A 2.5 year’s source apportionment study of black carbon from wood burning and fossil fuel combustion at urban and rural sites in Switzerland

. The contributions of fossil fuel (FF) and wood burning (WB) emissions to black carbon (BC) have been in-vestigated in the recent past by analysis of multi-wavelength aethalometer data. This approach utilizes the stronger light absorption of WB aerosols in the near ultraviolet compared to the light absorption of aerosols from FF combustion. Here we present 2.5 years of seven-wavelength aethalometer data from one urban and two rural background sites in Switzerland measured from 2008–2010. The contribution of WB and FF to BC was directly determined from the aerosol absorption coefﬁcients of FF and WB aerosols which were calculated by using conﬁrmed ˚Angstrom exponents and aerosol light absorption cross-sections that were determined for all sites. Reasonable separation of total BC into contributions from FF and WB was achieved for all sites and seasons. The obtained WB contributions to BC are well correlated with measured concentrations of levoglucosan and potassium while FF contributions to BC correlate nicely with NO x . These ﬁndings support our approach and show that the applied source apportionment of BC is well applicable for long-term data sets.


Introduction
Atmospheric aerosol particles affect chemical, microphysical, and radiative atmospheric processes. They are important when considering both the natural and the anthropogenic climate forcing (Forster et al., 2007). An abundant constituent in atmospheric aerosols is carbonaceous matter (CM), which is composed of black carbon (BC) and organic carbon (OC). BC is the light absorbing part of carbonaceous material, which has a wavelength independent imaginary part of the refractive index. It is commonly referred to as soot. Compared to other aerosol constituents BC has very different optical and radiative properties, contributing significantly to current global warming (Jacobson, 2001(Jacobson, , 2010Forster et al., 2007;Ramanathan and Carmichael, 2008).
Besides their effects on the Earth's radiation budget, fine carbonaceous particles have been found to cause serious health effects as they penetrate into the human respiratory system (Oberdorster et al., 2002;Jerrett et al., 2005;Kennedy, 2007). Epidemiologic studies associated BC in particular with respiratory health effects in children and with cardiovascular diseases (Peters et al., 2000;Gauderman et al., 2004).
The major emission sources of soot particles in large parts of Europe and Switzerland are diesel engines and incomplete biomass burning. Especially in winter, wood combustion from domestic heating has been found to be a major contributor to air pollution in residential areas (Szidat et al., 2007;Lanz et al., 2008). Given that wood burning (WB) is a CO 2 neutral energy source its impact on air quality is likely to become more and more relevant in the coming years.
Published by Copernicus Publications on behalf of the European Geosciences Union. Several commercial instruments are available for the determination of BC in particulate matter (PM). In this study, the aethalometer (AE, Hansen et al., 1984) was used for continuous measurement of BC. The working principle of an AE is the following: aerosols are collected on a quartz fibre filter and illuminated with light. The aerosol absorption coefficient b abs is then calculated from the measured light attenuation. Corrections of the measured light attenuation artefacts such as multiple scattering and so called "shadowing" are typically applied (Weingartner et al., 2003;Collaud Coen et al., 2010 and references therein). However, note that further possible systematic errors need to be considered for correct determination of b abs using a filter based method (Subramanian et al., 2007). The BC mass concentration is obtained from b abs divided by the mass specific aerosol light absorption cross section σ abs . Newer AE instruments operate at several wavelengths ranging from the near-ultraviolet (UV) to the near-infrared (IR). The wavelength dependence of the aerosol absorption coefficient b abs can be described by the power law b abs (λ) ∼ λ −α , where λ is the wavelength of the light beam and α is theÅngstrom exponent. The spectral dependence allows distinguishing carbonaceous aerosols from different sources. This is because of light absorbing OC which in contrast to BC exhibits a stronger absorption at shorter wavelengths (Andreae and Gelencser, 2006;Lukács et al., 2007;Moosmüller et al., 2009). For example, biomass burning aerosols are known to contain a significant number of light absorbing organic substances or brown-carbon and have a strong spectral dependence (α > 2) while emissions from diesel engines contain primarily BC and have a weak spectral dependence (α ∼ 1) (e.g. Kirchstetter et al., 2004;Clarke et al., 2007). Under certain conditions the range of Angstrom exponents for brown carbon and BC may possibly be wider (Lack and Cappa, 2010).
