The European aerosol budget in 2006

Introduction Conclusions References


Introduction
Aerosols have a large impact on the behaviour of our atmosphere as they influence the earth's radiation budget both directly through interaction with solar radiation (Hess et al., 1998;Haywood and Boucher, 2000;IPCC, 2007) and indirectly through altering the properties and life cycle of clouds (Rosenfeld et al., 2008;Kaufman et al., 2002).The aerosolclimate interactions are complex and the aerosol forcing is much less certain compared to the radiative effect of greenhouse gases.Hence, the combined direct and indirect aerosol effect may have masked the climate sensitivity towards an increase in greenhouse gases to an unknown extent (Anderson et al., 2003).
Exposure to particles has been associated with adverse health effects and particles are believed to be the most important air pollutant responsible for these health effect.Shortterm exposure has been associated with increased human morbidity and mortality (Brunekreef and Holgate, 2002;Dockery et al., 1993;Pope et al., 1995).Although most health studies have quantified relationships between the total aerosol mass (PM 10 or PM 2.5 ) and health effects, some studies indicate that also the chemical composition of aerosols (Hoek et al., 2002) and particle size (Stone and Donaldson, 1998) might play a significant role.To reduce the adverse health effects, air quality standards for particulate matter have been implemented in many countries.To design effective mitigation strategies, governments need to know the relationship between sources and concentrations of particulate matter.Within Europe, these relationships are traditionally obtained through source-receptor calculations (Seibert and Frank, 2004).Another way to investigate these relationships is a budget analysis as pointed out in this paper.
To better understand the relationship between the emission of aerosols and their precursors on the one hand, and the observed distribution of aerosols on the other hand, numerical models have been developed that describe the aerosol life cycle (Wilson et al., 2001;Bauer et al., 2008;de Meij et al., 2006).This presents an extremely challenging task as one needs to accurately model a host of sources, formation and transformation pathways as well as removal processes to assess aerosol composition, size distribution and mixing state.Together they determine the optical properties of aerosols as well as their ability to act as cloud condensation nuclei (CCN).Thus, to describe the full life cycle of aerosols one needs reliable (size-resolved) emission inventories and parameterisations to supply the necessary boundary conditions for the models (Vignati et al., 2010a;Dentener et al., 2006).Furthermore, one needs to represent the complex aerosol dynamics (Stier et al., 2005;Vignati et al., 2004;Wilson et al., 2001;Lee and Adams, 2010;Korhonen et al., 2008).Also, it is necessary to couple the aerosol dynamics to atmospheric chemistry to account for secondary aerosol formation, semivolatile species and the involvement of aerosols in numerous chemical cycles.Finally, one needs to consider size and composition resolved aerosol removal by wet and dry deposition processes.In the assessment of the aerosol budget, the key uncertainties arise from inaccurate emission estimates (Dentener et al., 2006;Vignati et al., 2010b) and uncertainties in the wet-removal process (Chin et al., 2000).
In a model intercomparison study, Textor et al. (2006) highlighted the poor agreement among models concerning aerosol processes, and specifically the wet removal of aerosols.The parameterisations in models are probably often tuned to produce a reasonable comparison with (satellite) observations.Unfortunately, model specific tuning often remains undocumented and arbitrarily tuning of models can have led to the huge diversity in the analysed simulations.Textor et al. (2006) showed that methodologies to analyse and compare different models are indispensable to improve our ability to model the aerosol distribution.One method that provides details about the processes that matter for aerosol modelling is a budget analysis.Budget analysis also helps to understand differences between models.An aerosol budget analysis with a bulk aerosol approach has been described in Kanakidou (2007).
The aim of this paper is threefold.First, we will present a description of the TM5 model coupled to the size-resolved aerosol module M7 and compare model results with observations.Second, we will analyse the European aerosol budget and quantify the aerosol import and export terms in the boundary layer and the free atmosphere.Thirdly, we will highlight some uncertainties that are associated with aerosol modelling.Specifically, we will address the wet-removal parameterisation in our model and focus on an anecdotical improvement of the fire-related emissions during a biomass burning episode in April-May 2006 in Eastern Europe.

Model and measurements
The quantification of the aerosol budget over Europe is performed with the global transport model TM5 (Krol et al., 2005) coupled to the aerosol dynamics module M7 (Vignati et al., 2004).To calculate the aerosol budget in the model, Europe is defined from 34 • N to 62 • N and from 12 • W to 36 • E. We examine the import, export, emission and deposition of aerosols as well as chemical processes that influence particulate matter.Below we describe the main characteristics of the model with a focus to the aerosol description as well as the observations used for evaluation.

TM5 model description
The global horizontal resolution of the offline chemistry transport model TM5 is 6 • longitude by 4 • latitude.The vertical grid comprises 25 hybrid σ -pressure levels ranging from surface up to in the stratosphere.As the region of interest is Europe, we used TM5's two-way nested zoom capability (Krol et al., 2005;Berkvens et al., 1999) to acquire a higher resolution over Europe.The zoomed region is defined from 12 • N to 66 • N and from 21 • W to 39 • E, with a resolution of 1 • × 1 • .Note that our definition of Europe to calculate the budget is only a part of this zoomed region.A transitional zone from 2 • N to 74 • N and from 36 • W to 54 • E, with a resolution of 3 • × 2 • is used to smoothen the transition (Krol et al., 2005).The vertical resolution remains the same.
The global time resolution is 90 min.TM5 uses operator splitting, in which each process (e.g.advection, chemistry) has a time step of 45 min.This time resolution is refined four times in the zoom region and twice in the transitional region.
The version used is subversion (SVN) revision 2887.

Aerosol module description
All aerosol processes including wet and dry removals are calculated in the model using the size resolved number and masses given by the aerosol dynamics module M7 (Vignati et al., 2004).

