Aerosols in the CALIOPE air quality modelling system : evaluation and analysis of PM levels , optical depths and chemical composition over Europe

The CALIOPE air quality modelling system is developed and applied to Europe with high spatial resolution (12 km× 12 km). The modelled daily-to-seasonal aerosol variability over Europe in 2004 is evaluated and analysed. Aerosols are estimated from two models, CMAQv4.5 (AERO4) and BSC-DREAM8b. CMAQv4.5 calculates biogenic, anthropogenic and sea salt aerosol and BSCDREAM8b provides the natural mineral dust contribution from North African deserts. For the evaluation, we use daily PM10, PM2.5 and aerosol components data from 55 stations of the EMEP/CREATE network and total, coarse and fine aerosol optical depth (AOD) data from 35 stations of the AERONET sun photometer network. Annual correlations between modelled and observed values for PM 10 and PM2.5 are 0.55 and 0.47, respectively. Correlations for total, coarse and fine AOD are 0.51, 0.63, and 0.53, respectively. The higher correlations of the PM 10 and the coarse mode AOD are largely due to the accurate representation of the African dust influence in the forecasting system. Overall PM and AOD levels are underestimated. The evaluation of the aerosol components highlights underestimations in the fine fraction of carbonaceous matter (EC and OC) and secondary inorganic aerosols (SIA; i.e. nitrate, sulphate and ammonium). The scores of the bulk parameters are significantly improved after applying a simple model bias correction based on the observed aerosol composition. The simulated PM10 and AOD present maximum values over the industrialized and populated Po Valley and Benelux regions. SIA are dominant in the fine fraction representing up to 80 % of the aerosol budget in latitudes north of 40 ◦ N. In southern Europe, high PM10 and AOD are linked to the desert dust transport from the Sahara which contributes up to 40 % of the aerosol budget. Maximum seasonal ground-level concentrations (PM10> 30 μg m−3) are found between spring and early autumn. We estimate that desert dust causes daily exceedances of the PM 10 European air quality limit value (50 μg m−3) in large areas south of 45 ◦ N with more than 75 exceedances per year in the southernmost regions.


Introduction
Atmospheric aerosols or particulate matter (PM) are highly variable in space and time due to the variety of their sources and their fast removal from the atmosphere (from days to weeks).Aerosols contribute to adverse human health effects (WHO, 2005), including increased morbidity and mortality arising from altered respiratory and cardiovascular function (Pope et al., 2009;Medina et al., 2009) and have a significant, yet uncertain effect on climate from regional to global scales (IPCC, 2007).To understand the wide-ranging effects of aerosols, it is important to characterize them with high spatial and temporal resolution.
The European Union (EU) directive on ambient air quality and cleaner air, which came into force in June 2008 and was transposed into the national legislation of each Member State by June 2010 (European Commission, 2008), introduced daily and annual PM 10 limit values of 50 µg m −3 and 40 µg m −3 , respectively, from year 2010.Also this directive introduces a PM 2.5 annual limit value of 25 µg m −3 to be assumed from year 2015.In this regard, the objectives proposed by the EU are usually less well attained in southern Europe than in northern countries (Yttry and Aas, 2006).Reports from countries around the Mediterranean Basin and Eastern Europe show high levels of atmospheric PM compared to Northern regions and some central European regions (Querol et al., 2009).Both anthropogenic (transport sector, industrial processes, power generation and biomass burning, among others) and natural emissions (African dust, resuspension, sea spray, forest fires, primary biological particles and biogenic secondary organic compounds, windblown dust), as well as orographic and climatic factors contribute to the ocurrence of (or cause) those enhanced levels.When PM 10 values are exceeded due to natural events (such as desert dust outbreaks or volcano eruptions), Member States shall inform the Commission, providing the necessary justification to demonstrate that such exceedances are due to natural events.
Air quality models are useful to understand and predict the dynamics and transport of pollutants.In recent years a number of experimental and operational air quality forecast systems have been developed around the world.Nowadays, in Europe, 23 modelling systems routinely simulate the air quality over Europe, 7 systems also operate in forecasting mode (Menut and Bessagnet, 2010): PREV'AIR, EU-RAD, EMEP-CWF, MATCH, MOCAGE, CHIMERE, and CALIOPE.
In the frame of the CALIOPE system (Baldasano et al., 2008a; http://www.bsc.es/caliope), the Barcelona Supercomputing Center-Centro Nacional de Supercomputación (BSC-CNS) operates an air quality forecasting system for Spain (at 4 km × 4 km horizontal resolution) and Europe (at 12 km × 12 km horizontal resolution) with WRF-ARW/HERMES-EMEP/CMAQ and desert dust forecasts with BSC-DREAM8b (Nickovic et al., 2001;Pérez et al., 2006a, b).In contrast to many other European modelling systems, CALIOPE includes a non-climatic representation of Saharan dust transport in its forecast mode.The forecasts are evaluated on a daily and hourly against ground-based and satellite observations to establish confidence in the modelling system predictions among users, identify problems and routinely improve the system (Baldasano et al., 2010).Pay et al. (2010) presented a full year evaluation of the CALIOPE system for the European domain for gaseous pollutants (O 3 , NO 2 and SO 2 ) and PM levels (PM 10 and PM 2.5 ) against EMEP ground-based measurements.The study shows that the skill scores of the system lie within the range of most European models and that while the dynamics of PM 2.5 and PM 10 are rather well reproduced, concentra-tions remain systematically underestimated by a factor of 2 on average.
The main objective of the present work is to complement the results of the PM evaluation presented in Pay et al. (2010) by ( 1) providing a detailed quantitative assessment of the capabilities of the CALIOPE system to simulate the daily aerosol distribution over Europe for year 2004 in terms of PM levels, aerosol optical depth (AOD) and chemical composition; (2) understanding the underestimation of the PM mass; (3) estimating and analysing the spatial and seasonal distribution of the different aerosol fractions over Europe based on model results and observations; and (4) estimating the exceedances of the EU limits due to natural desert dust.
The manuscript is organized as follows.Section 2 describes the CALIOPE system and the observational datasets used for the model evaluation.In Sect. 3 we use aerosol chemical composition from EMEP/CREATE and AOD from the AERONET network to identify the origin of the discrepancies in PM levels.The evaluation of the AOD is particularly useful for validating and analysing the capabilities of the modelling system to reproduce long-range transport of desert dust from North Africa.In Sect. 4 we estimate and analyse the spatial and seasonal distribution of the different natural and anthropogenic aerosol fractions over Europe by applying a simple bias correction to the model based on the results of the evaluation.African dust transport and its contribution over Europe are analysed in detail.Finally, Sect. 5 summarizes the main findings of the study.

Description of the CALIOPE system
CALIOPE is a state-of-the-art air quality modelling system that integrates an emission-processing model (HERMES-EMEP), a meteorological model (WRF-ARW), a chemical transport model (CMAQ) and a mineral dust model .The configuration used in this work is described in Pay et al. (2010).In the sections below, the models utilized to estimate aerosols (i.e.CMAQ and BSC-DREAM8b) are described in detail.