In the past multiple wavelength AEs were deployed to determine the contributions of traffic and WB to total CM. This was accomplished with a source apportionment model introduced by Sandradewi et al. (2008a). Sandradewi et al. (2008a) collected AE data in winter during a measurement campaign in an alpine valley where WB from domestic heating and traffic emissions were the dominating sources of CM. Linear regression of CM against the aerosol absorption coefficient of FF combustion aerosols in the infrared (950 nm) and the aerosol absorption coefficient of WB aerosols in the UV (470 nm) was proposed for source apportionment. The authors estimated an average contribution of WB to total CM of 88 %. In a recent study, Favez et al. (2009) applied the same approach to data sampled in urban Paris during a winter field campaign. The authors determined regression coefficients similar to Sandradewi et al. (2008a) and estimated the average contribution of WB to total CM to be 46 %.
In this study we applied the AE model to data collected from 2008 to 2010 at three measurement sites in Switzerland. To our knowledge this is the first long-term source apportionment study using this modelling approach. In a first step we focused on CM. Sensitivity tests for different regression models and for variousÅngstrom exponents were performed. It was found that the regression modelling approach is not suitable for our long-term datasets because of significant fractions of CM resulting from sources and processes other than FF and WB. Thus in a second step we focused on the contributions of FF combustion and WB to BC which was calculated directly from b abs by the use of site specific σ abs values. We determined the fractions of BC resulting from emissions of FF combustion and WB on a seasonal and a daily basis and compared our findings with measured concentrations of tracers for WB (levoglucosan and potassium) and with estimated elemental carbon emitted by WB as obtained from a receptor modelling study.
2 Experimental procedure

Sampling sites and instrumentation
Measurements were performed at one urban and two rural stations of the Swiss National Air Pollution Monitoring Network (NABEL) (EMPA, 2010) from 2008 to 2010 as listed in Table 1. The NABEL station Zurich-Kaserne (ZUE) is an urban background site located in a courtyard in the city centre of Zurich (47 • 22 ′ N, 8 • 32 ′ E, 410 m a.s.l.). The location is surrounded by roads with rather low traffic and is not affected by major emissions from industries. The stations Payerne (PAY) and Magadino-Cadenazzo (MAG) are rural sites. PAY is located in the western part of the Swiss Plateau one kilometre outside of Payerne, a small city with 8000 inhabitants. The site is surrounded by agricultural land (grassland and crops), forests and small villages (46 • 48 ′ N, 6 • 56 ′ E, 489 m a.s.l.). The MAG site is located south of the Alps in the Magadino plane close to the Lago Maggiore (46 • 09 ′ N, 8 • 56 ′ E, 204 m a.s.l.) and about two kilometres outside of Cadenazzo, a village with 2000 inhabitants.
Multiple-wavelength AEs (Magee Scientific, USA, model AE31) were deployed at all measurement sites for determination of BC. All instruments were equipped with PM 2.5 inlets. The AE continuously detects the aerosol attenuation coefficient of the collected aerosol particles b ATN (λ) at seven wavelengths λ (370, 470, 520, 590, 660, 880 and 950 nm) with a time resolution of 5 minutes. The calculation of the aerosol absorption coefficients b abs (λ) was done according to the data correction procedure by Weingartner et al. (2003) b abs (λ) = b abs (λ) C · R (ATN λ ) . (1) A constant factor C = 2.14 is applied to correct for multiple scattering of the incident light at the filter fibres in an unloaded filter. This constant factor was determined by Weingartner et al. (2003) in studies using pure soot particles with known absorption coefficient. In order to correct for increasing light attenuation due to accumulating particles in the filter (shadowing effect) an empirical function of the measured light attenuation at the different wavelength ATN λ is used The f λ in Eq.