Aerosol dynamics
During the chemistry step of TM5, M7 is called to simulate the aerosol microphysics.M7 distinguishes seven classes (modes) of aerosols of different size and solubility.The properties of M7's aerosol modes are listed in Table 1.There are four size classes for soluble aerosols and three for insoluble aerosols.Chemical compounds can be present in various modes.For each mode, one number tracer and several tracers for the chemical compounds are subject to transport in the model for a total of twenty-five tracers.M7 considers the modes as log-normal size distributions (von Salzen, 2006) with defined median radius (r) and spread (σ ).Although aerosols may not be spherical, they are assumed spherical in the model.The size distribution of a mode looks like (Seinfeld, 1986): Here, r is the size, N is the total aerosol number concentration, r is the median radius and σ is the geometric standard deviation.
A mode consists of a number concentration and several component masses that are internally mixed (N,[M]) (Stier et al., 2005).Given the modal component masses and their densities, the total volume (V ) can be calculated.This volume is represented by a log-normal distribution as in Eq. (1).To derive this distribution, M7 assumes a constant standard deviation, which allows the volume per aerosol ( V N ) to determine the median radius (r).The standard deviation is set to 2.00 for the coarse modes and to 1.59 for the other modes (Vignati et al., 2004).The median r is given by (Seinfeld, 1986): M7 handles the formation of new particles (binary nucleation) (Vehkamäki et al., 2002), the coagulation of particles and the condensation of sulphuric acid to existing particles.M7 ensures that the modes keep their inherent solubility by moving coated particles to the soluble (mixed) mode.The size classes are preserved by transferring mass of growing aerosols to the next mode.M7 diagnostically calculates the water attached to the soluble particles (Vignati et al., 2004).

Ammonium and nitrate
In Table 1, two compounds that are important for the aerosol budget over Europe (Putaud et al., 2004) are missing: ammonium (NH + 4 ) and nitrate (NO − 3 ).Observations show that nitrate is very abundant, especially in the western European cold season (Schaap et al., 2002;Mehlmann and Warneck, 1995).There is a temperature-dependent equilibrium between gas phase nitric acid (HNO 3 ) and nitrate, dissolving into and evaporating out of the aerosol.This equilibrium also depends on the available aerosol sulphate (SO 2− 4 ) and gas phase ammonia (NH 3 ).M7 is not designed to model semivolatile aerosol components.We use the Equilibrium Simplified Aerosol Model (EQSAM) version 3. (Metzger et al., 2002a,b) to calculate the partitioning of ammonium nitrate between the gas and aerosol phase with the total available sulphate (of all modes).EQSAM uses a bulk aerosol approach, so it does not define in which aerosol size category the ammonium nitrate gathers.We assume that the ammonium nitrate and the water absorbed by it resides in the soluble accumulation mode.Test simulations with ISORROPIA and GMXe (Nenes et al., 1998;Pringle et al., 2010) confirm that virtually all ammonium nitrate mass is in the soluble accumulation mode.EQSAM limits the relative humidity to 99% to exclude cloud formation (Metzger et al., 2002a).The aerosol associated water mass, can be sufficiently accurately reproduced by EQSAM with respect to other global model uncertainties (Metzger et al., 2002b).
To calculate the deposition velocities for the tracers in each mode, the lognormal distribution is used to distribute the mass and number tracers into twenty-three size bins.Each of these bins is subject to a bin-dependent deposition velocity, recalculated every three hours depending on e.g.atmospheric stability and surface type (Ganzeveld et al., 1998).
After accounting for dry deposition in each bin, the lognormal distribution is reconstructed.For simplicity, the modes remain log-normal with a fixed standard deviation.Hereby, aerosol mass moves from a size range with a slower deposition velocity to a size range with a faster deposition velocity.This introduces a bias that hard to avoid, but might accelarate loss by deposition.
Apart from surface deposition, the coarse mode aerosols exhibit a non-negligible fall velocity due to gravitational settling in the atmosphere.This sedimentation process is modelled by calculating 3-D fields of the fall velocities for each mode (Slinn and Slinn, 1980).Sedimentation removes preferably the larger particles, which results in smaller fall velocities for the aerosol numbers than for aerosol masses.The sedimentation process also changes the median radii of the M7 modes.

Wet deposition
Wet deposition is split in deposition from stratiform and convective precipitation.For stratiform precipitation, in-cloud scavenging and below-cloud scavenging is handled separately.Below-cloud scavenging of gases is linearly related to the surface rain flux, using a gas-to-droplet transfer coefficient based on the Reynolds and Sherwood numbers of the falling rain droplets.For aerosols, the scavenging parameterisation from Dana and Hales (1976) is used and calculated for each aerosol mode separately.In-cloud scavenging is modelled in two phases: the mass transfer of soluble gases and aerosol to the liquid phase and the formation of rain droplets (Roelofs and Lelieveld, 1995).In convective precipitation, aerosols are assumed to be removed very efficiently (similar to HNO 3 ).The removal rate is modelled as a simple function that depends on the convective precipitation at the surface (Vignati et al., 2010b).
TM5 assumes well-mixed grid cells.However, the time scale of wet removal can become faster than the mixing time scale of the grid cell.Therefore, the wet-deposition yield will increase for larger grid boxes.This is a significant issue as multiple resolutions are used within the same simulation.This issue is treated pragmatically by introducing a time paramter τ nomix , in which the in-cloud, belowcloud and cloudless fractions of a grid cell are treated quasiindependently (Vignati et al., 2010b).This way, the wet removal in large grid cells is slowed down, reducing the resolution dependence.Applying the wet removal on a fraction of the gridbox has always been a challenge for modellers, as outlined in a study on 210 Pb (Balkanski et al., 1993).

