A thermal–optical OC/EC analyzer (Sunset Laboratory Inc.) was used for the combustion and recovery of the OC and EC fractions. The instrument is specifically developed to separate the carbon content of ambient atmospheric samples collected on quartz fiber filters into OC and EC fractions by thermal evaporation and/or oxidation at separate temperature steps. As the carbon fragments pass through a manganese dioxide (MnO2) oven, they are quantitatively converted to CO2 gas and measured directly by a self-contained non-dispersive infrared (NDIR) detector system, placed upstream the instrument's outlet. At the end of each run, a known volume of methane (CH4) is injected into the oven and oxidized to CO2 to ensure accurate quantification of OC and EC concentrations.

When operated with standard analytical protocols (e.g., NIOSH (NIOSH, 1999), IMPROVE-A (Chow et al., 2007), EUSAAR_2, (Cavalli et al., 2010), the Sunset OC/EC analyzer evaporates the OC fraction under an inert helium (He) atmosphere, while the EC fraction is subsequently oxidized in a He/O2 mix (Cavalli et al., 2010, Chow et al., 2001). Under these conditions a fraction of the OC can undergo substantial pyrolysis (charring) during the O2-free step. When measuring concentrations, the amount of charring-derived EC is accounted for internally by the OC/EC analyzer. A tunable diode laser measures the light transmittance of the sample throughout the heating ramp cycle. Since the transmittance decreases as OC is pyrolized to EC, but increases as EC and the pyrolized OC are removed during the second heating cycle, the OC/EC split point for quantification is defined to be the point at which the laser transmittance returns to its initial value (Bauer et al., 2009). This widely accepted procedure for the determination of OC and EC concentrations may not be suitable for 14C analysis of the EC fraction, especially when analyzing ultra-small samples, because accurate measurements of 14C require a physical separation of the EC from the OC fraction and it is not possible to mathematically correct the 14C data for charring.

For 14C measurements, the Sunset OC/EC analyzer was modified and equipped with a solenoid valve and a gas flow sensor for controlling O2 flow, which allows the use of pure O2 (99.999 %) during oxidation (Zhang et al., 2012). The use of pure O2 reduces OC charring and allows the use of alternative TOA protocols. The O2 flow is controlled in the same way as the default carrier gases – He (> 99.999  %) and O2/He mix (10 % O2 in He) and the reference gas – CH4 (5 % CH4 in He), all of which are controlled by the instrument's gas flow program. The gas flow rate through the OC/EC analyzer is adjusted and stabilized at 60 mL min-1.

To achieve a clear physical separation of OC and EC for 14C-based source apportionment, we used the Swiss_4S protocol, which involves four consecutive steps: (S1) pure O2 at 375 ∘C for evaporation/oxidation of OC without premature EC evolution, (S2) pure O2 at 475 ∘C followed by (S3) in He at 650 ∘C, both of which aim to achieve complete OC removal before EC isolation and lead to better consistency with TOA protocols like EUSAAR_2, and (S4) pure O2 at 760 ∘C for desorption and recovery of EC (Zhang et al., 2012). Note that our analysis differed from the Zhang et al. (2012) method, in which OC is trapped only during (S1). Here, we trapped OC evolving during both (S1) and (S2) to achieve maximum OC recovery. The material evolving during (S3) is defined as a mixture of OC and EC and is not included in the analysis of either carbon fraction and therefore discarded.

Each run was completed by an injection of reference gas (5 % CH4 in He), allowing for a conversion of sample CO2-peak sizes to µg C cm-2 of sample filter. The OC/EC analyzer is purged with He prior to every run to ensure that there is no cross contamination from a prior run.

