2.1 Modeling framework

The assessment of the long-term evolution of air quality in the context of a changing climate is performed using a suite of deterministic models following the framework introduced by Jacob and Winner (2009). Global climate projections are obtained from a global circulation model (GCM) that feeds a global chemistry transport model and a regional climate model. Finally, the latter two drive a regional chemistry transport model. The setup used in this study is presented in detail in Colette et al. (2013, 2015).

The global circulation model is the IPSL-CM5A-MR large-scale atmosphere–ocean model (Dufresne et al., 2013). It provides input to the regional climate model and the global chemistry transport model with global meteorological fields. It uses LMDZ (Hourdin et al., 2006) as its meteorological model, ORCHIDEE (Krinner et al., 2005) as its land surface model, and NEMO (Madec and Delecluse, 1998) and LIM (Fichefet and Maqueda, 1999) as the respective oceanic and sea-ice models. The horizontal resolution of this global model is 2.5^∘ × 1.25^∘ with 39 vertical levels. For each scenario, the corresponding RCP is used for the anthropogenic radiative forcing. The Weather Research and Forecasting model (WRF, Wang et al., 2015) is used as the regional climate model (RCM). The regional climate simulations were part of EURO-CORDEX (Jacob et al., 2014) with a spatial resolution of 0.44^∘. The historic simulations were evaluated by comparison with experimental data (Menut et al., 2012; Kotlarski et al., 2014; Katragkou et al., 2015). The LMDZ-INCA (Hauglustaine et al., 2014) global CTM is used for the production of chemical initial/boundary conditions for the regional CTM. The LMDZ-INCA runs used in this study have been analyzed in Szopa et al. (2013) and Markakis et al. (2014), and intercomparisons of the same runs with other global chemistry transport models have been analyzed in Shindell et al. (2013) and Young et al. (2013) in the framework of the ACCMIP (Atmospheric Chemistry and Climate Model Intercomparison) experiment. Monthly climatological fields are used as the boundary condition inputs, as the background changes over long periods of time and it was not possible to store hourly global model output to create hourly varying boundary conditions. This induces an unavoidable inconsistency between meteorology and dust fields.

2.2 CHIMERE CTM

The CHIMERE offline regional CTM has been widely used for both future scenarios (Colette et al., 2015; Lacressonnière et al., 2016) and for research activities in France (Zhang et al., 2013; Petetin et al., 2014; Menut et al., 2015; Rea et al., 2015; Cholakian et al., 2018) and abroad (Hodzic and Jimenez, 2011). In this work, the 2013b version of the model was used for all simulations (Menut et al., 2013). The simulations were conducted using the EURO-CORDEX domain with a horizontal resolution of 0.44^∘ and nine vertical levels ranging from the surface to 500 mb. The aerosol module was run with a simple two-product scheme for the simulation of secondary organic aerosols (SOA, Bessagnet et al., 2008) and with the ISORROPIA module for the simulation of inorganic aerosols (Fountoukis and Nenes, 2007). It provides simulated aerosol fields including EC (elemental carbon), sulfate, nitrate, ammonium, SOA, dust, salt and PPM (primary particulate matter other than those mentioned above) considering coagulation, nucleation and condensation processes, as well as wet and dry deposition. The same unchanged land use data from GlobCover (Arino et al., 2008) with a base resolution of 300 m × 300 m have been used in different series of simulations. Dust emissions are taken into account inside the simulation domain based on the method proposed by Marticorena and Bergametti (1995).

The simulation domain has a 0.44^∘ resolution (Fig. 1). The analysis was performed using the subdomains presented in Fig. 1. The EUR subdomain only concerns the European continent (including the British Isles), and a land–sea mask was used to remove other parts of the domain. The MED subdomain was also produced using a land–sea mask, but this time it only contained the Mediterranean Sea. The MEDW and MEDE are the last two subdomains. They refer to the western and eastern Mediterranean areas, respectively. It is important to bear in mind that these two subdomains, contrary to the previous subdomains, contain both land and sea for the purpose of observing the effects of enclosing land on the Mediterranean area. Due to this setup, the sum of MEDW and MEDE is different to that of MED.

Figure 1Extension of the main domain and subdomains. Four subdomains are used in this study: EUR – containing only continental Europe (blue cells), MED – containing only the Mediterranean Sea (red cells), MEDW – western Mediterranean region (green rectangle), and MEDE – eastern Mediterranean region (yellow rectangle).

Download

2.3 Climate scenarios

Representative concentrations pathway scenarios (RCPs) designed for the fifth IPCC report (Meinshausen et al., 2011; van Vuuren et al., 2011b) are used in this study. Simulations using three of these CMIP5 RCPs (Taylor et al., 2012; Young et al., 2013) are selected: RCP2.6, RCP4.5 and RCP8.5, which consider 2.6, 4.5 and 8.5 W m⁻² of radiative forcing at the end of the 21st century, respectively. It is worth noting that RCP8.5 includes the least mitigation policies by far compared with the other two scenarios; therefore, RCP8.5 results in a high radiative forcing at the end of the century, with a temperature increase of between 2.6 and 4.8 ^∘C for Europe according to the EEA (European Environmental Agency¹). On the contrary, the RCP2.6 scenario considers a radiative forcing value that leads to a low-range temperature increase by 2100 (between 0.3 and 1.7 ^∘C). This means that this scenario has to consider ambitious greenhouse gas emission reductions as well as carbon capture and storage. The RCP4.5 is an intermediate scenario with less stringent climate mitigation policies, which results in a temperature increase somewhere between the two previously mentioned extreme scenarios.

