The regional EMEP model has been applied to calculate EC concentrations over Europe for the years 2002–2004 using a new EC emission inventory. The results are compared with measurements from the CARBOSOL and EMEP EC/OC campaigns. The model underestimates EC concentrations by 19% on average, and the spatial correlation is 0.80. For individual sites, the model bias varies from −79 to 77% and the average temporal correlation is 0.53, varying from 0.25 to 0.79. The model flattens the north-south EC gradient as it tends to overestimate EC for Nordic sites and underestimate EC for more southern sites. We have also studied the contributions of various processes to the model EC results. Using EC as a tracer of primary PM emissions from combustion sources we have made a preliminary evaluation of the anthropogenic EC (PM) emission. There are indications of a possible underestimation of EC emissions from traffic in some areas and both underestimation and overestimation of EC emissions from residential combustion for some European countries. The largest uncertainties probably lie in EC emissions from residential wood/fossil combustion and are associated with both emission factors and spatial and temporal variation. The need to develop accurate and time resolved wildfire emissions is emphasized. The effect of EC aging is shown be rather limited for most of Europe (1 to 4%). Changes in EC wet scavenging ratio have a noticeable effect on calculated EC (between 5 and 25% for most Europe and 30–40% in remote areas), but EC scavenging ratios are still poorly known.
 Elemental carbon (EC), or black carbon (BC), is an important component of atmospheric aerosols. The environmental relevance of EC comprises a large number of topics, from human health [e.g., Jansen et al., 2005, and references therein], which is mostly a local and regional issue, to climate forcing [e.g., Jacobson, 2001, 2002; Wang, 2004], which is a global issue. The main source of elemental carbon particles is combustion of carbon-based fossil and biomass fuels and wildfires. Because of its simplified chemistry, EC has been proposed as a tracer of primary particulate matter (PM) from combustion sources and thus is useful for qualifying EC/PM emission.
 Several studies of the global transport and distribution of concentrations of BC have been performed with climate models during the last decade. The earlier model calculations [Penner et al., 1993; Cooke and Wilson, 1996; Lohmann et al., 2000] were based on the emission inventories for bulk BC, whereas later studies addressed atmospheric BC particles in the fine size range [Liousse et al., 1996; Cooke et al., 1999; Wang, 2004], as submicron particles have the largest radiative effects. In these studies, comparison of model calculated BC concentrations with measurements showed a general agreement within a factor of 2 [Liousse et al., 1996; Cooke and Wilson, 1996; Cooke et al., 1999]. The models tended to overestimate observed BC concentrations in remote and oceanic areas, and underestimate urban BC concentrations, whereas both underestimation and overestimation was found for rural measurements. Large uncertainties in the inventories of BC emissions were given as one of the main reasons for discrepancies between modeled and measured concentrations [e.g., Cooke et al., 1999]. Revised and generally lower emission factors combined with more detailed accounting for the effects of control devices has resulted in considerably lower emissions of BC compared with the previous estimates [Bond et al., 2004; Streets et al., 2001]. Use of the BC emission inventory developed by Bond et al.  in global model runs [e.g., Carmichael et al., 2003; Park et al., 2005; Dentener et al., 2006] has yielded much lower calculated load and surface concentrations of BC compared to earlier studies and often resulted in the underestimation of BC compared to observations.
 In Europe, Schaap et al.  performed a model assessment of the concentrations of fine BC for 1995. In that work, the authors constructed an independent European emission inventory for submicron BC, combining emission factors from Streets et al.  with the PM2.5 emissions from the CEPMEIP inventory (Coordinated European Programme on Particulate Matter Emission Inventories, Projections and Guidance). Schaap et al.  were concerned that with the updated emission estimates, calculated BC concentrations were underestimates compared to observations in Europe. On average, the modeled BC concentrations for 1995 were found to be a factor of 2 lower compared to data from measurements, conducted from end of 1980s to 2001. However, because of the lack of appropriate measurement data for 1995 any consistent verification of calculation results was not feasible.
 Since then, more measurement data of EC concentrations have become available in Europe, which has facilitated verification of model EC calculations. In particular in the period between 2002 and 2004, long-term measurements of EC concentrations were carried out in frameworks of two recent measurement campaigns: within the EU CARBOSOL project (Carbonaceous Aerosol over Europe, http://www.vein.hu/CARBOSOL [Legrand and Puxbaum, 2007]) and the EMEP EC/OC campaign [Yttri et al., 2007].
 In this work we present a model assessment of EC air concentrations in Europe for the years 2002–2004. The work goes beyond Schaap et al.  in that it explicitly analyses different reasons to explain the model EC results in Europe. Calculated EC concentrations are compared with observations from the CARBOSOL and EMEP measurement campaigns and the possible reasons of disagreements are discussed. On the basis of the comparison results between calculated and observed EC we make a preliminary evaluation of the quality of anthropogenic EC (and PM) emissions. We also estimate the influence of emissions from wildfires on EC concentrations in Europe. Further, we analyze to what extent various processes contribute to the uncertainties in model calculated EC concentrations.
2. EC/BC Terminology
 The terms elemental carbon (EC) and black carbon (BC) are often used alternatively in atmospheric chemistry literature. The choice and use of these synonyms is operationally justified and reflects the method of determination and purpose of study. For instance, climate modelers, who are concerned with radiative forcing of atmospheric carbon particles, typically deal with BC, which is the light-absorbing carbon and is composed of both elemental and organic carbon (OC) [Chýlek et al., 1999]. In this work, we focus on EC, as a chemical component of atmospheric PM which is not OC. In both CARBOSOL and EMEP measurement campaigns, thermo-optical methods were used, which by definition determine EC concentrations [Gelencsér, 2004]. Furthermore, most of underlying EC measurements for the emission inventory used the thermal method with an optical correction [Kupiainen and Klimont, 2006]. Therefore the term “elemental carbon” (EC) will be used henceforth.
