Evaluation of a two-step thermal method for separating organic and elemental carbon for radiocarbon analysis

We thoroughly characterized a system for thermal separation of organic carbon (OC) and elemental carbon (EC) for subsequent radiocarbon analysis. Different organic compounds as well as ambient aerosol filter samples were introduced into an oven system and combusted to CO 2 in pure O2. The main objective was to test which combustion times and temperatures are best suited to separate OC and EC. The final separation step for OC was combustion at 360C for 15 min. Combustion at this temperature proved enough to remove several organic test substances from the filter (including high molecular weight humic acid) but did not remove substantial amounts of EC. For isolation of EC, OC first needs to be completely removed from the filter. This was achieved by water extraction of the filter, followed by combustion of the water insoluble OC at 360 C and combustion at an intermediate temperature step of 2 min at 450 C. This last step removed the most refractory OC together with some EC. Finally, the remaining EC was combusted to CO 2 at 650C. The recovery of black carbon after the intermediate 450C step was approximately 80 %. Several tests provided strong evidence that OC was removed efficiently during the intermediate temperature step: (i) brown carbon, indicative of refractory OC, was removed; (ii) the fraction modern of EC did not decrease significantly if the temperature of the intermediate step was further increased. Based on tests with various organic compounds, we estimated that charred organic carbon could contribute 4–8 % to an elemental carbon sample that was isolated according to our method.


Introduction
Radiocarbon ( 14 C) measurements allow to distinguish fossil from contemporary carbon sources of carbonaceous aerosol and have therefore become an important method of source apportionment (e.g., Currie, 2000;Szidat et al., 2006Szidat et al., , 2007Szidat et al., , 2009;;Gelencser et al., 2007;Schichtel et al., 2008;Minguillon et al., 2011;El Haddad et al., 2011;Genberg et al., 2011;Gilardoni et al., 2011;Glasius et al., 2011;Dusek et al., 2013).This method exploits the fact that the radioactive carbon isotope 14 C is not present in fossil fuels, whereas it is present at a typical contemporary level in carbonaceous material that derives from living biomass.The contemporary 14 C level is maintained by the production of 14 C from the absorption of thermal neutrons, generated by cosmic ray showers, by N 2 , mainly in the upper troposphere and stratosphere.The produced 14 C is quickly oxidized via CO to 14 CO 2 , which is taken up by the biosphere.Therefore a characteristic 14 C / 12 C ratio is established for the biosphere, which is closely linked to the atmospheric 14 C / 12 C ratio (Levin et al., 2010).In organic material that does not exchange CO 2 with the atmosphere (e.g., fossil fuel), the 14 C isotope decays with the characteristic half-life of 5730 years (Godwin, 1962).
The 14 C / 12 C ratio of a sample is always reported relative to a standard of oxalic acid, whose activity is related to the activity of the biosphere under natural circumstances in the year 1950.We report our 14 C data as fraction modern (F 14 C), following the nomenclature of Reimer et al. (2004) (1950) . (1) The 14 C / 12 C ratio of the sample and the standard are both corrected for machine background and normalized for fractionation to δ 13 C = −25 ‰.The standard is not corrected for decay since the year 1950.
The main sources of carbonaceous aerosol, fossil fuel burning, formation of biogenic secondary organic aerosol (SOA), and biomass burning, have distinct F 14 C values.In fossil fuels, which have been buried for much longer periods than the 14 C half-life, all 14 C has decayed and F 14 C is therefore 0. F 14 C of the current atmospheric or living biospheric carbon would ideally be around 1.However, atmospheric 14 C levels have been altered by anthropogenic processes, in particular by the nuclear bomb tests in the 1960s and emission of fossil CO 2 , which does not contain 14 C. Nuclear bomb testing almost doubled the 14 C levels in the atmosphere, but since the stop of the bomb tests F 14 C of atmospheric CO 2 has decreased steadily and is currently estimated at F 14 C ∼ 1.04 (Levin et al., 2010).As a consequence, primary and secondary particles from biogenic origin have F 14 C ∼ 1.04.Aerosols from wood combustion have a variable F 14 C > 1.04, depending on the age and origin of the wood.Most of the wood burned today was grown over the last 20-60 years, when 14 C levels were elevated in the atmosphere, following the nuclear bomb tests.Estimates based on tree growth models (e.g., Lewis et al., 2004;Mohn et al., 2008) give a range from 1.08 to 1.30 for biomass burning aerosols (Szidat et al., 2006(Szidat et al., , 2007(Szidat et al., , 2009;;Minguillón et al., 2011;El Haddad et al., 2011;Genberg et al., 2011;Gilardoni et al., 2011).
The carbonaceous aerosol is often operationally subdivided in carbon fractions that comprise classes of carbonaceous material with similar properties.One of the most common sub-divisions aims to separate organic carbon from more refractory carbon components that are usually combustion derived and often referred to as soot, elemental, or black carbon.Organic carbon is assumed to consist primarily of individual organic molecules that are not or only weakly light absorbing and cover a great range of volatilities.On the other hand, a universal definition of elemental carbon, black carbon or soot is not agreed on in the literature and what is meant by these terms is often dependent on the analytical method or the context of the research (e.g., Petzold et al., 2013).This ambiguity is mostly caused by the fact that a clear distinction between organic and elemental carbon does not exist, but rather a continuous transition through more and more complex aromatic structures (e.g., Pöschl, 2005;Hammes et al., 2007).These include tars, chars, amorphous-like carbon consisting of randomly oriented poly-aromatic layers (e.g., Gustafsson et al., 2001) and finally carbon nanospheres, which are quite thermally refractory and strongly light absorbing (e.g., Ishiguro et al., 1997).These carbon fractions show gradually lower H / C and O / C ratios (e.g., Hammes et al., 2007), higher thermal refractivity and generally increasing light absorption.Petzold et al. (2013) suggest the use of the term "EC", with respect to the refractory component determined by thermal(-optical) methods.
Despite a lack of a clear separation between these fractions, the division into refractory and organic carbon is still useful for source apportionment.The thermally refractory carbon in aerosols is more likely to be combustion derived, even if some of the less refractory parts of the EC continuum, such as chars, are included.If the separation is pragmatically set at a point in the EC continuum that includes only refractory combustion derived carbon into EC, performing 14 C analysis on EC and OC separately can help distinguish biomass burning and fossil fuel sources of aerosol carbon.Since EC does not contain any secondary carbon, which usually has a strong contribution from modern sources, EC usually has a lower F 14 C than OC in most ambient samples.
The main challenge remains to isolate combustion-derived EC as completely as possible from organic material for 14 C analysis.The most common solutions for isolating EC (e.g., Szidat et al., 2004;Zhang et al., 2012;Bernardoni et al., 2013) are based on the thermal method by Cachier et al. (1989).In these methods, organic material is volatilized and oxidized to CO 2 during a low temperature step (S1) in pure oxygen.Then one or more intermediate steps, possibly in different atmospheres, are used to remove the most refractory OC together with the less refractory part of the EC continuum.In this work we used one intermediate step in pure oxygen (S2) and finally, a high temperature step (S3) in pure oxygen is used to oxidize the most refractory carbon to CO 2 .The CO 2 isolated in step S1 is assumed to be representative of organic carbon (OC).The remaining material burnt in step S3 will be called "refractory elemental carbon" (EC r ) in this work.
In practice there are several artifacts that can lead to incomplete separation of OC and EC: (1) refractory organic material remaining after S1 and S2, for example biological particles; (2) charring and pyrolysis of organic compounds during S1 and S2, which produce material that is inseparable from the elemental carbon that was contained in the particles originally; (3) premature loss of refractory carbon during steps S1 and S2 through catalytic reactions.The first two mechanisms transfer organic material into the extracted refractory carbon, and therefore usually bias F 14 C (EC r ) towards high values.Premature loss of EC results in a lower yield of EC, which might make 14 C analysis difficult.Since soot from burning biomass seems to be slightly less refractory than soot from fossil fuel burning, it is more easily lost in step S2, which will affect the source apportionment of EC (e.g., Han et al., 2007;Song et al., 2012;Zhang et al., 2012).
The goal of this research is to thoroughly evaluate a separation method for OC and EC based on the THEODORE method that was first proposed by Szidat et al. (2004).Our method was developed roughly in parallel with two other recently published methods (Zhang et al., 2012;Bernardoni et al., 2013) and used to analyze a large set of aerosol filter samples collected in the Netherlands.The evaluation of the current method will be also be compared to these two other methodological studies.