2.4 Air pollutant emissions

The biogenic emissions input is obtained from the Model of Emissions of Gases and Aerosols from Nature (MEGAN v2.04; Guenther et al., 2006). It is worth mentioning that the emission factors and the leaf area index (LAI) values provided by this model are the same for all simulations; however, as many of the biogenic gases have a temperature-dependent nature, their emissions increase with higher temperatures. The MEGAN version used in CHIMERE takes six biogenic volatile organic compound (BVOC) emissions into account (isoprene, α-pinene, β-pinene, humulene, limonene and ocimene) and their dependence on temperature, solar radiation and the LAI.

Anthropogenic emissions are taken from the ECLIPSE-V4a global emissions projections (Amann et al., 2013; Klimont et al., 2013, 2017). This database covers the 2005–2050 time period with two main prospective pathways: the current legislation emissions (CLE) and the maximum feasible reduction (MFR) scenarios. These two scenarios show the effects of minimum and maximum mitigation efforts that can be expected by 2050, which gives us a spectrum of possible influences of anthropogenic emissions in future scenarios. For both scenarios, the atmospheric emissions of the main anthropogenic pollutants are available as global maps at a resolution of 0.5 ^∘.

Table 1The different scenarios used in this work.

Download Print Version | Download XLSX

The simulations used in this study are summarized in Table 1. As the goal of this study is to separately investigate the regional CTM drivers that can affect the results of future simulations, different series of simulations were performed, with each one allowing for the evaluation of climate, anthropogenic emission and boundary condition impacts on PM concentrations.

As for the climate impact study, RCP2.6, RCP4.5 and RCP8.5 scenarios are used in combination with constant anthropogenic emissions and identical boundary conditions. These scenarios are compared with the historic simulations.

In order to explore the changes induced by boundary conditions on the CTM outputs, RCP4.5 scenarios were conducted using two different sets of boundary conditions from the same global CTM and compared with historic simulations. The difference between these two sets of conditions is the fact that anthropogenic pollutants from the RCP emissions (emissions produced for RCP scenarios exclusively) are used as anthropogenic inputs for one of them (Szopa et al., 2006, 2013), whereas the other one relies on ECLIPSE-V4a emissions as inputs (Markakis et al., 2014).

The emission impact study uses the RCP4.5 climate forcing with CLE 2010, CLE 2050 and MFR 2050 anthropogenic emissions, with each one again being compared with historic simulations.

As seen in Table 1, we simulated different prospective periods in the different series of simulations. In the case of the climate impact studies, a 30-year-long historical simulation and 70-year-long future simulation were used for each future scenario. As for the other simulations, 10 years of historical simulations (1996 to 2005) were compared with 10 years of future scenarios, representing the 2050s (2045 to 2054). This latter period was chosen because it is centered on year 2050, for which CLE/MFR 2050 emissions are available. In the end, a 10-year-long simulation was performed during which all of these factors were simultaneously changed. We have to mention that it would have been numerically too strenuous to perform 70 years of simulations for emission/boundary condition drivers, bearing in mind that the climatic impact simulations already amount to 240 years of simulations.

A validation of our historic simulations is presented in Menut et al. (2012) and Kotlarski et al. (2014) for the meteorological parameters, whereas a validation of some chemical species is presented in the Supplement to this article (Sect. S1). This validation uses an annual profile of 10 years of historic simulations from 1996 to 2005 (with 2010 constant anthropogenic emissions) compared with an annual profile of all available measurements from EEA and air base stations between 2005 and 2015 (EEA, 2016).

3.1 Meteorological parameters

The RCP2.6 reaches its maximum radiative forcing (2.6 W m⁻²) in the 2040s (van Vuuren et al., 2011a), meaning that an increase in radiative forcing is seen in this scenario until this decade, whereas afterwards we observe a continuous decrease. Therefore, in the bulk of the RCP2.6 simulations over Europe, a temperature decrease appears from about 2050 onwards, followed by a new increase in temperature over the last 10 years simulated (Fig. 2). As for the RCP4.5 scenario, this maximum radiative forcing is reached in the 2070s (Thomson et al., 2011), whereas such a maximum is not reached for RCP8.5 until the end of the century. As a consequence, the temperature increase levels out after about 2070 in RCP4.5, while the temperature keeps increasing over the whole period in RCP8.5. These elements explain the larger similarities between RCP4.5 and RCP8.5 as well as their structural differences with RCP2.6. The time evolutions are similar for the European (EUR) and Mediterranean (MED) domains, albeit absolute temperatures are nearly 10 ^∘C warmer over the Mediterranean on average (Fig. S2 for 2-D temperature fields).

Figure 2Time series of temperature (K; a, b), precipitation (mm; c, d) and number of rainy hours (e, f) for EUR (a, c, e) and MED (b, d, f) subdomains for all climate change scenarios as well as historic simulations. The average for each scenario is shown by the corresponding point on right side of the plot. The solid lines show the rolling average of 30 years for future scenarios and 20 years for historic scenarios. Numbers in the legend show the p value of the linear regression for each scenario.

Download

In the following comparisons, annual average values over each subdomain and for each parameter and time series corresponding to 30 years of historic simulations and 70 years of future scenarios are presented. The uncertainties associated with each value account for the spatial variability of each parameter.

