Air quality over Europe : modelling gaseous and particulate pollutants

Abstract. Air quality over Europe using Models-3 (i.e., CMAQ, MM5, SMOKE) modelling system is performed for winter (i.e., January 2006) and summer (i.e., July 2006) months with the 2006 TNO gridded anthropogenic emissions database. Higher ozone mixing ratios are predicted in southern Europe while higher NO 2 levels are simulated over western Europe. Elevated SO 2 values are simulated over eastern Europe and higher PM 2.5 concentrations over eastern and western Europe. Regional average results suggest that NO 2 and PM 2.5 are underpredicted, SO 2 is overpredicted, while Max8hrO 3 is overpredicted for low mixing ratios and is underpredicted for the higher mixing ratios. However, in a number of countries observed and predicted values are in good agreement for the pollutants examined here. Speciated PM 2.5 components suggest that NO 3 is dominant during winter over western Europe and in a few eastern countries due to the high NO 2 mixing ratios. During summer NO 3 is dominant only in regions with elevated NH 3 emissions. For the rest of the domain SO 4 is dominant. Low OC concentrations are simulated mainly due to the uncertain representation of SOA formation.


Introduction
Air quality is a focus of attention because of its important role in many areas including human health, atmospheric reactions, acid deposition, and the Earth's radiation budget (e.g., Seinfeld and Pandis, 2006;Peng et al., 2005).Although air quality management strategies have been applied over recent years to reduce atmospheric pollutant concentrations, ozone and particulate matter pollution are still an issue.For this reason, simulating and forecasting gaseous and particle concentrations as accurately as possible is fundamental in air quality planning for more effective adaptation and implementation guidelines.
Air pollution is not just a local issue since the pollutants released in one country can be transported in the atmosphere, affecting air quality in the nearby countries.As such several research groups have started simulations of the gaseous and particulate matter concentrations over the whole of Europe.However, there are a limited number of such studies.In order to explain the European trends in ozone since 1990, Jonson et al. (2006) have used the EMEP regional photochemistry model for the years 1990 and 1995-2002.The increase in winter ozone, partially, and the decrease in the magnitude of high ozone episodes is attributed to the decrease in ozone precursor emissions while emission reductions have resulted in a marked decrease in summer ozone in major parts of Europe.Using the RegCM3/CAMx modelling system for simulating near surface ozone mixing ratios driven either by the ERA-40 reanalysis dataset or the global circulation model ECHAM5, Zanis et al. (2011) found that on a seasonal basis both ECHAM5 and ERA simulations exhibited a seasonally dependent bias, with winter and spring ozone values being generally underestimated; summer and autumn values were slightly overestimated.However, ozone peak mixing ratios in summer (i.e., higher than 80 ppbV) could not be captured.A decadal (i.e., 1991-2000) study also for ozone over Europe suggests that the selection of external meteorological forcing can be as important as the selection of adequate chemical lateral boundary conditions (Katragkou et al., 2010).A modelling set up for the whole Europe with CALIOPE air quality modelling system has been performed by Pay et al. (2010) suggesting satisfactory performance for ozone but poor performance for particles.Largely, this is caused by the inability of the models to correctly capture the concentrations Published by Copernicus Publications on behalf of the European Geosciences Union.E. Tagaris et al.: Air quality over Europe: modelling gaseous and particulate pollutants of organic matter (e.g., Chen and Griffin, 2005).Applying the CAMx modelling system over Europe Nopmongcol et al. (2012) found an underestimation trend for all pollutants examined (i.e., O 3 , NO x , NO 2 , CO, PM 10 ) except for SO 2 .Appel et al. (2012), using CMAQ, found that the model overestimated winter daytime ozone mixing ratios in Europe by an average of 8.4 % while in the summer slightly underestimated by 1.6 %.PM 2.5 is underestimated throughout the entire year.Due to the lack of speciated PM 2.5 data for the EU, they concluded that it is not clear what is driving this bias.Langmann et al. (2008), using the regional scale atmospheric climate chemistry/aerosol model REMOTE, found that the deviation between modelled and measured organic carbon concentrations can be mainly explained by missing formation of secondary organic aerosols (SOA) and deficiencies in emission data.As such, the authors suggest that an updated emission inventory needs to take into account the changing heating practices in Europe.The need for a more detailed treatment of the formation of SOA has also been pointed out by Sartelet et al. (2007) who simulated aerosols and gas-phase species over Europe with the POLYPHEMUS system.Although they found that hourly ozone, sulfate, and ammonium simulation was good, SO 2 and nitrate concentrations were overestimated.Modelling carbonaceous aerosol over Europe using the EMEP modelling system, Simpson et al. (2007) found that the contribution of biogenic secondary organic aerosol far exceeds that of the anthropogenic one.This modelling work confirms the difficulties of modelling SOA in Europe where a severe underestimation of the SOA components was found.The evaluation of the aerosol components in the CALIOPE air quality modelling system over Europe (Basart et al., 2012) also highlights underestimations in the fine fraction of carbonaceous matter (EC and OC) and secondary inorganic aerosols (i.e., nitrate, sulphate, and ammonium).The total amount of secondary inorganic aerosols was on average underestimated by 18-50 % in most regions of Europe (Pay et al., 2012).SO 2 was systematically overestimated by the CALIOPE system, which suggests that SO 4 formation in the modelling system was often limited by oxidant availability and not always by SO 2 .NO 3 concentrations were underestimated in about 60 % in winter and more than 100 % in summer.Bessagnet et al. (2008) implemented the isoprene chemistry for SOA formation over Europe in the CHIMERE model.They found a better agreement between long series of simulated and observed particulate matter concentrations, but a clear underestimation by the CHIMERE model was noted in wintertime possibly due to missing wood burning emissions.
The objective of this study is to simulate gaseous (i.e., O 3 , NO 2 , SO 2 ) mixing ratios and particle (i.e., PM 2.5 ) concentrations over Europe, assessing the magnitude of disparity for each country and estimating errors and biases between observed and predicted values for European countries with available monitoring data.The current analysis provides an opportunity to compare the modelling results with the re-sults obtained by other regional air quality models commonly used in Europe, noting at the same time, the European countries where the predicted values are in agreement with the observed values.