System description
The system used for thermal separation of OC and EC at IMAU is shown in Fig. 1.It can be divided into two subsystems: (1) the extraction line, where the combustion of the aerosol samples and the CO 2 formation occurs and (2) the CO 2 purification line, where the carbon dioxide is collected, purified and stored.These two parts are separated by a needle valve (D) that regulates the flow through the reaction tube (usually 30-50 mL min −1 ).The needle valve keeps the pressure in the extraction line high, at roughly 1150 mbar, whereas the pressure in the second part is kept below 20 mbar, in order to prevent condensation of O 2 in the cryogenic traps.
The first part of the system, the reaction tube, is made from quartz glass and surrounded by three ovens (C).The tube can be opened and a movable sample holder can be inserted.This filter holder can be moved either into the first or the second oven.The first oven is kept at a temperature below 400 • C and used for OC combustion.The second oven is for EC combustion.In the third oven, remaining incomplete combustion products are oxidized to CO 2 over a platinum catalyst at 650 • C.
The reaction tube is connected to two gas lines that provide pure helium (He) or oxygen (O 2 ).The first is an inactive gas used to clean the system before sample extraction, whereas O 2 is introduced for the combustion reaction and CO 2 formation.Before reaching the reaction tube, these gases are purified.For flushing of the extraction tube these gases can be directed by valve E directly to the vacuum pump.In this setting of valve E the CO 2 purification line is disconnected from the extraction line.
For collection of a CO 2 sample, valve E is switched and the gases are directed via valve F to the first cryogenic trap (1) and further via valve G to the glass line and vacuum pump.The cryogenic trap is cooled with liquid nitrogen and CO 2 and other gases with sufficiently low vapor pressure are collected, whereas O 2 and gases with higher vapor pressures are pumped away.At the end of the collection process, valve E is switched to disconnect the purification line from the extraction line and trap 1 is warmed to release the collected CO 2 .By switching valve F the released CO 2 is flushed with He through the Cu reduction oven to remove traces of NO x .The water trap after the reduction oven is cooled to   −70 • C using a dry ice-ethanol mixture.The CO 2 is collected in cryogenic trap 2. After closing of valve F and evacuation of the purification line, the CO 2 can be released from this trap into the glass line.The glass line consists of different components: a flask containing phosphorous pentoxide (P 2 O 5 ) to remove any remaining traces of H 2 O vapor, a calibrated volume with pressure sensor for a manometric determination of the amount of CO 2 , and sample tubes, where the CO 2 is stored until the 14 C analysis.