Emission
Emission data used in the model are those recommended for the AEROCOM (Dentener et al., 2006) model intercomparison studies and from the IPCC (IPCC, 2000).For biomass burning emissions, we use climatologic inventories from the global fire emission database (GFED 2) (Randerson et al., 2006;van der Werf et al., 2006) with prescribed height distribution (Dentener et al., 2006).In these data, it is predefined in which modes the aerosols are emitted.Aerosol mass emissions have an assumed lognormal distribution with a median radius (r) (Table 2) in their inventories.This median radius is used to calculate the total emitted aerosol number with a mode-dependent standard deviation.
Ammonia is emitted mainly by domestic animals and synthetic fertilisers.Other sources of ammonia are biomass burning, the oceans, crops, human population and pets and natural soils (Bouwman et al., 1997).
Carbonaceous compounds are emitted by fossil fuel, biofuel (Bond et al., 2004) and biomass burning (Randerson et al., 2006;van der Werf et al., 2006).These particles are only emitted in the Aitken mode.The black carbon is always emitted as insoluble particles, while 65% of the particulate organic matter (POM) is emitted in the soluble mode (Stier et al., 2005).Production of secondary organic aerosols is modelled as a direct emission.In reality, secondary organic aerosols from e.g.terpenes are involved in the formation and the early growth of new particles (Sihto et al., 2006(Sihto et al., , 2009) ) and they condensate on existing particles.In our model, however, these organic compounds are directly added as particulate organic matter mass in the soluble Aitken mode (Kanakidou et al., 2005).However, to avoid unrealistic situations, particles of 10 nm are added when there are no other emissions in the soluble aitken mode.It is assumed that the organic matter involved in nucleation has been coagulated efficiently.Furthermore, through coagulation of the Aitken mode particles, we mimic the condensation of these organic compounds to the accumulation mode.
For dust, we used pre-calculated AEROCOM data (Dentener et al., 2006).The emission sizes are variable and are pre-calculated as well.Dust is emitted both in the insoluble accumulation mode and the Insoluble coarse mode.Sea salt emissions are calculated online as function of the ten-meter wind speed as described in Vignati et al. (2010a) and Gong (2003).Sea salt is emitted in both the soluble accumulation mode and the soluble coarse mode.

Aerosol optics
We calculate the AOD in our model from the aerosol concentrations using Mie scattering theory (Mie, 1908;Barber and Hill, 1990).An aerosol contributes to the AOD, depending on its wet radius (r w ), its complex refractive index (m) and the wavelength (λ).
The size of the wet droplets (r w ) is calculated from the median wet radius (r w ) and the fixed standard deviation (σ ).The refractive index (m) of aerosol compounds, including water, are based on ECHAM-HAM (Kinne et al., 2003), OPAC (Hess et al., 1998) and Segelstein (Segelstein, 1981).
We compute for each time step, for each grid cell and for each aerosol mode an effective refractive index based on the chemical composition.We do not employ a simple volume-weighted mean of the refractive indices of the chemical compounds, which is known to give inaccurate results.Rather, we use proper effective medium theory from Maxwell-Garnett (Maxwell-Garnett, 1904) and Bruggeman (Bruggeman, 1935).
Mie-scattering calculations demand a significant computational burden and simplifying the Mie-scattering theory causes significant errors (Boucher, 1997).To tackle this problem, we pre-calculated a lookup table.The input parameters of this lookup table (refractive index and size) are sampled with forty times fifteen values for the refractive index (real × imaginary) and a hundred values for the size parameter r w λ .With interpolation, the discretisation error is expected to be low (at most a few percent).
By calculating the AOD at several wavelengths, we can determine the Ångström parameter (Russell et al., 2010) with the following general relationship between AOD and wavelength ( Ångström, 1929): Here, τ is the AOD, β is a prefactor, λ is the wavelength and α is the Ångström parameter.The prefactor β is a measure for the overall AOD and the Ångström parameter is a measure for the wavelength-dependence of the AOD.Verification on the Ångström parameter enables to check whether the aerosol distribution is dominated by fine (α>1.3) or coarse (α<1.3)particles.

In-situ measurements
The modelled concentrations are compared to in-situ measurements from the EMEP network (EMEP, 2008) (http://www.emep.int),which provides data for a host of air pollutants (Lazaridis et al., 2002;EMEP, 2008).We used the yearly averaged data for particulate matter, aerosol composition and aerosol precursors.The averaged data for 2006 were used for the stations that produced valid data during at least 10% of the time.For the vast majority of the used data points (94%), this valid-data percentage was above 50%.Verification of the model results against measurements for PM and its components is hampered by potential artefacts in the sampling.PM filter measurements are uncertain due to potential losses of ammonium nitrate and absorption of nitric acid and organic compounds (Vecchi et al., 2009).The abovementioned volatilisation and absorption artefacts cause the sampling of nitrate and ammonium to be difficult (Yu et al., 2006;Zhang and McMurry, 1992;Cheng and Tsai, 1997).
Correct sampling is only possible with denuder filter packs.However, these labour intensive methods are hardly used in Europe (Schaap et al., 2002).Hence, we use aerosol nitrate and ammonium data from inert filters, although we acknowledge that they are prone to losses at temperatures above 20 • C (Schaap et al., 2004b).Most data on nitrate and ammonium, however, are given as the sum of gas and aerosol concentration.Gas phase concentrations for ammonia and nitric acid obtained with a filter pack are used here when reported by EMEP.
Observations of sulphur and nitrogen compounds are reported as masses S and N rather than total mass.Sea-salt concentrations are evaluated with observed sodium (Na) concentrations.Throughout this paper, we will express any sulphur compounds, nitrogren compounds or sea salt as masses S, N and Na.For the conversion of total sea salt to sodium, we use a conversion factor of 0.306 (Millero, 2004).
Unfortunately, measurements of carbonaceous compounds of 2006 are very scarce.Therefore, we will use measurements from the EMEP EC-OC campaign in 2002-2003(Yttri et al., 2007)).In the EMEP campaign, organic carbon (OC) is measured, while TM5 simulates organic matter (POM).In the analysis, a factor 1.4 is used to convert the observations of organic carbon to organic matter to account for the noncarbon part of the organic matter.
There are also insufficient observations of mineral dust aerosol.Mineral dust is a mixture of many components, so it is very difficult to measure it reliably, especially when only a small fraction of the total aerosol mass is dust, which is the case in the majority of Europe.
We compare the modelled aerosol optical depth (AOD) to European observations from the Aerosol Robotic Network (AERONET) (Holben et al., 2001).These are measured by the sun-powered CMEL Electrique 318A specral radiometer that points systematically to the sun in a programmed routine (http://aeronet.gsfc.nasa.gov).
In the model, the AOD is calculated at the wavelengths (λ) at which the AERONET stations measure.We sample the observed AOD-values with 1.5-h intervals, where multiple measurements within one interval are averaged.This adapts the time resolution of the observations to that of the model.
We also analyse the ability of the model to simulate the Ångström parameter.The Ångström paramter is reported by AERONET as well.In our model, we use a function fit of Eq. ( 3) to obtain the Ångström parameter.

