Contributions of fossil fuel, biomass-burning, and biogenic emissions to carbonaceous aerosols in Zurich as traced by ¹⁴C

2.1. Aerosol Sampling

[5] At the urban background site Zurich (47°22′42″N, 8°31′52″E, 410 m a.s.l.) of the Swiss National Air Pollution Monitoring Network (NABEL) [Swiss Federal Laboratories for Materials Testing and Research (EMPA), 2000; Gehrig and Buchmann, 2003], aerosols were collected on quartz fiber filters (D = 150 mm, QF 20, Schleicher & Schuell) with a high-volume sampler (DA80, Digitel) and a PM₁₀ inlet during 12 August to 8 September 2002 and 17 February to 26 March 2003 [Szidat et al., 2004b, 2004c]. The station is situated in a parklike courtyard in the city center close to the main railway station. It is surrounded in the immediate vicinity by roads with rather low traffic as well as apartment buildings, small companies, and shops. The sampling sequences included daytime (usually between 8 a.m. and 8 p.m.), nighttime, and whole-day samples, representing 1–5 consecutive days. After sampling, filters were folded, wrapped in aluminum foil, packed into airtight plastic bags, and stored at −20°C until analysis.

2.2. Separation of Carbonaceous Particle Fractions

[6] In the system “Two-step heating system for the EC/OC determination of radiocarbon in the environment” (THEODORE), OC and EC were combusted from the filters in a stream of oxygen at 340°C and 650°C, respectively [Szidat et al., 2004a]. The separation scheme for different carbonaceous particle fractions is given by Szidat et al. [2004c]. Figure 1 sketches this analytical procedure, which is briefly described in the following. OC was separated from other carbon fractions at 340°C. For the investigation of WINSOC, a water extraction was performed to remove WSOC prior to the thermal separation. The latter fraction was not analyzed directly, but determined by subtraction (WSOC = OC − WINSOC). Thermal elimination of OC for isolation of EC is susceptible to charring resulting in a positive artifact for the EC determination. Water extraction prior to thermal separation minimizes this artifact by elimination of the water-soluble fraction, and the difference in measured EC is then denoted polymerizable WSOC. The fact that polymerizable WSOC has a much higher f_M than EC (see below and in Figures 3 and 4) shows that the water extraction is important for EC isolation and does not result in an artifact removal of true EC. Note that the concentrations of this fraction were derived by subtraction as was done for WSOC (polymerizable WSOC = mixture “EC + polymerizable WSOC” − EC). Because of this fact, polymerizable WSOC is not a subfraction of WSOC, because the first was determined as the removed fraction within EC separation and the latter within OC separation. Thus both fractions sum up to the full water-soluble OC fraction.

2.3. Measurements of ¹⁴C

[7] After combustion of the different carbonaceous particle fractions, evolving CO₂ was cryotrapped, determined manometrically, and transformed to filamentous carbon for ¹⁴C/¹²C determinations. These measurements were performed at the PSI/ETH compact accelerator mass spectrometry (AMS) system, which is based on a 500 kV pelletron accelerator [Synal et al., 2000]. Details of the target production, measurement parameters, and data evaluation for submilligram samples are reported elsewhere [Szidat et al., 2004a].

2.4. Definition of f_M

[8] f_M represents a ¹⁴C/¹²C ratio of a sample related to that present in the reference year 1950:

Therefore fossil material is characterized by

[9] On the other hand, f_M should be 1 for contemporary samples because of this definition. However, materials from the last 50 years show values >1, with a maximum of ∼2 in the early 1960s as a consequence of the nuclear bomb excess. This was shown by Levin et al. [2003] using time series of atmospheric ¹⁴CO₂. From these data, f_M for pure biogenic OC from 2002/2003 results as

[10] The uncertainty of f_M,biogenic includes natural variability of ¹⁴CO₂ and a conservative estimation of the influence of primary biogenic OC, which could be older than the year of sampling. In accordance with Lewis et al. [2004],

was set for biomass-burning (bb) contributions on the assumption that residential wood burning of 30–50-year-old softwood and hardwood was the major biomass-burning source for Zurich.