(2) are constants, here we used the mean values of the f λ 's found by Sandradewi et al. (2008b) during campaigns in summer and winter at an rural site in a Swiss alpine valley. Further systematic errors in filter based aerosol light absorption measurements are possible (Subramanian et al., 2007), but neglected here. This has however no effect on the presented results on the source apportionment of BC, because systematic errors in the aerosol absorption coefficients would be compensated by the obtained values for σ abs .
Parallel to the measurement of b abs (λ) at the three sites, daily PM 2.5 and at MAG daily PM 10 samples were collected at every twelfth day on quartz fibre filters (Pallflex Tissuquartz 2500QAT) using a high-volume sampler (Digitel DHA-80, 30 m 3 h −1 flow rate). Punches of these PM 2.5 and PM 10 filter samples were analysed for organic and elemental carbon (OC and EC) by applying the thermal optical transmission method (TOT). The OCEC analyzer (Sunset Laboratory Inc.) was operated with the EUSAAR2 temperature protocol (Cavalli et al., 2010). The BC mass concentration is calculated from b abs (λ) divided by the wavelength dependent aerosol light absorption cross section σ abs , denoted as σ abs (λ). Here we determined site specific σ abs (λ) from the slope of the linear regression of the daily means of b abs (λ) against the EC concentration. Linear regression models individually for different seasons indicate higher aerosol light absorption cross sections at the near-UV wavelengths (370 nm and 470 nm) during the cold season. However, the seasonal dependence of the light absorption cross section is not significant at the 95 % confidence level, the number of data pairs for each season is currently too small. Therefore, light absorption cross sections were determined for data from all year, -the obtained coefficients of determination were R 2 > 0.94 (PAY), R 2 > 0.68 (ZUE) and R 2 > 0.9 (MAG). Table 2 summarizes the determined site specific values for σ abs (λ).
At PAY, OC and EC concentrations are additionally available as 3-hourly mean values from a semi-continuous OCEC analyzer (Sunset Laboratory Inc.; thermal optical transmission method, EUSAAR2 temperature protocol). This instrument was also equipped with a PM 2.5 inlet.
Finally, measurements of potassium and levoglucosan in PM 10 are available from PAY, potassium concentrations are also available in PM 10 from MAG. The concentration of water soluble potassium was determined by ion-chromatography (Dionex IC 3000) after extraction of punches (2.5 cm diameter) of the daily PM 2.5 (PAY) and PM 10 filter samples (ZUE) in 40 ml of nanopure water during ≈15 h. Levoglucosan concentrations at PAY were determined by NILU as part of the intensive measurement periods of the European Monitoring and Evaluation Programme (EMEP) in fall 2008 and spring 2009. The applied method is described in Dye and Yttri (2005).
Potassium concentrations were determined in daily PM 10 samples collected at every fourth day from August 2008 to July 2009. Levoglucosan was determined in eight approximately weekly PM 10 samples collected with a low volume sampler (Rupprecht and Patashnik, model Partisol FRM2000) in fall 2008 and spring 2009.

The aethalometer model
The AE model aims to quantify the contribution of fossil fuel (FF) and WB aerosol to the BC concentration. The model has been described in detail by Sandradewi et al. (2008a). Briefly, it relies on two assumptions (a) that during winter FF combustion and WB emissions from domestic heating are the dominating sources of CM, and (b) that total ambient CM can be modelled by the light absorption of aerosols emitted by these two sources. The first assumption implies that the aerosol absorption coefficient b abs (λ) at a given wavelength λ can be expressed as the sum of the light absorption of aerosols emitted by these two sources: (3) b absFF (λ) and b absWB (λ) are the wavelength dependent aerosol absorption coefficient of BC from fossil fuel and wood burning emissions, respectively. Both quantities can be calculated from light absorption measurements if the spectral dependence (expressed by theÅngstrom exponent α) for both sources are known. With given α FF and α WB and two different wavelength λ 1 and λ 2 , the following equations apply: Light absorption measurements at λ 1 = 470 nm and λ 2 = 880 nm (or λ 2 = 950 nm) are used in this approach. This is due to the fact that BC from FF combustion has a weak dependence on wavelength whereas BC from WB shows enhanced absorption at shorter wavelength.
