgm-3) over all of continental Europe and the Mediterranean (not shown).

The seasonal changes in these PM_2.5 components in the 2090s compared with 2000s are shown in Fig. 5. Numerous climate sensitive processes influence the concentrations of inorganic aerosol in the atmosphere. Of the inorganic components of PM_2.5, sulphate displays the largest response to climate change under RCP8.5. Sulphate aerosols exhibit a large and statistically significant wintertime decrease (>2 µgm-3) over central Europe and an increase over ocean regions (Fig. 5a). A smaller significant summertime decrease in sulphate occurs over the Balkan regions with more widespread increases (notably in the Mediterranean) or no change elsewhere (Fig. 5h).

A key driver of winter changes in sulphate levels in the 2090s is changes in its deposition. Both wet deposition of sulphur dioxide (SO2) and dry deposition of SOx (SO2 + SO4) and increase prominently in winter over central Europe (Fig. A3a and b), whilst SO2 concentrations decrease significantly, and hence there is less sulphate formation. Greater washout of SO2 seems likely associated with higher precipitation over this region (Fig. 3c). In addition, the winter increase in mixing height may reduce sulphate aerosol levels at the surface (cf. similar spatial features Figs. 3b and 5a). Similar wet deposition (Racherla and Adams, 2006) and precipitation responses over continental land regions have been reported in previous studies (e.g., Allen et al., 2016). Other studies have noted winter sulphate decreases in Northern Hemisphere industrialised regions associated with oxidant limitation (Berglen et al., 2004; Shindell et al., 2009). As noted in Sect. 3.1, OH increases slightly in winter in the future (Fig. 2e), suggesting oxidant limitation does not worsen in the future. However, O3 is also important for in-cloud oxidation of SO2 to SO4 and lower O3 concentrations in winter suggest this could be important in limiting SO4 formation. In summer, the decreases in wet and dry deposition of SO2 and SOx respectively (Fig. A3e and f) over the Balkans coincide with lower sulphate concentrations (Fig. 5h), suggesting that deposition is not the main driver of this sulphate response. Under a warmer climate, an increase in sulphate loading could be expected as the oxidation of SO2 to sulphate is faster at higher temperatures. However, as both SO2 and OH (Fig. 2j) decrease in summer in this region, a reduced level of reactants may explain the sulphate decreases. Over the Mediterranean region, a combination of processes, such as faster oxidation with higher OH in eastern seas, seems likely to cause elevated sulphate aerosol concentrations in both seasons.

The response of nitrate aerosol to climate change in winter is mixed across Europe but with more widespread decreases (0–1 µgm-3) than increases; in summer, nitrate decreases in some hotspot emissions source locations (Fig. 5b and i). In the future, dry deposition of oxidised nitrogen increases over the European continent in winter; with a more mixed response in summer (Fig. A3c and g). Wet deposition of nitric acid (HNO3) is similarly influenced by climate-driven changes in precipitation in winter but only increases slightly over central and northern Europe and decreases elsewhere (Fig. A3d). Hence winter nitrate decreases across Europe seem influenced mainly by dry deposition increases than wet deposition responses. The response of this semi-volatile species to climate change may also be driven by enhanced partitioning into the gas phase with higher temperatures in both winter and summer reducing nitrate aerosol concentrations, as noted in other studies (Dawson et al., 2007; Pye et al., 2009). Ammonium concentrations also decrease (up to 1 µg m⁻³) mostly strongly in winter over the European continent (Fig. 5j) reflecting concomitant reductions in sulphate and nitrate aerosol loadings.

SOA shows by far the largest and most widespread summertime increase across Europe of all the PM_2.5 components (>5 µgm-3) in the 2090s driven by the response of its natural isoprene and monoterpene precursor sources to climate change (Fig. 5k). Increases are also found in winter across continental Europe that are largest over western Europe (up to 2 µgm-3). Primary PM_2.5 sources are also influenced by climate change. There are notable wintertime future increases in primary fine organic matter and fine elemental carbon concentrations in source regions which can be related to reductions in precipitation over certain regions (cf. Figs. 3c and 5e), but little change in summer.