3.1. EMEP Model Description
 The EMEP chemical transport model has been used to calculate concentrations of elemental carbon in Europe. The model calculation domain covers the whole of Europe, stretching over a large part of the North Atlantic and is resolved with grid cells of approximately 50 × 50 km2. In the vertical direction, the model employs 20 layers reaching a height of ca. 100 hPa. The model describes the emissions, chemical transformations, transport and dry and wet removal of gases and atmospheric particles. The meteorological fields from the weather prediction model HIRLAM are used to drive the simulations of transport and transformations of chemical species with the EMEP model. The description of the EMEP Unified model is given by Simpson et al. , Fagerli et al.  and on the EMEP website at http://www.emep.int. Here we briefly explain the treatment of atmospheric particles in the EMEP model.
 The chemical composition of atmospheric particles is described with seven components: sulfate, nitrate, ammonium, elemental and organic carbon, sea salt and mineral dust [Tsyro, 2005]. Formation of secondary organic aerosols (SOA) is not included in the standard EMEP model, but Simpson et al.  present model calculations of SOA, along with comparisons with measured OC, TC, levoglucosan and other components from the EMEP and CARBOSOL campaigns. The model distinguishes fine and coarse particles. Meteorology and land-use-dependent dry deposition velocities are calculated for two particle sizes, accounting for particle hygroscopic growth [Tsyro, 2006]. Size differentiated scavenging ratios are used in the model. EC hygroscopic properties and wet scavenging will be discussed in more detail in the next sections.
3.1.1. EC Hygroscopic Properties
 Freshly emitted anthropogenic EC is mostly hydrophobic. As it ages in the atmosphere, EC becomes hydrophilic because of the acquired coating of more hygroscopic species. The main processes involved in EC aging are condensation of inorganic and organic vapors (e.g., sulfuric and nitric acid, low volatility organic compounds), coagulation with soluble particles and oxidation of coating components on the surface of EC particles [Croft et al., 2005; Riemer et al., 2004]. Clearly, the rate and mechanisms governing the ageing process of EC in atmosphere vary in time and space depending on the chemical regime. In several modeling studies, EC aging rates corresponding to a half life of hydrophobic EC of about one day was found to give the best match between modeled and measured EC concentrations [Croft et al., 2005; Park et al., 2005; Hendricks et al., 2004].
 Different approaches have been used in global models to account for EC aging processes. In the work of Liousse et al. , EC is assumed to be hydrophilic from the instant it is emitted in the atmosphere. In other studies [e.g., Croft et al., 2005; Park et al., 2005; Hendricks et al., 2004], EC aging is described with a constant turnover rate, assuming transformation of hydrophobic EC to hydrophilic at timescales of 1–2 days (up to 6.7 days in the work by Cooke and Wilson ). The dependence of EC aging on the concentrations of sulfuric acid was first taken into account by Wilson et al. . In the work of Tsigaridis and Kanakidou , the aging rate of primary hydrophobic aerosol is calculated assuming that the ageing process occurs through oxidation of organic material that coats soot particles. Riemer et al.  used a mesoscale model to simulate soot ageing by coagulation and condensation in an industrialized environment, dominated by fossil fuel combustion. On the basis of the modeling results they constructed a simple parameterization, providing soot ageing rates dependent on the season, time of the day and the altitude.
 In the EMEP model, we have used for description of EC ageing the turnover rates for EC transformation from hydrophobic to hydrophilic based on the work by Riemer et al.  (Table 1).
Table 1. Turnover Rates of EC Transformation From Hydrophobic to Hydrophilic and Correspondent EC Aging Rates Used in Experiments AGE1 and AGE2 (τ is the Half Life Time of Hydrophobic EC)
Aging Rate, s−1
Three lowest model layers (below 250–300 m)
3.3 × 10−6
Model layers from 4th upward
1.4 × 10−4
1.0 × 10−5
3.1.2. Wet Scavenging
 Measured scavenging ratios for EC vary over a wide range for individual precipitation events because the efficiency of wet removal depends upon particle size and composition and upon the dynamic conditions of CCN formation and hydrometeor formation mechanisms [Penner et al., 1993]. For instance, formation of snow during the precipitation event will cause evaporation of cloud droplets and release of earlier scavenged aerosols, thus reducing scavenging efficiency. In several models the scavenging efficiency from convective clouds is assumed to be smaller than that from stratiform clouds as many convective events involve the formation of snow [Penner et al., 1993; Liousse et al., 1996; Cooke and Wilson, 1996; Koch, 2001; Wang, 2004]. On the other hand, nucleation scavenging can be efficient if the droplets formed on particles participate in riming [Scott, 1982]. The EMEP model does not presently distinguish between convective and stratiform clouds, and the scavenging ratio of Wv = 0.2 · 106 for hydrophilic EC has been used for the both types of precipitation. This value was adopted on the basis of the values of scavenging ratio derived from measurements of EC concentrations in precipitation during the CARBOSOL measurement campaign [Legrand and Puxbaum, 2007].
 The EC scavenging ratio of 0.2 · 106 is considerably smaller than the scavenging ratio for sulfate (which is 1.0 · 106 in the EMEP model) and probably lies at the lower range of scavenging ratios for internally mixed EC used typically in regional and global chemical transport models [e.g., Wilson et al., 2001; Iversen and Seland, 2002; Liao and Seinfeld, 2005]. Recent works has reported measured scavenging efficiencies for EC not being very much smaller than those for SO42− [Hitzenberger et al., 2000]. Except for the locations close to emission sources, the scavenging efficiencies for EC were about 20–30% lower on average than the scavenging efficiencies for SO42− and equaled the latter at very remote locations. To study the sensitivity of calculated EC concentrations to wet scavenging efficiency we have performed additional calculations with a scavenging ratio W = 0.7 · 106, which is about 30% smaller that the scavenging ratio for SO42− used in the model.