Measurement protocol
Before an experiment is started, a sample is inserted into a movable filter holder and placed in the reaction tube.After flushing with He and O 2 the sample is introduced into one of the ovens.For total carbon (TC) combustion the filter piece is directly inserted into oven 2 and heated to 650 • C for 15 min.The organic carbon (OC) combustion is done in oven 1, usually at 340 • C or 360 • C for 15 min, but a range of possible parameters is evaluated in this work.After this step, EC is isolated.The protocol for the EC isolation is the main focus of this work and will be discussed in detail in Sect.3.

Preparation of test filters
We performed experiments using known chemical substances to evaluate various aspects of the OC and EC separation.We used organic compounds and mixtures that represent typical components of the ambient organic aerosol: Glucose (alpha-D-Glucose, anhydrous, 96 %, Aldrich), Humic acid (Aldrich), a hydrocarbon standard mix (AccuStandard DRH-002S-R1), a fatty acid methyl ester (FAME) mix (Suppelco47885-U), and a non-volatile organic acid standard mix (Suppelco 46985-U).These organic compounds were applied to pre-heated quartz fiber filters of the same type that is used for collection of ambient aerosol samples.Small amounts (50-500 µL) of the three liquid organic mixes were directly pipetted onto the filter.The filters were analyzed after a short drying to evaporate the solvent (chloroform).Glucose was applied as an aqueous solution (50-100 µL) and the filter was dried before analysis over silica gel for several hours.In some experiments a commercial carbon black (Elftex 125, Cabot Corporation) was used as a surrogate for EC.The carbon black was pre-heated to 650 • C in air (not O 2 ) for 2 h to remove adsorbed organic material and a suspension of 0.2 g of carbon black in 100 mL of a 20 : 80 % ethanol -deionized water mixture was prepared.200 mL of this suspension was applied to the filters, which were then dried over silica gel for 24 h.

14 C measurements
The CO 2 collected from the combustion of various aerosol fractions was sent to the Centre for Isotope Research (CIO) at the University of Groningen for graphitization and Accelerator Mass Spectrometer (AMS) measurement.There CO 2 was reduced to graphite on a porous iron pellet (de Rooij et al., 2010) at a temperature of 600 • C in the presence of hydrogen gas at a molecular ratio of H 2 / CO 2 of 2.5.The water vapor formed during the reaction was cryogenically removed using Peltier cooling elements.The graphitization yield is usually virtually 100 % for samples larger than 30 µg.The iron pellet was pressed into a 1.5 mm target holder, which was introduced into the AMS system for subsequent measurement.
The high-throughput AMS system (van der Plicht et al., 2000) is dedicated to 14 C analysis, and measures simultaneously 13 C / 12 C and 14 C / 12 C ratios.Sample batches are usually analyzed together with varying amounts of reference materials ranging from 50-500 µg C. Two reference materials with known 14 C content are used: the HOxII standard (F 14 C = 1.3406) and graphite derived from a 14 C free CO 2 gas (F 14 C = 0).Contamination during the graphitization and AMS measurement causes deviations of the measured F 14 Cvalues from the nominal values.The magnitude of these deviations can be used to quantify the contamination with modern (F 14 C = 1) and fossil carbon (F 14 C = 0), which in turn are used for correcting the sample values (de Rooi et al., 2010).The modern carbon contamination is between 0.15 and 0.35 µg C, and the fossil carbon contamination is typically around 1 µg C (Prokopiou, 2010).

Filter handling
Ambient aerosol samples were also used as test filters.Three filter samples (CA2, CA12, and CA18) were used as test filters for EC recovery.They ware collected at the Cabauw observatory, a regional field site in the Netherlands in the following time periods: Filter CA2 -11 to 14 February 2011; Filter CA12 -27 February to 1 March 2011; Filter CA18 -6 to 12 April 2011.Other samples were used to evaluate the final combustion procedure and are more closely described in Table 1.All ambient samples are collected on Whatman quartz fiber filters (QMA 1851-150) using a high volume sampler (Digitel, DHA-80).Before loading, the filters were pre-heated at 800 • C for several hours, then wrapped in preheated (500 • C) aluminum foil and stored in sealed polyethylene bags at −18 • C until use.Then the filters were placed into the high volume sampler in a filter holder that was precleaned with ethanol.After sampling the filter was removed from the sampler immediately, wrapped again in clean aluminum foil and stored at −18 • C in a sealed plastic bag until analysis.During the transport to and from the field site (approximately 1 h) the filters were kept at ambient temperature.Blank filters used to determine the contamination introduced during handling and storage were treated exactly like the sample filters, except that they were kept in the high volume sampler for one minute only without switching on the sampler.
For analysis circular pieces with a diameter of 3 cm were cut from the filter.The tweezers and cutter were cleaned in acetone followed by ethanol and let dry for at least 10 min before handling the filters.For the separation of refractory carbon the water-soluble carbon was removed from the filter pieces by soaking the pieces in MilliQ water overnight (14 mL of water per filter piece) and drying them for 24 h over silica gel.A small fraction of insoluble aerosol material can be lost during the water extraction but this should have little influence on the 14 C content of the refractory carbon.However, it is recommended that the loss of insoluble material is monitored e.g., by measuring the absorption before and after water extraction, so that losses can be corrected if necessary.
3 Tests of the OC-EC separation method 3.1 Contamination from filter handling and thermal separation