Results and discussion
In this section, we first examine the modelled surface concentrations of the aerosol components and precursor gases in Europe, as well as AOD and Ångström parameter.Next, we compare these results with observations followed by the European budget of aerosol compounds and precursors.To address the model's ability to simulate the full aerosol burden we evaluate the modelled AOD and Ångström parameter with observations.Finally, we will address two main uncertainties, namely wet removal and biomass burning emissions.

Concentration distribution
The annual average concentration distribution of aerosol chemical compounds and their precursors are shown in Fig. 1.The AOD and the Ångström parameter are shown in Fig. 2.
Nitrogen and sulphur oxides show a hot-spot structure with high concentrations in industrialised regions with SO x (SO 2 , H 2 SO 4 ) more in Eastern Europe and NO y (NO, NO 2 , NO 3 , HNO 4 , N 2 O 5 , PAN) more in the Western Europe.We clearly model an ammonia hot spot in the Netherlands, which is due to high population of livestock (Buijsman et al., 1987).Nitric acid shows high concentrations over sea.In reality, nitric acid may react with sea salt and displace chloride (Glasgow, 2008;Schaap et al., 2004a).This reaction is not implemented in the model, since it has only a small effect on the aerosol distribution over land.
In contrast to primary gaseous pollutants, secondary inorganic aerosols have smoother distributions as they are of secondary origin.The ammonium concentration shows features of both nitrate, which peaks in north-western Europe and in the Po Valley, and sulphate, which shows highest concentrations in south-eastern Europe.This is because both nitric acid and sulphuric acid are neutralised by ammonium.
Primary anthropogenic carbonaceous aerosols show high concentrations in densely populated and industrialised regions.As for primary gaseous pollutants (NO y and SO x ), this results in a hot-spot structure.For black carbon and organic matter, the hot spots are located at different positions.However, the hot spot structures of NO y and black carbon only differ slightly (more NO y in England and more black carbon in Poland).
Mineral dust shows a sharp gradient from the Sahara to the north with significant concentrations over the Mediterranean countries.North of the Alps, mineral dust concentrations are low.As expected, sea-salt concentrations are highest above  open sea.Above land, significant sea salt-concentrations are only present in coastal areas.
The calculated annual mean AOD is highest in the south and south-eastern Europe with values above 0.15.Mineral dust and sulphate appear to be the most dominant contributors to the AOD, since the AOD is high at locations where mineral dust or sulphate is abundant (see Figs. 1 and 2).The Ångström parameter is low over the sea and over northern Africa, because sea salt and desert dust are mostly coarse mode aerosols.Over land, the fine anthropogenic aerosols dominate, resulting in high Ångström parameters.

Model evaluation
Model results of particulate matter compounds and precursor gases have been compared with in-situ observations.When comparing with size-segregated observations (e.g.PM 10 ), the log-normal distribution of M7 is used to calculate which fraction of the modelled aerosol mass is below the size limit.First, we will compare annually averaged concentrations.Later on, we will analyse a few time series of PM 10 for January 2006.We chose to evaluate January, because the surface concentrations are the most senstivie to synoptic events in winter.

Annual means
The most important results of the comparison of annually averaged aerosol and precursor concentrations are shown in Fig. 3. Figure 4 shows the results per country represented in a graphic way.
Total reduced nitrogen (ammonia and ammonium) is slightly overestimated.Aerosol ammonium is represented quite well, but ammonia is overestimated.It is well-known that modelling ammonia is challenging due to the importance of local effects (Dentener and Crutzen, 1994).Still, the Dutch ammonia hot spot is caught remarkably well in the model.Interestingly, TM5 overestimates ammonia while many other models underestimate ammonia concentrations (Schaap et al., 2004a).An issue is that the night-time concentrations estimated by the model are too high.These concentrations are very sensitive to the stability of the nocturnal boundary layer (van Loon et al., 2007).Especially TM5, as a global model, has resolution limitation for the nocturnal boundary layer, because it is a small-scale phenomenon (∼50 m).Therefore, the nocturnal boundary layer tends to be poorly defined in TM5.This is very important for ammonia, because the modelled emissions are assumed constant over the day (de Meij et al., 2006), which implies that ammonia is emitted into the stable boundary layer, causing night-time accumulation.In reality, ammonia emissions show a considerable diurnal variation with peak emissions during the day and even a net night-time surface uptake of ammonia (Wichink Kruit et al., 2007), which may be released during the next day.
The total oxidised nitrogen (nitric acid and aerosol nitrate) is represented better.There is some overestimation, but the spatial correlation is good (r = 0.89).When considering aerosol nitrate, about the same conclusions could be drawn.However, the modelled concentrations of nitric acid are far off (bad correlation and overestimation by a factor of two).We already addressed that the values above sea are modelled too high because the acid displacement reaction is not taken into account (see Sect. 3.1 and Schaap et al., 2004a;Glasgow, 2008).This issue may affect modelled concentrations in coastal areas, where many stations are located.Though the nitric acid concentration is the difference between total nitrate and aerosol nitrate (both well represented in the model), nitric acid has higher uncertainties because of higher uncertainties in sampling for nitric acid (see Sect. 2.3).Also, the nitric acid concentrations are often much lower than the aerosol nitrate concentrations, which means that the relative uncertainty becomes higher.
Sulphate is represented quite well.However, there is an overestimation by a factor of two for sulphur dioxide, the precursor of sulphate (not shown).A slow oxidation of sulphur dioxide may partly explain this discrepancy, but higher oxidation would also lead to higher surface sulphate concentrations.However, increased in-cloud oxidation at higher altitudes would be more consistent.The time scale of in-cloud oxidation of sulphur dioxide is very uncertain (Langner and Rodhe, 1991).Also, the emission heights of sulphur dioxide may play an important role.For instance, the sulphur dioxide emissions from AEROCOM are higher in the lower model levels than those of EMEP, which can cause a surface concentration discrepancy of a factor of two in eastern Europe (de Meij et al., 2006).Another possibility is that the emission rate of sulphur dioxide is too high in the model or there is an unaccounted or underestimated loss of sulphur dioxide that does not lead to sulphate production, e.g.dry deposition (Chin et al., 2000).
As we mentioned in Sect.2.3, we compare our modelled results with observations from the EC-OC campaign of 2002 and 2003.Black carbon is represented well, as shown also in Vignati et al. (2010b).There is a huge (factor 3 or more) underestimation of particulate organic matter, though there is a quite okay spatial correlation between observations and model results.Secondary organic aerosols (Volkamer et al., 2006) and resuspended aerosols (Sternbeck et al., 2002), which are rich in organic matter, are significantly underestimated by TM5.An earlier evaluation of organic matter (Vignati, personal communication, 2010) also shows such an underestimation.
From the comparison of modelled and observed total particulate matter, we can conclude that the aerosol spatial distribution is reproduced reasonably well.There is a slight underestimation of the coarse aerosols (PM 10 ) that is probably due to resuspension of aerosols, which is not included in the model, but may be important for local PM 10 concentrations (Sternbeck et al., 2002).Another factor can be an underestimation of secondary organic aerosols (Volkamer et al., 2006) or the unaccounted mass (e.g.dust or water) which is frequently present in PM 10 measurements.
The spatial variability of sea salt is represented very well.However, the absolute concentrations are significantly (50%) overestimated, probably due to uncertainties in emissions and the dry deposition parameterisation.A sea salt overestimation is also shown in Manders et al. (2010).