During routine operation, fine air particulate matter (PM2.5) samples, collected on quartz microfiber filters (Pallflex Tissuquartz, 2500 QAT-UP, Pall, Port Washington, NY) were introduced into the Sunset OC/EC analyzer as 1.5 cm2 filter punches (dimensions 1 cm × 1.5 cm, Sunset) on a flat quartz spoon. Under the Swiss_4S protocol, separate punches were used for OC and EC analysis. The EC punch was treated with Milli-Q (MQ) water prior to the analysis to remove the water-soluble OC and minimize charring as follows: a filter disk (23 mm diameter) was placed on a plastic filter holder (25 mm diameter, Sartorius GmbH, Germany) between two sealing rings with the laden side upwards, and 20 mL of MQ water was passed through from the top using a syringe (Zhang et al., 2012). After drying at 60 ∘C for up to 1 h, a 1.5 cm2 punch was taken from the disk for analysis.

Standards for calibration of the OC/EC analyzer (e.g., water-soluble OC, such as sucrose) are typically applied as solution, pipetted on a pre-baked filter punch, dried and analyzed through the desired TOA protocol. However, since EC is water insoluble, this procedure cannot be applied when running EC standards for 14C analysis or method verification. To introduce OC and EC reference materials alike, we developed and customized quartz boats (Jelight Company Inc., Irvine, CA, USA), which were placed into a specifically designed round-bowled (instead of flat) quartz spoon (Jelight Company Inc., Irvine, CA, USA). This enabled us to introduce solids into the OC/EC analyzer. Boats were pre-baked at 900 ∘C for 2 h and stored in a baked, closed porcelain crucible to avoid contamination. Spoons were baked in O2 inside the OC/EC analyzer for 10 min at 920 ∘C.

The outlet of the OC/EC analyzer was connected to a vacuum line (Fig. 1), which allows for the physical trapping of the CO2 released during the different steps of the analysis for subsequent 14C measurement. In addition to trapping the CO2 gas, the vacuum line is also used to remove non-condensables and O2. The vacuum line exhibits a very simple design and allows fast and efficient cooling, trapping and release of the gasses. In the following section, all numbers in brackets correspond to the vacuum line parts described in Fig. 1.

Prior to each OC/EC run, the vacuum line was evacuated to 1 × 10-4 torr (0.013 Pa) on the Pirani vacuum gauge (8) using a turbomolecular oil-free diaphragm pump system (9). If no trapping was in progress, the line was kept isolated from the Sunset analyzer and air by keeping valves V2 and V3 closed, while V1 was kept open so the exhaust from the Sunset was open to air (1). The rest of the line stayed open to the turbo pump (9), ensuring the vacuum in the line, which was monitored by both the Pirani vacuum gauge (8) (1 × 10-4 torr at vacuum) and the pressure transducer of the calibrated volume (6) (±0.1 torr at vacuum). While evacuated, we cleaned the inner surfaces of the break-seal tube (7) of any trapped exogenous C by heating along its length with a blowtorch (O2/natural gas).

Once the Swiss_4S TOA protocol started, the trapping procedure consisted of the following steps: a liquid nitrogen (N2(l)) dewar was placed so it covered the cryogenic trap (2) completely a minute prior to Swiss_4S step (S1)–for OC trapping or (S4)–for EC trapping. The cryogenic trap was then closed to the pump via V4. When CO2 was evolving from the OC or EC fractions (steps S1 and S2 or S4, respectively), the gas stream was re-directed to the vacuum line by first closing the exhaust (via V1), waiting for the pressure in the line to build to 1 psig (∼ 6895 Pa), as shown by the OC/EC analyzer's program, and opening the flow to the cryogenic trap (V2 is opened; pressure drops). After waiting for the pressure to reach 1 psig for a second time, the second exhaust (4) was opened via V3 and stayed open while the Swiss_4S protocol was in progress. During that time, CO2 accumulated in the cryogenic trap. At the end of S2 for OC or S4 for EC, the line was closed off from the OC/EC analyzer. First V3 was closed, and once the pressure again reached 1 psig, V1 was opened and V2 was immediately closed. After this step the CO2 from the desired fraction was in the cryogenic trap. Note that waiting for the pressure to build up before opening the line to either exhaust (1) (at the end of the trapping procedure) or exhaust (4) (at the beginning of the trapping procedure) is crucial in preventing backflow in the system, which could introduce ambient CO2 and O2 from the room air into the N2(l)-cooled sample trap (2). However, the pressure buildup causes the O2 carrier gas to condense in the trap. Once either exhaust is open, the reduced pressure prevents the O2 from condensing in the trap. Releasing the pressure allowed us to avoid the need for an additional chemical trap for O2 trapping, as compared to the setup in Bern.