Photochemical model: CMAQ
The Models-3 Community Multiscale Air Quality Modeling system (CMAQ; Byun and Ching, 1999;Binkowski, 1999;Byun and Schere, 2006) is used to study the behaviour of air pollutants from regional to local scales.The CMAQ version 4.5 used in this study has been extensively evaluated under various conditions and locations (Jiménez et al., 2003;Roy et al., 2007;Appel et al., 2007Appel et al., , 2008)).Following the criteria of Jiménez et al. (2003) the Carbon Bond IV chemical mechanism is applied (CBM-IV, Gery et al., 1989).The production of sea salt aerosol (SSA) is implemented as a function of wind speed and relative humidity (Gong, 2003;Zhang et al., 2005) through the AERO4 aerosol module.The AERO4 module distinguishes among different chemical aerosol components namely nitrate, sulphate, ammonium, elemental carbon and organic carbon with three subcomponents (primary, secondary anthropogenic and secondary biogenic), sodium and chlorine.Unspecified anthropogenic aerosols and aerosol water are additionally kept as separate components.Unspecified aerosols consist of noncarbon atoms associated with organic carbon as well as PM emissions that are not explicitly speciated in the HERMES-EMEP emission model, which includes traces elements, primary ammonium and other unidentified mass in the speciation profiles (Appel et al., 2007).Aerosols are represented by three size modes (Aitken, accumulation and coarse mode), each of them assumed to have a lognormal distribution (Binkowski and Roselle, 2003).Secondary inorganic aerosols (SIA) are generated by nucleation processes and consists of nitrate, ammonium and sulphate aerosols.Secondary organic aerosol (SOA) can be formed from aromatics (anthropogenic organic aerosols) and terpenes (biogenic organic aerosols; Schell et al., 2001).An assumption of the CMAQv4.5 model is that organics influence neither the water content nor the ionic strength of the aerosol particles (Binkowski and Roselle, 2003).The aerosol microphysical description is based on a modal aerosol model (Binkowski and Roselle, 2003) using the ISORROPIA thermodynamic equilibrium model (Nenes et al., 1998).For a more complete description of the processes implemented in CMAQ, the reader is referred to Byun and Schere (2006).
AOD at 550 nm from CMAQ outputs is calculated using a simple approach described by Malm et al. (1994) and Binkowski and Rosselle (2003).The method is known as the "reconstructed mass-extinction method".The extinction coefficient (β ext ), which is a function of wavelength (λ), is a sum of the attenuation by scattering (β s ) and absorption (β a ).The model-predicted AOD is calculated by summing the product of the extinction and layer thickness ( Z i ) over the number of layers (N) in the column: We estimate the model-predicted β ext = β s + β a using this semi-empirical approach for extinction calculation which depends on aerosol mass and humidity as follows: (2) where the brackets in the above equations indicate mass concentration in mg m −3 and all coefficients in the equation represent the specific extinction cross section at 550 nm (in m 2 mg −1 ).When implementing the aforementioned equation, the term in brackets is determined by adding ammonium mass (NH + 4 ) plus sulphate mass (SO 2− 4 ) plus nitrate mass (NO − 3 ).Organic carbon (OC) is taken as the sum of all non-light absorbing organic species.Light absorbing carbon is elemental carbon (EC).PM fine mass is taken as the unspeciated portion of PM 2.5 and the fine fraction of sodium (Na + ) and chloride (Cl − ), and the PM coarse mass represents all coarse mode aerosols that includes the unspeciated portion of PM 10 and the coarse fraction of sodium (Na + ) and chloride (Cl − ).The mass concentration for each of these species is directly obtained from CMAQ.The relative humidity correction factor, f (RH), takes into account that the growth and phase change of hygroscopic particles affect their light-scattering effiency (Malm et al., 1994).f (RH) is parameterized from data published by Tang et al. (1981) as a function of the relative humidity from the WRF-ARW meteorological model and it varies between 1 (at low RH) and 21 (at RH = 99 %).

Mineral dust model: BSC-DREAM8b
The Dust REgional Atmospheric Model (DREAM; Nickovic et al., 2001) was designed to simulate and/or predict the atmospheric cycle of mineral dust.The updated model version BSC-DREAM8b used here is described in Pérez et al. (2006a, b).It includes an 8-bins size distribution within the 0.1-10 µm radius range according to Tegen and Lacis (1996).
The BSC-DREAM8b model has demonstrated its capabilities in a number of validation studies performed so far using data from observation networks such as the European Lidar Network EARLINET, the AERONET/PHOTONS sunphotometer network, satellite and ground-level PM levels (e.g.Pérez et al., 2006a;Amiridis et al., 2009;Papanastasiou et al., 2010).Such validations outline the good skills of the model concerning both the horizontal and vertical extent of the dust plume in Europe and the Mediterranean Basin.Additionally, the model has been validated and tested against measurements at source regions for SAMUM-I (Haustein et al., 2009) and BODEX campaigns (Todd et al., 2008).
In BSC-DREAM8b, the AOD is calculated from the column mass loading by: where for each size bin k: ρ k is the particle mass density, r k is the effective radius, M k is the column mass loading and Q ext (λ) k is the extinction efficiency which is calculated using Mie scattering theory.

Simulation
The CALIOPE system is run at 12 km × 12 km grid resolution covering the European domain (Fig. 1).The simulation consists of 366 daily runs covering an annual cycle (2004).CALIOPE uses an inventory of anthropogenic emissions of SO 2 , NO x , NMVOC, CO, PM 10 , PM 2.5 , and NH 3 derived from the 2004 annual EMEP emission database (EMEP, 2007).Raw emission data are processed by the High-Elective Resolution Modeling Emission System (HERMES, see Baldasano et al., 2008b) in order to include a comprehensive set of emissions in the air quality model.The inventory distinguishes the source categories following the Selected Nomenclature Air Pollution (SNAP).Disaggregation of EMEP (50 km resolution) data is performed in space (12 km × 12 km) and time (1 h).Natural PM emissions (such as wildfire emissions) and primary biogenic emissions (such as pollen, bacteria, fungal, and fern spores, viruses, fragments of animals and plants, etc.) containing organic compounds are not included in the emission model.
The Advanced Research Weather Research and Forecasting (WRF-ARW) Model v3.0.1.1 (Michalakes et al., 2004;Skamarock and Klemp, 2008) is used to provide the meteorology to the chemical transport model.WRF-ARW v3.0.1.1 is configured with a grid of 479 × 399 points and 38 σ vertical levels (11 characterizing the PBL).The model top is defined at 50 hPa to resolve properly troposphere-stratosphere exchanges.WRF-ARW initial and boundary conditions at intervals of 6 h are based on the Final Analyses of the National Centers of Environmental Prediction (FNL/NCEP; at 1 • × 1 • ) at 12:00 UTC.
The CMAQ horizontal grid resolution corresponds to that of WRF-ARW.Its vertical structure was obtained by a collapse from the 38 WRF-ARW layers to a total of 15 layers steadily increasing from the surface up to 50 hPa with a stronger concentration within the PBL.CMAQ boundary coditions for gas-phase species are based on the global climate chemistry model LMDz-INCA2 (Szopa et al., 2009) developed by the Laboratoire des Sciences du Climat et l'Environnement (LSCE).Monthly mean data for the year 2004 are interpolated in the horizontal and vertical dimensions to force the major chemical concentrations (i.e.O 3 , NO, NO 2 , HNO 3 , HCHO, H 2 O 2 , PAN and CO) at the boundaries of the domain (Piot et al., 2008).A detailed description of the INteractive Chemistry and Aerosol (INCA) model is presented in Hauglustaine et al. (2004) and Folberth et al. (2006).
The initial state of dust concentration in the BSC-DREAM8b model is defined by the 24-h forecast from the previous-day model run.The FNL/NCEP (at 1 • × 1 • ) at 00:00 UTC are used as initial conditions and boundary conditions at intervals of 6 h.The resolution is set to 1/3 • in the horizontal and to 24 layers extending up to approximately 15 km in the vertical.The domain of simulation covers northern Africa, the Mediterranean Sea, southern Europe and Middle East.An offline coupling is applied to the calculated concentrations of PM over the European domain from CMAQ outputs (Jiménez-Guerrero et al., 2008).

