The models used in this study are the Comprehensive Air quality Model with extensions, CAMx, version 5.40 (http://www.camx.com) and the Weather Research and Forecasting Model (WRF-ARW), version 3.2.1 (http://wrf-model.org/index.php). The coarse model domain covered all of Europe with a horizontal resolution of 0.250∘×0.125∘. A second, nested domain with 3 times higher resolution (0.083∘×0.0417∘) covered Switzerland. The meteorological fields were calculated for 2006 and used for all emission scenarios (see Table 1). We used 6 h ECMWF data (http://www.ecmwf.int/) to provide initial and boundary conditions for the WRF model. There were 31 terrain-following σ layers up to 100 hPa in WRF, of which 14 were used in CAMx. The lowest CAMx layer was 20 m above ground and the model top corresponded to about 7000 m a.s.l. The initial and boundary concentrations for the coarse domain were obtained from the MOZART global model data for 2006 (Horowitz et al., 2003). The boundary conditions were kept constant for all future emission scenarios. The choice of background ozone is crucial for air quality simulations and for predicting the effect of emission reductions (Andreani-Aksoyoglu et al., 2008). A recent analysis of various ozone observational data in Europe showed that annual mean ozone concentration increased in the 1980s and 1990s (Logan et al., 2012). Summer ozone levels started decreasing slowly in the 2000s, but there were no significant changes in other seasons. Logan et al. (2012) indicated the inconsistencies in various data sets leading to different trends. It is therefore difficult to choose a realistic background ozone values for the model domain and for the period of interest. In view of this, we kept the background ozone levels constant for simulations in the period between 2005 and 2020 (Wilson et al., 2012; Logan et al., 2012). For the 1990 simulation, background ozone mixing ratios were set about 5 ppb lower in each model layer for all boundaries and for each hour, based on the positive trend in the 1990s reported by the long-term measurement studies (Cui et al., 2011, Logan et al., 2012). Seasonal variation was also taken into account. Photolysis rates were calculated using the TUV photolysis pre-processor (Madronich, 2002). The required ozone column densities were extracted from TOMS data (NASA/GSFC, 2005). Dry deposition of gases in CAMx is based on the resistance model of Zhang et al. (2003). For surface deposition of particles, CAMx includes diffusion, impaction and/or gravitational settling. CAMx uses separate scavenging models for gases and aerosols to calculate wet deposition. The gas-phase mechanism used in this study was CB05 (Carbon Bond Mechanism 5) (Yarwood et al., 2005).

We performed CAMx simulations for 1990 (the reference year for the Gothenburg Protocol), 2005 (the reference year for the revised Gothenburg Protocol), 2006 (for model validation) and 2020 (the target year for the revised Gothenburg Protocol) with different emission scenarios as described in the next section. For all of these simulations, however, the 2006 meteorology was used.

In order to determine the changes in pollutant concentrations in the past (since 1990) and in the future (until 2020), the annual average ozone and PM2.5 for each scenario were compared with those in the reference year 2005. Dry and wet deposition of nitrogen species were summed over the entire year for each scenario and compared with 2005. AOT40 for forests was calculated for the daytime hours (08:00–20:00) from the beginning of April until the end of September in all scenarios. SOMO35 was calculated by summing the daily maximum of the 8 h running average over 35 ppb for the whole year.

We prepared six emission scenarios (see Table 1). The gridded (0.125∘×0.0625∘) TNO/MACC data (http://www.gmes-atmosphere.eu/) for 2006 were used as the basic anthropogenic emission inventory (Denier van der Gon et al., 2010). The European values in both domains were replaced by the high-resolution Swiss emission data for grid cells located within the Swiss national boundary (INFRAS, 2010; Heldstab and Wuethrich, 2006; Kropf, 2001; Heldstab et al., 2003; Schneider, 2007; Kupper et al., 2010). The output of the CAMx simulation using the meteorological data and emissions for 2006 was used for model evaluation.