Modelling setup
Meteorological fields are derived using the Penn State/NCAR Mesoscale Model (MM5) (Grell et al., 1994).Since most meteorological models, such as MM5, are not built for air quality modelling purposes, it is necessary to address issues related to data format, unit conversion, temporal and spatial domains, and numerical grids.For this purpose, the Meteorology Chemistry Interface Processor (MCIP) (Byun et al., 1999) is used to provide the meteorological data from the MM5 outputs needed for the emissions and air quality models.
Gridded yearly averaged anthropogenic emissions for the year 2006 over Europe are provided by TNO (Netherlands Organization for Applied Scientific Research) at a 0.1 × 0.1 degree resolution (http://www.tno.nl) in the framework of the AQMEII exercise (http://aqmeii.jrc.ec.europa.eu/).The available data include annual total emissions of CH 4 , CO, NH 3 , NMVOC, NO x , PM 10 , PM 2.5 , and SO 2 for both area and point sources in ten (10) Standardized Nomenclature for Air Pollutants (SNAP) categories (i.e., power generation, residential-commercial and other combustion, industrial combustion, industrial processes, extraction distribution of fossil fuels, solvent use, road transport, other mobile sources, waste treatment and disposal, agriculture) (Table 1).According to this emission inventory, the United Kingdom, Spain, Germany, Ukraine, France, and Italy have the highest NO x emissions, while Ukraine, Spain, and Poland have the highest SO 2 emissions (only a part of the Russian Federation and Turkey belongs to the domain examined).In general, road transport and energy sector-utilities-refineries are the major sources for NO x emissions, while SO 2 emissions originate mainly from the energy sector-utilities-refineries.Emissions are processed by the Sparse Matrix Operator Kernel Emissions (SMOKE v2.6) modelling system (http://www.smoke-model.org/index.cfm) to convert their resolution to the resolution needed by the air quality model using monthly, weekly, and hourly time profiles provided by TNO (2011).However, TNO has reported that the temporal profiles are a generalisation, not regularly updated and not country specific and could affect emissions over time for air quality modelling.The Biogenic Emission Inventory System, version 3 (BEIS3) is used for processing biogenic source emissions.Gridded land use data at 1 km resolution provided by the US Geological Survey (USGS) (http://edc2.usgs.gov/glcc/glcc.php).BEIS3 default summer and winter emission factors and the MM5 meteorological fields are used to create hourly model-ready biogenic emissions estimates.The Community Multiscale Air Quality (CMAQ) v4.7 Modelling System with the Carbon Bond mechanism (CB05) is used here for the regional air quality modelling (Byun et al., 2006;Foley et al., 2010) for winter (i.e., January 2006) and summer (i.e., July 2006) months.CMAQ is a multipollutant, multiscale air quality model for simulating all atmospheric and land processes that affect transport, transformation, and deposition of atmospheric pollutants on both regional and urban scales.The modelling domain covers almost all of Europe with 177 × 217 grid cells of 35 km × 35 km spatial resolution and 14 vertical layers (Fig. 1).Although a finer grid resolution could affect modelling results, other studies have found that it does not always enhance model performance (e.g., Queen et al., 2008).The default boundary and initial conditions based on Gipson (1999) for gaseous and particulate species have been used.Boundary conditions have a very minor impact on pollutants concentrations, since European land is far away from the domain borders, except along the eastern border.Russian Federation, Ukraine, and Turkey are close to the eastern borders, but these countries are not included in our analysis.Moreover, a spin up time of 10 days was used to minimise errors due to the initial conditions.In the version of CMAQ used, several new pathways for secondary organic aerosol (SOA) formation have been implemented (Edney et al., 2007;Carlton et al, 2008).The CB05 is a condensed mechanism of atmospheric oxidant chemistry that provides a basis for computer modelling studies of ozone, particulate matter (PM), visibility, acid deposition, and atmospheric concentrations of toxic gases (Yarwood et al., 2005).The core CB05 mechanism has 51 species and 156 reactions.The CB05 has been evaluated against smog chamber data ( Jeffries et al., 2002;Carter, 2000) and the results are discussed in detail by Yarwood et al. (2005).
Since an extensive evaluation and discussion of meteorology used has been presented by Vautard et al. (2012), here we focus on discussing the gaseous and particulate pollutant concentrations.Briefly, Vautard et al. (2012) found that the seasonal cycle of the 10 m wind speed is well reproduced although it is overestimated over Europe.The spatial distribution of surface wind speed is fairly well simulated.Wind speed is well simulated along the vertical profile, but markedly overestimated at lower altitudes over Europe.It was also found that the Planetary Boundary Layer (PBL) height at noon is simulated quite well.However, at 18:00 UTC and particularly in the summer months, the modelled PBL height is much lower than the observed.Biases of monthly means of the 2 m temperature are generally small.The diurnal cycle of the 2 m temperature is also fairly well reproduced while the typical vertical temperature profile bias is between ± 1 K. On average, the temperature is slightly underestimated while relative humidity above the surface is overestimated.