3.1.3. Model Experiments
 To study the effect of EC hygroscopic properties on calculated EC concentrations a series of model calculations have been performed (Table 2), in which EC was assumed to be either hydrophilic (PHIL), or hydrophobic (PHOB), or ageing (AGE1 and AGE2). In the experiments AGE1 and AGE2 we have assumed that 80% of emitted EC is hydrophobic and 20% is hydrophilic [e.g., Cooke et al., 1999]. The turnover rates for EC transformation from hydrophobic to hydrophilic and correspondent EC ageing rates used in the model simulations are as in Table 1. The hydrophobic fraction of EC was assumed not to be washed out from clouds, but hydrophobic EC could be scavenged by rain in the subcloud layers. In addition, no hygroscopic growth was assumed for the hydrophobic EC particles, which affected their dry deposition rate. For the hydrophilic fraction of EC we have used a scavenging ratio of Wv = 0.2 · 106 in AGE1 run and Wv = 0.7 · 106 in AGE2 run.
Table 2. Assumptions on EC Hygroscopic Properties Used in the Model Sensitivity Experiments
internally mixed, hydrophilic
0.2 · 106
initially hydrophobic, aging to hydrophilic
0.2 · 106
externally mixed, hydrophobic
as in AGE1 but more soluble
0.7 · 106
3.2. Emission Data
3.2.1. Anthropogenic Emissions
 Anthropogenic emissions of gases (SO2, NOx and NH3) used in the model calculations are from the EMEP emission database [Vestreng et al., 2006]. For PM2.5 and PM10, we have used emission estimates made at IIASA for the use within Clean Air For Europe (CAFÉ) programme (http://www.iiasa.ac.at/rains). For the countries not included in the IIASA's inventory, PM emissions from the EMEP emission database have been employed. Grid distribution of gaseous and PM emissions is based on a gridding methodology, described in the work of Tarrasón et al. . Further, the emissions of all pollutants are disaggregated by month and day of week, using the factors derived from data provided by the University of Stuttgart (IER). These factors are specific for each pollutant, emission sector and country. For PM2.5 and PM10 emissions monthly and weekly factors for NOx are currently used. Simple day-night factors are also applied. Emissions from high sources are distributed vertically (more details are given by Simpson et al. ).
 To make a chemical speciation of PM2.5 emissions we have used data from the inventory of submicron carbonaceous particles presented by Kupiainen and Klimont . The total European submicron EC emissions were estimated to be about 774 Kt in 2000 (here Europe includes whole of Turkey and European part of Russian Federation and emissions from forest fires are not included).
 These EC emission data were used to derive fraction coefficients for EC in PM2.5 emissions. Emissions of coarse EC have been derived from coarse PM (PM10–PM2.5) emissions, applying EC fractions estimated from the measurement data used by Kupiainen and Klimont . For the few countries which were not comprised in the EC emission inventory, EC fractions for the countries with expected similar economies have been applied. For more discussion on the carbonaceous emissions see also Simpson et al. .
Table 3 provides the total (fine + coarse) EC emissions for the year 2000 used in the model simulations, given for the SNAP 1 emission sectors (Source-Nomenclature for Air Pollution, [Richardson, 1999]). The main sources of EC emissions were residential/commercial combustion (40.4%), road transport (30.0%) and off-road mobile sources and machinery (16.4%). The spatial distribution of the EC emissions from all anthropogenic combustion sources used in model calculations is shown in Figure 1a.
Table 3. EC10 Emissions by SNAP 1 Sectors in Europe (EMEP Domain), Used in the Model Calculations (Kt) and Contribution From Different Sources to the Total EC10 Emissions in 2000
SNAP 1 Sector
Fraction of Total EC, %
Combustion in energy industries
Nonindustrial combustion plants/residential and commercial combustion
Combustion in manufacturing industry
Extraction and distribution
Other mobile sources and machinery
3.2.2. Wildfire Emissions
 In this work we have accounted for wildfire emissions using the Global Fire Emission Database (GFED2) [van der Werf et al., 2006]. The GFED database provides monthly emissions from biomass burning on a global scale (1 × 1°) over the period of 1997–2004. The total EC emissions from wildfires within the model calculation domain are 75 Kt in 2002 and 42 Kt in 2003. The annual wildfire EC emissions from the GFED2 database in 2002 and 2003, aggregated in the EMEP modeling grid, are shown in Figures 1b and 1c. Figure 1 shows that EC emissions from anthropogenic sources predominate over wildfire emissions in most of Europe. However, in some areas (e.g., in Baltic countries, north and northwest of Russia, Belarus, Kazakhstan, Portugal and Mediterranean countries), wildfire emissions can be comparable to anthropogenic EC emissions even on a yearly basis. Further, Figures 1b and 1c reveal a considerable year-to-year variability in the amount and location of wildfire emissions. It shows large EC emissions from wildfires in Baltic countries, Byelorussia, western Russia and Kazakhstan in 2002, while fires in Portugal and in Balkan countries were large sources of EC in 2003, which is in a general agreement with officially reported fire situations [European Commission Joint Research Centre (EC JRC), 2004, 2005; Goldammer et al., 2003; International Forest Fire News (IFFN), 2003, 2004]. As a first approximation, the wildfire emissions have been equally distributed between model vertical layers up to the altitude of about two boundary layer heights, which is loosely consistent with the maximum emission heights for European forest fires suggested in the work of Dentener et al. .
 Model calculated EC concentrations have been compared with observation data from two recent campaigns: (1) measurement campaign in the framework of EU CARBOSOL project and (2) EMEP EC/OC campaign, coordinated by the Norwegian Institute for Air Research (NILU).