Next, the O2 that accumulated in the trap during the periods of increased pressure in addition to the sample CO2, was pumped away, together with other non-condensables: N2(l) was left in place (2), while V4 was opened and closed, after which V7 was opened until vacuum was established again. These steps were repeated until the pressure on the pressure gauge (6) was below 300 torr (the upper limit of the pressure transducer) and then the line was finally opened fully to the turbo pump (9). The repeated closing and opening of V4 and V7 was necessary, because the turbo pump cannot operate at high pressures.

After the line pressure at the vacuum gauge (8) was less than 1 × 10-4 torr, the cryogenic trap was closed to the pump via V4, and the N2(l) was replaced with a dry ice–ethanol slurry to release the sample CO2, but not water. The sample was transferred to the calibrated volume (5), where the pressure, corresponding to the carbon mass of the sample, was recorded. Finally, the measured CO2 was transferred to the pre-cleaned 4 mm O.D. Pyrex tube (7) and sealed with the blowtorch for future graphitization and 14C analysis.

Our sample CO2 was graphitized and analyzed for its 14C content at the KCCAMS facility. Graphite for AMS measurements is typically produced by reduction of CO2 gas over a catalyst. The production of high quality, uniform graphite targets, suitable for the AMS ion source, is important for the optimal performance of any AMS system. The quality of the graphite influences the level of background contamination, the mass fractionation, and the beam current (Santos et al., 2007c, Kim et al., 2010).

In this study, blanks and standards were chosen to resemble regular fine particulate matter aerosol samples collected over periods of 1–7 days (usually ∼ 2–50 µg C) and were much smaller than regular sized 14C samples (∼ 300–1200 µg C). A modified protocol was used to maximize yields for our ultra-small samples and to improve target processing and measuring (Santos et al., 2007a). Cryogenically purified CO2 was transferred into Pyrex tube reactors of 1.6 cm3 volume and reduced to graphite at 450 ∘C using hydrogen gas over pre-cleaned iron powder as a catalyst. A magnesium perchlorate trap was used to remove water from the reaction. Graphite targets were analyzed for their 14C content using a modified compact AMS system (NEC 0.5 MV 1.5SDH-1) (Beverly et al., 2010).

During the processing of samples for 14C analysis, contamination with extraneous carbon (Cex) always occurs independently of sample size. Its presence can significantly affect the 14C signature of smaller samples and puts a practical lower limit on the minimum 14C sample size that can be reliably measured (Santos et al., 2007a, 2010; Ziolkowski and Druffel, 2009). In some cases, the mass and 14C signature of the Cex can be evaluated directly, and the measured 14C data of unknown samples can be corrected for it, using a mass balance approach. However, when the mass of Cex is too small to directly measure, the extraneous carbon contamination cannot be quantified using this approach. In this study, we adopted an indirect mathematical approach (Santos et al., 2007a), which treats Cex as a mixture of two components; (a) modern carbon – mostly introduced during sample handling by tools and in equipment, and (b) fossil carbon – originating from 14C-free carbon in the iron powder used to catalyze the reduction of CO2 to graphite and/or in our OC/EC isolation and trapping setup. The mass of the Cex (mex) can then be expressed as (Eq. 1): mex=mmodern+mfossil, where mmodern is the mass of the modern component and mfossil the mass of the fossil component of Cex.