Model evaluation
Comparison between predicted and observed gas and particle concentrations is performed for January and July 2006 using observation data from AirBase, the European air quality database (http://www.eea.europa.eu/data-and-maps/data/airbase-the-european-air-quality-database-2).AirBase is the air quality information system maintained by the European Environment Agency (EEA) through the European topic centre on Air and Climate Change.It contains air quality data delivered annually, offering a reciprocal exchange of information and data from networks and individual stations measuring ambient air pollution within the Member States.Model evaluation is conducted here, for species with sufficient monitoring data all over Europe such as sulfur dioxide (data from 35 countries, in our domain there are 1928 stations for winter and 1883 for summer months), nitrogen dioxide (data from 35 countries, in our domain there are 2591 stations for winter and 2508 for summer months), ozone (data from 35 countries, in our domain there are 1954 stations for winter and 1977 for summer months) and particulate matter < 2.5 µm (data from 30 countries, in our domain there are 266 stations for winter and 267 for summer months).Hourly average data for NO 2 and SO 2 as well as daily average data for maximum 8 h ozone and PM 2.5 are used in our analysis.Unfortunately, comparison with observed PM 2.5 components could not be performed since speciated PM 2.5 data are not readily available for EU; this has also recently been pointed out by other researchers (Appel et al., 2012).