Time series January 2006
To evaluate the ability of TM5 to capture synoptic events, we compare time series of modelled and observed total PM 10 for January 2006.Four EMEP stations provide hourly data of PM 10 for this month.The results are shown in Fig. 5.
We clearly calculate less variability than observed.TM5, as a global model, is unable to simulate local effects of short durations.This is most clearly visible in Narbeth (GB) where TM5 cannot follow the rapid changes in PM 10 that are observed.Vredepeel (NL) and Vavihill (SE) are represented quite well.The timing of the peaks is roughly correct.Only the magnitudes of two peaks are significantly underestimated, namely the Vredepeel (NL) peak around day 27 and the Vavihill (SE) peak around day 15.TM5 is clearly unable to simulate aerosol concentrations on Ayia Marina (CY).Here, resolution plays a big role, because Cyprus is an island as small as a TM5 grid box.Therefore, TM5 models dust storms at Ayia Marina (CY) that are not observed at the station.Only the broad peak at the beginning is visible in the observations, though with a much smaller magnitude.

The aerosol budget
For the analysis of the aerosol budget, we split the atmosphere over Europe into two parts: the boundary layer (surface up to 850 hPa) and the free atmosphere (850 hPa up to top of atmosphere).The budget is split into sources and sinks and the processes: emission (E), chemical reactions (C), aerosol condensation (A), dry deposition (D), wet deposition (W) and the horizontal (H) and vertical (V) transport terms.The vertical transport term denotes the transport of tracer mass between the boundary layer and the free troposphere.Nitric acid also has a stratospheric boundary condition determined by its relationship with stratospheric ozone (Santee et al., 1995).The gain or loss due to this boundary condition is counted as vertical transport (V) for the free atmosphere.
Figure 6 visualises the transport terms and all budget terms are listed in Tables 3 and 4. Note that the lifetimes, especially in the boundary layer, are low because export terms, including vertical export, are regarded as sinks as well.All budgets close with an accumulation or depletion (difference between sources and sinks) of less than one percent of the budget.
Although only 15% of the air mass is in the boundary layer, the calculated burdens in the boundary layer are com-parable to those in the free atmosphere.However, the horizontal fluxes in the boundary layer are much smaller than in the free atmosphere (see Fig. 6), indicating that the boundary layer budget is dominated by emission, deposition and vertical transport.Wet deposition is the major sink of all aerosols, except for mineral dust and sea salt, which exhibit efficient dry deposition because of their large size (see Tables 3 and 4).The numbers in these tables are raw model results, only rounded to whole gigagrams per year, while the uncertainties are much larger.Given the uncertainties, we will round percentages in the interpretations to multiples of 5%.
We model a net export of all anthropogenic aerosol compounds from Europe and a net import of natural aerosol (sea salt and mineral dust).The boundary layer over Europe exports anthropogenic aerosols in all four directions, while in the free atmosphere, Europe imports aerosols and gases from the west due to the jet stream.However, the net horizontal export in the free atmosphere is comparable to the export in the boundary layer as a large part of the tracers in the jet stream are not deposited in Europe but pass through the European domain.
Out of the emitted carbonaceous compounds (BC and POM), about 50% reaches the free atmosphere and 20% is exported in the boundary layer.The other 30% is removed, mainly by wet deposition.Dry deposition plays only a limited role (less than 5% of emission).The carbonaceous compounds that reach the free atmosphere are removed partially (65%) in clouds (wet removal).The rest is exported out of the domain.Note that biomass burning can Table 4.European annual budget table of nine tracers for the free atmosphere (850 hPa up to top of atmosphere), with (B)urden, (L)ifetime; the sinks: (C)hemical reactions, (A)erosol condensation, (D)ry deposition, (W)et deposition, (H)orizontal transport and (V)ertical transport; and the sources: (E)mission, (C)hemical reactions, (A)erosol condensation, (H)orizontal transport and (V)ertical transport.Vertical transport includes stratospheric interchange.A value of 0 in the table means a value below 0.5 (rounded down to 0), while an empty spot means that the process does not take place at all.Burdens (Gg) are averages of monthly samples.Lifetimes (days) are burdens divided by total sinks.Fluxes (Gg yr −1 ) are annual totals.NO y includes NO, NO 2 , Peroxyacytyl nitrate (PAN), NO 3 , HNO 4 and N 2 O 5 , but no nitric acid or aerosol nitrate.SO x includes SO 2 and H 2 SO 4 , but no aerosol sulphate.Nitrogen compounds are expressed as masses N, sulphur compounds as masses S and sea salt as masses Na. inject non-negligable amounts of carbonaceous components into the free atmosphere (Dentener et al., 2006).Mineral dust is the only component that has a net negative vertical flux in Europe, from the free atmosphere to the boundary layer.Figure 6 shows that the major transport pathway of dust lies in the free troposphere.Table 3 shows that the emission term for dust in the defined European domain is relatively small compared to the transport term.These features for mineral dust are in line with common understanding that during sand storms mineral dust is transported to elevated altitudes by strong convection.Outflow and transport towards Europe occurs above a marine boundary layer causing import at higher altitudes.Sea salt does not exhibit these features as a big part of its emission source (open sea) is within the budget regeion.Therefore, it has a net positive vertical flux, like the anthropogenic tracers.
About 25% of the emitted ammonia is absorbed by aerosols in the boundary layer.Only 10% reaches the free atmosphere, of which most (80%) gets absorbed by aerosols there.Notable is that, in contrast to aerosols, ammonia is removed much more by dry deposition (55% of emission) than by wet deposition (5% of emission).Transport of ammonia out of Europe is negligible.
About 15% of the nitric acid produced by chemistry is taken up by aerosols.Like for ammonia, dry deposition (45% of production) is a larger sink for nitric acid than wet deposition (25%).Export of nitric acid is small (10% of production).Only about 5% of the nitric acid produced in the boundary layer enters the free atmosphere, while a sizable amount of nitric acid is produced in the free atmosphere.In this atmospheric domain, there is remarkably little absorption of nitric acid by aerosols, which is due to the acidic aerosol environment.We clearly model a high sulphate burden in the free atmosphere compared to ammonium and ammonia (see Table 4).About 35% of the nitric acid is exported and the other 65% is removed by wet deposition.