 In the CARBOSOL campaign, weekly average concentrations of fine EC were measured at 6 stations located along the west-east transect across Europe, extending from the Azores to central Europe, for the period from July 2002 to October 2004 [Pio et al., 2007]. The samples were collected on quartz filters preheated for several hours at 500–700°C. Analyses of samples for OC/EC were made using a thermo-optical method [Pio et al., 1994]. The methodology used in the present study was intercompared with other methods, revealing that in general values of EC lie between those obtained with the IMPROVE method by Chow et al.  and the NIOSH protocol used in the Sunset Laboratory instrument [Schimdt et al., 2001]. In this work, we have primarily focused on comparison of model results with measurements at three surface sites (Table A1).
 During the EMEP OC/EC campaign, carried out in the period from July 2002 to June 2003, PM10 sampling was done at 14 stations distributed over 13 European countries (Table A2 and Figure 2b). The particle sampling was performed using CEN approved or equivalent PM10 gravimetric samplers, collecting one 24h sample every week. PM10 were collected on preheated quartz-fiber filters, preconditioned and postconditioned at NILU in accordance with the EMEP Manual for Sampling and Chemical Analysis [Norwegian Institute for Air Reseach, 1995]. All analyses of PM10 for EC mass were performed at the NILU laboratory, using the thermal optical transmission (TOT) instrument from Sunset Lab Inc., which corrects for charring during analysis. The instrument was operated according to a temperature program slightly modified from that of the NIOSH protocol. Measurement results from EC/OC campaign are discussed in detail by, e.g., Kahnert  and Yttri et al. . As the EMEP model calculates EC concentrations representative of regional background, data from two urban sites (Ghent and San Pietro Capofiume) have been excluded from comparison with calculated EC.
5. Base Case Simulations (AGE1)
5.1. EC Concentrations Over Europe
 Model calculated EC concentrations, averaged over 2002–2003, range between 0.3 and 1.0 μg/m3 in most of Europe (Figure 2a). The highest EC concentrations (1–2 μg/m3) are associated with large emission sources of primary particles in the Netherlands and Belgium and a number of big cities (e.g., Paris, Milan, Moscow), while the lowest EC (below 0.3 μg/m3) are calculated for the northern parts of Europe and parts of Spain. Comparison with observed EC concentrations from the EMEP campaign, averaged over the period from July 2002 to July 2003 (Figure 2b), shows that the model reproduces well the general features of EC distribution in Europe.
 As calculated with the model, the contribution of EC to PM2.5 mass varies from 3 to 7% across Europe on average for 2002–2003. The highest EC contribution to PM2.5, reaching 7–15% of PM2.5 mass, is calculated for areas with high primary PM emissions, e.g., from traffic and combustion in residential/commercial sectors in urban agglomerations and also from ship emissions in the Mediterranean Sea and the English Channel. Because SOA is not included in PM2.5 in these calculations, these results give upper estimates for the EC fraction in PM2.5. Further, the model calculations show that most of EC mass is associated with fine particles, which comprises over 90% of the total EC mass.
5.2. EC From Forest Fires
Figure 3 illustrates a large interannual variability of EC concentrations from wildfires, showing calculated annual mean EC for 2002 and 2003. In 2002, the largest EC concentrations due to fire emissions are calculated for western and central parts of Russia, Kazakhstan and Portugal, where annual mean EC concentrations reach 50–80 ng/m3 and even exceed 200 ng/m3 in areas of the most severe fires. The situation is quite different in 2003, when the highest EC concentrations from forest fires (up to 50 ng/m3) are calculated for Portugal, the Adriatic coast and Kazakhstan.
 On the annual basis, EC concentrations due to wildfires are from 1 to 10% of anthropogenic EC concentrations over most of Europe, however reaching 10–20% in Portugal and 20–30% in western Russia, Baltic countries and Kazakhstan in 2002. In 2003, which was a harsh fire season in southern Europe, the calculated forest fire contribution to the annual mean EC concentrations is from 5 to 15% in Balkan countries and exceeds 20% in Portugal. Although EC concentrations from wildfires are on average rather low compared to anthropogenic EC, fires can cause severe pollution episodes in the locations directly affected by the smoke plume. During such episodes, calculated EC concentrations due to forest fires are almost as high as EC concentrations from anthropogenic emissions (see also section 6.2).
5.3. Comparison With Observations
5.3.1. Overall Model Performance for EC
 On average, the model underestimates observed EC concentrations from the EMEP campaign by 19%, as shown on scatterplot in Figure 4. The spatial correlation between modeled and measured EC is quite high (R = 0.80), indicating the model capability to realistically reproduce the regional EC gradients.
Table 4 provides a summary of statistics of model performance for EC compared with observations at the CARBOSOL and EMEP sites. For the whole campaign periods, the model's relative bias ranges from −69% at Bragança to +77% at Birkenes. The model tends to overestimate EC concentrations for more northern sites in Nordic countries, UK, the Netherlands and Germany, whereas it tends to underestimate EC concentrations for stations located in central and southern parts of Europe (Czech Republic, Slovakia, Hungary, Italy, Portugal) compared to measurements. As a consequence, the model somewhat flattens the north-south EC gradient. The average temporal correlation between calculated and measured daily EC concentrations (weekly EC for Aveiro and K-Puszta) is 0.53. However, correlation coefficients range widely from the lowest R = 0.25 at Penicuik (GB46) and R = 0.36 at Aspvreten (SE12) to as high as R = 0.77 at Birkenes (NO01) and R = 0.79 at Mace Head (IE31). The lower correlations found for several sites are probably due to unaccounted effect of local emission sources or due to inaccurately described meteorology.
Table 4. Verification Statistics for Calculated EC Concentrations (in PM10 Unless Commented) Compared With Measurements at Surface Layer Sitesa
Obs and Mod are the mean observed and modeled EC concentration (μgC m−3), R is the coefficient of temporal correlation between calculated and measured EC concentrations, Bias is the relative bias, calculated as (Mod-Obs)/Obs × 100%.