Surface PM and chemical composition from EMEP/CREATE Networks
In this study, simulated PM concentrations are compared with ground-based measurements from the European Monitoring and Evaluation Programme (EMEP) monitoring network for year 2004.EMEP stations are located at a minimum distance of approximately 10 km from large emission sources (Larssen et al., 1999).EMEP stations are assumed to be representative of regional background concentrations (Torseth and Hov, 2003).A total of 17 and 26 stations for PM 2.5 and PM 10 respectively have been used to evaluate the model predictions.Details on the location of the EMEP stations used for this comparison can be found in Appendix Table A1.
Modelled aerosol concentrations of chemical species are compared with ground-based measurements provided by EMEP and the FP5/GMES project CREATE at 55 rural background stations.The uncertainty of PM measurements strongly depends on the method and chemical composition of the collected aerosol.Usually the collected filters are weighted at 50 % relative humidity (Tsyro, 2005).Error sources leading to a bias are potential losses of semivolatile compounds (particularly ammonium nitrate and carbonaceous aerosols) from the filters at temperatures higher than 20 • C as well as gas condensation on filters which could introduce positive measurements artefacts.SIA components can be measured with an uncertainty of about ±10 % for major species (Putaud et al., 2004).In this context, European legislation (2008/50/EC) establishes that measurement groups have to demonstrate that the uncertainty of PM 10 and PM 2.5 meets the quality objective of 25 %.However legislation does not establish any quality objective for uncertainty of chemical species yet.SIA components collected within CREATE are measured by ion chromatography in which the error is usually within 10 % (Putaud et al., 2000).
One source of underestimation of SIA components is related to the size fraction of aerosol components in both measurements and the model.The observed concentrations are available as total mass, without any discrimination of size since EMEP sites typically used filter-packs to measure sulphate, nitrate and ammonium and captured particles that are approximately in the PM 10 fraction.Conversely, the CMAQv4.5 model considers speciated inorganic aerosol only in the PM 2.5 fraction.One exception is marine sulphate aerosol which is present also in the coarse fraction.The comparison of carbonaceous matter is conducted using the PM 2.5 fraction of the modelled and measured values.
In the present work, measured PM 10 , PM 2.5 and their chemical aerosol components (i.e.sulphate, nitrate, ammonium, sea salt and carbonaceous matter) are available on a daily (except at Birkenes station for EC and OC that is available on a weekly basis).The main characteristics and location of EMEP/CREATE stations are described in Ap-pendix A1 and their locations are displayed in Fig. 1.The CALIOPE system is evaluated at 55 stations: 54 for sulphate (SO 2− 4 ), 28 for nitrate (NO − 3 ), 16 for ammonium (NH + 4 ), 10 for sodium (Na + ), 6 for chloride (Cl − ) and 3 for carbonaceous matter (EC and OC).The measurements are well documented and available on the EMEP web page (http://www.emep.int).

Aerosol optical depth from AERONET Network
The optical properties of the aerosol in the entire atmospheric column are routinely observed within the Aerosol Robotic Network (AERONET, Holben et al., 1998;Smirnov et al., 2000).The network imposes standardization of instruments, calibration, processing and distribution.
Sun photometer measurements of the direct (collimated) solar radiation provide information to calculate the columnar AOD.The typical uncertainty in the direct-sun AOD measurements ranges from 0.01 to 0.02 with higher errors in the ultraviolet spectral range (Dubovik et al., 2000).Additionally, direct-sun AOD processing includes the Spectral Deconvolution Algorithm (SDA) retrievals (O'Neill et al., 2003).The SDA algorithm yields fine (sub-micron) and coarse (super-micron) AOD at a standard wavelength of 500 nm (AOD fine and AOD coarse , respectively).The amplitude of the errors of the derived parameters varies as the inverse of the total AOD.In addition to measurement errors, there are errors in the AOD retrieval due to the uncertainty in the assumed values of the spectral curvature in each mode (O'Neill et al., 2001) which are most critical in coarse mode dominated conditions.
All operational stations within our target domain that collected data on at least 30 h in 2004 have been selected for the model evaluation.Appendix A2 and Fig. 1 describe and show the location and the main characteristics of the 35 selected AERONET sites within our study region.Appendix A2 lists additional information including type of site, observation periods, percentage of cloud-screened data and the availability of SDA products.
Quality-assured direct-sun data in the 440-870 nm wavelength range is used to calculate the AOD at 550 nm obtained by the Ångström's equation adjustment.The contribution of each aerosol fraction is analysed using the SDA retrieval products (AOD fine and AOD coarse ).

Model evaluation
There is a number of metrics that can be used to examine performances of air quality models (US EPA, 1984, 1991;Cox and Tikvart, 1990;Weil et al., 1992;Chang and Hanna, 2004;Boylan and Russell, 2006;Dennis et al., 2010).Correlation coefficient (r), root mean squared error (RMSE), mean bias (MB) and mean normalized bias error (MNBE) are commonly used by the modelling community.For the evaluation of PM concentrations, Boylan and Russell (2006) suggested the mean fractional bias (MFB) and the mean fractional error (MFE).Additionally, they propose model performance goals (the level of accuracy that is considered to be close to the best a model can be expected to achieve) and criteria (the level of accuracy that is considered to be acceptable for modelling applications) that vary as a function of concentration and extinction.A model performance goal is met when both MFE and MFB are less than or equal to 50 % and ±30 %, respectively, and a model performance criterion is met when both MFE and MFB are less than or equal to 75 % and 60 %, respectively.The model-to-data statistics MB, MNBE, RMSE, MFE and MFB are selected for the present study, together with the measured and modelled mean and the correlation coefficient.A description of these statistics is included in Appendix A3.Comparisons are made for annual and monthly basis as well as 4 seasonal periods: winter (DJF) corresponding to December, January and February, spring (MAM) corresponding to March, April and May, summer (JJA) corresponding to June, July, August and autumn (SON) corresponding to September, October and November.Measured PM 10 , PM 2.5 and their chemical aerosol components (i.e.sulphate, nitrate, ammonium, sea salt and carbonaceous matter) from EMEP/CREATE networks are used for the model comparison on a daily basis (except Birkenes station for EC and OC).Moreover, since AERONET data are acquired at 15-min intervals on average, all AERONET measurements within ±30 min of the model outputs have been extracted and used for the model comparison on an hourly and daily basis.
Since differences in the comparison between simulations and observations are rather small among sites located in the same region, statistical indicators have been averaged over the regions indicated in Fig. 1 by colours and defined in Appendix A1 and A2.