The budget terms of the inorganic secondary aerosols (ammonium, nitrate and sulphate) are relatively similar.Almost half of the inorganic aerosol material absorbed or produced in the boundary layer is removed by wet deposotion (ammonium: 40%, nitrate: 55%, sulphate: 40%).Horizontal export in the boundary layer is close to one quarter (ammonium: 25%, nitrate: 15%, sulphate: 30%).The rest, about a third, enters the free troposphere (ammonium: 40%, nitrate: 30%, sulphate: 25%).In the free troposphere, a little over half is removed by wet deposition (ammonium: 55%, nitrate: 60%, sulphate: 55%) and the rest is exported.Aerosol nitrate is more concentrated in the boundary layer, while ammonium is equally distributed between boundary layer and free atmosphere and sulphate is more abundant in the free atmosphere.This can be explained by production in the free atmosphere, which is done efficiently for sulphate in clouds (60% of production in the free atmosphere).Together with dry oxidation followed by condensation, the production of sulphate is responsible for the majority of the sulphate source in the free atmosphere (80%), so only a small part comes from the boundary layer.For ammonium, this production percentage is 45% and for nitrate only 5%.This low nitrate production in the free atmosphere is, as explained above, due to the acidic environment.Notable is that ammonium does have a significant horizontal export term (25% in boundary layer and 50% in free atmosphere), while ammonia has not.Ammonia is only abundant in the Netherlands and only a very small part will make it to the European borders without being absorbed by aerosols.
Sulphate is produced by oxidation of sulphur dioxide, partly in clouds (45% in boundary layer and 60% in free atmosphere).The sulphur dioxide oxidised in clouds directly produces sulphate in the aerosol phase ("C" as sulphate source and SO x sink in Tables 3 and 4), while the dry oxidation of sulphur dioxide produces sulphuric acid, which quickly condenses on aerosols ("A" as sulphate source and SO x sink in Tables 3 and 4).Out of the emitted sulphur dioxide in the boundary layer, about 10% is oxidised and another 5% is exported and 20% enters the free atmosphere.The rest is removed by dry (55% of emission) and wet (10% of emission) deposition.There is a considerable amount of sulphur dioxide that is injected directly into the free atmosphere, mainly by volcanic emissions (Andres and Kasgnoc, 1998;Halmer et al., 2002).Together with what is transported up from the boundary layer, this sulphur dioxide is oxidesed to sulphate for 35%, 20% is removed by wet deposition and 45% is exported out of Europe.Sulphur dioxide production by oxidation of dimethyl sulphide ("C" as SO x source in Tables 3 and 4) is small.Nitric acid, and thus aerosol nitrate, originates from other nitrogen oxides in the atmosphere (NO y ).Note that NO y does not include nitric acid.Out of the emitted nitrogen oxides in the boundary layer, 55% is removed by chemical processes.Besides oxidation to nitric acid, nitrogen oxides are also removed by reactions with organic chemicals (not shown).Dry deposition removes 15% of the emitted nitrogen oxides in the boundary layer, 20% goes to the free atmosphere and only 5% is exported.A small amount of nitrogen oxides are directly injected into the free atmosphere by lightning (Pickering et al., 1998) and aircraft emissions.In    the free atmosphere, 75% of the nitrogen oxides is removed chemically and 25% is exported.

Optical analysis
In Table 5, Fig. 7 (upper panels) and Fig. 8, we compare the calculated optical data with AERONET observations.We clearly underestimeate AOD systematically, though the temporal variability is captured adequately by the model.Generally, the AOD in summer is much higher than in winter, and the relative underprediction by TM5 is less in summer and early autumn (factor less than two) than in other months with a factor often above two (see Fig. 9).Also note that more data points are available in summer than in winter.
As the AOD is severely underestimated, we can conclude that besides the small underestimation of the PM 10 surface concentrations, the total column burden appears to be underestimated by a factor of two.Part of the underestimation may be due to emissions that are not included in the model.de Meij et al. (2006) shows that calculated AODs at low relative humidities are underestimated, indicating that the AEROCOM emissions are too low.Another reason for the underestimation of the AODs is that the vertical distribution is not well represented.The underestimated emissions include non-combustion aerosols such as emissions from agricultural activities or traffic abrasion and resuspension (Sternbeck et al., 2002).However, this will mainly affect surface PM 10 concentrations rather than AOD.We will show in Sect.3.5.2that the severe underestimation of the AOD in eastern Europe is largely explained by inadequate biomass burning emission used by the model.The underestimation of the AOD may also be related to too high wetdeposition rates (Chin et al., 2000).Wet deposition is a dominant term in the budget (Tables 3 and 4) and we will address this further in Sect.3.5.1.
At those stations that are located within or near major source regions of black carbon, part of the underestimation of AOD values may derive from biases introduced by the homogeneous sphere approximation, which is employed in Mie computations.For instance, externally mixed black carbon aggregates absorb twice as much radiation in the atmosphere as volume-equivalent homogeneous spheres (Kahnert, 2010a,b).Model computations that account for inhomogeneous mixing of BC with soluble aerosol components predict absorption cross sections that are a factor of 1.5 higher than those computed with a homogeneous mixture approximation (Bond et al., 2006).Such morphological effects are neglected in our effective medium and Mie computations and may contribute to the AOD bias.
There is a reasonable temporal correlations between observed and modelled Ångström parameter (Table 5).The yearly averages agree very well.Also, the spatial variability among station is represented very well.There is one exception (Hornsund, Svalbard), which is due to the low AOD values there, making the Ångström parameter very sensitive to errors.