The TNO/MACC emission inventory was scaled with the annual data from the Centre for Emission Inventories and Projections (CEIP) of the European Monitoring and Evaluation Programme (EMEP), and the International Institute for Applied Systems Analysis/Greenhouse Gas and Air Pollution Interactions and Synergies Model (IIASA/GAINS) was used to prepare gridded, hourly emissions for 1990, 2005 and 2020. CEIP manages a database of annual emissions before 2009 based on data submitted by participating countries (http://www.ceip.at/webdab-emission-database/emissions-as-used-in-emep-models/). IIASA uses the GAINS model to predict national emission projections until 2020 on the basis of the assumed economic development of each country (http://gains.iiasa.ac.at/gains/EUN/index.login). The emissions for a given emission scenario were calculated by scaling the raw data using annual emission totals for each country, species and SNAP (Selected Nomenclature for Air Pollution) category. For scenarios 1990 and 2005, the annual emissions for each SNAP category were extracted from the EMEP/CEIP database, which contains the historic emissions submitted by the EMEP member states. Data for PM2.5 and PM10 are only available for 2000 and later, so the 1990 data were calculated from the 2005 data, using GAINS simulations. For the 2020 scenarios, the 2005 data were scaled to 2020 using GAINS CIAM4/2011 simulations. The baseline scenario (BL) assumes that emissions will continue to be regulated by the current legislation. The MTFR (maximum technically feasible reductions) scenario uses the lowest expected emissions for most of the source categories. The MID scenario uses moderate emission reductions that are between those of BL and MTFR.

The relative changes in emissions between 2005 and 2020 for various scenarios in Switzerland (CH) and an average of 27 Union European countries (EU) are shown in Fig. 1. The emissions for the revised Gothenburg Protocol (2020 revision) are included in the figure, although there was no GAINS scenario available at the time of this work. After its publication, however, the reductions specified by the revised Gothenburg Protocol were found to be very close to those for the baseline (2020 BL). In general, emission reductions increase with increasing ambition, i.e. they are lowest in BL and highest in MTFR. The relative changes for Switzerland are usually lower than those for the EU countries (due to the larger emission reductions that had previously been imposed in Switzerland) except for PM2.5, for which all reductions are comparable.

The biogenic emissions were calculated using the method described in Andreani-Aksoyoglu and Keller (1995) for each CAMx domain using the temperature and shortwave irradiance from the WRF output, the global USGS land use data and the GlobCover 2006 inventory. For each European country the deciduous and coniferous forest fractions were split into tree species according to the method reported in Simpson et al. (1999). Inside the Swiss border the global data were replaced by data based on land use statistics (100 m resolution) and by forest data (1 km resolution) taken from the national forest inventory (Mahrer and Vollenweider, 1983). Currently this biogenic emission inventory is being improved by extending the number of species and trees, using the best available land use data and including updated temperature and irradiance dependencies (Oderbolz et al., 2013).

The results from the lowest layer of both model domains were compared with various observations in 2006. The comparison of modelled meteorological parameters with observations is given in Fig. S1 of the Supplement. In general the agreement between measurements and model results was good, with high correlation coefficients (0.76–0.98) and low mean bias error, MBE (0.00023 for specific humidity, -1.13 for air temperature, 0.57 for wind speed). These values fulfil the desired accuracy suggested by Cox et al. (1998), which is 2 ∘C for temperature, and 1 m s-1 and 2.5 m s-1 for wind speeds < 10 m s-1 and > 10 m s-1, respectively.

The predicted concentrations of ozone and PM2.5 in the European domain were compared with measurements at the rural background stations of the European Air quality database AirBase (http://acm.eionet.europa.eu/databases/airbase/). Table 2 gives the overall statistical parameters for all of the year 2006 (only those stations below 500 m a.s.l. and with 80 % of data available were used for the statistical analysis, and these sites cover a large part of Europe between -8.8 and 27.7 degrees from west to east, and between 37.3 and 60.5 from south to north). Mean annual O3 and PM2.5 are slightly over- and underestimated, respectively. In the case of ozone, although the temporal variation is captured, the maximum concentrations in summer are underestimated as reported by other studies (Fig. 2, upper panel). For instance, evaluation of several air quality models for 2006 within the Air Quality Model Evaluation International Initiative (AQMEII) showed that the models have a predominant tendency to underestimate (in some cases significantly) the peak daily mixing ratio in summertime as well as to overestimate night-time mixing ratios, with the exception of central Europe (Solazzo et al., 2012a). Time series show that the model reproduced the temporal variation of PM2.5 quite well, except for January–February, when unusually high concentrations were recorded in Europe (Fig. 2, lower panel). The underestimation of PM2.5 is partly due to the severe meteorological conditions prevailing during that exceptionally cold inversion period. It is also possible, however, that the contribution of wood burning to emissions was underestimated. As reported by Solazzo et al. (2012b), none of the models used in AQMEII study consistently matched PM2.5 observations for all locations throughout the entire year. Results of the AQMEII study suggest that while the models do relatively well in simulating the inorganic aerosol species, large uncertainty remains in the simulation of other components such as secondary organic aerosols and unspeciated PM2.5. Elimination of the sources of PM bias in the models is still challenging.