Test procedure
Since organic material is ubiquitously present in the environment, it is virtually impossible to measure small carbon samples without introducing contamination.If the quartz fiber filters on which the aerosols are sampled or the filter holder are exposed to ambient air they quickly adsorb volatile organic carbon.Any contact with tools or surfaces increases the contamination despite pre-cleaning.Therefore, the contamination that cannot be eliminated by our filter handling protocol has to be quantified as accurately as possible and subtracted from the carbon concentration of our samples.
We assessed the contamination introduced by the thermal separation system in two ways.The first way is to directly introduce a pre-cleaned filter piece (7.07 cm 2 ) into the analytical system and measure the amount of CO 2 after all the extraction steps detailed in Sects.2.1 and 2.2.A second, indirect way of inferring the contamination is by combusting standard materials with known 14 C content and measuring the deviation from the expected values.Two standard materials (see Sect. 2.4) a HOxII standard and a 14 C free graphite powder were put directly onto the sample holder and combusted at 650 • C for 15 min.Figure 2 shows that F 14 C measured on anthracite samples deviates from the nominal value of 0 because of contamination during the extraction.Since the contamination adds a roughly constant amount of C to each extraction, the experimentally determined F 14 C of small amounts of standard material deviates more strongly from the nominal value than for larger amounts.The actual contamination can be parameterized as a sum of two components: a modern contamination (M mc ) with F 14 C (mc) = 1 and a fossil contamination (M fc ) with F 14 C (fc) = 0. Then the experimentally determined mass (M m ) after subjecting the standard material to combustion and purification can be written as follows: where M st is the carbon mass of the standard that was introduced into the ovens.This parameterization has a physical foundation, since contamination with aged, partially decayed 14 C sources is rare in a laboratory environment and most sources of contamination are either modern carbon or fossil fuel derived.From the F 14 C values measured in small standard samples we can calculate the mass of modern contamination by a simple mass balance equation: where M m is the experimentally determined carbon mass, F 14 C (m) the experimentally determined fraction modern, and F 14 C (st) the nominal fraction modern of the standard material.Substituting F 14 C (st) = 0 for the anthracite standard and considering that F 14 C (fc) = 0 and F 14 C (mc) = 1, this equation allows to estimate the mass of modern contamination M mc = F 14 C (m) •M m .Once the modern contamination is known, M fc can be derived from measurements of the HOxII standard by substituting F 14 C (st) = 1.3406,F 14 C (fc) = 0 in Eq. ( 3) and combining with Eq. ( 2): (4)

Results
The direct determination of the contamination from blank filter analysis yields usually around 1-2 µg of carbon/sample for organic carbon (OC) and extracted refractory carbon (EC r ), which includes all the contamination introduced by filter cutting and handling.This contamination is not increased by subjecting a clean filter to the water extraction procedure, described in Sect.2.2.Field blank filters yield on average the same amount of OC and a slightly higher amount of EC r of 2 µg C/sample.The latter does not depend strongly on the filter area used.For a typical filter piece of 3 cm in diameter this amounts to approximately (0.35 ± 0.15) µg C cm −2 for OC.For EC r extraction, twothree filter pieces are usually analyzed to collect enough material for 14 C analysis and therefore the contamination of EC r is approximately 0.1 and 0.2 µg C cm −2 for three and two filter pieces, respectively.
The organic fraction of one field blank filter was analyzed for 14 C by pooling several extracts of the same filter.This fraction had F 14 C (OC,blank) = (0.797 ± 0.015), which is close to typical ambient F 14 C (OC) values.The 14 C content of EC r was analyzed by pooling CO 2 extracted from 5 different blanks that for test purposes were kept in the filter stacks of the sampler not for one minute but for varying amounts of time up to several days without being exposed to the air stream.The time in the sampler had no effect on EC r concentrations, which indicates that EC r on the blank filter is only introduced by filter handling and thermal extraction.The F 14 C (EC r ,blank) for these 5 samples is (0.543 ± 0.014), which is significantly higher than typical values of F 14 C (EC r ,blank) from ambient aerosol samples (e.g., Szidat et al., 2006).The blank correction will therefore shift F 14 C (EC r ) to lower values.
Twelve independent measurements of the anthracite standard throughout the time period 2010-2013 yielded a modern contamination M mc with an average and standard deviation of (1.92 ± 0.83) µg C per extraction, excluding one outlier of 6 µg.However, an improved cleaning method for the reaction tube used from 2011 onwards resulted in a lower M mc to (1.36 ± 0.20) µg C per extraction for the 5 standards measured in this time period from 2011-2013.The fossil contamination is on average (1.37±0.93)µg C per extraction for 12 samples between 2011-2013 and is not influenced by the change in cleaning procedure of the reaction tube.The propagated measurement uncertainties (arising from the uncertainties in 14 C measurements and carbon amount) for the average M mc are smaller than 5 %.For the average M fc the propagated measurement uncertainties are around 30 %.
Figure 3 shows the handling blank in µg C cm −2 for TC, OC and EC as white bars.For EC the two white bars represent the lower limit of ∼ 0.1 µg cm −2 , if 3 filter pieces are used for EC extraction and ∼ 0.3 µg cm −2 , if one filter piece is used.This is compared to the concentration of carbon on a typical small sample, which is shown in black bars.Using materials with known 14 C content, the contamination of TC could be assessed by an alternative method and divided into modern contamination (M mc ) and fossil contamination (M fc ) as detailed in Eqs. ( 2)-( 4).These are shown as additional grey bars for TC only.The sum of M fc and M mc is slightly lower than the total carbon handling blank.The blanks are also compared to a filter sample with a carbon mass concentration in the lower range of typical filter samples used for radiocarbon analysis in our group.For samples with a total carbon concentration ∼ > 20 µg cm −2 , the amount of carbon on a blank filter is less than 5 % of the carbon amount on a typical ambient filter.Therefore, the blank correction does not introduce large uncertainties to the data.