Uncertainty analysis
In this section, we will investigate the uncertainties related to two key processes: wet removal and emission strengths.With sensitivity simulations we explore possible explanations for the underestimation of the AOD.

Wet deposition
Wet deposition is an important sink, especially in the free atmosphere (see Table 4).A too fast wet-deposition rate may therefore explain why the AOD is underestimated while the surface concentrations look reasonable.We performed three additional simulations for May, June and July 2006 with all in-cloud scavenging rates (both for stratiform and convective precipitation) scaled down to respectively 50%, 10% and 0%.It appeared that halving (50%) the in-cloud scavenging hardly made any difference in the simulated AOD values (about 10% higher AOD after spin-up).This clearly indicates the high efficiency of wet deposition in TM5.As expected, completely ignoring it (0%) resulted in unrealistically high values for the AOD (factor 6 after three months and ever rising).We will analyse the 10% wet-removal simulation, which showed a clear improvement, for the analysis period June and July 2006.
Figure 7 (lower panels) shows that in the 10% simulation, the large underestimation of the AOD has been turned into a slight overestimation.The Ångström parameter is still represented quite well, which indicates that the aerosol size distribution is little affected.Although the modelled AOD and Ångström parameter now agree in Hornsund, the temporal correlation between model and observations at that station remains very poor (not shown).In general, the temporal correlations remain roughly similar (not shown).
We also investigated the changes in surface concentrations that result from a reduction of in-cloud scavenging to 10%.Sulphate and sea salt concentrations rise significantly (about 50%), while other compounds change only very little.From Fig. 3, it is clear that we already overestimate sea salt by 50%.Also the agreement of surface concentrations of sul-phate, good in the unperturbed simulation as shown in Fig. 3, slightly deteriorates by reducing the wet-removal rates.
It appears that a scaling factor of 10% on the wetdeposition rates results in slightly too high surface concentrations and AOD values.For AOD, we expect a slight underestimation because of non-implemented emissions of biogenic volatile compounds and resuspension.The signal of the surface concentrations also indicates that with a 10% scaling of the in-cloud scavenging, the wet removal is underestimated.The aerosol budget changes mainly in the free troposphere.
Based on an analysis of the months June and July, we estimate that the wet-deposition flux is roughly halved in favour of the net export.
We refrain from a further tuning of the wet deposition here, because a sound parameterisation should be based on the physical and numerical considerations (e.g.grid-size dependency) that are associated with both stratiform and convective wet removal.We have shown, however, that a poor representation of wet deposition may be a major cause of the general underestimation of the AOD and may have masked other model deficiencies.

Forest fires
Emission inventories exhibit multiple uncertainties.We already addressed the lack of resuspension in our model and the possible underestimation of secondary organic aerosols.Besides these missed sources, there is also a significant uncertainty in biomass burning emissions.Our model uses climatologic GFED fire emissions, while real biomass burning emissions tend to exhibit large variability between the years and between the seasons in a year (van der Werf et al., 2006).An extreme case occurred in spring 2006, when there were strong forest fires in western Russia (Sofiev et al., 2009;Saarikoski et al., 2007).These events take place every spring, but in 2006 they were particularly strong.Apart from that, the last week of April and the first week of May, the mean wind direction in eastern Europe (20 • E-30 • E) was easterly, which coincides with the fires.This transported the smoke towards Europe, so that it was recorded in the time series of the AOD in eastern European stations.
Figure 10 (upper two graphs) shows a comparison between modelled and observed AOD for Dunkerque and Minsk, in which we can see that Minsk exhibits a clearly separate population of points that belong to the period of the forest fires.For Dunkerque, the points of this period are more mixed with the rest of the dataset.The high observed AOD values during the event were not reproduced by the model.Because easterly circulation is associated with fair weather, the possible too high wet-deposition rates in the model is not likely to play a significant role.
To reproduce the high AOD values at the eastern European stations, we repeated the simulation with the European emission data from the Fire Assimilation System (FAS) that was used in combination with the dispersion model SILAM (Sofiev et al., 2009).The FAS-SILAM PM 2.5 emissions in the area specified below are 4.3 Tg for the considered monthly period, while the climatologic GFED emissions in that area were only 8.7 Gg per month (500 times less).Moreover, the GFED emissions are temporally spread over the entire months, while FAS reported them with daily resolution.As the majority of the emissions occurred during the days with easterly winds, it was evidently important to apply a daily time resolution of emissions to capture the specific transport conditions during the event.
The FAS-SILAM emission data consist of daily 2D fields of PM 2.5 emissions in Europe (11 • W-73 • E, 34 • N-80 • N).We assumed that 10% of this PM 2.5 is black carbon and 90% is organic matter, which is a rough estimation based on observations in Saarikoski et al. (2007).This assumption may influence the results as the optical properties of black carbon and organic matter are different.Also, the injection height of these emissions can be important (Chen et al., 2009).We assumed the following distribution injection heights following Dentener et al. (2006): 20% between surface and 100 m, 20% between 100 m and 500 m, 20% between 500 m and 1 km and 40% between 1 km and 2 km.We performed a simulation, replacing the original climatologic emissions from 15 April to 14 May.
Figure 10 (lower two graphs) show the results for the two AERONET stations.There is a drastic improvement for Minsk, which indicates that this event is caught by the model including these emissions.There is also a small improvement in the results for Dunkerque as well, which means that Dunkerque is affected by these emissions through long range transport.The improvement in the model results clearly illustrates that episodic fire events at the eastern edge of Europe in combination with certain transport patterns may have a significant impact throughout the European domain.