Ground level PM 2.5 and PM 10 concentrations
Before evaluating PM composition, it is important to ensure that reactive gases are well reproduced, since a major fraction of ambient PM arises from atmospheric gas-to-particle conversion (Meng et al., 1997).In Pay et al. (2010) the CALIOPE model output for gas and particulate phase concentrations are compared with ground-based measurements from the EMEP monitoring network for year 2004 over Europe.The evaluation showed that the modelling system is able to reproduce reasonably well daily variations of gas phase pollutants (SO 2 , NO 2 and O 3 ) as well as their spatial distribution and seasonal cycle.
Table 1 and Fig. 2a, b and f, g depict the annual and seasonal statistical results of the evaluation of PM and the all-European time series of the model and observations calculated on a daily basis.The model presents a systematic negative bias for PM 2.5 and PM 10 (annual MB = −6.45µg m −3 and −10.23 µg m −3 , respectively), although it is able to re-produce the daily variability through the year with correlation of 0.47 and 0.55, respectively (Fig. 2a, b).
The modelled coarse fraction includes the contribution of unspeciated anthropogenic mass, sea salt and desert dust.One source of uncertainty in the PM 10 comparison comes from the assumption of dry and inert coarse mode aerosols in CMAQv4.5 with the AERO4 aerosol module.Several studies in Spain and the Mediterranean (e.g.Rodríguez et al., 2002;Querol et al., 2004Querol et al., , 2009) ) suggest that between mid-spring and mid-autumn most nitrate is present as calcium and/or sodium salts in the 2.5-10 µm fraction.Other potential forms of nitrate and sulphate, such as salts linked to the reaction of acid pollutants with dust (Rodríguez et al., 2002;Querol et al., 2004Querol et al., , 2009) ) are not considered in the simulated PM 10 .Higher PM 10 correlations are observed in summertime over southern European stations affected by Saharan dust outbreaks (see Fig. 3).The modelled desert dust contributions in the PM 2.5 and PM 10 fractions are on average 20 % and 25 %, respectively, increasing to 33 % and 35 % in the Mediterranean stations.Although the model calculates a considerable contribution of dust on both PM 2.5 and PM 10 , a higher annual correlation is observed in the PM 10 fraction.Pay et al. (2011) shows that despite the total dust mass is reasonably well captured by the BSC-DREAM8b model, it tends to overestimate the fine dust fraction (<2.5 µm).
Overall, the skills achieved by CALIOPE are in the range of those shown in other European modelling studies (see Sect. 4 of the work of Pay et al., 2010).However, the MFE and MFB for PM 2.5 and PM 10 do not meet the performance criteria proposed by Boylan and Russell (2006).Many studies have recognized the difficulty of models to simulate the mass of PM over Europe (e.g.Matthias, 2008;Pay et al., 2010).The underestimation of PM 10 is, among others, the result of a lack of fugitive dust emissions, resuspended matter, a possible underestimation of primary carbonaceous particles, the inaccuracy of SOA formation, the difficulty of representing primary PM emission from wood burning and other sources not considered in the emission inventory as pollutant sources over North Africa (Rodríguez et al., 2011) and a more general lack of process knowledge on aerosol removal and dispersion and transport processes.

PM chemical composition
Chemical composition measurements can help to identify model limitations in simulating the physical and chemical processes leading to the formation of SIA (namely, sulphate, nitrate and ammonium), SOA and SSA.For the sake of comparison with EMEP stations different size modes are lumped to obtain the total sulphate (Aitken, accumulation and coarse modes), nitrate (Aitken, accumulation and coarse modes), ammonium (Aitken and accumulation modes), chloride and sodium (accumulation and coarse modes) and carbonaceous matter concentration (i.e.organic and elemental carbon).Table 1, Fig. 2c-e, Fig. 2h-j and Fig. 3 show the The overall annual variability of the modelled sulphate concentrations (Fig. 3) agrees fairly well with measurements (r = 0.49), achieving better results in summer (r = 0.60) when sulphate concentrations reach maximum levels due to enhanced photochemistry, low air mass renovation and recir-culation at regional scale (Rodríguez et al., 2002;Querol et al., 2009).On average, sulphate levels are underestimated by 18 %, less than PM 2.5 , with winter having the largest bias.In winter, significant underestimations are present at stations in Eastern Europe (E.Eu region), where the mean bias per station ranges from −0.5 to −2.5 µg m −3 (Fig. 3).In contrast, the simulated sulphate exceeds the measurements by 0.3 µg m −3 and 0.4 µg m −3 at stations in the region of France (S.Fr and C.Fr) and near the Northern Atlantic (N.At),   respectively (Fig. 3).The highest correlations are found in the Iberian Peninsula and the Western Mediterranean Basin, ranging between 0.48 and 0.81 (Fig. 3).Pay et al. (2010) found a positive mean bias for SO 2 which suggests that SO 2− 4 formation in the model is often limited by oxidant availability and not always by SO 2 availability.Winter underestima-tion of sulpahte is a common issue in most European models representing a direct couplet of sulfur chemistry with photochemistry and has been previously detected with CMAQv4.5 over Europe (Matthias, 2008).This feature may be explained by a lack of model calculated oxidants or missing reactions (Kasibhatla et al., 1997).On average, modelled nitrate concentrations reproduce the daily variability of the observations through the year (r = 0.58), presenting higher concentrations in winter and lower levels in summer due to its thermal instability (Querol et al., 2009).Nitrate formation is limited by the availability of nitric acid over land.The underestimation of nitrate is around 50 %, peaking in winter and spring (MB = −1.2µg m −3 ) and may be partly attributed to an underestimation of NO 2 as outlined and discussed in Pay et al. (2010).The largest underestimations are located over Eastern Iberian Peninsula-Western Mediterranean (E.IP-W.Med; except at Montseny station) and the Central Mediterranean (C.Med) area with a MB of −1.8 µg m −3 and Eastern Europe (E.Eu region, except at Illmitz and Sniezka stations) with MB of −1.5 µg m −3 (Fig. 3).As in the case of sulphate, the model presents the best correlation coefficients over the Western Iberian Peninsula ranging from 0.40 to 0.65 (Fig. 3), and a MB below 1 µg m −3 .In contrast to calcium-and sodiumnitrate, ammonium-nitrate mostly occurs in PM 2.5 , whereas calcium-and/or sodium-nitrate mostly occur in the coarse PM 2.5−10 mode which are known to be significant in Spain from mid-spring to mid-autumn (Rodríguez et al., 2002;Querol et al., 2004Querol et al., , 2009)).The fact that CMAQv.4.5 (AERO4) does not consider the formation of coarse calciumand/or sodium-nitrate; it could be related with the underestimations observed in Eastern Spain, particularly in summer (Fig. 3).Moreover, correlations between the simulated and the observed values are rather low in summer in Spain (Fig. 3).
In air masses with a continental signature, nitrate and sulphate are associated with ammonium (Schaap et al., 2004;Querol et al., 2009).Atmospheric ammonia is first neutralized by sulphuric acid to form ammonium sulphate.Remaining ammonia may then combine with nitric acid to form ammonium nitrate (Seinfeld and Pandis, 1998).94 % of NH 3 total emissions are attributed to agriculture and livestock (EMEP, 2007).Livestock sources vary during the year since volatilization of NH 3 from the animal waste is a function of temperature (Gilliland et al, 2003).Seasonality in NH 3 emission is expected since field application of fertilizers occurs during specific seasons (Asman, 2001).In the CALIOPE system over Europe, annual emissions of NH 3 are derived from the 2004 annual EMEP emission database (EMEP, 2007).However, detailed agricultural registers are not generally available in many countries (e.g.Menut and Bessagnet, 2010).The evaluation of ammonium shows that the annual trend is correctly reproduced (r = 0.62, see Table 1, Figs. 2e and 3) and it is underestimated by 37 %, with the highest bias occurring in winter (MB = −0.68 µg m −3 ).
For sea salt, the modelling system reproduces the daily variability of sodium and chloride with correlations of 0.67 and 0.45, respectively (Table 1 and Fig. 4).It should be noted that the results are strongly biased towards measurements obtained in northern Europe, since 9 out of 11 stations are located there (see Appendix A1).There are 4 coastal stations where the model simulates slightly higher chloride and sodium concentrations than those measured while at continental areas sea salt tends to be underestimated.Correlation for sodium is higher than for chloride since sodium is considered as inert in the atmosphere.One source of uncertainty in AERO4 is the consideration of coarse mode aerosols as dry and inert.This approach does not allow for several aerosol processes, such as the replacement of chloride by nitrate in mixed marine/urban air masses (Kelly et al., 2010) especially intense in summer when nitric acid is released by the thermal instability of ammonium nitrate (Querol et al., 2004).Moreover, degassing of Cl − is not implemented in the model, and heterogeneous reactions are not taken into account.The coarse mode in the AERO4 module in CMAQv4.5 is implemented in a non-interactive way.That is, fine particles do not coagulate with coarse particles, nor do coarse particles coagulate with each other.
Both OC and EC can contribute significantly to PM 2.5 and PM 10 in urban and kerb sites, and their mass fraction might be higher than the sum of the inorganic components (Putaud et al., 2004).Unfortunately, a detailed investigation of the organic aerosol cannot be done for the present study since there are only very few measurements available that cover a longer time.Therefore OC and EC could only be evaluated at three EMEP/CREATE stations: Birkenes (NO0001), Melpitz (DE0044) and Montseny (ES1778) in 2004 on an annual basis.The results of the carbonaceous aerosol are far from being representative and currently subject to reevaluation by the EUSAAR project.The observed values are approximately a factor of 4 higher than the modelled values.This factor is higher than that obtained by Matthias (2008) at Birkenes in 2001 (factor of 3) partly because carbonaceous particles from biomass burning emissions were taken into account in the aforementioned study.There are other studies that demonstrate that SOA are underestimated by current models by a factor of 6 (Volkamer et al., 2006).The large uncertainties are associated with (1) probable underestimation of primary carbonaceous emission (Cooke and Wilson, 1996;Bond et al., 2004;Tsyro et al., 2007;Monsk et al., 2009) and (2) the state-of-the-science concerning SOA formation pathways (Eder and Yu, 2006;Edney et al., 2007;Appel et al., 2008).
From all anthropogenic emissions used in CALIOPE primary traffic emissions provide the highest contribution of OC and EC (80 % of PM 2.5 ), followed by combustion in energy and transformation industries (60 % of PM 2.5 ).Additionally, the absence of some natural PM sources (such as wildfire emissions) and primary biogenic emissions (such as pollen, bacteria, fungal and fern spores, fragments of animals and plants, viruses, etc.) contributes to OC underestimation.Wildfire emissions during 2004 were important in southern Europe during summer (European Commission, 2005).Additionally, the traditional 2-product SOA model adopted by CMAQv4.5 (Binkowski and Roselle, 2003) does not include SOA formation from isoprene and sesquiterpenes.The absence of the isoprene-SOA route on SOA may impact significantly in southern Europe during summer where the predominant vegetation types favour isoprene as the main biogenic volatile organic compounds (VOC; Keenan et al., 2009).
As a summary, the modelling system reproduces reasonably well the daily variability of the main aerosol components in Europe.Their concentrations are in most cases strongly underestimated.The most important underestimations are observed for total carbonaceous material (i.e.OC and EC).The total amount of SIA is on average underestimated by 18-50 %.SSA is underestimated inland and overestimated at coastal sites of northern Europe.