The average contribution of modern carbon can be quantified by measuring blanks (14C-free materials) of different sizes through each step of the sample processing and is expressed as (Eq. 2): mmodern=mmeas.×FMblankFMref, where mmeas. is the mass measured on the graphitization line, FMblank is the measured 14C signature of the blank, expressed as fraction modern carbon, and FMref is the fraction modern of a normal-sized, modern reference materials (in this study we used 1 mg C of OX-I). This latter term is directly measured by the AMS system, and reported normalized to six time-bracketed aliquots of the reference material, and corrected for isotope fractionation (Santos et al., 2007b). Similarly, we quantified the average contribution of the fossil component of Cex by measuring the 14C content of modern standards (FMstd) with different masses, with known fraction modern content (Eq. 3): mfossil=mmeas.×[1-FMstd]FMref. Finally, the 14C content of a sample can be corrected for the mass and 14C contribution of Cex and expressed as (Eq. 4): FMsample=FMcons.×FMmeas.FMref-mmodernmmeas.1-mmodernmmeas.-mfossilmmeas., where FMcons. is the canonic fraction modern value (the standard multiplier) of the reference material, used to produce the six time-bracketed graphite targets measured in a single batch or wheel, and FMmeas. is the 14C content of the unknown sample, directly measured by the AMS system. To determine the uncertainty in our calculation of FMsample, we mathematically propagated the uncertainty, applying a 50 % error in mfossil and mmodern, based on long-term measurements of mex variance in small samples (Santos et al., 2007a). Further detailed description of this approach for correcting ultra-small mass samples and its use at the KCCAMS/UCI can be found in Santos et al. (2007a).

Here, we evaluate the contribution of modern and fossil carbon, introduced both during the TOA analysis with the Swiss_4S on our vacuum line (Cswiss_4S) and during the graphitization procedure (Cgraph). To assess the overall mass and 14C signature of Cex, we measured a set of blanks and standards through all steps of the analysis for both OC and EC. To quantify the relative contribution of Cex introduced during either processing step, we measured another set of blanks and standards, which only underwent graphitization. The Cswiss_4S contributions were estimated as the difference between Cex and Cgraph.

To evaluate the mass of the Cex and its associated 14C signature for proper background corrections and to assess the performance of our OC/EC analyzer and vacuum line, we measured a set of 14C fossil (blanks) and modern (standards) reference materials, described in Table 1. They were chosen to have strongly contrasting 14C contents and (when possible) to consist of international reference materials. We chose adipic acid as a blank and oxalic acid (OX-I) as a standard for OC. Coal (POC #3 USGS (United States Geological Survey) coal) and rice char (Oryza sativa L.) were used as blank and standard for EC, respectively. Both OC and EC reference materials were introduced into the OC/EC analyzer as solids. The EC reference materials were pre-cleaned using the ABA (acid–base–acid) method (Santos and Ormsby, 2013).

To quantify the efficiency of our setup at trapping each carbon fraction, we measured seven samples with a range of sizes from each reference material, with the exception of OX-I (n=6). For better assessment of the Cex and its 14C content, more samples from the same reference materials were included from two additional AMS runs. A detailed description of all reference material samples can be found in Appendix A.

A second set of similar-sized 14C reference materials (coal, n= 2–4 and OX-I, n= 3–5) were graphitized in parallel with each batch of samples (separately for each of the three AMS runs) to quantify any contamination introduced only during the graphitization procedure and to distinguish it from the contamination introduced by the OC/EC analyzer and vacuum line.

In addition, we analyzed the 14C content of five mixed OC and EC in ambient particulate matter on quartz fiber filters, which were previously collected and analyzed for their 14C content under the Swiss_4S protocol at the University of Bern, Switzerland. Samples were collected during winter with high-volume air samplers in Switzerland (Bern: 8 January 2009, Chiasso: 11 February 2012, Massongex: 13 February 2012 and Solothurn: 13 February 2012; for details see Zotter et al., 2014) and Poland (Zabrze: Winter 2012) (Appendix B).