Equations (1) to (3) can be used to calculate b absFF (880 nm) and b absWB (470 nm). In the approach by Sandradewi et al. (2008a), the mass concentration of total carbonaceous matter was regressed against b absFF (880 nm) and b absWB (470 nm) for determination of the contribution of fossil fuel combustion and wood burning to CM, with the parameters C 1 and C 2 relating the aerosol absorption coefficient to the total carbonaceous mass concentration.
In this study, the contributions of FF and WB to total BC (BC FF and BC WB ) are also directly calculated by assuming that the light attenuation cross sections for aerosols from FF combustion and WB (σ absFF (λ) and σ absWB (λ)) can be represented by the average site specific σ abs (λ) as indicated in Table 2 and BC WB = b absWB (470 nm)/σ abs (470 nm).
This simplification is justified by the absence of a seasonal cycle of the determined σ abs (λ) and the high correlations of measured aerosol absorption coefficients and EC leading to rather small uncertainties of the average σ abs (λ) ( Table 2). Therefore, the varying impacts of sources and processes seem to have a small or negligible influence on σ abs (λ). Thus this simplification seems not to introduce significant uncertainty or bias. Note that the derived BC FF and BC WB are consistent with elemental carbon concentrations because the aerosol light absorption cross sections were calculated from EC analysis using the thermal optical transmission method.

Light absorption measurements
For PAY, MAG and ZUE theÅngstrom exponent α was calculated over all seven wavelengths and for different time intervals. α was thus determined from power law fits of b abs (λ)/b abs (950 nm) as a function of the wavelength.   Kirchstetter et al., 2004;Bergstrom et al., 2007) but very similar to values observed at populated and polluted sites during winter (Favez et al., 2009;Yang et al., 2009). The diurnal patterns show at all stations lowest α during daytime and especially during the morning rush hours. This indicates the influence of particulate matter emitted by road traffic. Figure 2a and b include OC/EC ratios derived from 3-hourly concentration measurements at PAY. The mean OC/EC ratio at PAY varies from 4 to 8 during summer with the daily minimum between 06:00 a.m. and 09:00 a.m. and the maximum from 12:00 p.m. to 03:00 p.m. During winter the variability of the OC/EC ratio is reduced, the observed OC/EC ratios vary between 4 and 6.
During summer,Ångstrom exponents less than 1 are observed at all measurement sites. Such low α values have been reported before (Gyawali et al., 2009 and references therein). Gyawali et al. (2009) show that values of α < 1 may occur for large aerosol particles and/or aerosol particles of certain shape and mixing state. For example, the authors attributed α < 1 to aerosol particles that consist of a collapsed soot core and a coating shell of organic and inorganic secondary aerosol. Especially in MAG where the lowest α values were observed during summer it is known from filter analyses that organic matter contributes on annual average 40 % to the total PM 10 mass (M. Gianini, personal communication, 2010). This is high compared to rural sites north of the alps.
At PAY, MAG and ZUE α values are in summer ≈0.9 ± 0.1 during the morning rush hours. At the same time the OC/EC ratio at PAY is at a daily minimum. This observations point to an impact of road traffic emissions, we therefore attribute α ∼ 0.9 to BC from road traffic emissions. As discussed earlier, previous studies often reported values of α ∼ 1 for diesel soot experiments. However site specific traffic values may differ from the laboratory results. Our measurement stations are neither located nearby roadways nor directly situated close to other primary emission sources. Sampled traffic related particulate matter may have likely undergone some aging and may be collapsed or coated by the time the sampling site is reached. Here we choose theÅngstrom exponent α FF = 0.9. The absorption coefficients α WB = 1.9 is taken from literature (Sandradewi et al., 2008a). Sandradewi et al. (2008a) estimated the contribution of FF and WB to carbonaceous matter by multiple linear regression according to Eq. (6). This approach implies that no other sources than FF combustion and WB and no relevant CM formation process (e.g. formation of secondary organic aerosols SOA) are present. This assumption might be adequate during winter and for certain locations as shown in  . 3. Long-term pattern of BC FF and BC WB concentrations at PAY, MAG and ZUE (stacked). The scatter plots show the relationship between total daily BC and the daily EC concentration as determined with the thermal optical transmission method at every 4th day from 8 August to 9 July for the corresponding measurement site. Sandradewi et al. (2008a). The approach was also applied by Favez et al. (2009), both studies found similar model parameters C 1 and C 2 for the short measurement campaigns performed during winter.