The frequency distributions of modelled and measured ozone and PM10 values in 2006 are shown in Fig. 3. Comparison of the model results from the nested domain with measurements in Switzerland suggests that the model performance is better at rural sites. At the rural site, Chaumont, for example, the shape of the measured and modelled distributions of O3 is similar: both have the highest number of points approximately in the middle of the graph. At the urban site, Zurich, on the other hand, the discrepancy between the measurements and model results at low concentrations are clearly seen. Because of the finite model resolution, NOx concentrations are usually underestimated at urban sites, where local emissions are relatively high and variable. This leads to overestimation of ozone at night and in the morning. In addition to the model horizontal resolution, its representation of the inversion layer at night and the mixing layer during the day also plays an important role in the prediction of pollutant concentrations. In the case of PM10, the measured and modelled concentrations also show a very similar distribution at the rural site Chaumont, indicating very good model performance, whereas the high concentrations at the urban background site, Zurich, were underestimated.

The modelled concentrations of particulate species in the nested domain were compared with AMS (aerosol mass spectrometer) measurements of particulate nitrate, sulfate, ammonium and organic aerosols (Lanz et al., 2010) at Payerne in June 2006 (Fig. 4). Although the model calculates PM2.5 and the AMS measures only particles smaller than 1 µm, the results may be compared, because the difference between PM1 and PM2.5 measurements is very small as shown in Aksoyoglu et al. (2011). Elemental carbon (EC) data are obtained from Aethalometer equivalent black carbon (BC) measurements. The model performance for aerosol components in this study is significantly better than that in our previous study, which used the MM5 meteorological model with an earlier CAMx version (Aksoyoglu et al., 2011). The modelling of organic aerosols, however, is still quite challenging, mainly due to limited knowledge about the processes involved in secondary organic aerosol (SOA) formation. The CAMx model used in this study includes an SOA model based on a theory of the gas–particle partitioning of various precursors, such as anthropogenic and biogenic volatile organic carbon (VOC) species. The oligomerization process, which leads to an increase in aerosol concentrations, is also included. The model performance for organic aerosols is reasonably good for relatively low concentrations. It becomes worse, however, when the formation of secondary organic aerosols increases. The total modelled PM2.5 (sum of inorganic and organic species) concentrations match the observations quite well, with one exception on 14–16 June, which was due to underestimation of increased levels of organic aerosols. Models that take into account the volatility distribution and atmospheric aging of OA might give more realistic results (Bergström et al., 2012).

The modelled annual average PM2.5 concentrations vary between 5 and 40 µg m-3 for the reference year 2005 in Europe (Fig. S2 of the Supplement). Our results suggest that PM2.5 concentrations decreased significantly in Europe between 1990 and 2005. The relative changes range from -20 % in Scandinavia to more than -60 % in the eastern part of the domain; they are between -40 and -45 % in central Europe (Fig. 5). There have been long-term measurements of PM10 throughout Europe since the late 1990s, but measurements of PM2.5 at some European sites are available only after 2000 (Tørseth et al., 2012). The available data, however, show average changes between 2000 and 2009 of -18 and -27 % for PM10 and PM2.5, respectively. Recently Cusack et al. (2012) reported that PM2.5 concentrations in various parts of Europe decreased by 7–49 % between 2002 and 2010. The average trends of -0.4 µgm-3yr-1 for PM10 and PM2.5 at several European sites reported by Barmpadimos et al. (2012) correspond to a decrease of about 40–45 % between 1998 and 2010. The PM10 measurements at various sites in Switzerland indicate a large decrease (20–56 %) between 1991 and 2008 (Barmpadimos et al., 2011). This supports our model results (see Fig. 5), because most of the change in PM10 was in the PM2.5 fraction (Barmpadimos et al., 2012). Combining the PM10 trends from Barmpadimos et al. (2011) with the modelled PM10 for 1990, 2005 and 2006 in this study at four stations shows the interannual variability in the observed trends together with the emission-induced changes modelled in this study (Fig. S3 of the Supplement).