Results and discussion
High ozone mixing ratios are predicted in southern Europe where meteorological conditions enhance ozone formation (Fig. 2).The daily average maximum 8 h ozone (Max8hO 3 ) mixing ratio during July is simulated up to 75 ppbV while a large portion of the domain has values higher than 50 ppbV.Higher NO 2 mixing ratios are simulated over western Europe (i.e., Belgium, the Netherlands, Germany, and northern France), northern Italy and the United Kingdom for both seasons.Belgium and the Netherlands have elevated NO 2 values since their small area results in a high emission rate per acre, however, they are not ranked as one of the countries with high NO x emission rates.Road transport and industry are responsible for the elevated NO x emissions in northern Italy, while road and non-road transport energy sector and industry are responsible for the high NO x emission in the United Kingdom.NO 2 mixing ratios are higher during January compared to July for two reasons: energy sector and industry emit more NO x during winter and NO 2 photolysis is unfavourable during winter.Elevated SO 2 mixing ratios are simulated over eastern Europe, with higher values in Poland and the North Balkan Peninsula.Since power generation and industry are mainly responsible for SO 2 emissions, SO 2 values are very location dependent and also show higher values during winter.Elevated PM 2.5 levels are simulated over eastern and western Europe (i.e., daily average concentrations up to 30 µg m −3 during winter).NO 3 is dominant during winter over western Europe and in a few eastern countries due to the high NO 2 mixing ratios (Fig. 3).During summer NO 3 is dominant only in regions with elevated NH 3 emissions (i.e., the Netherlands and northern Italy).For the rest of the domain SO 4 is dominant.Low OC concentrations are simulated in general.Representation of secondary organic aerosol (SOA) formation is uncertain (e.g., Chen et al., 2005;Kroll et al., 2006) and low OC has been noted in the CMAQ approaches (Foley et al., 2010).NH 4 follows SO 4 and NO 3 spatial distribution plots for both seasons, since atmospheric SO 2 is oxidized to sulfuric acid which reacts with ammonia to form ammonium sulfate, while gas-phase NO x oxidizes to nitric acid which reacts with ammonia to form ammonium nitrate.Spatial distribution plots presented here for gaseous pollutants and PM 2.5 are similar with those presented by another study (Pay et al., 2010).Using the WRF-ARW meteorological model, the HERMES-EMEP emission processing model, a mineral dust dynamic model (BSC-DREAM8b), and CMAQ chemical transport model, they provide annual simulations for 2004 over Europe.Both studies have found high ozone mixing ratios over the Mediterranean and the nearby land; elevated NO 2 mixing ratios over western Europe (i.e., Belgium, the Netherlands, Germany, northern France), southern United Kingdom, and northern Italy; elevated SO 2 levels over eastern Europe (i.e., Poland) and northern Balkan Peninsula; higher PM 2.5 concentrations over western and eastern countries (i.e., Belgium, northern France, Poland, southern Romania) and northern Italy.
Model performance for ozone shows a mixed trend: Max8hO 3 is overestimated for low mixing ratios (about 50 ppbV) while it is underestimated for the higher mixing ratios.This trend is in agreement with the CMAQ application performed by Appel et al. (2012) for Europe where daytime ozone mixing ratio is overestimated in winter and underestimated in summer months.The overestimation tendency for the lower mixing ratios gives a much higher mean estimated value during winter compared to the observed one (Table 2), which is diminished during summer where higher ozone mixing ratios are recorded.Both observed and predicted ozone mixing ratios are similarly spread out around mean values during winter (similar simulated standard and mean absolute deviations).During summer the spread of simulated mixing ratios is less and the mean simulated mixing ratios are closer to the observation data; this is related to the overestimation