Test of applicability
In a first try to test this source apportionment approach, we calculated total CM concentration from Equation (6) by using the values for C 1 and C 2 from (Sandradewi et al., 2008a;Favez et al., 2009) and b absFF (λ 1 ) and b absWB (λ 2 ) as derived from our measurements using Eqs. (3) to (5). We found systematic differences between calculated CM and CM determined from measured OC and EC (denoted as measured CM). During summer the calculated CM was less than 50 % of the measured CM which might result from large contributions by sources and processes other than FF combustion and WB during summer (e.g. SOA formation). During winter the calculated CM was about 25 % larger than measured CM.
In a second try we performed regression modelling (with and without intercept, i.e. C 3 ) and with varyingÅngstrom exponents α for estimating the contribution of FF combustion and WB to CM (Eq. 6). The modelling was performed for all available AE data as well as for winter data only.
This approach leads to a satisfactory agreement between measured and modelled CM, however, the standard error of the estimated C 1 , C 2 (and where applicable C 3 ) is around ±30 % allowing no meaningful quantification of source contributions. In addition the sensitivity of C 1 and C 2 on the chosenÅngstrom exponents for aerosols from FF combustion and WB is high leading to a further increase in uncertainty.
Error estimates in the AE model are only sparsely discussed in literature. For example, Sandradewi et al. (2008a) and Favez et al. (2009) give no information about the uncertainty of the estimated parameters of the regression models. Also subsequent studies that include an intercept in the regression approach give only limited information about model errors (Sandradewi et al., 2008c;Favez et al., 2010).
Our investigations imply that the simple two sources approach expressed by Eq. (6) is not adequate for the longterm datasets from the three considered measurement sites because of significant contributions to CM from additional sources and processes.

Source apportionment of BC
As an alternative to the approach described in Sect. 3.2.1, we determined the contribution of FF combustion and WB to BC (BC FF and BC WB ) at PAY, MAG and ZUE. As mentioned in Sect. 3.1 we used α FF = 0.9 and α WB = 1.9 to determine b absFF and b absWB . BC FF and BC WB are directly determined from b absFF (880 nm) and b absWB (470 nm) using the site specific values for σ abs (λ) (see Table 2). Figure 3 shows the calculated BC FF and BC WB concentrations at the PAY, MAG and ZUE sites from April 2008 to October 2010 as daily mean values. For each measurement site the relationship between total BC and the EC concentrations In general, the total BC concentration is in winter at all stations substantially higher than in summer. On the one hand this can be explained by different meteorological conditions, e.g. in winter frequent temperature inversions lead to a reduced vertical mixing of the air and to an accumulation of air pollutants within the boundary layer. On the other hand emissions from WB have an additional impact on total BC concentrations during the cold season.
At the rural measurement site PAY the average BC concentration during winter (DJF mean) is 0.78 ± 0.05 µg m −3 with 33 ± 12 % of the total BC resulting from WB emissions. In summer (JJA mean) the BC concentration is on average 0.44 ± 0.03 µg m −3 with WB contributions of 6 ± 10 %. A similar seasonal pattern for BC can be observed in MAG. This rural site south of the alps shows highest BC concentrations during winter with a mean of 2.29 ± 0.14 µg m −3 where 30 ± 11 % of total BC are estimated as BC WB . During summer the mean BC concentration is 0.83 ± 0.05 µg m −3 , almost all of the BC during summer is found to result from emissions of FF combustion. The seasonal differences in BC concentrations are much lower at the urban background site ZUE compared to the rural sites. During summer the average BC concentration is 1.19 ± 0.37 µg m −3 compared to 1.54 ± 0.48 µg m −3 during winter. BC WB accounts for 10 ± 8 % of total BC during summer and 24 ± 11 % during winter. In 2002and 2003, Szidat et al. (2006 performed 14 C analyses of elemental carbon in PM 10 collected at the measurement site ZUE. The authors reported contributions of biomass burning to EC as 6 ± 2 %, 12 ± 1 %, and 25 ± 5 % for PM 10 samples from summer, spring and winter, respectively. These results are in good agreement with the WB contributions determined in this study.