Discussion
The contamination introduced by the filter handling and analysis is reasonable compared to expected sample size.The contamination is on the order of 1 µg/extraction for OC and EC and probably difficult to reduce even further without special clean room facilities.The sum of M fc and M mc , which is more indicative of the contamination introduced by the thermal extraction system, is lower than the total carbon handling blank.Figure 2 indicates that contamination from filter handling contributes roughly 1/3 to the total blank.

Test procedure
Several studies show that when heating different kinds of soot in He-O 2 mixtures or air, a small mass loss can start already at fairly low temperatures of around 300 • C (e.g., 32  Sharma et al., 2012;Su et al., 2004).To test for loss of EC at temperatures that are typically used to combust OC we used commercial carbon black samples (Elftex), which is often regarded as an appropriate surrogate for diesel soot (e.g., Yezerets, 2005, and references therein).The carbon black was applied to quartz fiber filters as described in Sect.2.3.Before analysis, the filters were heated at 100 • C in oxygen to remove any ethanol that might have adsorbed on the filter during the application of the carbon black suspension.The filters were then exposed to 340 • C in pure O 2 for up to two hours and the amount of CO 2 evolved during this time was determined manometrically.Blank filters spiked with only deionized water and ethanol were analyzed in the same way and the carbon amount found on the blank filter (around 1 µg cm −2 ) was subtracted from all analyzed filters.

Results
Figure 4 shows the loss of carbon black material at 340 • C in O 2 as a function of oxidation time for two types of quartz fiber filters, namely circular Whatman QMA 1851-150 (Filter 2) as used for the ambient samples in this study, compared with a square Whatman QMA No: 1851-865 (Filter 1), which was already several years old.The loss of carbon black increases linearly with oxidation time and the rate is roughly half for filter 2. Impurities in the filter material can catalyze the loss of soot and this example shows that the quartz fiber filter material itself can make a difference in the rate of oxidation of soot.On filter type 1, which was used in all subsequent studies with this system, approximately 1.5 h are necessary to oxidize 10 % of the carbon black at 340 • C.
Considering that the filter is only heated for 15 min for OC determination, the loss of carbon black in this time period is less than 3 %.In ambient aerosol the concentration of EC is usually significantly smaller than the concentration of OC.
Assuming that typically 20-30 % of total carbon consists of EC the contribution of prematurely evolved EC to the extracted OC should be below 1 %, if carbon black is a suitable model for atmospheric EC.

Discussion
Several studies (e.g., Sharma et al., 2012;Han et al., 2007) show that most carbon black samples are somewhat more thermally refractory than diesel soot, which in turn seems to be more refractory than soot from wood burning, at least with respect to the thermal-chemical oxidation method (Song et al., 2012).Therefore, the carbon black experiments can give a reasonable estimate for the loss of more refractory types of soot, but it must be kept in mind that some types of soot or char can be less refractory than carbon black.However, it is reasonable to assume that even for less refractory types of soot the contribution of prematurely evolved EC to the OC sample will not exceed a few percent.This is supported by indirect evidence gained by combustion of an ambient filter sample at increasing temperatures of 300, 320, 340, 360, and 380 • C (not shown).If prematurely oxidized EC significantly contributed to the OC that is combusted at these temperatures, this contribution should increase at higher temperatures.This would cause a decrease of F 14 C (OC) with oven temperature, since the F 14 C (EC) is usually considerably smaller than F 14 C (OC) .Even though the amount of extracted OC increased with oven temperature and was 30 % higher at 380 • C than at 300 • C, F 14 C (OC) remained at 0.809 ± 0.013 %, independent of the oven temperature.A similar result was found by Szidat et al. (2004), who also concluded that the influence of prematurely evolved EC on F 14 C (OC) is negligible.

Test procedure
Charring of organic compounds during thermal desorption is often monitored by optical methods, such as a laser beam, to detect an increase in the light absorption through a filter.This is especially useful when the heating of a filter occurs in an inert gas, where charring is substantial and loss of EC is negligible.When combusting a filter in pure O 2 , charring is significantly reduced, especially after water extraction.A small loss of EC could partially offset the small increase in light absorption caused by charring.It can therefore be difficult to conclusively quantify charring by optical methods, especially on highly loaded filters such as often used for 14 C analysis.We therefore used representative organic test materials described in Sect.2.3 to investigate their charring characteristics directly.The test compounds were applied to pre-cleaned quartz fiber filters.The filters were introduced into oven 1 at 340 • C for 15 min and the amount of evolved CO 2 was determined manometrically.The combustion step at 340 • C was repeated a second time to verify that all OC was oxidized during the first combustion step.Then the filter was introduced into oven 2 at 650 • C and the evolved CO 2 was measured.Since only organic and no refractory carbon was present on the filter originally, the CO 2 measured at 650 • C originates from the char that was formed during the combustion of the organic test material.