of the lower mixing ratios.At regional scale, according to the grouping used by the United Nations Statistics Department for northern, western, eastern, and southern Europe (http://unstats.un.org/ unsd/methods/m49/m49regin.htm#europe; Fig. 1), the model overestimates ozone mixing ratios in all regions during winter; during summer it overestimates ozone mixing ratios in northern and southern Europe and underestimates them in eastern and western Europe where higher mixing ratios have been recorded (Table 2).A consistent bias for NO 2 , SO 2 , and PM 2.5 estimations is noted: NO 2 and PM 2.5 are underestimated while SO 2 is overestimated for both seasons in all regions.The same biases have also been noted by a previous  study (Pay et al., 2010).The consistent underestimation trend for PM 2.5 has been found also by Appel et al. (2012) using CMAQ modelling system for Europe.At the regional scale, we underestimate NO 2 mixing ratios more in southern Europe and overestimate SO 2 mixing ratios more in eastern Eu-rope where higher underestimation in PM 2.5 concentrations is found.
Modelling results averaged over large areas (i.e., Europe or European regions) are important for several issues (e.g., model performance, sensitivity studies, and strategic plans for pollutants reduction), but of equally importance are (P (i) − O (i) ) 2 ) (where P (i) stands for predicted data and O (i) for observed data).
Max8hO 3 is overestimated for all countries during January since low ozone mixing ratios are observed in winter.The observed to predicted ratio is between 0.6 and 0.8 for almost all countries (Table 3).During July, the mean observed values are closer to the mean predicted values for all countries (the ratio is 1.0 ± 0.1 for the majority of the countries).Few countries (i.e., Estonia, Finland, Ireland, Latvia, Norway, and Romania) show similar disagreement for both months.MB for Max8hO 3 mixing ratios is positive for all countries for January since the model overestimates low mixing ratios that recorded during winter (Fig. 4).The calculated MB for the majority of the countries is more than 10 ppbV.Both negative and positive MB values for Max8hO 3 mixing ratios are obtained for July.For many countries (i.e, Bulgaria, Croatia, F.Y.R.O.M., Germany, Hungary, Italy, Lithuania, Luxembourg, the Netherlands, Poland, and the United Kingdom) the calculated MB is 0 ± 5 ppbV for the summer month while higher positive MB values are calculated for Ireland, Latvia, Malta, Portugal, and Spain.Spatial distribution plots with the Normalised MB for the daytime ozone mixing ratios are presented by Appel et al. (2012).Although an exact comparison with Appel et al. (2012) could not be performed, since we present MB for Max8hO 3 mixing ratios averaged per country for January and July 2006 vs. their Normalised MB for daytime ozone mixing ratios averaged per monitoring station for the three winter and the three summer months of 2006, the general trend is similar: better agreement between observed and predicted mixing ratios during summer and model high overestimation during winter.However, Appel et al. (2012) found a tendency to underestimate winter daytime ozone mainly in Czech Republic and Poland.Higher RMSE are calculated for the southern European countries compared to the northern European countries for both months.Lower RMSE are calculated in almost all countries for July compared to January.The countries that seem to have the best agreement between observed and predicted Max8hO 3 mixing ratios for all parameters examined here (i.e., Observed mean Predicted mean : 1 ± 0.1, MB: 0 ± 5 ppbV, RMSE: ≤ 12 ppbV) in July are: Lithuania, Hungary, Poland, Bulgaria, Germany, the Netherlands, and Croatia, while no country was found to agree well with all the previous parameters for January.The mean simulated NO 2 mixing ratios are close to the observed mixing ratios in Belgium, the Netherlands, Spain, and the United Kingdom for both months (Table 3).However, the disagreement is greater during July in almost all countries.MB for NO 2 values