 Cold and warm seasons differ with respect to the main sources of EC emissions. While emissions from mobile sources are distributed relatively equally over the year, the emissions from residential and commercial combustion are very low in the warm period: in the EMEP model the monthly time factors vary from about 0.2 in summer to 2.2 in winter on average for European countries. Therefore a separate for cold and warm periods analysis of the model results has been made as it might give an indication about the quality of emission data from different sources. Here, we have defined “summer” as a period from April through September and “winter” as a period from November through March. The model performance in terms of bias and temporal correlation appears rather mixed in the summer and winter seasons for the individual stations (Table 4). The main findings from the seasonal analyses are outlined below.
 The model results and observations show a general agreement about EC seasonal variation, with higher concentrations in the winter period than in the summer period for all of the sites (except PT01) (Table 4). The model tendency to flatten observed north-south EC gradient exists in the both seasons, but it is more pronounced in winter than in summer period.
 For the winter period, different features of the model performance for EC can be distinguished for northern and southern sites:
 For three Nordic sites (Birkenes, Aspvreten and Virolahti), the model considerably overestimates EC concentrations. In the Nordic countries, the emissions from residential/commercial combustion sector dominate total EC emissions (Table 5), especially in winter. Thus the model results suggest a possible overestimation of the EC emissions from this sector, or a problem with the model's representation of dispersion in the Nordic winter climates. However, the good temporal correlation found for these sites, as shown below, would tend to suggest emissions rather than a dispersion problem. The preliminary conclusion about a possible overestimation of residential/commercial combustion emissions for the Nordic countries is in agreement with the results presented by Simpson et al.  for levoglucosan, which can be regarded as a marker for wood burning emissions. It is shown there that the model tends to overestimate measured levoglucosan concentrations at the Nordic sites, that indicates that emissions from wood burning sources in Nordic countries may be overpredicted.
Table 5. Annual Mean Contributions From the Main Sectors to Fine EC Emissionsa
These numbers refer to anthropogenic sources excluding wildfires; remaining sources include industrial combustion and processes and open burning of waste and agricultural residues.
 In contrast, for the southern sites (Bragança and Aveiro in Portugal and Ispra in Italy), the model considerably underestimates EC concentrations in the winter periods, notably much larger than in summer. The contribution from the residential/commercial combustion sector to EC emissions is considerable in Portugal, though less so in Italy (Table 5). As shown by Simpson et al. , the model underestimates levoglucosan concentrations at Bragança and Ispra, which may indicate an underestimation of wood-burning emissions in those countries (or at least in the vicinity of the sites), or unaccounted local emissions. It should be pointed out that inaccuracies in modeled EC can also be related to the spatial distribution of residential/commercial combustion used in the calculations, which is based on the population distribution, so that large emissions are placed in urban areas. This is probably not a good approximation, especially for wood/fossil fuel burning for domestic heating, which largely occurs in rural areas or in city outskirts.
 In the summer period, EC concentrations are underestimated by the model by between 30 and 60% at seven of the stations. This is likely related to the uncertainties in emissions from mobile sources, which are the main EC source in summer. This EC underestimation can be related to the emissions amount as well as to their spatial distribution. We have looked at model results for NO2 for the sites with available measurements, for which traffic and off-road machinery are also the major mission sources. Table 4 shows that the model underestimates observed EC by 40% at Illmitz (AT02), by 58% at Kosetice (CZ03) and by 42% a Stara-Lesna (SK04). It also underestimates NO2 concentrations by from 30 to 50% at these sites (not shown here). Since the NO2 and EC are both underestimated for these sites, this may be an indication of a more general problem concerning the description of emissions from mobile sources in Austria, Czech Republic and Slovakia. There is also a possibility that the problems can be with vertical or horizontal dispersion of the emissions. It should also be mentioned that some minor sources of EC emissions in warm seasons, e.g., agricultural residue burning and backyard waste burning, might be underestimated as neither reliable statistics nor European measurements on that are available.
 Further, model EC overestimation at Kollumerwaard (NL09) is most likely due to uncertainties in EC emissions from mobile sources as contribution from residential/commercial combustion sources in the Netherlands is rather small. Since the model considerably underestimates NO2 concentrations at several Dutch sites (not shown), the EC overestimation at NL09 may indicate a problem with the emission factors used for either PM1 emissions or EC share in PM emissions in this region.
 The temporal correlation between calculated and measured EC concentrations is about the same in the winter and summer periods (R = 0.46 and 0.43 respectively). In winter, the highest temporal correlations (from 0.47 at CZ03 to 0.95 at IE31) are found for sites situated in the countries with dominating (or large) contributions from residential combustion sources (at sites NO01, IE31, FI17, PT01, CZ03, AT02), although the model representation of temporal variation of residential/commercial combustion emissions is presently associated with considerable uncertainties. Furthermore, emissions from residential wood burning could experience strong day-to-day variations caused by temperature changes, which is not yet taken into account in the model. Then good EC correlation in winter mainly indicates that the meteorological variability (dispersion, transport) is fairly well captured in the model. In the summer period, temporal correlations between calculated and measured EC are worse than in winter at most of these sites, indicating a poorer description of summer dispersion in the model.
6. Process Uncertainties (Sensitivity Tests)
6.1. Sensitivity Experiment for Wood Burning Emissions
 As discussed in section 5.3, for most of the sites where EC concentrations were underestimated (overestimated) by the model, concentrations of levoglucosan were similarly underestimated (overestimated) compared with observations [Simpson et al., 2007], suggesting a probable underestimation (overestimation) of the contribution from EC emissions from wood burning at those sites. We have performed a sensitivity experiment where PM2.5 (and thus EC) emissions from wood burning were increased or decreased by scaling factors, loosely based upon the ratios of model calculated to measured levoglucosan concentrations [Simpson et al., 2007]. The scaling factors for wood burning emissions used for different countries are shown in Table 6. Unfortunately, levoglucosan/OC ratios in wood burning emissions for different stoves, wood type and load vary widely (a constant ratio of 7.7% were used in model calculations used for this exercise), so such scaling factors are only an indication of possible problems with the inventory. These issues and examples of the applications of similar scaling to OC have been extensively discussed by Simpson et al. .