The 2 m temperatures are generally higher for the Mediterranean area than for the European continent; however, the increase in temperature is more pronounced for EUR than for MED (+1.69, +2.63 and +3.62 ^∘C for EUR and +1.06, +1.72 and +2.91 ^∘C for MED for RCP2.6, RCP4.5 and RCP8.5, respectively). In the other two Mediterranean subdomains temperature changes are higher for MEDE than for MEDW (+1.25, +2.05 and +3.38 ^∘C for MEDE and +0.73, +1.77 and +2.88 ^∘C for MEDW for RCP2.6, RCP4.5 and RCP8.5, respectively). While the European subdomain shows a more important temperature increase in winter (+1.86, +3.44 and +4.42 ^∘C for RCP2.6, RCP4.5 and RCP8.5, respectively), the Mediterranean subdomains show a larger increase in summer (for MED +1.40, +2.06 and +3.22 ^∘C for RCP2.6, RCP4.5 and RCP8.5, respectively). Maps of differences in temperature for the different scenarios and seasons are presented in the Supplement (Fig. S2). It should be noted that the average temperature changes discussed here agree with the literature (Knutti and Sedláček, 2012; Vautard et al., 2014).

As seen in Fig. 2, some of the parameters behave differently in the two subdomains. For example, for the amount of precipitation, an increase is simulated in EUR for RCP8.5, whereas a slight decrease is simulated for RCP2.6 after the 2050s, which makes precipitation stronger in RCP8.5 than in RCP2.6 on average for the future period. In the MED area, the opposite behavior is generally noted: precipitation is stronger in RCP2.6 than in RCP8.5. On the contrary, there is a steady decrease in the total duration of rain hours (sum of hours rained each year in each scenario – a threshold that is fixed for each subdomain using the average of 25th percentile for the whole duration of simulations). Therefore, rain events are expected to become more intense (Vautard et al., 2014). Regarding the total duration of rain hours in the Mediterranean region, the same result is obtained as for the EUR region, except for the RCP2.6 scenario where an increase in the number of rainy hours is simulated. This increase corresponds to the western basin of the Mediterranean Sea. For the eastern basin, all scenarios show a decrease in this parameter (Fig. S3 for MEDW and MEDE).

The same type of comparison is performed for wind speed (WS; 10 m wind), relative humidity (RH) and planetary boundary layer height (PBLH) in Fig. 3 for the EUR and MED areas. RH remains rather constant without a significant trend over EUR and MED, with the exception of the RCP2.6 scenario for MED which shows a significant decrease. WS mostly shows nonsignificant trends, with the exception of the RCP2.6 scenario over the MED domain. Similar results for WS changes have also been seen in other studies (Dobrynin et al., 2012; de Winter et al., 2013). Finally, PBLH increases are significant for RCP8.5 over EUR and for RCP2.6 over MED.

Figure 3Time series of wind speed (WS; m s⁻¹; a, b), relative humidity (RH; c, d) and planetary boundary layer height (PBLH; in meters; e, f) for EUR (a, c, d) and MED (b, d, f) subdomains for all climate change scenarios as well as historic simulations. The average for each scenario is shown by the corresponding point on right side of the plot. The solid lines show the rolling average of 30 years for future scenarios and 20 years for historic scenarios. Numbers in the legend show the p value of the linear regression for each scenario.

Download

These meteorological parameters undergo interactions between themselves and show correlations with each other, as they are driven by circulation patterns. The values of correlations between different meteorological parameters examined in this work are shown in Supplement (Sect. S4). The main points that should be taken into account regarding cross-correlations between meteorological parameters are the positive correlations between wind speed and the PBLH, between wind speed and precipitation and also the anti-correlation between wind speed and RH. All of these correlations are above 0.6 as an absolute value.

Also it should be noted that if the GCM used to provide the boundary conditions to the regional climate model was changed, the results seen here might be different (Olesen et al., 2007; Teichmann et al., 2013; Kerkhoff et al., 2015; Lacressonnière et al., 2016).

3.2 PM₁₀ concentrations

The simulated PM₁₀ concentrations are shown in Fig. 4 for the EUR and MED subdomains. The differences between future (2031–2100) and historical (1976–2005) simulations are presented, with historical simulations being used as a reference. Compared with historical simulations, we observe a decrease in PM₁₀ for all scenarios, for both EUR (0±0.95 %, $-\mathrm{2.57}\pm \mathrm{0.90}$ % and $-\mathrm{4.40}\pm \mathrm{0.87}$ % for RCP2.6, RCP4.5 and RCP8.5, respectively) and MED (0.9±0.09 %, $-\mathrm{5.65}\pm \mathrm{0.11}$ % and $-\mathrm{8.10}\pm \mathrm{0.12}$ % for RCP2.6, RCP4.5 and RCP8.5, respectively). The uncertainties shown here and in the rest of the document refer to spatial 1-sigma intervals. The reasons for these changes in PM₁₀ are discussed in the next subsection by analyzing individual PM components.

Figure 4PM₁₀ time series for the EUR and MED subdomains for all climate change scenarios and historic simulations. The average for each scenario is shown by the corresponding point on right side of the plot. The solid lines show the rolling average of 30 years for future scenarios and 20 years for historic scenarios. Numbers in the legend indicate the p value of the linear regression for each scenario.

Download

Alternatively, we also calculate linear trends for the future periods. A statistically significant positive trend is observed for RCP2.6 in the future, whereas it is significantly negative for RCP8.5, and nonsignificant for RCP4.5 (p values are given in Fig. 4 for linear trend lines).