Sea salt and dust, that contribute a small fraction of their mass to PM_2.5, are sensitive to changes in wind speed and transport patterns. Sea salt aerosol loadings increase slightly in the future (up to 0.5 µgm-3 across Europe) whilst desert dust from the Sahara strongly decreases (by more than 2 µgm-3) in both seasons in the future (Fig. 5f, g and m, n). Both components primarily affect oceanic and maritime regions. These modified distributions in sea-salt and dust aerosols can be explained by changes in wind speed under RCP8.5 in the 2090s. Wind speeds generally increase by up to 1 ms-1 in the Atlantic in both seasons and by >2 ms-1 polewards of 60° N in winter, aiding sea-salt aerosol formation; whilst wind speeds are reduced over the Mediterranean in both seasons, hindering Saharan dust transport (by up to 1 ms-1; Fig. A4c and f). Similar responses were seen in a study by Turnock et al. (2022) under SSP3-7.0, which found enhanced sea salt aerosols in maritime parts of northern Europe and a reduction in fine dust aerosol across North Africa.

Primary organic matter and elemental carbon replaces sulphate (and to a lesser extend nitrate) to become the dominant component of wintertime PM_2.5 in the 2090s over central northern Europe (Fig. 6a and d). Across the Iberian Peninsula SOA replaces dust as the dominant winter PM_2.5 component (Fig. 6a and d). In summer, across all of continental Europe, SOA remains the dominant PM_2.5 constituent in the future, whilst over certain oceanic locations (e.g., the Atlantic), the spatial dominance of Saharan dust decreases, and that of sea-salt increases in the future (Fig. 6b and e). Considering annual average PM_2.5, the most evident changes are an increase in SOA as the dominant PM_2.5 component driven by the reduction in sulphate over Central Europe, the reduction in nitrate and sulphate over the North Sea and the reduction in Saharan dust over the Iberian Peninsula (Fig. 6c and f).

The finer-scale 5km×5km simulations show significant surface O3 decreases across the UK of ∼2–4 ppbv in winter in the 2090s compared to the 2000s with the largest decreases over eastern parts of the UK (Fig. 7a). Surface O3 decreases of a similar magnitude occur in summer, except in the southern UK (including London) where there are O3 increases of up to 1 ppbv in future (Fig. 7d). These patterns of change are similar to those seen in the European 50km×50km resolution model domain in Fig. 2a and f; and largely due to background O3 reductions. The magnitudes of winter and summer mean O3 simulated over the UK at the finer resolution are about 2 ppbv (∼5 %) higher than at the coarser resolution (Table A2). However, the reductions between present-day and future climate (∼5 ppbv in DJF and 3 ppbv in JJA over the UK) at the two resolutions are only very slightly different (0.1/0.4 ppbv in DJF/JJA; Table A2). The magnitudes of the O3 decreases are consistent with estimates of annual-average reductions from global models by Zanis et al. (2022), as reported in RS (2021), although Colette et al. (2015) report smaller summer O3 decreases, likely due to the larger areal extent of their UK region that includes surrounding oceans.

Surface PM_2.5 concentrations simulated within the 5km×5km UK domain increase moderately in winter over much of the UK (∼1–2 µgm-3) and show smaller (up to 1 µgm-3) but mixed responses for summer with slight increases over southernmost UK in the future (Fig. 7b and e). These patterns of changes again resemble those simulated over the UK with the 50km×50km European domain (Fig. 2c and d). However, these changes are not statistically significant at the 95 % confidence level. In both seasons, the magnitudes of PM_2.5 simulated over the UK at the finer 5 km resolution are slightly lower (up to 2.1 µgm-3; ∼10 %) than at the coarser 50 km resolution, and the changes between present-day and future (∼0.8 µgm-3 in DJF and 0.1 µgm-3 in JJA over the UK) from the two resolutions for winter/summer differ marginally (∼0/0.3 µgm-3; Table A2).