Table 6. Scaling Factors for Wood Burning Emissions From the Residential Sector, Based on Preliminary Results for Levoglucosan Presented in the Work of Simpson et al. a
Contributions from wood burning emissions to the total national EC emissions are shown in Table 7.
 Scaling of EC wood burning emissions has increased the spatial correlation between calculated and measured at EMEP sites EC concentrations from R = 0.80 to R = 0.87 for the whole measurement period. For the winter period, when scaling of wood burning emissions actually gives effect, the spatial correlation increased to R = 0.85 (Figure 5a) compared to R = 0.73 for the base emission run (Figure 5b). On the other hand, the general model EC underestimation has increased from 19% to 26% (and from 18% to 26% in the winter period) in the case of scaled emissions from wood burning compared to the base emission case. The scatterplots in Figure 5 help to understand these results. In the base case run, results are rather scattered about 1:1 line and model underpredictions and overpredictions of EC are counterbalanced for different sites. In the scaled emissions case, most sites lie closer around the 1:1 line, but a few sites (PT01, SK04, IT04) are still significantly underpredicted. This results in a larger overall model underestimation of EC in the latter case.
 The effect of wood burning emissions scaling on calculated EC concentrations is illustrated for some selected sites in Figure 5. Compared with observations, the most significant improvements from wood burning emissions scaling in calculated EC have been achieved for Avspreten (SE12) and Aveiro (Figures 5a and 5b), where model biases have decreased from +77% and −49% to only −4%, and for Bragança (PT01), where the model EC underestimation decreased from 69% to 56%. Also, temporal correlations between modeled and observed EC have improved considerably at these stations: increasing from 0.37 to 0.60, from 0.39 to 0.51 and from 0.36 to 0.51 respectively. At Birkenes (NO01), Kollumerwaar (NL09) and K-Puszta (KPZ) (Figure 5c), EC concentrations calculated in the case of scaled wood burning emissions are much closer to measured values than in the base case (while the temporal correlations have not changed significantly for these sites).
6.2. Effect of Wildfire Emissions
Table 7 compares the model performance for two simulations: with emissions from vegetation fires included and excluded. The sites where pronounced influences of wildfire emissions on EC concentrations have been calculated (Nordic, east European and Portuguese) are highlighted in bold. For the whole measurement period, the effect from accounting for wildfires on calculated EC concentrations is rather small. Still, the temporal correlations between calculated and measured EC concentrations become somewhat better when fire emissions are included in calculations. Presumably, this temporal effect would have been even more pronounced if we had forest fire emission data with better temporal resolution (monthly fire emissions have been used).
Table 7. Effect of Accounting for EC Emissions From Wildfires on Model Performance for ECa
Surface Layer Station
Sites experienced the influence of wildfires are highlighted in bold.
 The effect of forest fire emissions on calculated EC is more evident for the summer period. Accounting for forest fires has reduced the model underestimation of summer EC concentrations by between 4 to 15%. It has also improved the temporal correlation at most of the sites affected by fire emissions, e.g., from 0.46 to 0.56 at Aspvreten (SE12), from 0.20 to 0.47 at Viroilahti (FI17) and from 0.52 to 0.71 at Waldhof (DE02). However, at both Portuguese sites, Bragança (PT01) and Aveiro, and also at K-Puszta (HU02) and Birkenes (NO01) temporal correlation somewhat decreased, which is a likely consequence of using monthly averaged fire emissions.
 Some examples of pollution episodes associated with forest fires captured well by the model are shown in Figures 6–8. The model calculates enhanced EC concentrations at the Nordic sites (FI17, SE12, NO01) for August–September 2002, which could be related to severe peat fires in central and western Russia from the end of July to the beginning of September 2002 and forest fires in Ukraine and Belarus in September 2002 [IFFN, 2003; Cesari et al., 2005]. The fire smoke plume, moving from Russia over Scandinavia on 12 August 2002, is clearly seen in Figure 6a, which shows calculated contribution from fire emissions to EC concentrations. Niemi et al.  studied three strong PM2.5 episodes, which occurred on 12–15 August, 26–28 August and 5–6 September over large areas in Finland and were apparently caused by the emissions from forest and peat fires in Russia and other eastern European countries. These episodes were also observed in Sweden. The model reproduces well the wildfires pollution episodes documented by Niemi et al. , manifested by enhanced EC concentrations at FI17, SE12 and NO01 on 12 and 18 August 2002 (Figures 7a–7c).
 At the Portuguese site Bragança (PT01), the model reproduces elevated EC concentrations due to wildfires in August and beginning of September 2000 (Figure 7f). Exceptionally severe wildfires occurred in Portugal in the year 2003, particularly in August. As the EMEP measurement campaign ended up in July 2003, those fires were not recorded at Bragança. However, comparison with CARBOSOL measurements at the other Portuguese site Aveiro (AVE), taken through the whole 2003, shows that the model captures the fire smoke pollution episodes (Figures 6b, 7e, and 8b). In particular, it calculates elevated EC concentrations for the week 31 July to 7 August and a concentration peak for the week 8–14 August, which correspond well with two recorded critical fire periods: between 27 July and 4 August and between 7 and 12 August [EC JRC, 2004].
 Summarizing, our results show the importance of accounting for fire emissions for accurate calculations of EC concentrations in Europe. However, there is a strong need to develop time resolved (e.g., daily) wildfire emission data in order to predict accurately the EC episodes caused by forest fires.