is negative for all countries except Belgium for both months as well as for the Netherlands and the United Kingdom in January with values up to 4.2 ppbV (Fig. 5).The largest negative MB is estimated for Norway (< −15 ppbV) in January while estimates for Spain, Estonia, Denmark, Portugal, Finland, Luxembourg, Poland, Germany, and Latvia along with the United Kingdom, the Netherlands, and Belgium have MB 0 ± 5 ppbV.Austria, F.Y.R.O.M., and Norway is estimated to have much higher negative MB in January compared to July, while Serbia, Germany, and Luxembourg is estimated to have much higher negative MB in July compared to January.High RMSE is calculated in Romania, Austria, Italy, F.Y.R.O.M., Norway, and Bulgaria in January while Serbia is estimated to have the highest RMSE in July.The Netherlands has the best agreement between observed and predicted NO 2 mixing ratios for all parameters examined here (i.e., Observed mean Predicted mean : 1 ± 0.1, MB: 0 ± 5 ppbV, RMSE: ≤ 12 ppbV) in January.However, model performance for Belgium and the United Kingdom is quite good for the same month.Belgium has the best agreement in July between observed and predicted NO 2 mixing ratios.
The mean simulated SO 2 mixing ratios are close to the observed mixing ratios in F.Y.R.O.M. for both months (Table 3).In few countries (i.e., Czech Republic, Greece, Luxembourg, and Sweden) the average simulated SO 2 mixing ratios are close to the observed mixing ratios in January, while for Finland and Serbia the average mixing ratios show good agreement in July.Three countries (i.e., Croatia, Latvia, and Switzerland) are simulated to have mean SO 2 mixing ratios much lower than the mean observed mixing ratios for both months.However, in three countries (i.e., Greece, Czech Republic, and Sweden) although mean observed mixing ratios are similar to the mean predicted mixing ratios during January they are much lower in July.Estonia shows reverse behaviour for the simulated months i.e., mean observed mixing ratio is much higher compared to the mean predicted in January and much lower in July.MB in SO 2 mixing ratios are estimated to have both positive and negatives values for both months (Fig. 6).The highest positive MB are calculated for Norway, Bulgaria, Bosnia and Herzegovina, and Romania for both months, while the highest negative MB are calculated for Croatia and Serbia in January and for Croatia in July.More than half of the countries have MB in the range 0 ± 2 ppbV for both months.There is no significant change in the MB between the two seasons with the exception of Serbia.High RMSE is calculated for Spain, Romania, Mean observed PM 2.5 concentrations are higher compared to the predicted concentrations for all countries except France in January (Table 3).For Belgium and Luxembourg mean observed and predicted concentrations are in good agreement for both months while good agreement is also noted for Finland, France, and the United Kingdom in January.MB for PM 2.5 concentrations is estimated to be negative for all countries for both months except France in Jan-uary (Fig. 7).The highest (negative) MB and RMSE are calculated for Switzerland, Poland, Bulgaria, Hungary, Czech Republic, Italy, Austria, Slovakia, and Bosnia and Herzegovina in January.However, they are much lower in July.The United Kingdom and Finland in January and Luxembourg in July are estimated to have the lowest MB (i.e., up to -1µg/m 3 ) as well as the lowest RMSE (i.e., up to 5.4µg/m 3 ).Luxembourg, Finland, and the United Kingdom seem to have the best agreement between observed and predicted PM 2.5 concentrations for all parameters examined here (i.e., Observed mean Predicted mean : 1 ± 0.1, MB: 0 ± 5 µg m −3 , RMSE: ≤ 12 µg m −3 ) in January as well as Luxembourg in July.Spatial distribution plots of the seasonal Normalised MB for PM 2.5 concentrations for 2006 are presented by Appel et al. (2012).Both studies underestimate PM 2.5 concentrations, and predict lower values of the parameters examined in each study (i.e., MB and Normalised MB) during summer months.