Evidently, PM₁₀ has a seasonal variation which is shown in Fig. 5; box plots of PM₁₀ for all four seasons and all four subdomains are shown in Fig. 6 (Fig. S5 for PM_2.5). Interestingly, for EUR, the general PM₁₀ decrease noted above for the Hist, RCP2.6, RCP4.5 and RCP8.5 scenarios is reversed in the summer period. We see that elevated concentrations of PM₁₀ are simulated over the Mediterranean area for all seasons, reaching their maximum in spring (Fig. 6). Another interesting result in Fig. 5 is the increasing concentrations of PM₁₀ over eastern Europe in summer, and over the Scandinavian and eastern European regions both in summer and winter, in RCP4.5 and RCP8.5. This increase for the summer period is due to BVOC emission increases, which are discussed in Sect. 3.4.2. However, the same effect is not seen in RCP2.6 in either of these seasons, highlighting a structural difference between RCP2.6 and the other two scenarios. These results are discussed in the light of PM₁₀ components and meteorological parameter covariance in Sect. 3.4. A more general explanation for these structural differences can be found in the nature of the three scenarios, caused by the previously mentioned discriminated changes in meteorological parameters (Sect. 3.1).

Figure 5PM₁₀ seasonal average concentrations in the historical simulation (first column). Relative differences between the climate simulations (RCP2.6, RCP4.5 and RCP8.5) and the historical simulation (Hist) (second, third and fourth columns, respectively). Rows represents the different seasons (winter, spring, summer and fall, from top to the bottom). Please note that the scale used differs between seasons for the sake of readability.

Download

Figure 6Seasonal box plots for all scenarios for all seasons for PM₁₀ (historic simulations are black, RCP2.6 is green, RCP4.5 is blue and RCP8.5 is red) for the different subdomains. Scales are different for different subdomains.

Download

3.3 Distribution of chemical PM₁₀ components

Figure 7 shows the PM₁₀ concentrations and concentration changes for all of different scenarios and subdomains, as well as the contributions of all of the different PM₁₀ components (Fig. S6 for PM_2.5). For each PM component, the relative differences between future scenarios and historical simulations are also reported. Major differences can be seen in the distribution of the different PM components: for PM₁₀ the major contributors are salt and dust particles in our domains of interest, whereas their contribution to PM_2.5 is lower. Consequently, secondary inorganics (${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$, ${\mathrm{NO}}_{\mathrm{3}}^{-}$ and ${\mathrm{NH}}_{\mathrm{4}}^{+}$) and carbonaceous aerosol (black carbon, primary organic aerosol, biogenic secondary organic aerosol) are the major contributors to PM_2.5 (Fig. S6).

Figure 7Concentrations and relative changes of PM₁₀ components (same color code as in Fig. 6) for all subdomains averaged over the whole period of simulations in climate change scenarios and for the different subdomains. Error bars show the confidence interval calculated for the annual averages for each subdomain. The changes show (future − historic)∕historic ⋅ 100. Tables report the percentage of each component in each scenario. The concentration of PM₁₀ in total is noted in the legend above each figure.

Download

Another interesting point from this study is that primary aerosol species such as primary organic aerosols (POA), anthropogenic secondary organic aerosols (ASOA), PPM and black carbon (BC) only change slightly under a future climate (when emissions are kept constant), over both Europe and the Mediterranean Basin (a maximum of ±5 % change for most of them). It should be noted here that this article deals with annual averages and not extreme events; regional climate changes can have strong effects on primary pollutant peaks as is shown in Vautard et al. (2018). The evolution of these species is again discussed in Sect. 4.2 with respect to anthropogenic emission changes.

The most important changes in the future scenario outputs (with respect to historic simulations) are found to be a decrease in nitrates (for EUR $-\mathrm{11.2}\pm \mathrm{0.8}$ %, $-\mathrm{21.4}\pm \mathrm{0.7}$ % and $-\mathrm{28.8}\pm \mathrm{0.8}$ % for RCP2.6, RCP4.5 and RCP8.5, respectively) and an increase in the organic aerosol concentration (for EUR $+\mathrm{15.1}\pm \mathrm{1.2}$ %, $+\mathrm{22.7}\pm \mathrm{1.3}$ % and $+\mathrm{34.9}\pm \mathrm{1.3}$ % for RCP2.6, RCP4.5 and RCP8.5, respectively). A slight increase in sea-salt particles (for EUR $+\mathrm{0.8}\pm \mathrm{0.05}$ %, $+\mathrm{1.4}\pm \mathrm{0.06}$ % and $+\mathrm{0.2}\pm \mathrm{0.06}$ % for RCP2.6, RCP4.5 and RCP8.5, respectively) and in sulfates (for EUR $+\mathrm{4.50}\pm \mathrm{0.62}$ %, $+\mathrm{3.0}\pm \mathrm{0.6}$ % and $+\mathrm{1.6}\pm \mathrm{0.6}$ % for RCP2.6, RCP4.5 and RCP8.5, respectively) is also found. An interesting result regarding sulfates is that the concentration of this species displays an increase in all scenarios compared with historical simulations, although the scenario order for this increase is the inverse of that for the temperature increase (i.e., RCP2.6 > RCP4.5 > RCP8.5). This phenomena is discussed later in this section. While a decrease in dust particles is observed for RCP4.5 and RCP8.5 (for EUR $-\mathrm{5.6}\pm \mathrm{2.4}$ % and $-\mathrm{5.3}\pm \mathrm{2.2}$ %, respectively), these particles remain almost constant in RCP2.6. A decrease is seen for the ammonium particles in all scenarios (for EUR $-\mathrm{2.1}\pm \mathrm{0.2}$ %, $-\mathrm{6.4}\pm \mathrm{0.2}$ % and $-\mathrm{10.6}\pm \mathrm{0.2}$ % for RCP2.6, RCP4.5 and RCP8.5, respectively). However, the main driving force for the decrease in the concentration of total PM₁₀ (seen in Fig. 4) for the RCP4.5 and RCP8.5 scenarios is the decrease of nitrates. The increase in the concentration of other species, especially BSOA, is compensated for by the decrease in nitrate concentrations in the case of the two abovementioned scenarios with a seasonal dependence. As for RCP2.6, as the decrease in nitrates is generally lowest among all future scenarios, the slight increase in BSOA, sulfates, dust and salt particles drives the increase in the PM₁₀ concentrations (Fig. 7).