Results
Figure 5 shows that charring is small for all investigated compounds during flash heating in O 2 .Less than 2 % of glucose, of humic acid, of the hydrocarbon standard mix, and of the fatty acid methyl ester (FAME) mix is found as refractory carbon after the first OC combustion step.Charring is slightly higher for organic acids, of which 3.5 % are found as refractory carbon.However, these are normally only a small fraction of the total organic carbon in ambient aerosols.Moreover, they are largely removed by water extraction of the filter prior to EC analysis.In general, our results imply that 1-2 % of the water insoluble OC is found as EC r .Even this small amount can have a measurable effect on F 14 C (EC r ) , because EC r constitutes usually a small fraction of TC.Assuming that elemental carbon is around 20 % of total carbon, then around 5 % of the analyzed EC r can be charred organic carbon.In the extreme case of completely fossil EC (F 14 C (EC) = 0) and biogenic OC (F 14 C (OC) ∼ 1), this would increase F 14 C (EC) by 0.05 (absolute).If F 14 C (OC) is lower and F 14 C (EC) is higher than in this extreme case, the overestimate is smaller (e.g., 0.03 for F 14 C (OC) ∼ 0.8 and F 14 C (EC) ∼ 0.2).

Discussion
Removal of organic carbon by flash heating in O 2 was originally developed to minimize the charring artifact (Cachier et al., 1989) and from our tests it is clear that this approach is very successful for water insoluble compounds, but less successful for water soluble carbon.WSOC usually also accounts for the majority of charring during heating under inert atmospheres (e.g., Yu et al., 2002).Zhang et al. (2012) used laser transmission measurements to monitor charring during the OC combustion steps.They combusted the water extract of an ambient aerosol sample, which showed even higher charring than the organic acids investigated in this work.However, for water extracted ambient samples they also concluded that the contribution from charred OC to the extracted EC can be reduced to below 5 % for ambient samples.This indicates that the water insoluble test compounds chosen for our charring test are representative of the ambient water insoluble aerosol.This makes them a useful substitute for evaluating charring in OC-EC separation systems that lack optical monitoring.Cachier et al. (1989) found a higher percentage of charred humic acids remaining on the filter (∼ 30 %) and somewhat higher amounts of charred glucose (∼ 10 % for filter loadings comparable to our work).The reason for this difference is not clear.

Test procedure
During the charring tests we found that, except for a small charred fraction, all organic test compounds used in this study were removed after combustion at 340 • C for 15 min in pure O 2 , even high molecular weight humic acid.However, the continuous transition from OC to the most refractory carbon in form of elemental carbon nanospheres in ambient aerosols is difficult to capture using simple test compounds.Therefore ambient samples are also used to test the efficiency of separating OC and EC in ambient aerosols.The extraction methods usually utilize a practical compromise by excluding compounds with intermediate refractivity from the analysis.A short intermediate step S2 in O 2 at higher temperature should remove a mixture of the most refractory organic material and less refractory elemental carbon from the filter (e.g., Szidat et al., 2006;Zhang et al., 2012).However, the use of too harsh oxidation methods that allow only the most refractory part of EC to be recovered have the disadvantage that biomass burning EC is removed to a somewhat larger extent than fossil EC.Biomass burning soot is less refractory than soot from liquid fuels (e.g., Song et al., 2012;Zhang et al., 2012) and has been found to evolve together with chars during thermal treatments rather than with diesel soot.Preferential removal of biomass burning soot thus artificially decreases F 14 C (EC r ) which leads to an underestimate of the fraction modern of EC.
We therefore investigate how different choices of parameters for extraction step S2 (with temperature for S1 remaining constant at 340 • C) affect (1) the recovery of EC and (2) the efficiency of OC removal.A water-extracted filter was kept in the first oven for 15 min for OC combustion (S1).Then it was introduced into the second oven at a higher temperature T 2 for 1-3 min and the CO 2 that evolved during this extraction step was pumped away.Then the filter was taken out of the oven system, while the CO 2 resulting from the combustion step S1 was purified and processed.For the last step, the filter was introduced into oven 2 at 650 • C and EC r was measured.
The complete removal of OC is difficult to assess directly.However, two indirect estimates are used that can indicate the extent of OC removal.First, refractory OC is often associated with highly polymeric, char-or tar-like substances.These substances are often subsumed under the name "brown carbon" (BrC), which is characterized by its light absorption properties (e.g., Andreae and Gelencsér, 2006;Pöschl, 2005).The recovery of EC was estimated by comparing the light absorption of a water extracted filter piece to the absorption of a filter piece after organic carbon was removed during S1 and S2.Absorption was measured using an integrating sphere photometer (Hitzenberger et al., 1996;Wonaschütz et al., 2009).BC and BrC concentrations were determined by measuring the light absorption of the filter pieces at 3 different wavelengths.BrC is distinguished from BC by exploiting the different wavelength dependence of light absorption following Wonaschütz et al. (2009).If brown carbon is removed after S1 and S2, this gives a good indication that the extraction protocol was effective in removing most refractory OC.
A second method is to measure the mass and F 14 C of the refractory carbon EC r as a function of increasing temperature and/or duration of step S2.If a decrease in recovered mass is 35 "1"min" 400C" "2"min" 400C" "3"min" 400C" "1"min" 450C" "2"min" 450C" "3"min" 450C" "1"min" 500C" "2"min" 500C" "3"min" 500C" associated with a strong decrease in F 14 C (EC r ) , this can be an indication that fairly "modern" organic material is still being removed when the temperature of step S2 is increased.Preferential removal of biomass burning EC is usually associated with a much more gradual decrease in F 14 C (EC r ) .