Aerosol optical depth
The evaluation of AOD is particularly useful for evaluating the capabilities of the modelling system to reproduce European regional transport and the long-range transport of desert dust from North Africa.
The modelling system is quantitatively compared against direct-sun AOD measurements (AOD) and the AOD mode products from the SDA retrieval (AOD fine and AOD coarse ).Modelled AOD fine includes ammonium, sulphate, nitrate, SSA, OC, EC, unspeciated fine fraction from CMAQ and desert dust from BSC-DREAM8b.Modelled AOD coarse is the sum of the super-micron fraction of desert dust from BSC-DREAM8b and SSA and unspeciated coarse fraction from CMAQ.Table 2 and Fig. 5 show the annual and seasonal statistics, and the temporal evolution of model and AERONET measurements for AOD, AOD fine and AOD coarse calculated on an hourly and daily basis.
The model reproduces the AOD hourly and daily variability with average correlations of 0.51 and 0.56, respectively (Table 2, Figs.5a and 6) and underestimates the hourly and daily AOD by 41 and 38 %, respectively (Table 2, Figs.5a  and 6).
If we take a closer look to the desert dust component and its role on the model skills, its influence is remarkable.When considering only CMAQ outputs the AOD underestimation rises up to 61 % for hourly values and 58 % for daily values and the annual correlation decreases to 0.39 for the hourly values and to 0.40 for daily values.The differences between CMAQ-alone and CMAQ+BSC-DREAM8b are larger in the Iberian Peninsula and Mediterranean sites due to the proximity to the African continent.Correlations remarkably improve from 0.16 to 0.59, 0.39 to 0.58, 0.31 to 0.49 and 0.05 to 0.50, for the Western Iberian Peninsula, Eastern Iberian Peninsula-Western Mediterranean, Central Mediterranean and Eastern Mediterranean, respectively for hourly values.Averaging over the entire study region, desert dust contributions represent about 35 % of the AOD and over the Iberian Peninsula and the Mediterranean about 51 % on average.These results are consistent with the aerosol characterization presented in Basart et al. (2009).Because the evaluation of the aerosol model performance is hampered by the lack of routine data above the surface, satellite observations offer new opportunities for model evaluation.In Fig. 7, the seasonal averages for 2004 of the collection C005 AOD data (cloud free) at 550 nm from MODIS/Aqua sensor (Levy et al., 2003;Remer et al., 2005) are qualitatively compared with the modelled aerosol fields.Additionally, superimposed to MODIS values, the AERONET seasonal mean values are included showing moderate differences between both observational datasets.Among other causes, this is partly because the satellite product is obtained from 2 images per day in contrast to more frequent AERONET measurements during the day.
In general, the CALIOPE system reproduces the main seasonal AOD patterns observed in MODIS despite the important underestimations at hot spots located in Northern Italy, Eastern and North-western Europe as shown in the comparison with AERONET data.The dust transport from North Africa is well captured with maximum AOD values in the Eastern Mediterranean in spring shifting to the Central and Western Mediterranean in summer.
The aerosol fields obtained with CALIOPE are similar to those shown in other European modelling studies (e.g.POLYPHEMUS; Tombette et al., 2008;CHIMERE, Péré et al., 2010).In another study with CMAQ (Matthias, 2008) 3 are applied to sulphate, nitrate and ammonium and EC + OC in the corrected-modelled series.
AERONET observations.In contrast to our results, the best scores in this study were achieved in northern sites where desert dust contributions are smaller, which again highlights the importance of desert dust in southern latitudes.The evaluation of the POLYPHEMUS system (Tombette et al., 2008), which also did not consider desert dust, against 19 AERONET stations underestimated the hourly AOD average from 0.02 to 0.07 and correlations ranged from 0.40 to 0.87 obtaining the best results in the northern European sites.In the evaluation of the CHIMERE model against 13 AERONET stations presented in Péré et al. (2010), correlations lied within the range of 0.50-0.74.The values obtained in the present work lie within the range of these studies with correlations of 0.40-0.72 depending on the region.We take a closer look now at the fine and coarse components of the AOD.The annual and seasonal statistics for AOD fine and AOD coarse are presented in Table 2 and Fig. 6 and the temporal series are shown in Fig. 5b and c, respectively.Desert dust contributions to AOD fine and AOD coarse are on average 22 % and 88 %, respectively.The largest discrepancies between model and data are associated to AOD fine (MB = −0.13 and RMSE = 0.19, Table 2), although the hourly variability is reasonably well captured (r = 0.52, see Figs. 5b and 6).The best scores are found at SMHI site in the Nordic region (r = 0.70, RMSE = 0.04 and MB = −0.03)where no desert dust is present in the fine mode AOD and where the frequent background situation is associated with low values.The largest error in AOD fine is found for Northern Italy (RMSE = 0.29 and MB = −0.20).The correlation remains moderately high though (∼0.52) indicating that the modelling system is able to reproduce reasonably well the background AOD and the occurrence of enhanced (or peak) concentrations (or episodes).Northern Italy includes one of the most polluted regions in Europe, the Po Valley (Mélin and Zibordi, 2005) and the model indicates almost no contribution from dust (less than 4 %) in agreement with Gobbi et al. (2007).In this area, the air stagnation in a mountainsurrounded valley favours photochemical reactions that produce SIA and fine mode growth (Gobbi et al., 2007) together with frequent humid conditions in winter which favours nucleation and growth of aerosols.
The daily variability of AOD coarse is well reproduced (r = 0.63, see Fig. 5c).The best scores are found in the Mediterranean Basin (0.6 < r < 0.8, RMSE < 0.09 and MB ∼ −0.04, on average in hourly mean values for the entire Basin) particularly in summertime coinciding with the maximum AOD coarse values and maximum activity in the desert dust sources (Middelton and Goudie, 2001).The largest discrepancies are found in Nordic countries (r = 0.06, RMSE = 0.03 and MB = −0.02)and Eastern Europe (r = 0.18, RMSE = 0.10 and MB = −0.06).In general, the highest AOD coarse values are linked to the presence of desert dust.However, some coarse events of coarse particles in North-Western and Eastern Europe are not captured by the modelling system (not shown here).Theses episodes may be linked to the occurrence of the coarse sodium and calcium nitrate because high levels of sulphate neutralize ammonium (Querol et al., 2009).
The ESCOMPTE experiment (Mallet et al., 2003), which investigated the microphysical and optical properties of aerosols around Marseille in the summer 2001, showed that 90 % of the light extinction was due to anthropogenic aerosol and only 10 % was due to natural aerosol (SSA and desert dust).44 % of the anthropogenic extinction was due to ammonium sulphate, followed by 20 % from EC and 21 % from organic matter.Nitrate aerosol had a weak contribution of about 5 %.In our study, the Western Mediterranean sites (i.e.Avignon, Carpentras, Toulon and Villefrange) in summer, show a 28 % contribution of natural aerosol (SSA and desert dust) and a 72 % contribution of anthropogenic aerosol to the extinction.In our model, 54 % of the anthropogenic extinction is due to ammonium sulphate followed by nitrate (∼13 %) and EC (∼6 %) and OC (∼9 %).These results reveal a significant underestimation of total carbonaceous material species which can be partly explained by the influence of local emissions and natural sources as indicated in the Sect.3.2.The uncertainties associated to the emission inventory for black carbon may be an important cause for the discrepancy (Baldasano et al., 2008b).