In the Mediterranean area, however, the predictions appears to be quite different: in general, the aerosol burden over the Mediterranean area (MED) is higher than that over the European area (average of 9.8, 9.5 and 9.4 µg m⁻³ for EUR compared with 17.4, 16.7 and 16.6 µg m⁻³ for MED for RCP2.6, RCP4.5 and RCP8.5, respectively, Fig. 7). This is mainly due to sea salt (approximately 9 µg m⁻³ in all scenarios) and dust (nearly 4 µg m⁻³ in all scenarios). Note that the MED region, by definition, only contains grid cells over sea, which explains the large sea-salt contribution. Conversely, for this area, the concentrations of aerosols that depend on continental emissions are considerably lower (−25 %, −33 % and −32 % for BC, POAs and ammonium, respectively, in MED compared with EUR for the RCP4.5 scenario for PM₁₀). Nitrates also show a significantly lower concentration in this region (from 1.03 µg m⁻³ for EUR to 0.21 µg m⁻³, 0.39 µg m⁻³ and 0.25 µg m⁻³ for RCP4.5 in PM₁₀ for MED, MEDW and MEDE, respectively). Sulfur emissions from maritime shipping lead to high sulfate concentrations over the Mediterranean area, especially over the eastern Mediterranean (2.5 µg m⁻³ in MEDE compared with 1.99 µg m⁻³ in EUR for the RCP4.5 scenario for PM₁₀). Finally, it is worth noting that the BSOA fraction is lower over the Mediterranean (−36 %, −31 % and −23 % of BSOA compared with EUR in MED, MEDW and MEDE, respectively). The relative changes in the Mediterranean domain compared with historic simulations are close to those for the European subdomain for most species, although they show different intensities for most components. An interesting behavior is seen for BSOA, where the increase in the concentration of this component becomes more homogenous between the three future climatic scenarios in the Mediterranean Basin with respect to continental Europe. The reason for this behavior is the origin of BSOA from advection over the Mediterranean subdomain. Also, sulfates, while showing the same general behavior as in the European subdomain, show lower changes between future and historic simulations over the Mediterranean Basin, resulting in a decrease in concentration in the RCP8.5 scenario.

3.4 Dependence of PM₁₀ components on meteorological parameters

In order to explain the evolution of PM components under future climate scenarios, they are correlated here with different meteorological parameters. The parameters tested here are temperature, wind speed, precipitation, RH, PBLH and shortwave radiation. Because of the variations in the seasonality of the different PM components, the analysis of their dependencies on meteorological parameters must be conducted for each season separately. Figure 8 shows the seasonal changes for nitrates, sulfates, BSOA and dust particles for all subdomains. The 2-D concentration fields for nitrates, BSOA, sulfates and dust particles are shown in Fig. 9 for the season when each component shows its highest concentration (Fig. 8). The correlations for all seasons and for the five selected meteorological parameters with PM₁₀ components are shown in Figs. 10 and 11 for the EUR and MED subdomains for different years. This comprises 30 pairs of values for the historic period and 70 for the future period. Here, a Pearson correlation coefficient of greater than 0.6 is considered to represent a significant relationship. Correlations between the different meteorological parameters are shown in Figs. S7 and S8. Figure S9 shows the 2-D correlations for the same species as Fig. 8, with the parameter that correlates best with them for selected seasons when their concentration is highest (i.e., nitrates, BSOA, sulfates and dust with temperature, temperature, RH and PBLH for winter, summer, summer and spring, respectively).

Figure 8Seasonal absolute concentrations of BSOA, nitrate (NO₃), sulfate (${\mathrm{SO}}_{\mathrm{4}}^{-\mathrm{2}}$) and dust particles (same color code as in Fig. 5) for the different subdomains. Each panel shows one of the species mentioned above for four scenarios and each subpanel shows one season.

Download

Figure 9Historic concentrations and relative changes in future scenarios for biogenic SOA (BSOA), nitrates (${\mathrm{NO}}_{\mathrm{3}}^{-}$), sulfate (${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$) and dust particles for selected seasons in which the concentrations were the highest. Figures in the left column show the average concentration of historic simulations, the other three columns show the relative difference of RCP2.6, RCP4.5 and RCP8.5 future scenarios, respectively, to the historic simulations. Each row shows one season, and the scales are different for each row of simulations.

Download

3.4.1 Inorganic PM components

Particulate nitrate concentrations appear to be strongly anti-correlated with temperature (Fig. 10). Hence, in most regions, they show a decrease in the future scenarios, mainly due to the higher temperature predicted in those scenarios, which might lead to a shift in the nitrate gas–aerosol partitioning towards the gaseous phase and more volatilization of already formed nitrate aerosols. Especially during the winter season, anti-correlations are seen with wind speed, precipitation, and PBLH, whereas a correlation is found with surface radiation. This fits with a switch from anticyclonic conditions – characterized by cold continental weather, clear skies and high solar radiation, large vertical stability and a low PBLH – to marine conditions. Continental conditions during this season indeed favor enhanced nitrate concentrations, whereas marine conditions are related to lower concentrations.