Conclusions
Size-resolved aerosol simulations with the TM5 model coupled to the M7 module have been conducted for the year 2006 with a focus on the European domain (34 • N-62 • N, 12 • W-36 • E).The main conclusions can be summarised as follows: -Comparison of the simulated aerosol distribution with surface observations over Europe shows a reasonably good agreement with spatial correlations of simulated PM mass of 0.75.As expected, spatial correlations are lowest (r = 0.64) and biases are highest for PM 10 , possibly due to neglected resuspension of aerosols.Total ammonium (r = 0.61) is overestimated in the high concentration range, due to the overestimation of NH 3 in emission regions.
-A three-dimensional budget analysis is carried out to enable model intercomparison and assessement of important uncertainties.From our budget, we can conclude that Europe is a net exporter of anthropogenic aerosols, and an importer of natural aerosols (sea salt and mineral dust).For instance, it is calculated that about half of the emitted anthropogenic carbonaceous aerosols are exported from Europe.Dust is the only aerosol component that exhibits a negative vertical flux over the European domain.Notable is that the export rate of gaseous pollutants (e.g.nitrogen oxides) is considerably lower than for anthropogenic aerosols because of dry deposition.
-A comparison to AERONET AOD measurements shows a serious underestimation of the modelled AOD values.We showed that a significant downscaling of the wet-removal rates in the model is required to bring the model closer to the observations.This, however, significantly raises the modelled surface sulphate and sea salt concentrations, while other components are little affected.The modelled Ångström parameter is little affected, which indicates that the aerosol size distribution remains roughly the same.
-We have shown that, apart from uncertainties with the wet removal, large uncertainties arise from inaccurate emission inventories.We were able to significantly improve the modelled AOD at Minsk in April and May 2006 by replacing the GFED climatologic fire emissions by a tailored fire emission inventory that is based on daily fire counts.
Based on this study, future model developments will target at improving the aerosol wet-deposition parameterisation in the TM5 model and the aerosol emission inventories.TM5 employs multiple resolutions at the same time, which calls for a fundamental approach of resolution-dependent processes like the wet removal of aerosols.Fire emissions, but also the emissions of aerosol precursors such as NH 3 exhibit day-to-day variability and diurnal emission patterns that should be taken into account to enable a sound comparison to observations.Finally, it is recommended to continue intermodel (Wilson et al., 2001;Bauer et al., 2008;de Meij et al., 2006) comparisons based on budget analysis as presented in this paper or similar techniques (Textor et al., 2006).

Fig. 1 .
Fig. 1.Modelled surface concentrations of the aerosol tracers and precursor gases.Note that the colour scale used for mineral dust is logarithmic.Nitrogen oxides include NO, NO 2 , Peroxyacytyl nitrate (PAN), NO 3 , HNO 4 and N 2 O 5 , but no nitric acid or aerosol nitrate.Sulphur oxides include SO 2 and H 2 SO 4 , but no aerosol sulphate.All values are averaged over 2006.

Fig. 4 .
Fig. 4. Graphical overview of the comparison between modelled and observed average concentrations over 2006 for total ammonium, total nitrate, sulphate, black carbon, organic matter, sea salt and total particulate matter at EMEP ground stations.Values of stations per country are averaged.The size of the circles represent the number of stations in that country.The country in the upper left box of the sulphate and sea salt maps, is Iceland.

Fig. 6 .
Fig. 6.Transport diagram showing fluxes from north, east, south and west for the boundary layer and the free atmosphere; and the exchange between the two layers.These values are net fluxes integrated over the year 2006.The legend at the top maps the colours to the tracers and defines the value (in Tg yr −1 ) to which the black reference bars at the upper left corner correspond.Nitrogen oxides include NO, NO 2 , Peroxyacytyl nitrate (PAN), NO 3 , HNO 4 and N 2 O 5 , but no nitric acid or aerosol nitrate.Sulphur oxides include SO 2 and H 2 SO 4 , but no aerosol sulphate.Nitrogen compounds are expressed as masses N, sulphur compounds as masses S and sea salt as masses Na.For blackand-white print: the bars represent from left to right (for east/west transport from bottom to top): nitrogen oxides, sulphur oxides, ammonia, nitric acid, ammonium, nitrate, sulphate, black carbon, organic matter, mineral dust and sea salt.

Fig. 7 .
Fig. 7. Comparison between modelled and observed AOD (440 nm) and Ångström parameter (440-870 nm) at AERONET stations.Above: Original results (average over whole 2006).Below: simulation with in-cloud scavenging rates scaled down by a factor of 10 (average over June and July 2006).

Fig. 8 .
Fig. 8. Graphical overview of the comparison between modelled and observed AOD (440 nm) and Ångström parameter (440-870 nm) at AERONET stations.The island in the upper box is Svalbard (Norway).

Fig. 9 .
Fig. 9. Comparison between modelled and observed AOD per month.Valid data of all AERONET stations are averaged.

Table 1 .
The seven modes in M7, their solubilities, size ranges and chemical compounds.

Table 2 .
Implemented aerosol emissions with the predefined M7 modes and median emission radii.

Table 3 .
European annual budget table of nine tracers for the boundary layer (surface up to 850 hPa), with (B)urden, (L)ifetime; the sinks: (C)hemical reactions, (A)erosol condensation, (D)ry deposition, (W)et deposition, (H)orizontal transport and (V)ertical transport; and the sources: (E)mission, (C)hemical reactions, (A)erosol condensation, (H)orizontal transport and (V)ertical transport.Burdens (Gg) are averages of monthly samples.Lifetimes (days) are burdens divided by total sinks.Fluxes (Gg yr −1 ) are annual totals.NO y includes NO, NO 2 , Peroxyacytyl nitrate (PAN), NO 3 , HNO 4 and N 2 O 5 , but no nitric acid or aerosol nitrate.SO x includes SO 2 and H 2 SO 4 , but no aerosol sulphate.Nitrogen compounds are expressed as masses N, sulphur compounds as masses S and sea salt as masses Na.

Table 5 .
Comparison between modelled and observed optical parameters.Listed are temporal correlations and averages of time series of the AOD 440 nm and the Ångström parameter 440-870 nm at 18 European AERONET stations.