Model bias correction
Several studies have demonstrated the benefit of adjusting site-specific air quality model predictions using observational data to reduce systematic model error (e.g.Hogrefe et al., 2006;Djalalova et al., 2010).Bias-adjustment strategies range from the relatively simple mean bias and multiplicative ratio adjustments used by McKeen et al. (2005) to the more complex Kalman filter techniques (Manders et al., 2009;Kang et al., 2010;Sicardi et al., 2011).In this section we provide an estimation of the spatial and seasonal distribution of the different aerosol components over Europe with our model results including an a posteriori correction.The evaluation consistently reveals a similar degree of underestimation of the fine fraction on both surface levels and columnintegrated values (AOD) and the correction applied is based on the results of the evaluation of the chemical aerosol composition presented in Sect.2.3.1.We calculate spatially homogeneous multiplicative correction factors per aerosol species and apply it to the mass concentrations in the model.The main limitation of this method is the application of a spatially homogeneous correction factor for a large region such as Europe.However, as shown in the model evaluation results, the seasonal variability has a stronger impact than the geographical differences in the formation of the secondary atmospheric aerosols.As described below the modelled bulk parameters (PM and AOD levels) significantly improve after correcting the bias of each aerosol species individually.
For carboneous compounds, a spatially homogeneous mean annual multiplicative correction factor of 4 is estimated.For the SIA (i.e.sulphate, nitrate and ammonium) seasonal correction factors are calculated from the modelled and observed mean daily values.We employ a weighted sum of the squared difference between the modelled and observed values.We minimize this sum to identify the magnitude of the aerosol concentration being in optimal agreement with the observations.Table 3 shows the spatially homogeneous multiplicative correction factors obtained for each chemical species.The bias corrected simulation is then evaluated against the bulk PM and AOD observations from EMEP and AERONET (Table 4).
The corrected PM 2.5 and PM 10 results present a much better agreement with the observations if correction factors are applied (Fig. 2 and Table 4).The annual underestimations are reduced from 53 to 19 % for PM 2.5 and from 58 to 35 % for PM 10 .Correlations increase from 0.47 to 0.61 for PM 2.5 and from 0.57 to 0.62 for PM 10 .MFE and MFB for the corrected PM 2.5 fall within the performance goal and the corrected PM 10 within the performance criterion proposed by Boylan and Russell (2006).Correlation coefficient RMSE, MFB and MFE of PM 2.5 and PM 10 with correction factors are significantly better than those obtained in Pay et al. (2010) by applying a factor of 2 to total PM 2.5 and PM 10 .
Underestimations are reduced from 57 to 32 % for the hourly AOD fine and from 41 to 13 % for the hourly AOD.Correlation coefficients increase from 0.52 to 0.56 for the hourly AOD fine and from 0.51 to 0.56 for the hourly AOD (Table 2 and Figs. 5 and 6).Similar results are obtained for daily AOD and AOD fine (see Table 4).Finally, Fig. 7 highlights the closer agreement between the model corrected seasonal AOD distribution and the MODIS/Aqua satellite product in comparison with the uncorrected model estimates.The corrected simulation provides more realistic information for the analysis of the aerosol spatial and temporal patterns over Europe.

Spatial and seasonal distribution of PM concentrations and AOD
Figure 8 depicts the corrected-modelled annual mean of PM 2.5 , PM 10 , AOD and AOD fine .The highest mean aerosol concentrations (AOD > 0.3 and PM 10 > 30 µg m −3 ) found over the Po Valley and the Benelux region (Belgium, The Netherlands, Luxembourg) are dominated by fine anthropogenic aerosols.A second maximum is found over South-eastern Europe (AOD ∼ 0.2) and southern Europe (AOD > 0.2 and PM 10 > 20 µg m −3 ).Aerosol concentrations decrease towards the North and North-western Europe reaching the lowest values in the northern region (AOD < 0.15) in agreement with the modelling study of Schaap et al. (2004).Low values are also found over the major European mountain chains, i.e. the Alps, Massif Central, the Pyrenees and the Carpathians with AOD < 0.1 and PM 10 < 10 µg m −3 .We distinguish some common features in the columnintegrated and surface distributions such as the high values over the North of Italy and the Netherlands.There are also significant contrasts as for instance the maritime areas in the North Atlantic region and Eastern Europe.Differences over the North Atlantic region are associated to the relatively low SSA extinction (Dubovik et al., 2002)

in comparison
Table 4. Seasonal and annual statistics obtained with CALIOPE and the seasonal correction factors applied to sulphate, nitrate, ammonium and EC + OC over Europe for 2004 at the EMEP stations for PM 2.5 (in µg m −3 ) and PM 10 (in µg m −3 ) and at the AERONET stations for AOD fine and AOD.Winter: January, February and December; spring: March, April, May; summer: June, July, August; autumn: September, October, November.The number of data points indicates the number of pair measurement-model used to compute the statistics.The calculated statistics are: measured mean for available data, modelled mean for the whole year, correlation coefficient (r), mean bias (MB), mean normalize bias error (MNBE), root mean square error (RMSE), mean fractional bias (MFB) and mean fractional error (MFE).