Conclusions
Application of CMAQ modelling system over Europe for January and July 2006 using the TNO gridded anthropogenic emissions database for the year 2006 shows an overestimation trend for low ozone mixing ratios (less than 50 ppbV) while the higher mixing ratios are underestimated.However, spatial distribution plots are reasonably estimated (e.g., higher ozone mixing ratios in southern Europe).Simulated values for NO 2 , SO 2 , and PM 2.5 suggest a consistent bias using regional average values: SO 2 is overestimated while NO 2 and PM 2.5 are underestimated.Speciated PM 2.5 components give low OC concentrations as a result of the uncertain representation of SOA formation.Statistical analysis (i.e., mean concentrations, MB, RMSE) for each country found that observed and predicted Max8hO 3 mixing ratios are in good agreement in Lithuania, Hungary, Poland, Bulgaria, Germany, the Netherlands, and Croatia in July.However, no country was found to be in good agreement in January where low ozone mixing ratios are recorded.The Netherlands has the best agreement between observed and predicted NO 2 mixing ratios in January, although model performance in Belgium and the United Kingdom is quite good for the same month.Belgium was the only country to have good agreement between observed and predicted NO 2 mix-ing ratios in July.Czech Republic, Greece, Luxembourg, and Sweden had the best agreement between observed and predicted SO 2 mixing ratios in January as well as Finland and Serbia in July.Luxembourg, Finland, and the United Kingdom are estimated to have the best agreement between observed and predicted PM 2.5 concentrations in January as well as Luxembourg in July.
There are numerous reasons why a bias or an error may exist.This could be related to inaccuracies in emission inventories.Emissions of air pollutants originate from a variety of small and large individual sources (e.g., power plants, industries, motor vehicles) of varying temporal and spatial characteristics.They are subject to significant uncertainties given that they are based on datasets of limited spatiotemporal coverage and that countries do not always estimate emissions in a uniform and transparent manner.However, biases and errors could also be related to discrepancies in the meteorological data and the source locations; incommensurability of grid cell averaged predictions with point measurements, horizontal grid resolution, topographic effects that are not accounted for in the model; or the lack of detail in some of the model parameterizations.An in depth analysis on these issues is required to identify the sources that cause these biases and errors.This will need to consider that small countries are affected by transported pollutants from neighbouring countries.Assessing (i) the effect of precursor emissions using inverse modelling techniques, (ii) the effect of a finer resolution domain, iii) the effect of other chemical mechanism or even a different air quality model will provide more information for the sources and the magnitude of the uncertainty.

Fig. 1 .
Fig. 1.Modelling domain and the regional European grouping used by the United Nations Statistics Department.

Fig. 3 .
Fig. 3. Simulated PM 2.5 component daily average concentrations for January and July 2006.

Fig. 4 .
Fig. 4. Mean Bias and RMSE for daily average Max8hO 3 mixing ratios for the European countries (dots show the distribution of O 3 monitoring stations).

Fig. 5 .
Fig. 5. Mean Bias and RMSE for hourly average NO 2 mixing ratios for the European countries (dots show the distribution of NO 2 monitoring stations).

Fig. 6 .
Fig. 6.Mean Bias and RMSE for hourly average SO 2 mixing ratios for the European countries (dots show the distribution of SO 2 monitoring stations).

Fig. 7 .
Fig. 7. Mean Bias and RMSE for daily average PM 2.5 concentrations for the European countries (dots show the distribution of PM 2.5 monitoring stations).

Table 2 .
Statistical analysis for hourly average NO 2 and SO 2 concentrations and daily average Max8hO 3 and PM 2.5 concentrations over Europe (see Fig.1for the definition of the European regions).

Table 3 .
Observed mean /Predicted mean concentration ratios for the European countries and the number of monitoring stations (in parenthesis).