Figure 10Correlation coefficient between all meteorological parameters tested and BSOA, (NO₃), sulfate (${\mathrm{SO}}_{\mathrm{4}}^{-\mathrm{2}}$) and dust particles for all seasons and the EUR and MED subdomains. D, M, J and S represent winter, spring, summer and fall, respectively (first letter of the first month of each season). Uppercase letters mean that the correlation between the two parameters is statistically significant, whereas lowercase letters show the contrary. Color coding for different scenarios is the same as in previous figures.

Download

Figure 11Correlation coefficient between all meteorological parameters tested and ammonium (${\mathrm{NH}}_{\mathrm{4}}^{+}$), salt, PM₁₀ and PM_2.5 particles for all seasons and the EUR and MED subdomains. D, M, J and S represent winter, spring, summer and fall, respectively (first letter of the first month of each season). Uppercase letters mean that the correlation between the two parameters is statistically significant, whereas lowercase letters show the contrary. Color coding for different scenarios is the same as in previous figures. SWRD represents shortwave radiation.

Download

The correlation coefficient between nitrates and temperature is the lowest in summer. If we remain in our synoptic-scale framework, hot summer days favor pollution build up and accumulation, but decrease the partitioning in favor of the particle phase, so these effects compensate for each other. In spring, RH shows a high correlation with nitrate alongside temperature as higher RH favors nitrate partitioning into the particulate phase, in particular if it exceeds the deliquescence point of ammonium nitrate (RH > 50 %). These hypotheses are supported by the correlations presented in Sect. S4, which show the correlations between the different meteorological parameters. For the MED region, a strong anti-correlation is still observed, especially for winter and spring. However, the (anti)correlations with other meteorological parameters are less pronounced and not necessarily in the same direction as for EUR, as the distinction between continental and maritime conditions is not valid for this region. Overall, the analysis suggests that the major point to be taken into account for particle nitrates is their high anti-correlation with temperature (seen in Fig. 10). This confirms the results reported by Dawson et al. (2007), Jiménez-Guerrero et al. (2012) and Megaritis et al. (2014), who conducted sensitivity studies on individual meteorological parameters.

Sulfates are the second most abundant species in Europe after sea salt, and the third most important species after sea salt and dust over the Mediterranean. They show an increase in all of the future scenarios compared with historical simulations, but in the inverse order of the degree of severity projected for temperature increase, i.e., the increase is strongest for RCP2.6 (5 %) and less pronounced for RCP8.5 (1.2 %) as seen in Fig. 7. The spatial distribution of this species varies quite strongly between the RCP2.6 scenario and the RCP4.5 and RCP8.5 scenarios (Fig. 9). For instance, in RCP2.6, sulfate increases with respect to historic in the southwest part of the domain, whereas it decreases for the same area in RCP4.5 and RCP8.5. The 2-D correlation maps of sulfates with RH show a strong correlation between these two parameters, especially for the Mediterranean and the Atlantic, but also for the EUR subdomain in winter and spring periods (Fig. S9). The positive relationship between sulfates and RH could be related to two different processes. First, during the winter season, the major pathway of ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ formation from SO₂ proceeds via aqueous chemistry, and large-scale RH values are a tracer of sub-grid-scale cloud formation (Seinfeld and Pandis, 2016). On the contrary, during summer, over the MED region, gas-phase ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ formation via SO₂ oxidation by OH is dominant, and increased future RH levels, along with increasing temperatures, may lead to increased OH levels (Hedegaard et al., 2008). However for summer and fall, the PBLH shows a higher correlation with sulfate, which is shown in Fig. 10. Another parameter that shows a strong anti-correlation with sulfate concentrations is the wind speed in the spring and winter periods, which can be explained by the correlation between the PBLH and wind speed (Fig. 10).

Ammonium concentrations show a steady decrease in all future scenarios and in all subdomains. The correlations of ${\mathrm{NH}}_{\mathrm{4}}^{+}$ with meteorological variables appear to be a combination of those simulated for ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ and ${\mathrm{NO}}_{\mathrm{3}}^{-}$ with which ${\mathrm{NH}}_{\mathrm{4}}^{+}$ forms inorganic aerosol. When looking at correlations, a relationship is seen for ammonium with RH and wind speed, as well as a strong correlation with the PBLH (Fig. 11).