2.5. Determination of Organic and Inorganic Tracers

[11] Levoglucosan was determined in the aqueous extracts of the quartz fiber filters by HPLC with electrochemical detection (ED40, pulsed amperometry, gold working electrode). As eluent a sodium hydroxide gradient ranging from 30 to 40 mM was used. The eluent flow was 1 mL min⁻¹. The analytical column was a CarboPac PA 10. All instruments were from Dionex Corp. The detection limit of the method was 10 ng mL⁻¹. Field blanks for levoglucosan were negligible. The determination of cellulose combined enzymatic saccharification followed by enzymatic determination of the formed glucose and was performed according to the procedure given by Puxbaum and Tenze-Kunit [2003]. Potassium was measured with standard ion chromatography after water extraction of the filters and corrected for sea-salt and mineral dust contributions [Szidat et al., 2004c]: non-sea-salt (nss) potassium and calcium was determined assuming constant K⁺/Na⁺ and Ca²⁺/Na⁺ ratios for sea salt of 0.0218 and 0.044, respectively; excess potassium (exK⁺; i.e., potassium corrected for sea-salt and mineral dust contributions) was then calculated using the crustal average nssK⁺/nssCa²⁺ ratio of 0.71. The procedure to trace biomass-burning emissions using exK⁺ benefits from K⁺/Na⁺ and K⁺/Ca²⁺ ratios >10 for wood combustion [Fine et al., 2001; Schauer et al., 2001]. Isoprene and other trace gases were determined with gas chromatography–mass spectrometry (GC-MS). Biogenic isoprene denotes isoprene concentrations corrected for anthropogenic releases assuming a constant anthropogenic isoprene/1,3-butadiene mass ratio [Reimann et al., 2000; Szidat et al., 2004b].

3.1. General Results

[12] For February/March 2003, aerosol sampling began with a dry and cold period and a closed snow cover (Figure 2). During the first two weeks, an exceptional inversion episode dominated, with PM₁₀ concentrations above 100 μg m⁻³. This episode ended beginning of March 2003, when the weather turned to springlike conditions, and the snow cover has melted. During this month, PM₁₀ concentrations reached an average value of about 30 μg m⁻³, which were comparable to the level of August/September 2002. Consequently, the February/March 2003 campaign was divided into a winter (before 1 March) and a springlike regime. The meteorological conditions during the summer campaign were already described previously [Szidat et al., 2004b]. Briefly, this was a warm period with maximum temperatures of up to 30°C and sequences of several dry days followed by distinct precipitation events.

[13] Seasonally averaged f_M determinations of different carbon fractions are presented in Figure 3 (for f_M values of the whole data set, see Figure 4). The mean f_M(OC) amounted to 0.76 ± 0.09, 0.79 ± 0.08, 0.81 ± 0.05 for summer, winter, and springlike conditions, respectively. These averages are equal within uncertainties suggesting overall constant ratios of emission sources independent of the season. This result is unexpected, as the contemporary carbon content is supposed to be lower for winter because of reduced SOA formation from biogenic gaseous precursors as well as reduced biogenic primary emissions [Tsigaridis and Kanakidou, 2003], and will be explained below.