6.3. Effect of EC Hygroscopic Properties
6.3.1. EC Aging
Figure 9a compares annual mean EC concentrations in 2003 calculated in the experiments PHYL (hydrophilic EC) and AGE1 (ageing EC) (Table 2). Calculated EC concentrations from AGE1 run are somewhat higher then those from PHYL run, as in AGE1 a portion of EC is hydrophobic and is removed inefficiently from the atmosphere. In most parts of Europe calculated EC concentrations are about 1–3% larger when EC ageing is taken into account as compared to results for hydrophilic EC. The largest effect of accounting for EC ageing on calculated EC concentrations (3 to 6%) are found in the areas with largest annual precipitation amount (e.g., northern slopes of Alps and Caucasus and the eastern coast of Adriatic Sea), as well as remote areas such as over Greenland.
Table 8 provides verification statistics of calculated EC compared with EMEP measurements, averaged over all sites, from model runs for hydrophilic (PHYL), aging from hydrophobic to hydrophilic (AGE1) and hydrophobic (PHOB) EC. Accounting for EC ageing has only a slight effect on model performance compared to the PHYL run at the considered stations. On average, model EC underestimation decreases from 21% in the case of hydrophilic EC to 19% in the case of aging EC (AGE1 case). In the extreme simulation case of hydrophobic EC the model still underestimates EC, though the underestimation is reduced now to 8%. For the individual sites, the increase in calculated EC concentrations varies from 0.9% at Bragança to 4.4% at Aspvreten as a result of accounting for EC aging compared to PHYL case. The spatial and temporal correlations between calculated and observed EC concentrations remain practically unchanged for PHYL, AGE1 and PHOB experiments.
Table 8. Verification Statistics for EC Concentrations Calculated in Different Test Runs Compared With Measurements at All Sites From the EMEP EC/OC Campaigna
PHIL W = 0.2 · 106
AGE1 Wphil = 0.2 · 106
AGE2 Wphil = 0.7 · 106
PHIL, hydrophilic EC; AGE, aging EC; PHOB, hydrophobic EC (W, scavenging ratio used for hydrophilic EC fraction).
Relative bias, %
6.3.2. Wet Scavenging Ratio
 The calculated relative effect of increasing scavenging ratio for soluble fraction of EC from 0.2 · 106 to 0.7 · 106 on annual mean EC concentrations for 2003 is shown in Figure 9b. The increase of scavenging ratio by a factor of 3.5 has resulted in a decrease of calculated EC concentrations by between 5% and 25% in most of Europe and up to 30–40% in the remote areas in Scandinavia, northern Russia and Kazakhstan. This is because EC concentrations close to the source are mainly determined by the emission strength and the efficiency of removal by advection and to a smaller extent by wet scavenging. On the other hand, in remote locations concentrations are rather sensitive to wet removal rates [Liousse et al., 1996]. Compared to the EMEP observations, the average model EC underestimation in this experiment (AGE2) is 29% (in Table 8) compared to 21% in the run AGE1. The increase of scavenging ratio has not changed the spatial correlation between calculated and measured EC.
6.3.3. Result Discussion
 In our model calculations, accounting for EC ageing from hydrophobic to hydrophilic has been shown to have only a limited effect on model EC results compared to observations. Partly, this is biased by the location of the measurement sites considered here. Most of them are located in central Europe (Figure 2b), where the effect from EC aging has been shown to be the smallest (Figure 9a). In these areas, anthropogenic gaseous emissions are relatively large compared to primary PM (and EC) emissions and therefore the process of EC coating with hygroscopic components is quite efficient (see estimated turnover rates for hydrophobic EC for industrialized environment in Table 1). Rodhe and Grandell  estimated EC residence time with respect to its wet removal to be about 5 days in the European atmosphere. Thus, in the pollution conditions prevailing in Europe, emitted EC remains airborne for a long enough time to become hydrophilic before it is scavenged by precipitation.
 However, the effect of accounting for EC aging compared to assuming internally mixed EC can be larger in different conditions. For instance, this effect can be greater in areas with relatively large primary PM emissions, where EC aging will be slower as a smaller amount of hygroscopic species is available for EC coating. Consequently in such areas, emitted EC will remain hydrophobic and airborne for a longer time, and the effect of accounting for EC ageing on modeled EC will be greater compared to hydrophilic EC calculations. Therefore, for more accurate characterization of EC hygroscopic properties, EC aging rates should be calculated in the model dependent on the presence of condensable vapors and other hydroscopic aerosols rather than using a constant ageing rate.
 Furthermore, the effect of EC aging on calculated EC concentrations appears to be larger in the areas with more precipitation, e.g., over the Alps and along the eastern coast of the Adriatic Sea (Figure 9b). In the conditions of more frequent precipitation, less hydrophobic EC will be transformed to hydrophilic form in the case AGE1, resulting in a greater difference in model results from ageing EC and hydrophilic EC calculation cases. In contrast, in dryer areas with more rare precipitation events (e.g., in Spain, parts of France in 2002) emitted EC will have plenty of time to become hydrophilic.
 Finally, surface EC concentrations in polluted areas (and at the stations considered) are determined to a larger degree by short-range emissions and dispersion that by wet deposition. On the other hand, EC concentrations in the remote areas will be more affected by wet scavenging as more EC gets lifted up to the cloud levels during the time of transport. Our sensitivity tests on the effect of EC aging on EC air concentrations are in agreement with results presented by Cooke and Wilson . They found the largest differences in calculated EC concentrations due to changes in scavenging efficiency (up to a factor of 3) for very remote areas, e.g., over the Pacific Ocean, whereas only minor differences were calculated over Europe.
6.4. Uncertainties Related to EC Measurements
 In the sections above, a series of experiments were presented, in which we tried to reconcile EC concentrations calculated with the EMEP model with observed EC concentrations. We have also discussed uncertainties in modeled EC concentrations and tried to explain the discrepancies between calculated and measured EC in terms of uncertainties in emissions and modeling. However, uncertainties in EC measurement data need to be considered as well.
 The definition of EC depends on the analytical method applied to quantify it. The results obtained with different sampling procedures and analytical techniques can differ considerably. Differences of up to a factor of four in EC concentrations obtained with different methods were reported in the work by ten Brink et al. . Thermo-optical methods for determining EC concentrations are among the most commonly applied and were used in both CARBOSOL and EMEP measurement campaigns.