4.1 Boundary conditions

Among PM₁₀ components, dust, nitrate, BSOA, POA and sulfates show the highest impact on future PM concentrations when boundary conditions change, with other species showing only minor changes or no change at all (Fig. 12). Among these species, dust particles show the strongest dependence on boundary conditions: an increase of $+\mathrm{77}\pm \mathrm{2}$ %, $+\mathrm{30}\pm \mathrm{10.7}$ %, $+\mathrm{9}\pm \mathrm{1.9}$ % and $+\mathrm{51}\pm \mathrm{15.2}$ % for EUR, MED, MEDW and MEDE, respectively, for future RCP4.5 with respect to historic boundary conditions. Their simulated dependence on regional climate was indeed much smaller ($-\mathrm{9}\pm \mathrm{0.3}$ %, $-\mathrm{4}\pm \mathrm{1.4}$ %, $+\mathrm{3}\pm \mathrm{0.6}$ % and $-\mathrm{4}\pm \mathrm{1.8}$ % for EUR, MED, MEDW and MEDE, respectively, for the same period). It is important to bear in mind that the concentration of dusts in the European subdomain in our simulations is 0.8 µg m⁻³ on average with an important spatial variability, with concentrations dropping significantly as we move further north (Fig. 9). Therefore, the low relative changes simulated for the Mediterranean subdomains have to be considered as a sign of the high absolute sensitivity to the scenario. The reason for the important changes in these species may be due to wind intensity and humidity in source regions or land use changes caused by climate change outside our domain. There are many uncertainties regarding the future changes of dust concentrations (Tegen et al., 2004; Woodward et al., 2005). Changes in climate drivers such as precipitation, wind speed, regional moisture balance in source areas, and land use changes, either resulting from anthropogenic changes or for climatic reasons, can have important effects on dust emissions (Harrison et al., 2001). Projection of changes in land use resulting from both sources are highly uncertain, which results in strong uncertainties in the projections of dust concentration changes in future scenarios (Tegen et al., 2004; Evan et al., 2014). Furthermore, the Mediterranean Sea is located on the southern border of the domain used in this study; therefore, it should be noted that although the results of dust concentration changes seen in this study are consistent with existing literature, the model might not be capable of consistently capturing the relationship between boundary condition changes and the southern parts of the Mediterranean due to the position of the domain. This is not the case for the European subdomain.

Nitrates and BSOA are more affected by climate change impacts than by boundary condition input changes in all subdomains ($-\mathrm{13}\pm \mathrm{2.6}$ %, $+\mathrm{14}\pm \mathrm{0.5}$ %, $-\mathrm{2}\pm \mathrm{0.8}$ % and $+\mathrm{12}\pm \mathrm{0.7}$ % for RCP4.5 compared with $-\mathrm{6}\pm \mathrm{2.5}$ %, $-\mathrm{26}\pm \mathrm{0.5}$ %, $-\mathrm{23}\pm \mathrm{0.9}$ % and $-\mathrm{25}\pm \mathrm{0.7}$ % for boundary condition changes for nitrates and $+\mathrm{12}\pm \mathrm{1.3}$ %, $+\mathrm{23}\pm \mathrm{1}$ %, $+\mathrm{24}\pm \mathrm{1}$ % and $+\mathrm{12}\pm \mathrm{1.1}$ % for RCP4.5 compared with $+\mathrm{6}\pm \mathrm{1.3}$ %, $+\mathrm{14}\pm \mathrm{1.3}$ %, $+\mathrm{7}\pm \mathrm{1}$ % and $+\mathrm{14}\pm \mathrm{1.4}$ % for BSOA for EUR, MED, MEDW and MEDE, respectively).

Figure 13Emission scenario comparisons. Each panel shows one subdomain, the upper subpanels show the PM₁₀ components for each scenario, the lower subpanels shows the percentage difference for each scenario ((future − historic)∕historic ⋅ 100).

Download

Contrary to the species discussed above, sulfates and POA are more sensitive to boundary conditions than to climate effects. Sulfates show a $-\mathrm{2}\pm \mathrm{1.2}$ %, $-\mathrm{7}\pm \mathrm{1.6}$ %, $-\mathrm{6}\pm \mathrm{0.9}$ % and $-\mathrm{7}\pm \mathrm{1}$ % decrease, whereas POA shows a $-\mathrm{9}\pm \mathrm{0.7}$ %, $-\mathrm{26}\pm \mathrm{0.1}$ %, $-\mathrm{11}\pm \mathrm{0.1}$ % and $-\mathrm{19}\pm \mathrm{0.1}$ % decrease related to boundary conditions for EUR, MED, MEDW and MEDE, respectively. Climate effects on these particles were smaller.

Ammonium shows a negligible change with regards to boundary conditions in EUR, and an increase in most Mediterranean subdomains ($+\mathrm{12}\pm \mathrm{0.3}$ %, $-\mathrm{1}\pm \mathrm{0.1}$ % and $+\mathrm{12}\pm \mathrm{0.2}$ % for MED, MEDW and MEDE), whereas the change in this species was mostly negative for climate impact results.