Surface NO2 increases by ∼2 ppbv in winter over the southern UK and the English Channel, but changes elsewhere in the UK are small (Fig. 7c and f). Smaller and mixed NO2 responses are seen in summer. These differences are generally not statistically significant. Like the other two air pollutants, the patterns of NO2 changes are very similar to those simulated at the coarser resolution, with the magnitudes slightly lower (∼1 ppbv; ∼5 %; Table A2) as found for PM_2.5 changes at the coarser resolution.

In summary, differences in horizontal resolutions employed by EMEP4UK model do not seem to influence the patterns, and only minorly influence the magnitudes, of changes in simulated concentrations of surface O3, PM_2.5 and NO2 between present-day and future. To provide context to our results, the impact of model structural uncertainty on O3 and PM_2.5 climate change projections over the UK region is also assessed, using results from three to five global-scale models from Zanis et al. (2022) as presented in RS (2021). In this study, the impact of climate change was evaluated for a baseline SSP3-7.0 anthropogenic emissions scenario for simulations with SSTs for present-day compared to SSTs for SSP3-7.0 for the 2090s. The summer mean surface O3 response to climate change over the UK is -2.4 ± 1.1 ppbv across the five models (Table A2). These multiple global-scale earth system model results suggest structural uncertainty in projected surface O3 mixing ratios is greater than the uncertainty associated with spatial resolution identified using EMEP4UK e.g., that shows a summer O3 response of -2.8 ppbv at 50 km and -3.2 ppbv at 5 km resolution (Table A2). Annual-average PM_2.5 concentrations across the three models reduces slightly but the standard deviation is substantial (-0.14 ± 0.18 µgm-3, Table A2). The structural uncertainty in annual-mean PM_2.5 projections associated with the three global models used in Zanis et al. (2022) appears considerably larger than the uncertainty associated the spatial resolution of the EMEP4UK model over the UK (which exhibits an annual-mean PM_2.5 response to climate change of 1.4 µgm-3 at 50 km resolution and 1.1 µgm-3 at 50 km resolution). However, the underlying emissions used in Zanis et al. (2022) are very different to this study, and the UK regional extent is somewhat larger (and includes the surrounding ocean) than used in this study, hence only a qualitative comparison is possible.

To explore the changes in key pollutants at an urban scale, concentrations at 56 receptor sites representing UK reference air quality measurement network locations across London are considered. Statistics based on these 56 locations for the ADMS-Urban model, are compared these with those from 10 model grid cells that span these locations from the EMEP4UK model simulations at 5km×5km resolution. Surface O3, PM_2.5 and NO2 distributions across London from both models are shown in Fig. 8, and median values are summarised in Table A3.

The seasonal cycles of O3 across London, from both the regional and urban street-scale models, have very similar amplitudes of ∼20 ppbv; with the lowest median mixing ratios in winter of ∼10 ppbv and highest in spring/summer of ∼30 ppbv (Fig. 8a and b). Differences in magnitudes of present-day values between the two models are relatively small (less than 1.5 ppbv; Table A3) and differences in future changes are also small (0.1–0.9 ppbv). To assess model performance of simulated seasonal cycles relative to observations; monthly O3, PM_2.5 and NO2 distributions were calculated using data from the Hood et al. (2018) study, which employed the same set-up as used here, enabling a direct comparison for the year 2012. The amplitudes of the median values of surface O3 seasonal cycles for both models are also ∼20 ppb in 2012, which is an overestimate compared to observations of ∼16 ppbv, arsing from a smaller spring/summer peak (Fig. A5, Table A4). Differences in magnitudes between the two models are also small (1–3 ppbv) for 2012 (Table A4).