Beside the seasonal averages, the contributions of FF combustion and WB to total BC have also been evaluated on a higher temporal resolution. Diurnal cycles of BC FF , BC WB and total BC have been calculated as 3 h mean values and are shown in Fig. 4.
In summer mean concentrations of BC in PAY are generally below 0.5 µg m −3 , both at weekdays and weekends, all of the BC is identified as BC FF . BC concentrations are generally lowest during noon and afternoon which can be attributed to an increasing boundary layer height and corresponding aerosol dilution. During weekdays a maximum of total BC concentration occurs during morning rush hours. In winter the total BC concentrations at PAY is 0.8-1 µg m −3 during weekdays with slightly elevated concentrations during rush hours. A BC concentration of ≈0.5 µg m 3 can be attributed to FF combustion during night and at weekends. At weekdays in winter the contribution of BC FF is ≈1 µg m −3 during the morning and evening rush hours. The BC concentration in winter attributed to WB is both at weekdays and weekends relatively constant during the day and increases to 0.4 µg m 3 in the evening.
In MAG the diurnal pattern of total BC is in summer similar to the BC pattern in PAY. Total BC can exclusively be attributed to BC FF . During weekdays BC concentrations dominate during the morning rush hours. In winter, BC concentrations in MAG vary from 2.5-4.5 µg m −3 during weekdays and from 1.4-3.2 µg m −3 at weekends. During weekdays BC FF is ≈3.5 µg m −3 during the late morning and the evening rush hours. At weekends and at night BC FF varies from 1.0-2.0 µg m −3 . During weekdays and weekends BC WB is lowest during the day, in the evening concentrations are increasing.
In ZUE the BC concentrations are in summer predominantly resulting from traffic emissions, the contribution of BC WB to total BC is negligible except for the evening hours at the weekends. This fraction can be attributed to local emissions from public fire and barbecue places close to the measurement site. In winter the total BC concentration in ZUE varies from 0.8-1.3 µg m −3 during weekends and from 1. occur during the morning and the evening. BC WB concentrations vary from 0.3-0.5 µg m 3 , both at weekdays and weekends where the higher concentrations occur during the evening hours. In summary, the diurnal cycle of BC FF follows at all sites the expected pattern with highest concentrations during the rush hours when road traffic density is at maximum. On the other hand, the contribution of BC WB is only significant during winter. The diurnal cycles show at all sites increased concentrations during evenings and nights which can be explained by the typical operating time pattern of domestic heating appliances. The absolute concentrations and the found diurnal cycles of BC WB and BC FF indicate that the applied source apportionment approach gives reasonable results.
In the following we try a further verification of our approach by relating BC FF and BC WB to independent tracers for FF combustion and WB, respectively.

Comparison with auxiliary data
For PAY the determined BC for FF combustion BC FF and for WB emissions BC WB are plotted against two markers for WB related aerosols, the water soluble fraction of potassium and levoglucosan. Figure 5a shows the relationship between daily potassium concentration and daily BC FF and BC WB at PAY. Daily BC WB and potassium are positively correlated (R 2 = 0.77) while there is a much lower correlation between BC FF and potassium (R 2 = 0.14). In Fig. 5b  over roughly a one week period. The calculated BC WB correlates well with levoglucosan, resulting in a high coefficient of determination R 2 = 0.67. There is no linear dependence between BC FF and levoglucosan (R 2 = 0.01).
In addition, BC FF and BC WB were compared to EC source contributions as obtained in a PM 10 source apportionment study (M. Gianini, personal communication, 2010) using a receptor modelling approach (positive matrix factorization PMF; Paatero and Tapper 1994). Figure 5c shows BC FF and BC WB versus the daily contribution of "wood combustion EC" for PAY as obtained by PMF. The correlation between BC WB and "wood combustion EC" is good resulting in a coefficient of determination R 2 = 0.69 while there is only little correlation between BC FF and "wood combustion EC" derived by PMF.