On average, 91 % of the carbon in each sample was cryogenically trapped following the Swiss_4S TOA on the Sunset OC/EC analyzer and attached vacuum line (Fig. 2a). Here, the OC content was directly estimated as the total carbon content of the reference materials, based on its molecular weight. The EC content could only be assessed with our TOA setup, and for each EC run the measured EC content was compared to the average EC content of the reference materials (n=7).

Sample recovery, or yield, during this step reflects the homogeneity of each reference material, the accuracy of weighing samples into the boats, and the efficiency of the cryo-trapping procedures. In general, the OC materials were more pure than the EC materials and consequently showed a better linear relationship between expected and recovered yields (Table 1).

About 5 % of each sample was lost on the graphitization line during the second cryogenic trapping of CO2 (Fig. 2b). One sample was excluded from this analysis, because only 58 % of sample CO2 was recovered – implying improper handling. Sample recovery during this step was influenced by CO2 lost on the break-seal and vacuum line inner surfaces, the efficiency of the cryo-trapping, as well as differences in the calibration of the pressure gauges.

The amount of Cex introduced during each step of the analysis was estimated indirectly by measuring it as the sum of the modern and fossil 14C background contamination (Table 2). We found no difference in the amount of Cex introduced during the analysis of OC compared to EC.

The amount of modern carbon introduced during the analysis was 0.8 ± 0.4 µg C (Fig. 3b, d). A comparison of the results for samples that only underwent combustion and graphitization, but not TOA and cryo-trapping, indicated that the Swiss_4S TOA with the OC/EC analyzer plus the attached vacuum line introduced a smaller amount of modern carbon (46 % of modern Cex) than the graphitization procedure (Table 2). Our modern carbon blank estimate was based on the analysis of three separate AMS runs and the exclusion of one coal replicate, which showed an excess modern carbon amount of 1.5 µg C and was not comparable to the rest of the blanks.

The total amount of fossil Cex introduced by our method was estimated to be 0.67 ± 0.34 µg C (Fig. 3a, c), primarily (80 % of fossil Cex) due to graphitization, with much smaller contribution from the TOA and the cryo-trapping procedure (Table 2). Note that to have all 14C results from small reference materials fall within ±2σ of the expected value, an error of 50 % was then imposed into their background subtractions and propagated into their final uncertainties, as described in Sect. 2.4. This approach accounts for any long-term variability on blank measurements on ultra-small samples, as shown in Santos et al. (2010).

The comparison of 14C measurements of the OC and EC content of PM2.5 filter samples at UC Irvine and the University of Bern showed a very good agreement, with samples measured in Bern being slightly enriched in 14C compared to UC Irvine (Fig. 4). A Pearson's major axis type II regression, minimizing both x and y residuals simultaneously with all data given equal weight, showed a strong linear relation with a slope of 1.24 ± 0.17 and a R2 value of 0.85. It is worth noting that the EC samples are much smaller in size and consequently have a higher uncertainty level, which can explain the standard deviation of the slope. The steepness of the slope we observed is primarily driven by the smallest measured EC sample, which consequently has highest uncertainty. If this sample is excluded, the new Type II regression slope is equal to 1.07 ± 0.01 and is much closer to the 1:1 line with R2=0.95.

Further, to obtain a better comparison between the two labs, we performed a paired sample t-test, assuming unequal variances in the measurements of the two labs and two tails. The results showed that the null hypothesis (“means are equal”) cannot be rejected at the 5 % significance level for OC, EC and OC and EC together. The p-value or probability of observing the given result was 0.79.

The fraction modern values shown in Fig. 4 were corrected for background contamination during the sample preparation at each lab, but no field blank corrections were applied. This inter-laboratory comparison confirmed that the Swiss_4S TOA protocol can be used successfully with our proposed modifications (cryo-trapping and graphitization procedures) to reproduce OC and EC 14C measurements of unknown aerosol samples.