Changes in particulate matter concentrations result not only from changes in primary PM emissions but also from changes in precursor emissions such as nitrogen oxides (NOx), non-methane volatile organic compounds (NMVOC), sulfur dioxide (SO2) and ammonia (NH3). As seen in Fig. 1, the European emission reduction of NH3 for the 2020 scenarios is much smaller than the reduction of other precursor emissions. We compared the predicted annual mean PM2.5 concentration for 2020 with that of the reference year (2005) and found that a considerable reduction in PM2.5 would be obtained in Europe under the BL scenario (Fig. 6). The decrease in PM2.5 would vary from 30 to 40 % in central Europe, and up to 50 % in some local polluted areas in eastern Europe. The predicted reductions using the MID and MTFR scenarios are about 50–60 %, with the largest changes being predicted in the Balkan countries (only MTFR is shown in the lower panel of Fig. 6.) In Switzerland the predicted reductions in PM2.5 are 30 and 40 %, using BL and MTFR scenarios, respectively.

The average ozone mixing ratios for the reference year (2005) are shown in Fig. S4 of the Supplement. The model results – based on the assumption that the background ozone levels increased by 5 ppb between 1990 and 2005 – suggest that the average annual ozone increased between 1990 and 2005 in a large part of Europe in spite of the large reductions of precursor emissions (Fig. 7). The increase in ozone was predicted especially for England, the Benelux countries and around Ukraine. In an earlier sensitivity study, we reported that these areas have VOC-limited regimes for ozone production (Aksoyoglu et al., 2012); a reduction of precursor emissions leads to an increase in ozone levels in such regions.

The impact of the choice of boundary conditions for 1990 on the results shown in Fig. 7 was investigated through a sensitivity test in which we increased the background ozone by 5 ppb. The results suggested that the change in the annual mean ozone (Fig. 7) would be 1–2 ppb lower in central Europe. Long-term observations, however, show a significant positive trend in the annual mean ozone especially between 1990 and 2000 (Cui et al., 2011; Logan et al., 2012). A positive trend of 0.32 ppb yr-1 reported by Cui et al. (2011) for annual mean ozone measured at Jungfraujoch between 1990 and 2008 supports our choice of a 5 ppb increase in the background ozone between 1990 and 2005.

The predicted O3 increase is about 1–2 ppb (3–9 %) over the Swiss Plateau, whereas observations indicate larger changes between 10 % at rural areas and 40–50 % at urban sites (Table 3). On the other hand, modelled peak ozone values are lower in 2005 than in 1990 (see Table 4). Measurements also show a decrease in peak ozone levels except in Basel (suburban) and Zurich (urban). The simulation of ozone trends is quite challenging, as has been shown in other model studies (Colette et al., 2011; Wilson et al., 2012). As seen in the example for Zurich, the frequency distribution of ozone mixing ratios in 1990 and 2005 is clearly different (Fig. 8). The most frequent ozone levels are shifted toward higher levels in 2005 and the change is larger in the measurements.

The relative change in annual average ozone mixing ratios between 2005 and 2020 is shown in Fig. 9a and b for the European and Swiss domains, respectively. For both BL and MTFR scenarios, the predicted decrease is small (< 4 ppb, < 10 %) in central Europe, whereas ozone is expected to increase further in England and the Netherlands, due to reduced titration with NO. On the other hand, no further increase is expected around Ukraine between 2005 and 2020 as predicted for the period between 1990 and 2005 (see Fig. 7, upper panel). A decrease of about 5–7 % is predicted over the Alpine regions and the southern part of the Alps, while ozone is predicted to increase by about 1 ppb (3 %) at urban sites (Fig. 9b). One has to keep in mind, however, that the background ozone levels in these simulations were assumed to stay constant between 2005 and 2020, based on the study of Logan et al. (2012).

The modelled AOT40 and SOMO35 results for the reference year (2005) are shown in Fig. 10. AOT40 values range between 5 and 30 ppmh, with elevated levels in southern Europe. The SOMO35 values show a similar spatial distribution, lying between 1000 and 5000 ppbd. In Switzerland, the modelled AOT40 is 10–15 ppmh and 20–30 ppmh in the north and south, respectively (Fig. 11, upper panel). We predicted SOMO35 values between 2400 and 2800 ppb d for northern Switzerland and between 4000 and 4800 ppbd for the southern part of the Alps (Fig. 11, lower panel). These results match the AOT40 and SOMO35 values derived from measurements in 2005 very well (Table 3). Compared to an EMEP model study which reported average AOT40 and SOMO35 of 35.1 ppmh and 5303 ppbd, respectively, for Switzerland in 2005 (Gauss et al., 2012), our results are lower and in better agreement with the measurements.