3.2. Source Apportionment of EC

[14] Concentrations of EC originating from biomass burning (EC_bb) were deduced from f_M(EC) by

where EC_tot is the concentration of total EC. Consequently, the concentration of EC from fossil sources (EC_fossil), is given by

[15] The contribution of biomass burning to EC is then EC_bb divided by EC_tot and amounts to 6 ± 2%, 12 ± 1%, and 25 ± 5% during summer, springlike, and winter conditions, respectively (Table 1). Therefore EC was dominated by fossil emission sources, in contrast to OC (see above). The higher fraction of biomass burning for winter was corroborated by measurements of potassium and levoglucosan (Table 2), two widely used tracers for biomass burning [Sheffield et al., 1994; Andreae and Merlet, 2001; Nolte et al., 2001]: levoglucosan and EC_bb concentrations showed similar seasonal variations with average summer/winter ratios of 0.11 and 0.06, respectively. Note that these ratios become 0.49 and 0.26, respectively, if normalized to the PM₁₀ concentrations, because of the highly elevated PM values during the inversion episode in winter. However, levoglucosan emission factors may vary by a factor of 5 or higher [Andreae and Merlet, 2001; Nolte et al., 2001], resulting in large uncertainties for source apportionment (see also Table 3). The potential of potassium as biomass-burning tracer is limited as well because of interferences from mineral dust [see also Szidat et al., 2004c]. Thus these two conventional tracers lack the accuracy of ¹⁴C for the source apportionment of EC because of the variability of emission ratios.

3.3. Source Apportionment of OC

[16] For source apportionment of OC, an advanced model is applied to discriminate between biomass-burning and biogenic emissions within the contemporary carbon fraction (Figure 5). The biomass-burning fraction of OC (OC_bb) is estimated from EC_bb using an average EC/OC emission ratio (EC/OC)_ER,bb:

Table 3 shows that (EC/OC)_ER,bb amounts 0.16 ± 0.05 for residential wood burning of softwood and hardwood in fireplaces, which we assume as the major source of biomass burning in Switzerland and western Europe. Excess contemporary OC is defined as biogenic OC (OC_biogenic), comprising primary and secondary components:

where OC_tot is the total OC concentration. The concentration of fossil OC (OC_fossil) is then

Finally, OC_{anthropogenic} is defined as sum of all man-made emissions:

[17] As the uncertainty of (EC/OC)_ER,bb has a large impact on the combined standard uncertainties of OC_bb, OC_{anthropogenic}, and OC_biogenic, the values in Table 3 will be discussed in detail in the following. (Note that the uncertainty of (EC/OC)_ER,bb has only little influence on the uncertainty of OC_fossil and no influence on the ones of EC_bb and EC_fossil.) The relative standard deviation (r.s.d.) of (EC/OC)_ER,bb of 32% comprises two main components of uncertainty among the studies listed in Table 3: wood burning conditions and procedures of the OC/EC determination. Especially, the use of literature data from different OC/EC separation methods always has to be considered carefully [Schmid et al., 2001], and it is not clear how much this method-dependent variability contributes to the uncertainty of (EC/OC)_ER,bb. Therefore the significant outlier in Table 3 may have been caused by substantially deviating methods. Within Table 3, some older data [e.g., Dasch, 1982] were not considered, when direct sampling of combustion aerosols was conducted without dilution with clean air, as this procedure does not reflect ambient atmospheric conditions [Bond et al., 2004]. Moreover, Table 3 does not account for waste burning. Agricultural, forestry, and private waste fires of wood have a certain occurrence in Switzerland (see Tables 6 and 7), but they are supposed to occur mainly in April/May and September-November [Liousse et al., 1996]. Especially in the end of February 2003, when a large contribution of EC_bb was observed, substantial activities of open-field waste burning of wood are very unlikely because of the low temperatures and the closed snow cover. Particle emissions of waste incineration facilities, on the other hand, are very small compared to the total burden in Switzerland (Tables 6 and 7). Consequently, the impact of all these sources on this work can be neglected.