In the future, O3 mixing ratio decreases (Sect. 4.1) are largest in November and December, up to ∼3 ppbv for both models. Surface O3 over London has a springtime peak, and the amplitude of this decreases significantly in future by up to 5 ppbv in April in both regional and urban-scale model simulations. This is in agreement with global model results for northern Europe by Schnell et al. (2016), who also noted changes in climate-sensitive BVOC precursor emissions could also impact seasonal cycles. In all months except January, July and September (and August for ADMS-Urban) surface O3 distributions simulated by the two models differ significantly between present-day and future.

Both models simulate an autumn/winter peak in the seasonal cycles of PM_2.5 concentrations across London (median values ∼20 µgm-3 in November and ∼15 µgm-3 in December/January) which significantly increases in the future by as much as 5 µgm-3 in November (Fig. 8c and d; Table A3). For the year 2012, observations show the wintertime peak in PM_2.5 concentrations occurs later in February/March and this timing is captured by the two models, but the maxima are substantially underestimated (by 13 µgm-3; Fig. A5; Table A4), which Hood et al. (2018) attribute to underestimated regional contributions. There are fewer sites available for evaluation of PM_2.5 concentrations and over 50 % of these are near road sites which may be more challenging to simulate than background and rural locations. In spring and summer, median surface PM_2.5 concentrations (7–11 µgm-3) only marginally change in the future. Comparing the two models, differences in both present-day values (up to 2 µgm-3 between November–February) and present-day and future changes (less than 1.1 µgm-3) are small; in agreement with results for 2012. A previous street-scale modelling study for London using the ADMS-Urban model found different results because they neglected the processes affecting background O3 (Athanassiadou et al., 2010).

NO2 seasonal cycles also exhibit an autumn/winter peak (median value of ∼30 ppbv in November) for both models with median values between 15–26 ppbv in other seasons (Fig. 8e and f; Table A3). Unlike for O3 and PM_2.5, median monthly NO2 values are consistently 3–5 ppbv higher (∼15 %) for the ADMS-Urban compared to the EMEP4UK model. This difference is likely to be due to the inclusion of near-road locations in the ADMS-Urban model, where NO2 concentrations are strongly influenced by NOx emissions from the nearest road. For 2012, seasonal cycle peak and amplitude values for NO2 are similar to those for the present-day period; median values are also higher for ADMS-Urban and in good agreement with observations (Fig. A4; Table A4). In the future, the largest increases in NO2 mixing ratios, exceeding 4 ppbv for both models, occur in October, November and February and are statistically significant in both models.

Overall, the magnitudes of the median monthly concentrations, and changes between present-day and future, are in good agreement between both models for all three air pollutants, although NO2 values are higher. However, considerably larger spatial variability across the 56 locations is simulated by the urban street-scale (ADMS-Urban) model compared to the 10 fine-scale EMEP4UK regional model grid cells, and this is most prominent for NO2. This is also the case for the year 2012, where the spatial variability simulated using the ADMS-Urban model is similar to (for NO2) or smaller (for O3 and PM_2.5) than for the observations. There are also fewer months with significant differences between present-day and future distributions simulated for all three air pollutants simulated by the street-scale model as compared to the high-resolution regional model. The sensitivity of the street-scale model outputs to the number of sampling sites employed was examined by randomly sampling 10 of the 56 locations to produce distributions for London. Small changes between the results for 10 compared to 56 locations were found. This finding adds confidence to the conclusion that the street-scale simulation exhibits larger spatial variability in air pollutant concentrations, that agrees well with the observed spatial variability for 2012, due to its ability to explicitly represent road emissions sources as compared to the grid-box representation in the regional model that can lead to the dilution of emissions.