Also for MAG BC FF and BC WB were plotted against potassium (Fig. 6a). Daily BC WB and potassium are positively correlated (R 2 =0.72). The correlation between BC FF and potassium is clearly lower (R 2 = 0.47). In Fig. 6b, BC FF and BC WB were compared to EC source contributions from PMF for MAG. Here, the correlation between BC WB and "wood combustion EC" results in a coefficient of determination R 2 = 0.69. There is again a clearly lower correlation between BC FF and "wood combustion" derived from PMF, (R 2 = 0.32).
Note that for PAY and MAG, the findings for potassium and the modelled contribution of "wood combustion EC" are based on long-term measurements, the used data were collected during a whole year.
For the time period of the ZUE measurements neither potassium concentrations nor PM 10 source apportionment results are available. However, in contrast to the rural sites PAY and MAG, the urban background site ZUE shows typically a high correlation between BC and nitrogen oxide NO x (NO x = NO + NO 2 ). Figure 7a and   resulting in a high coefficient of determination R 2 = 0.60 (summer) and R 2 = 0.83 (winter). There is no correlation between BC WB and NO x .
The comparisons show that there is in general a good agreement between the derived BC WB and BC FF with concentrations of WB and FF combustion indicators. These findings give confidence that the applied source apportionment approach for BC is well suited for long-term data sets. Similar to the above described comparisons, Sandradewi et al. (2008c) found rather high correlations between the optical absorption of aerosols from FF combustion and WB and the fractions of EC that are of fossil and non-fossil origin. The latter was derived from analysis of 14 C in EC. The results from Sandradewi et al. (2008c) are therefore also indicating the high potential of data from multi-wavelength aethalometers for identification of the fractions of BC from FF combustion and WB.

Conclusions
The AE is a robust and easy to use instrument for continuous optical determination of BC concentrations with temporal resolution of a few minutes. In this study, the measured BC concentrations are consistent with elemental carbon concentrations because the aerosol light absorption cross sections were calculated from EC analyses using the thermal optical transmission method.
We deployed multi-wavelength AE instruments at two rural (PAY and MAG) sites and one urban background site in Switzerland. The measurements were performed for up to 2.5 years. We found average black carbon concentrations of 0.43 µg m −3 at PAY, 0.8 µg m −3 at MAG and 0.99 µg m −3 at ZUE in summer and 0.8 µg m −3 at PAY, 3.03 µg m −3 at MAG and 1.34 µg m −3 at ZUE in winter.
Recent studies give reason that AE data may be used for source apportionment (Sandradewi et al., 2008a;Favez et al., 2009). But in these studies, the contribution of fossil fuel combustion and wood burning to the total carbonaceous aerosol was determined by analysis of AE data predominantly collected in winter during short-term measurement campaigns. Here we conclude that the proposed modelling approach is not applicable for long term datasets. This is likely due to significant fractions of the carbonaceous aerosol resulting from other sources and processes than FF combustion and WB.
In this study we focused on source apportionment of BC instead of total carbonaceous matter. The modified two sources approach fits very well to the measured BC concentrations. Separation of total BC into BC FF and BC WB was successful for all seasons and measurement sites. In winter, the determined mean fraction of BC WB to total BC was 33 %, 30 % and 24 % at PAY, MAG and ZUE respectively. These results are noticeable with respect to air quality control as wood combustion only contributed 3.9 % to the total energy consumption in 2008 in Switzerland (Kaufmann, 2009).
It is interesting to note that the calculated contribution of BC WB is in excellent agreement with results reported for ZUE based on 14 C analyses (Szidat et al., 2006). Also, the obtained WB contributions to BC at PAY and MAG correlated well with measured concentrations of levoglucosan and water soluble potassium as well as with results from PM 10 factor analytical modelling. In ZUE there is a good correlation between the obtained BC from FF combustion and NO x . The latter findings support our approach and show that multiwavelength AE data are suitable for source apportionment of BC.