Data
Obs to its higher impact on PM 10 values.Over the continent, differences between PM surface concentrations and AOD are partly due to the dilution conditions related to the PBL height due to changes in the vertical mixing and to the moisture content affecting the optical properties of aerosols (Gupta and Christopher, 2009).On the other hand, aerosols in the free troposphere are transported from other regions, significantly contributing to the total AOD (e.g.Amiridis et al., 2005).
Figures 9 and 10 present the mean seasonal PM 10 and AOD together with the respective fractions of anthropogenic (nitrate, sulphate, ammonium and OC + EC) and natural aerosols (SSA and desert dust).Despite SO x emissions are rather constant throughout the year, a clear seasonal trend is observed in the sulphate contribution to PM 10 and AOD during the warm season due to high temperature and low relative humidity that favour the oxidation of SO 2 and the formation of sulphate under the presence of hydroxyl radical (e.g.Querol et al., 2009).In summer, sulphate AOD reaches average values above 0.12 in Eastern Europe, with a maximum of 0.28 over Romania.High surface concentrations of sulphate are calculated in summer over the Aegean Sea (∼5 µg m −3 ) due to the shipping traffic emissions and over the Balkans (above 5 µg m −3 ) due to emissions from power plants.Nitrate levels also present a strong seasonal variability with the highest values during the cold months (Figs. 9 and 10).In winter the nitrate contribution to PM 10 and AOD reaches up to 40 % over the entire European continent.In summer the highest contributions are localised in Central Europe.Ammonium concentration shows only variations throughout the year (Figs. 9 and 10).The highest contributions to PM 10 and AOD (∼15 % and ∼20 %, respectively) are observed in winter and the lowest contributions (<10 %) in summer.Maximum surface concentrations of nitrate (∼4.6 µg m −3 ) and ammonium (∼3 µg m −3 ) are simulated over most European countries in winter, with the highest levels over the Po Valley (>8 µg m −3 and >5 µg m −3 for nitrate and ammonium, respectively).Their maximum contribution to AOD is found in the Po Valley (∼0.10) and to a lesser extent in the Benelux area (∼0.08).
EC is a primary pollutant so its spatial variability is relatively high.Major sources of EC include diesel engines, particulate heavy-duty trucks, and combustion process (including biomass and fossil fuel), thus high levels are associated with urban areas and maritime routes.Conversely, OC is emitted directly, or in a large proportion formed, from the condensation of low-volatility organic compounds.Thus, the spatial variability of OC is between that of purely primary and secondary pollutants.Major primary sources of OC include diesel and gasoline-burning engines, biomass burning and some industrial processes, so OC will be found in urban and rural background environments.OC + EC are concentrated at hot spot locations where they are found in the fine fraction with maximum PM 2.5 values of ∼6 µg m −3 and AOD fine values up to 0.01.Maximum OC + EC contributions are observed in summer (Figs. 9 and 10).OC + EC estimations should be taken carefully due to the few measurements available over Europe for the model evaluation results as well as the considerable uncertainties associated to the calculations of EC and OC.Also, SSA contributes weakly to the AOD with an average mean annual value of less than 0.01.At ground-level, the mean contribution of SSA to PM 10 in the Mediterranean Sea reaches up to 40 % (∼10 µg m −3 ) and up to 80 % (15 µg m −3 ) in the North Atlantic.
Sulphate and nitrate contribute up to 80 % of the PM 10 and AOD in latitudes beyond 41 • N (Figs. 9 and 10).Maxima SIA and carbonaceous matter are mainly concentrated at hot spot localizations in Europe where they are found in the fine fraction.In contrast to anthropogenic sources, which are mainly located in the European continent, mineral dust sources affecting Europe's PM levels are mostly found in North African deserts.One distinct feature of the CALIOPE modelling system with respect to other European systems is the inclusion of the influence of African dust on a non-climatological basis by means of the offline coupling with the BSC-DREAM8b model.Desert dust exhibits a strong seasonal variability throughout the year.Maximum desert dust transport to Europe occurs from spring to early autumn according to the model results (Figs. 9 and 10), while minimum desert dust contributions are found in winter over the entire region.In spring, an increase of dust outbreaks and high desert dust contributions (>40 %) are observed in the Central and Eastern Mediterranean regions.In summer, the maximum desert dust contribution is shifted towards the Western Mediterranean and the Southern Iberian Peninsula.In the Southeastern Iberian Peninsula, desert dust contributes about 60-70 % of the PM 10 and AOD levels which reach 40 µg m −3 and 0.2, respectively (Figs. 9 and 10).
The simulated seasonal patterns are in agreement with observational studies using ground-based (e.g.Papayannis et al., 2008;Basart et al., 2009) and satellites observations (e.g.Barnaba and Gobbi, 2004;Antoine and Nobileau, 2006).Long-range transport of Saharan dust across the United Kingdom appears to be much less frequent than for Southern and Central Europe (Ryall et al., 2002) as a consequence of typical atmospheric circulation patterns.Dust particles can penetrate Northern Europe reaching the United Kingdom and the North Sea in winter (contributions to PM 10 ∼ 15 % in northern United Kingdom in winter see Fig. 9).
African dust outbreaks have been long recognized to cause exceedances of the European air quality limit values (Rodríguez et al., 2001;Barnaba and Gobbi, 2004;Querol et al., 2004;Papanastasiou et al., 2010).Table 5 shows the number of exceedances of the daily PM 10 air quality limit value established by the European Commission (50 µg m −3 ) for each EMEP station in 2004 as well as the number of exceedances obtained with the model.The number of exceedances is underestimated by the CALIOPE modelling system (248 observed days versus 72 simulated days exceeding the daily PM 10 limit value).In those stations mostly affected by desert dust outbreaks such as the Spanish EMEP sites, these differences are reduced (with 87 observed versus 37 simulated days exceeding the daily PM 10 limit value) indicating the improvement from the inclusion of desert dust in model simulations.
In Fig. 11 we provide a spatial estimate of the number of days in which the daily European air quality limit value (50 µg m −3 ) is not met due to natural dust events.Figure 11a provides an estimate based only on desert dust levels.Large regions from the Iberian Peninsula to Greece (in the North Mediterranean arc, <45 • N) exceed the limit value while the number of daily exceedances increases southwards reaching up to about 75 days per year in the southernmost areas.When we include the other aerosol components (Fig. 11b), a significant number of exceedances appear in well-known European hot spots.In particular, the limit value is exceeded more than 75 days in the Po Valley, Eastern Europe and in the Benelux regions.Additionally in those regions most affected by the presence of desert dust (in the North Mediterranean arc, <45 • N), the number of exceedances notably increases with respect to the desert-dust-alone case achieving up to 100 days in mountainous areas in the South-Eastern Iberian Peninsula.Note that Fig. 11b may only represent a lower end estimate of the total number of exceedances given that the PM 10 mass is significantly underestimated in the modelling system (about 6 µg m −3 on average).