Results
The effect of the duration and temperature of S2 on the recovery of black carbon are shown in Fig. 6 for filter CA18.A duration of S2 of 1 min causes loss of black carbon that does not exceed 20 %.There is little change in black carbon content between a filter that has been exposed to 450 and 500 • C for 1 min, which suggests that 1 min of heating might not be enough for the filter temperature to fully reach 500 • C. When the duration of S2 is increased to two minutes an increase in temperature from 400 to 450 • C causes little extra loss of BC.The amount of BC remaining after S2 of 450 • C is very similar for exposure times of 1 and 2 min, suggesting that little additional BC is removed by increasing the oxidation time.However, raising the temperature further to 500 • C strongly decreases the recovered BC to around 60 %.If the duration of S2 is increased to 3 min, then the recovery at 400 and 450 • C decreases by only 8 % and remains comparable, whereas the recovery at 500 • C decreases sharply to 50 %.This indicates that a temperature 500 • C is too high, since substantial amounts of BC are removed.A time step of 1 min is likely too short for the filter to reach the desired temperature, therefore a step of 2 min was tried for further experiments.
F 14 C (EC) of filter CA12 was measured as function of S2 temperature with S2 duration of 2 min.Figure 7a shows that the amount of remaining EC r (black bars) decreases only slightly if the S2 temperature is raised from 400 to 450 • C, but only 50 % of the original EC remains after 2 min at 500 • C. F 14 C (EC r ) (grey bars) remains virtually unchanged for S2 temperatures of 400 and 450 • C and then decreases from 0.36 to 0.31 once the S2 temperature is raised to 500 • C. F 14 C of the material removed by each successive temperature increase can be calculated according to the mass balance: where F 14 C (2−1) is the fraction modern of the material that was removed by raising S2 temperature from T 1 to T 2 ; F 14 C (EC1) , F 14 C (EC2) are the fraction modern of the refractory carbon recovered after T 1 and T 2 ; M EC (T 1 ) and M EC (T 2 ) are the mass of the refractory carbon recovered after T 1 and T 2 .These are shown in Fig. 7a as patterned bars.It is obvious that the removed carbon has a similar fraction modern as the EC r itself, which suggests that the largest amount of organic carbon was removed already at S2 of 400 • C and the higher temperature steps remove mainly EC.The slight decrease in F 14 C with increasing S2 temperature results more likely from differential oxidation of biomass burning EC and fossil fuel EC, than from remaining organic carbon.If organic carbon was left in the sample a larger decrease in F 14 C (EC) between 400 and 450 • C would be expected than at a temperature that is high enough to burn significant amounts of black carbon.
Figure 7b shows similar results for a 1 min S2 time step.Filter CA2 was used for these experiments, since not enough material was left of filter CA12 for further tests.An increase in temperature of S2 from 400 to 500 • C does not cause a strong decrease in the recovered EC.A similar trend was observed for the black carbon concentrations shown in Fig. 6.This supports our conclusion that the filter itself probably does not reach 500 • C within 1 min.However, even though not much extra material is removed during 1 min at 500 • C, the fraction modern of the recovered EC decreases significantly.Following mass balance Eq. ( 6), F 14 C of the carbon removed by raising T to 500 • C is 0.67, very close to the F 14 C of water insoluble organic carbon of filter CA2 (0.62).This shows that a time period of 1 min for the intermediate step S2 is probably not sufficient to remove all organic carbon from the filter.
The final decision was to use a S2 temperature step of 2 min at 450 • C, because this combines reasonable BC recovery rates with removal of most refractory OC.For T 1 we decided to use a temperature of 360 • C, since this gives a higher recovery rate of OC on some highly loaded samples.Varying the temperature of step 1 did not change F 14 C (OC) and showed only minimal loss of carbon black material, which both indicates that this slightly higher temperature should not strongly affect EC recovery and consequently F 14 C (EC r ) .
This temperature protocol was tested on 4 ambient filter samples with respect to removal of brown carbon and EC  recovery.Two of the samples had marine air mass origin, whereas the other two had continental air mass origin.The two samples with marine air mass origin contain little water insoluble brown carbon (Table 1), whereas the continental samples contain significant amounts.After OC removal the brown carbon is below detection level for all test filters, which is further indication for satisfactory removal of the organic fraction.The recovery of BC is approximately 80 % for three of the filters, which is close to the target value for which the S2 step was chosen.However, for filter CA21 the recovery is slightly above 100 %.Filter CA21 was sampled for a week during fairly high aerosol concentrations and has therefore a very high aerosol loading (TC = 219 µg cm −2 ).It is possible that the saturation of the integrating sphere method is reached and a small loss of BC cannot be detected.However, is also possible that 2 min at 450 • C are not enough to remove such large amounts of OC that is coating the BC.However, since all the brown carbon is nevertheless removed, this does not seem very likely.Alternatively, a fairly high amount of brown carbon that is relatively susceptible to charring could produce measurable amounts of black carbon during S1 and S2.The F 14 C (EC r ) values of filter samples with a BC recovery of ∼ 1 should be treated with caution, and possibly the extraction procedure should be repeated either with higher temperature or longer duration of S2.
In the last column of Table 1 we show F 14 C (EC t ) , the fraction modern estimated for the total elemental carbon at 100 % recovery, assuming that biomass burning EC is lost to a greater degree than fossil EC during S2.The relationship between F 14 C (EC) and optical recovery is taken from the supplementary material in Zhang et al. (2012).