4.2 Anthropogenic emissions

Changes in anthropogenic emissions in CLE and MFR 2050 inputs are compared with CLE 2010 emissions for different species in the Supplement (Fig. S11). As expected, a decrease is seen for most species in CLE 2050 emissions, but to a higher extent in the MFR 2050 scenario. A simple comparison between CLE and MFR 2050 emission scenarios is shown in Fig. 13. Bear in mind that this figure shows the effects of climate change (RCP4.5) and emission change at the same time, as does every value presented in this paragraph. For CLE simulations, sulfates show a decrease of $-\mathrm{28}\pm \mathrm{1.2}$ %, $-\mathrm{29}\pm \mathrm{1.3}$ %, $-\mathrm{34}\pm \mathrm{0.9}$ % and $-\mathrm{18}\pm \mathrm{0.9}$ % for EUR, MED, MEDW and MEDE, respectively, with respect to historic simulations, whereas MFR scenarios show a $-\mathrm{60}\pm \mathrm{1}$ %, $-\mathrm{51}\pm \mathrm{1.1}$ %, $-\mathrm{55}\pm \mathrm{0.8}$ % and $-\mathrm{52}\pm \mathrm{0.7}$ % decrease for the same order of subdomains. The reason for this decrease is the decrease in the emissions of SO₂ (SO₂ emissions reduction of −30 %, −53 %, −52 % and −42 % for CLE and −60 %, −53 %, −57 % and −68 % for MFR for the same order of subdomains). Particulate nitrates also show a strong decrease with the decrease of precursor emissions, presenting $-\mathrm{48}\pm \mathrm{2.5}$ %, $-\mathrm{32}\pm \mathrm{0.4}$ %, $-\mathrm{36}\pm \mathrm{0.8}$ % and $-\mathrm{28}\pm \mathrm{0.6}$ % for CLE and $-\mathrm{79}\pm \mathrm{2.2}$ %, $-\mathrm{61}\pm \mathrm{0.4}$ %, $-\mathrm{74}\pm \mathrm{0.8}$ % and $-\mathrm{77}\pm \mathrm{0.5}$ % for MFR for EUR, MED, MEDW and MEDE, respectively, (NO_x emission reduction of −60 %, −38 %, −48 % and −30 % for CLE and −84 %, −38 %, −76 % and −76 % for MFR for the same order of subdomains). Ammonium displays the same behavior, showing $-\mathrm{36}\pm \mathrm{1}$ %, $-\mathrm{55}\pm \mathrm{0.2}$ %, $-\mathrm{56}\pm \mathrm{0.1}$ % and $-\mathrm{18}\pm \mathrm{0.1}$ % for CLE and $-\mathrm{68}\pm \mathrm{0.9}$ %, $-\mathrm{87}\pm \mathrm{0.2}$ %, $-\mathrm{87}\pm \mathrm{0.1}$ % and $-\mathrm{64}\pm \mathrm{0.1}$ % for MFR for EUR, MED, MEDW and MEDE, respectively. Other components such as BC and POA show a strong decrease, as their concentrations depend directly on the amount of anthropogenic emissions. Interestingly, BSOA concentrations also show a strong decrease related to changes in anthropogenic emissions (contrary to the increase for climate change alone). Values of $-\mathrm{7}\pm \mathrm{1.3}$ %, $-\mathrm{34}\pm \mathrm{1.1}$ %, $-\mathrm{39}\pm \mathrm{1.1}$ % and $+\mathrm{4}\pm \mathrm{1.3}$ % for CLE and $-\mathrm{36}\pm \mathrm{1.1}$ %, $-\mathrm{64}\pm \mathrm{1}$ %, $-\mathrm{65}\pm \mathrm{0.9}$ % and $-\mathrm{35}\pm \mathrm{1.1}$ % for MFR for EUR, MED, MEDW and MEDE are seen, respectively, for this species. The fact that the decrease in anthropogenic emissions overshadows the increase in BSOA when looking at climate and boundary condition effects is an important message to take away from these simulations. The exact mechanism for this effect is not clear, although it could be due to the general decrease of seed aerosol in these scenarios modifying the gas–particle equilibrium for SVOCs formed from monoterpenes and isoprene oxidation. Changes in oxidant levels due to a decrease in anthropogenic emissions in addition to a direct decrease in anthropogenic VOCs may be other reasons for this change (i.e., less organic aerosol mass available for oxidation products to condense on).

For the comparison of driver impacts (Fig. 12), only CLE 2050 simulations for emission impact scenarios are used, as CLE 2050 emissions are used in the simulation where all drivers change (presented in Sect. 4.3). Almost all species show strong dependence on emission changes, except dust and salt particles. Quantitative effects vary for the Mediterranean subdomains: the effect of emission changes becomes less pronounced for species without maritime emission sources (such as ${\mathrm{NH}}_{\mathrm{4}}^{+}$), whereas they stay high for species like POA, BC and ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ which can be emitted by shipping lines.

4.3 Cumulative impacts

To provide a more complete view of probable atmospheric composition (under the hypotheses of the scenario) in the 2050s, the “All” scenario (Fig. 12) shows the combined effects of all drivers changing at the same time. There are many uncertainties affecting future scenarios as can be surmised; however, with regard to drivers that are explored here, this scenario shows what a more realistic future air composition might look like. As seen in Fig. 12, and with regard to total PM₁₀ and PM₁₀ components, the changes in emissions set the tone for the future, meaning that a reduction in anthropogenic emissions overshadows the climate and the boundary condition drivers for most of species. This highlights that mitigating air pollution with respect to air quality in the future depends greatly on the reduction of anthropogenic emissions.

For PM₁₀ for the period from 2046 to 2055 using RCP4.5, the different drivers indicate a decrease of −15.6 %, −6.7 %, −10.5 % and +4 % for EUR, MED, MEDW and MEDE, respectively, mainly driven by anthropogenic emissions. Due to the boundary condition changes, PM increases of +5.3 %, +6.8 %, +1.2 % and +15.1 % for EUR, MED, MEDW and MEDE, respectively, are observed, mainly because of the dust concentration increase. The climate impacts on PM₁₀ concentrations for the same period are −2.9 %, −0.5 %, +0.6 % and −1.6 % for EUR, MED, MEDW and MEDE, respectively. The total change for the period from 2046 to 2055 using RCP4.5 (meaning in simulations where all drivers change at the same time) is a decrease which is seen for all subdomains for PM₁₀ (−11.8 %, −1 % and −9.2 % for EUR, MED MEDW, and MEDE, respectively) except for MEDE where an increase of +9.1 % is observed. Thus, for most of the domains, the effect of emission reductions on PM₁₀ concentrations around 2050 is reduced to a certain extent by modifications of boundary conditions and regional climate.