 However, many thermal evolution protocols for distinguishing between OC and EC have been developed for each of the analytical methods, what introduces further uncertainties into the measured EC concentration. As reported by Chow et al. , EC concentrations determined with the NIOSH protocol were typically less than half of those determined with the IMPROVE protocol.
 In the same work it was also found that lower EC concentrations were obtained for the pyrolysis correction (i.e., pyrolitical generation of EC from OC) determined by transmittance method, than by the reflectance method for both thermal protocols. Furthermore, the presence of inorganic compounds (e.g., ammonium bisulfate) was shown to affect OC charring in thermal optical analysis [Yu et al., 2002]. Novakov and Corrigan  reported that the presence of potassium and sodium also changed the thermal evolution of EC.
 Given significant and often nonquantified uncertainties in measured EC concentrations, one should be particularly careful validating model EC results with observations. A valuable finding in the work of ten Brink et al.  was that although the differences between different methods were large, they were apparently of a systematic nature. This means that EC concentration data obtained with different methods can be useful to look at the temporal variation of EC. Furthermore, EC concentrations determined with the same analytical method, as was done in both CARBOSOL and EMEP measurement campaigns (albeit with different methods in each campaign), are also applicable for studying EC spatial variations. So, although uncertainties on the measurements of EC are still large, the existing available data provide useful information on the temporal and spatial variations of EC over Europe.
 In recent years, new up-to-date EC emission inventories have been developed and more EC measurements have become available in Europe, which facilitates the model assessment of EC concentrations in Europe and the validation of model results. The regional EMEP model has been applied to establish the relation between the emissions of EC, derived from the new EC emission inventory of Kupiainen and Klimont , and ambient concentrations of EC in Europe. The calculated EC concentrations have been compared with observation data from CARBOSOL and EMEP EC/OC measurements campaigns, covering 2002–2004.
 Modeled EC concentrations are on average 19% lower than EMEP observations, and the spatial correlation between calculated and measured EC is 0.80. The model somewhat flattens the north-south EC gradient as it tends to overestimate EC for Nordic sites and underestimate EC for more southern sites, and this model tendency is more pronounced in winter than in summer. The average temporal correlation coefficient between calculated and measured EC is 0.53, varying from as high as R = 0.79 at Mace Head (IE31) and R = 0.77 at Birkenes (NO01) to as low as R = 0.26 at Penicuik (GB46).
 A series of model experiments have been carried out in order to study the uncertainties in modeled EC and to explain the discrepancies between EC calculations and observations. The main conclusions are summarized below.
 1. For anthropogenic emissions, seasonal analysis and sensitivity tests suggest that (1) EC emissions from residential combustion, in particular from wood burning, are likely to be overestimated in Nordic countries; (2) EC emissions from residential combustion, in particular from wood burning, are likely to be underestimated in central and southern Europe (i.e., Portugal, Italy, Hungary); and (3) EC emissions from mobile sources may be underestimated in several European countries (i.e., Portugal, Austria, Czech Republic, Hungary, and Slovakia).
 2. For wildfire emissions, (1) on average, the calculated effect of fire emissions on EC concentrations is rather limited in Europe, and calculated contributions from forest fires to the annual mean EC range between 1 and 10% in most of Europe, but can be as important as anthropogenic EC during the pollution episodes caused by forest fires. (2) Accounting for emissions from forest fires decreases model EC underestimation and increases temporal correlation of calculated EC compared with measurements for several sites.
 3. A description of EC hygroscopic properties is as follows: (1) Compared to results for hydrophilic EC, accounting for EC ageing has only a limited effect on modeled EC in most parts of Europe, resulting in an increase of calculated EC concentrations by 1% to 4% for the measurement sites considered, and (2) EC wet scavenging ratios have a moderate effect on model calculated EC concentrations, e.g., increase of scavenging ratio from 0.2 · 106 to 0.7 · 106 resulted in the decrease of annual mean EC by 10%. The effect is larger in remote areas (30–40%). However, EC scavenging ratios are still rather uncertain as the values derived from different measurements range widely;
 The modeling results presented are indicative of considerable uncertainties around EC (and PM) emission inventories for some European countries. There are indications of a possible underestimation of EC emissions from traffic in some areas and both underestimation and overestimation of EC emissions from residential combustion for some European countries. The largest uncertainties probably lie in EC emissions from residential wood/fossil combustion and are associated with both emission factors and with emissions spatial and temporal variation. In addition, some minor sources (e.g., agricultural residue burning and backyard waste burning) might be underestimated as neither reliable statistics nor European measurements are available. Also, more accurate and time resolved information on emissions from wildfires is urgently needed. Furthermore, the extension of the monitoring networks measuring EC is essential for a more accurate representation of the expected EC gradient across Europe and the EC temporal variations and to facilitate model validation. Finally, regarding results presented in this work, one should bear in mind that there are considerable uncertainties in measured EC concentrations associated with analytical methods applied. This can affect the conclusions from validation of model EC results with observations.
 In this work, model calculated EC concentrations have been compared with observation data obtained from two measurement campaigns: (1) measurement campaign in the framework of EU CARBOSOL project and (2) EMEP EC/OC campaign, coordinated by the Norwegian Institute for Air Research (NILU). The list of the measurement sites and their characteristics are provided in Tables A1 and A2.
Table A1. Details for CARBOSOL Surface Measurement Sites (October 2002 to July 2004)
Table A2. Sampling Sites in the EMEP EC/OC Campaign (July 2002 to June 2003)
San Pietro Capofiume
 The work was supported by the EMEP project under UNECE, the Nordic council of Ministers NORPAC project, and the EU project CARBOSOL. Z. Klimont was supported by the European Network of Excellence ACCENT (Atmospheric composition Change) of the European commission.