The diurnal variation in surface O3, PM_2.5 and NO2 concentrations simulated at 56 locations across London with the street-scale model is shown in Fig. 9. The diurnal cycle of surface O3 mixing ratios is much more pronounced in summer than in winter due to longer daylight hours and higher temperatures increasing its photochemical formation, with daytime median values exceeding 30 ppbv between 12:00 and 17:00 LT, and nighttime median values below 15 ppbv for present-day (Fig. 9b). In winter, the diurnal variation of surface O3 is much flatter than in summer, with a slight afternoon peak (median value ∼15 ppbv at 12:00–13:00 LT) and evening to nighttime median levels of 6–13 ppbv (Fig. 9a). Across the 56 locations, the largest variability occurs in summer during afternoon to early evening hours (up to 10 ppbv for the interquartile range). The close agreement between the simulated ADMS-Urban and observed diurnal cycles for 2012, that capture the key features described above, is evident in Fig. A6. The greater ability of the ADMS-Urban urban model to capture O3 diurnal cycles for London as compared to the regional EMEP4UK model has been highlighted by Hood et al. (2018). A statistically significant decrease in winter surface O3 in the future is apparent for early morning and afternoon hours. In summer, surface O3 is significantly higher by up to 3 ppbv in the morning but significantly lower by up to 4 ppbv in the afternoon and evening in the future. This leads to a small shift in the diurnal cycle of summertime surface O3, with the peak occurring about one hour earlier, although the amplitude of the diurnal cycle remains very similar.

For surface PM_2.5 concentrations the diurnal variation in both seasons is relatively small with median concentrations between 11–16 µgm-3 in winter and ∼8 µgm-3 in summer for present-day (Fig. 9c and d). The spatial variability across the 56 locations is largest for winter daytime hours (up to 4 µgm-3 for the interquartile range). The underestimate in simulated PM_2.5 concentrations compared to observations, as noted in Sect. 4.2, is apparent in these diurnal cycles (Fig. A6). Higher median wintertime levels (by up to 3 µgm-3) are evident at all times of day in the future, with statistically significant differences between present-day and future for most hours (Fig. 9c). Additionally in winter, PM_2.5 increases are considerably larger at night-time leading to reduced diurnal variability in the future, with similar spatial variability. Summertime median levels are slightly higher during daytime with significant increases between 10:00 and 18:00 LT of about 1 µgm-3 in the future.

The diurnal variation of NO2 mixing ratios is similar to that of PM_2.5 concentrations in winter, highlighting similar anthropogenic emissions sources, but is more pronounced with median values around 20 ppbv between 02:00–05:00 LT and increasing at 07:00 LT and from 15:00 LT, reaching 42 ppb at 18:00 LT for present day (Fig. 9e and f). Summer NO2 mixing ratios additionally exhibit an early afternoon dip, likely related to the surface O3 peak. In both seasons there is substantial spatial variability, notably for daytime hours, across the 56 locations (up to 20 ppbv for the interquartile range) driven by differences in dispersion reflecting the differing proximities of the sites to road NOx emissions sources (as noted in Hood et al., 2018). Observed and modelled NO2 diurnal cycles for 2012 agree well (Fig. A6). In the future, larger significant winter NO2 increases (median values up to 7.6 ppbv) in the early morning lead to a flatter distribution, although significantly higher NO2 also occurs in early afternoon (Fig. 9e and f). In the summer, significant higher mid-afternoon NO2 values also cause a flattened distribution in the future. The spatial variability for NO2 across the 56 locations is similar for the two time periods. Fewer hours show statistical differences between present-day and future for NO2, as compared to O3 and PM_2.5, again due to the dominance of unchanged road emissions in NO2 concentrations.

The broad changes in the diurnal cycles of surface O3, PM_2.5 and NO2 concentrations are consistent with the changes in the described in Sect. 4.2 for the seasonal cycles of these air pollutants. Here, the effect of night-time/early morning changes in winter for the three air pollutants, and changes either side of peak levels for summer O3 is additionally highlighted Distributions of winter O3, and PM_2.5 and NO2 in both seasons, flatten in the future.