[18] For validation of the advanced source apportionment model sketched in Figure 5, OC_bb concentrations as obtained from equation (7) are compared in Table 4 with OC_bb,lev derived from levoglucosan (lev) concentrations using

where (lev/OC)_ER,bb is the average levoglucosan to OC emission ratio from residential wood burning of softwood and hardwood in fireplaces. This ratio amounts 0.15 ± 0.09 (Table 3), representing a smaller number of studies and a larger variation of results than (EC/OC)_ER,bb. The differences between OC_bb and OC_bb,lev are insignificant for all seasons, but the uncertainties for the single determinations are high because of the uncertainties of both emission ratios. It would be interesting to see if the discrepancy between OC_bb and OC_bb,lev during winter is real, possibly due to seasonal variations of emissions and atmospheric lifetimes for levoglucosan and EC_bb. This analysis is however not possible on the basis of the current data set and needs more investigation. However, results from Table 4 serve as an independent quality control for the advanced ¹⁴C model and the outcome of Table 1, because the determinations of OC_bb according to equations (7) and (11) are based on independent tracers.

[19] Table 1 summarizes average contributions of OC from anthropogenic (separated for biomass burning and fossil fuel usage) and biogenic emissions for summer, winter, and springlike conditions. While anthropogenic emissions dominate the OC concentration in winter with ∼73%, biogenic emissions are the largest source in summer with a contribution of ∼60%. The contribution of biomass burning follows a seasonal trend as well, varying from ∼10% for warm to ∼41% for cold periods. In contrast, the fossil fuel contribution, ∼30%, is quite independent of season, resulting in a constant average f_M(OC) for winter, springlike, and summer conditions.

[20] There is a high correlation between OC_fossil and EC_tot with nearly the same slope in summer and winter (with OC_fossil = 1.2 × EC_tot and R² = 0.84, see Figure 6). Various authors [e.g., Castro et al., 1999; Lim and Turpin, 2002] made use of this by interpreting the lowest OC to EC ratios (i.e., OC_tot/EC_tot according to our nomenclature) in such a correlation as the constant primary OC to EC ratio. In that approach, winter data are often used in order to neglect biogenic SOA. However, our results show that winter aerosol may be influenced by biomass-burning emissions, which cannot be detected by OC/EC determinations without isotopic information. Therefore the approach of a constant primary OC to EC ratio is only justified in the absence of biomass-burning emissions, which has to be evaluated carefully. The correlation between OC_fossil and EC_fossil indicates a similar relation (with OC_fossil = 1.5 × EC_fossil and R² = 0.80, see Figure 7). The quality of the correlation shown in Figure 7 is identical to that of Figure 6 on the basis of an F test of both standard deviations about regression (95% confidence). Thus the quality of correlation between OC_fossil and EC_tot was not significantly biased by the biomass-burning contribution of 25% within EC_tot during winter. On the other hand, both slopes are significantly different on the basis of a t test (95% confidence). The value of Figure 7 may give a realistic estimation of an ambient OC/EC ratio from fossil fuel sources including the influence of different emission patterns and atmospheric sinks.

[21] The OC fraction from biogenic emissions amounts to ∼27% for winter. Unfortunately, only 2 winter samples showed significant fractions of OC_biogenic applying equation (8), while the remaining 6 samples revealed a value below detection limit (95% confidence), which was caused by the high uncertainty of (EC/OC)_ER,bb. Thus the abundance of OC from biogenic sources during winter needs further confirmation. However, OC_biogenic was significantly detected in all samples under springlike conditions. Cellulose measurements reveal that organic matter comprises only 2, 5 and 6% primary particles from plant debris for summer, winter and springlike conditions, respectively (Table 5). This suggests that SOA from biogenic VOC precursors remains as a major source for the high biogenic OC fraction even in winter (see below). These findings are reflected in the f_M values for WSOC, which were ∼50% higher than for WINSOC for all seasons (Figure 3). In summer, WSOC is mainly attributed to SOA [Pun et al., 2000; Kanakidou et al., 2004]. Consequently, high f_M(WSOC) values suggest that biogenic SOA exceeds anthropogenic SOA by far even in the urban environment of Zurich for this season. In winter on the other hand, water-soluble products from biomass burning [Mayol-Bracero et al., 2002] contribute substantially to the high f_M(WSOC) values.