A comparison of simulations for 1990 and 2005 suggests that AOT40 and SOMO35 have decreased in Switzerland since 1990 (Fig. 12), although average annual ozone mixing ratios increased (Fig. 7). This indicates that peak ozone values decreased due to emission reductions, as shown in Table 4. Although measurements also show a decrease at rural sites, they suggest that AOT40 and SOMO35 increased significantly at urban sites (Table 3). This discrepancy between the model results and observations indicates the sensitivity of these indicator parameters to threshold values. Overestimation of ozone concentrations by regional models at night in polluted urban areas is a common problem. This alone, however, cannot be responsible for the discrepancy between measured and modelled AOT40, because AOT40 is the sum of ozone concentrations above 40 ppb and is calculated only during the daytime. The difference between the modelled and measured frequency distributions of ozone mixing ratios above 30–40 ppb is relevant to an understanding of the changes in AOT40 and SOMO35 (Fig. 8). The discrepancy between the modelled and measured relative change in damage indicators is most likely due to the background ozone levels, but this needs further analysis.

Assuming a constant background ozone after 2005, AOT40 and SOMO35 were predicted to decrease substantially by 2020 (Figs. S5–S8 of the Supplement). One must keep in mind, however, that these indicators depend strongly on the threshold values, which might be affected by the background ozone and its evolution in the future.

The atmospheric deposition of pollutants raises serious concerns for ecosystems. In Switzerland, emissions of air pollutants such as sulfur dioxide and nitrogen oxides have been substantially reduced in the last couple of decades. While sulfur emissions are now stabilized at lower levels than in the past, nitrogen oxide emissions are still rather high. In this section, therefore, we focus on nitrogen deposition.

In general, the main nitrogen sources are emissions of nitrogen oxides from combustion processes and ammonia from agricultural activities. The deposition of atmospheric nitrogen species constitutes a major nutrient input to the biosphere, which enhances forest growth. Despite this, increased nitrogen input into terrestrial ecosystems represents a potential threat to forests. Enhanced nitrogen deposition can cause soil acidification, eutrophication and nutrient imbalances, causing a reduction in biodiversity. The deposition of atmospheric nitrogen compounds occurs via dry and wet processes. NO2, NH3, nitric acid (HNO3) and nitrous acid (HONO) are the most important contributors to nitrogen dry deposition. Nitrogen wet deposition results from the scavenging of atmospheric N constituents.

The predicted annual deposition of total nitrogen in Europe varies between 5 and 45 kgNha-1yr-1 in 2006 (Fig. 13, upper panel), and it is mainly dominated by dry deposition (Fig. S9 of the Supplement). Dry deposition is generally largest over regions with large ambient NH3 concentrations over the Netherlands and Belgium, as also reported in the literature (Flechard et al., 2011). We also predict high nitrogen dry deposition around the Po Valley in northern Italy. The modelled total nitrogen deposition varies between 10 and 45 kgNha-1yr-1 in northern Switzerland (Fig. 13, lower panel). Elevated levels can also be seen in the south (10–20 kgNha-1yr-1). On the other hand, they are lower at high-altitude sites (about 5 kgNha-1yr-1). These numbers are in the same range as those based on measurements at various locations in Switzerland (Schmitt et al., 2005). In a recent study, Roth et al. (2013) reported an average N deposition on 122 plots in Switzerland of 18.3 kgNha-1yr-1 for the year 2007.

Deposition of oxidized and reduced nitrogen species for 2006 is shown in Figs. 14 and 15 for the European and Swiss domains, respectively. The calculated deposition of reduced nitrogen compounds is higher than that of oxidized species. Deposition of reduced N species – especially NH3 dry deposition – is high in central Switzerland, where the ammonia emissions are the highest in the country. The combination of high ammonia concentrations and land use favourable for dry deposition leads to the highest deposition of ammonia in the nested domain in a few grid cells in central Switzerland.

A comparison of the simulations for 1990 and 2005 suggests that nitrogen deposition decreased mainly in the eastern part of the European domain, while it increased in the Iberian Peninsula (Fig. 16, upper panel). In Switzerland, the decrease in nitrogen deposition was mainly over the Alpine regions and the southern part of the country (Fig. 16, lower panel). The decrease in nitrogen deposition is mainly related to the oxidized fraction, due to large reductions in NOx emissions that occurred in the past.

The future simulations assuming the BL 2020 scenario suggest that the oxidized nitrogen deposition will decrease further by about 40 % in all of Europe until 2020, whereas deposition of reduced nitrogen compounds will continue to increase by about 20 % especially in the southern and eastern part of Europe (Fig. 17). This would lead to a 10–20 % decrease in the total nitrogen deposition in most of the model domain, with a 10 % increase in the eastern part of Europe.