Discussion
The amount and F 14 C as well as optical properties of EC r suggest that an S2 time removal step of 2 min around 450 • C is a good choice for separating organic and refractory carbon.Two other methods published in the literature (Zhang et al., 2012;Bernardoni et al., 2013) use a similar approach but different combustion temperatures and atmospheres.Each is evaluated with different methods, but some main conclusions are similar.The black carbon recovery by the presented method (∼ 80 %) is similar to the one reported for the Swiss S4 protocol (Zhang et al., 2012) and slightly higher than in the method proposed by Bernardoni et al. (2013).The contamination of EC r with charged OC is estimated to be < 5 % in this work and Zhang et al. (2012).This similar performance in different test procedures gives additional confidence in these methods.
The method presented in this work does not use a He step, which the other methods find necessary to remove a non light absorbing carbon fraction that co-evolves with BC in O 2 up to very high temperatures.At present it is not clear whether this carbon fraction is truly organic, such as pollen or other refractory bio-polymers or whether it might be elemental carbon that is potentially not so strongly light absorbing, such as the more amorphously oriented carbon layers seen by Gustafsson et al. (2001).
A future inter-comparison of these three and also other methods (e.g., Zencak et al., 2007;Heal et al., 2011) would shed more light on the issue how strongly the differences in the extraction protocols affect F 14 C (EC r ).Such an intercomparison has been done for organic and total carbon (Szidat et al., 2013), but only very limited data on EC were available at that time.

Conclusions
We tested a thermal method designed to separate organic and combustion derived refractory carbon for radiocarbon analysis of aerosol samples.For OC extraction the sample is combusted in pure oxygen at 360 • C for 15 min.For EC extraction a water-extracted filter sample is first subjected to 360 • C for 15 min, followed by a 2 min intermediate combustion step at 450 • C to completely remove OC.After OC removal EC is extracted at 650 • C for 15 min.
The recovery of EC after the OC removal steps was quantified by measuring the recovery of black carbon (BC) using an optical method.The BC recovery was around 80 %, which is comparable to two similar methods (Zhang et al., 2012;Bernardoni et al., 2013).The effectiveness of the OC removal steps are strongly supported by the successful removal of brown carbon in 4 ambient test samples and by the removal of several organic test compounds, including high molecular weight humic acid.
Charring of water insoluble organic test materials was found to be minor.Less than 2 % of the organic test compounds were found as EC after the OC combustion step.From this we estimate that around 5 % of EC could originate from charred OC.This is in close agreement with Zhang et al. (2012), who estimated charring of ambient samples using optical methods.

FiguresFigure
Figures Figure 1.The EC--OC separation system consists of two sub--systems: (1) the extraction tube (green box), where organic and elemental carbon are oxidized to CO2 and (2) the purification lin (blue box), where CO2 is collected and separated from interfering gases.By switching three-way valve E to the low vacuum pump the purification line can be separated from the extraction tube.

Figure 1 .
Figure 1.The EC-OC separation system consists of two subsystems: (1) the extraction tube (green box), where organic and elemental carbon are oxidized to CO 2 and (2) the purification line (blue box), where CO 2 is collected and separated from interfering gases.By switching three-way valve E to the low vacuum pump the purification line can be separated from the extraction tube.

Figure 2 .Figure 2 .
Figure 2. The experimentally determined F 14 C of the anthracite standard (nominal F 14 C = 0) as a function of sample size.The blue diamonds correspond to samples measured after a new protocol for cleaning the reaction tube was introduced, the red squares before the new protocol.An outlier is shown in green.

Figure 3 .Figure 3 .
Figure 3.The carbon concentration found on a typical lightly loaded filter sample compared with the directly measured handling blank (white bars), and the average modern and fossil contamination derived from analysis of anthracite and HoxII standards.The modern and fossil contamination could only be determined for total carbon.The two white bars for RC give an upper and lower limit of the handling blank, depending on the number of filter pieces used in the extraction.

Figure 4 .Figure 4 .
Figure 4. Fraction of carbon black (CB) that was lost at 340 °C in pure oxygen as a function of oxidation time for two different filter types from the same manufacturer (Filter 1: Whatman QMA 1851--865and Filter 2: Whatman QMA 1851--150).The open squares correspond to carbon black coated with ammonium sulfate

Figure 5 .Figure 5 .
Figure 5. Charring of several test compounds given as % of the substance found as RC after heating at 340°C for 15 min in pure oxygen.See appendix A for a description of the hydrocarbon and organic acid mixes.

Figure 6 .
Figure 6.Black carbon recovery as a function of duration and temperature of separation step S2

Figure 6 .
Figure 6.Black carbon recovery as a function of duration and temperature of separation step S2.

Figure 7 .
Figure 7. Fraction modern and mass fraction (EC r /TC) of the extracted refractory carbon (EC r ) remaining after a separation step S1 (15 min at 360 • C) and S2.The duration and temperature of separation S2 are given on the x axis.The difference in EC r /TC and F 14 C (EC r ) found after extraction step S2 at lower and higher temperature are given as shaded bars.

Table 1 .
Zhang et al. (2012)and black carbon (BC) detected in 4 ambient filter samples before OC removal and after OC removal (OC rem.) using S1 at 360 • C for 15 min and S2 at 450 • C for 2 min in pure oxygen.mar-Filterwith back trajectories starting over the ocean, cont -Filter with back trajectories starting over the land.Also shown is the BC recovery, F 14 C of the recovered EC (EC r ) and the extrapolated F 14 C for 100 % recovery (EC t ).For the extrapolation we take the average slope of F 14 C with optical recovery (0.31) from the supporting material inZhang et al. (2012).