Summary and conclusions
A detailed aerosol characterization over Europe for year 2004 using the CALIOPE modelling system has been presented in this work, together with a detailed model evaluation of PM levels, AOD and chemical composition to quantify the model skills and analyze the causes for discrepancy.
The evaluation shows that the modelling system can reproduce reasonably well the daily variability of the main aerosol components in Europe.Aerosol levels are in most cases underestimated.On the one hand, the larger underestimation in PM 10 compared to PM 2.5 suggests missing sources of coarse PM in the modelling system.On the other hand, the evaluation of the chemical composition highlights important underestimations of the modelled fine fractions particularly for carbonaceous matter (200 %) and SIA (18-50 %).
We calculated spatially homogenous multiplicative correction factors for carboneous matter and SIA components that minimize the differences between the modelled and observed values.The results of the corrected simulation highlight the fact that besides increasing the total mass budget (underestimations are reduced by 34 % and 23 % for PM 2.5 and PM 10 , respectively), the correction by species and seasons improves the simulation of the variability of the bulk parameters with respect to the observed values (the annual correlation increases from 0.47 to 0.61 for PM 2.5 , and 0.57 to 0.62 and PM 10 , respectively).Also the corrected modelled AOD is significantly improved when compared to MODIS.
Aerosols over Europe are dominated by (1) local anthropogenic emissions, (2) the proximity to the African desert and (3) atmospheric dynamics at synoptic scale.The mean annual PM concentration decreases towards the North and North-Western Europe reaching the lowest values in the northernmost regions.The highest aerosol concentrations are found over the industrialized and populated areas of the Po Valley and the Benelux regions.A second maximum is detected over Eastern and Southern Europe.High values over southern Europe are linked to the transport of large particles from the Sahara desert.Maximum values are found in spring in the Eastern Mediterranean and in summer over the Iberian Peninsula and the Central-Western Mediterranean.SIA are concentrated in the fine fractions and they are the main contributors to the PM 10 and AOD values in the European continent reaching up to 80 % for latitudes north of 40 • N.
We have shown the importance of the desert dust transported from North Africa in the aerosol budget over Southern European countries.The presence of mineral dust from African deserts causes exceedances of the daily PM 10 European air quality limit value (50 µg m −3 ) in latitudes south of 45 • N reaching up to more than 75 days per year in the southernmost regions.
Despite the rather satisfactory performance of the modelling system, several aspects are now under further implementation in the framework of the CALIOPE project.Windblown dust should be taken into account especially in dry and arid regions as well as PM resuspension from paved roads in urban areas (Pay et al., 2011).Biomass burning and natural NO x are emissions currently not included in the CALIOPE modelling system and could contribute to the NO 2 underestimation.Another source of uncertainty comes from the chemical speciation of primary pollutants in the EMEP emission inventory.While in the United States, the Environmental Protection Agency (USEPA) is the common repository of profiles to speciate primary pollutants from different anthropogenic activities, in European modelling exercices there is not a common strategy in this issue.There are different chemical speciation profiles according to different sources of information.Most European emission models (such as Simpson et al., 2012) are based on profiles of Non-Methane Volatile Organic Carbon (NMVOC) species from Passant (2002) specifically developed for the United Kingdom.Other European studies such as Bessagnet (2008) used also the profiles from SPECIATE (http://www.epa.gov/ttnchie1/software/speciate/) which is the USEPA's repository of NMVOC.However, others such as Bieser et al. (2011) used SPECIATE for PM speciation and Passant (2002) for NMVOC.In this sense, it would be recommendable to work in a common European methodology for the chemical speciation of primary emissions.
We expect further progress with the new version of CMAQ (CMAQv5.0)featuring a new aerosol module, AERO5, containing substantial updates in SOA formation and dynamic interactions of fine and coarse aerosol.

S. Basart et al.: Aerosols in the CALIOPE air quality modelling system
Appendix A Table A1.Coordinates, altitude and the aerosol chemical species measured of the 55 selected EMEP stations are included.The code is composed by 2-letter country code plus 4-digit station code.Zone of location is defined in Fig. 1.

Fig. 5 .
Fig. 5. Modelled (black lines), corrected-modelled (red lines) and measured (grey lines) mean daily time series (right) and scatter plots of mean hourly values (left) for AOD, AOD fine and AOD coarse at the AERONET stations.The scatter plots include the 1:1, 1:2, 2:1, 1:5 and 5:1 reference lines.Correction factors of Table3are applied to sulphate, nitrate and ammonium and EC + OC in the corrected-modelled series.

Fig. 6 .
Fig. 6.Spatial distribution of mean bias (left panels) and correlation coefficient (right panels) at all stations for AOD, AOD fine and AOD coarse in hourly basis.The four columns represent the winter and summer seasons for 2004 for each parameter.

Fig. 7 .
Fig. 7. Seasonal AOD average (from top to bottom) of CMAQ + BSC-DREAM8b without correction factor (left panels), CMAQ + BSC-DREAM8b with correction factors of Table 3 for sulphate, nitrate and ammonium and EC + OC (central panels) and MODIS/Aqua AOD (rigth panels) with AERONET seasonal mean values in coloured circles.

Fig. 11 .
Fig. 11.Number of the days exceeding the EU PM 10 daily limit value (>50 µg m −3 ) for (a) BSC-DREAM8b and (b) BSC-DREAM8b + CMAQ derived aerosol.Correction factors are applied to sulphate, nitrate, ammonium and EC + OC in CMAQ model outputs.

Table 1 .
Seasonal and annual statistics obtained with CALIOPE over Europe for 2004 at the EMEP/CREATE stations for SIA (i.e.sulphate, nitrate, and ammonium) and SSA (i.e.chloride and sodium).Winter: January, February and December; spring: March, April, May; summer: June, July, August; autumn: September, October, November.The number of data points indicates the number of pair measurement-model used to compute the statistics.The calculated statistics are: measured mean for available data, modelled mean for the whole year, correlation coefficient (r), mean bias (MB), mean normalize bias error (MNBE), root mean square error (RMSE), mean fractional bias (MFB) and mean fractional error (MFE).

Table 2 .
Seasonal and annual statistics obtained with CALIOPE over Europe for 2004 at the AERONET stations for AOD, AOD fine and AOD coarse .Winter: January, February and December; spring: March, April, May; summer: June, July, August; autumn: September, October, November.The number of data points indicates the number of pair measurement-model used to compute the statistics.The calculated statistics are: measured mean for available data, modelled mean for the whole year, correlation coefficient (r), mean bias (MB), mean normalize bias error (MNBE), root mean square error (RMSE), mean fractional bias (MFB) and mean fractional error (MFE).

Table 3 .
Seasonal and annual multiplicative correction factors for SIA (sulphate, nitrate and ammonium) and EC + OC obtained minimizing the weighted sum of the squared difference between the modelled and measured chemical aerosol surface concentrations on a daily basis over Europe for 2004.Winter: January, February and December; spring: March, April, May; summer: June, July, August; autumn: September, October, November.

Table 5 .
Number of exceedances of the PM 10 daily European limit value (50 µg m −3 ) from EMEP observations, corrected CMAQ+BSC-DREAM8 modelled values and BSC-DREAM8b modelled values.Correction factors of Table 3 are applied to nitrate, sulphate, ammonium and EC + OC.