Journal of Geophysical Research: Atmospheres

Semicontinuous aerosol carbon measurements: Comparison of Atlanta Supersite measurements

Authors


Abstract

[1] An intensive field campaign of the Atlanta Supersite Experiment was carried out at a ground-based measurement site on Jefferson Street in midtown Atlanta, Georgia, from 3 August to 1 September 1999. This paper examines the semicontinuous particulate organic and elemental measurements that were made as a part of the experiment. Measurements were made using a Rutgers University/Oregon Graduate Institute in situ thermal-optical carbon analyzer, Rupprecht and Patashnick 5400 ambient carbon particulate monitor, Radiance Research particle soot absorption photometer, Aerosol Dynamics flash vaporization carbon analyzer, and Magee Scientific AE-16 Aethalometer. The “intersampler precision” with which semicontinuous particulate total carbon (TC), organic carbon (OC), and elemental carbon (EC) were measured is 7, 13, and 26%, expressed as pooled coefficients of variation of 2, 3, and 4 instruments, respectively. Correlations between pairs of OC measurements are moderate (R2 = 54–73%), and correlations between pairs of EC measurements are high (R2 = 74–97%). Differences in reported OC concentrations are small compared to differences in EC concentrations; intersampler EC concentration differences result from differences (1) in the operational definitions of OC and EC, (2) in the calibration of optical instruments, and (3) because EC values are closer to limits of detection. This agreement between semicontinuous samplers is quite good, especially in light of previous particulate carbon comparisons. Reasons for measurement differences and benefits of automated time-resolved measurements are discussed.

1. Introduction

[2] The importance of atmospheric carbonaceous aerosol to human health, visibility, and radiative forcing of climate has furthered the development of measurement methods for particulate carbon, from gas chromatography-mass spectrometry for individual organic compound analysis to combustion for total carbon analysis [Turpin et al., 2000; Jacobson et al., 2000]. Due to the complexity of the carbonaceous particulate matter [Rogge et al., 1993; Saxena and Hildemann, 1996; Turpin, 1999], analytical methods that fractionate particulate carbon into total organic and elemental (or black) carbon are widely used in aerosol characterization studies. The definitions of organic (OC) and elemental carbon (EC) are somewhat method-dependent. Inorganic carbonate carbon, found in minerals such as calcite (CaCO3) and dolomite (CaMg(CO3)2), is less frequently measured. It is frequently assumed, based on available data, that carbonate carbon is a small fraction of total ambient particulate carbon. Carbonate carbon is typically analyzed by sample acidification and detection of the evolved carbon dioxide.

[3] Two general operational-definitions of OC and EC exist: (1) OC combines with oxygen, hydrogen, and other elements, and therefore volatilizes and decomposes upon heating in an oxygen-free environment, and EC is an agglomerate of primarily carbon atoms that only combusts in the presence of oxygen; or (2) elemental carbon comprises all visible-light absorbing carbon. Solvent extraction, thermal (e.g., selective volatilization), and optical (i.e., visible light attenuation) methods have been used to measure “OC” and “EC”. Solvent extraction methods assume that solvent-insoluble carbon corresponds to an upper limit for EC; likewise, solvent-soluble carbon provides a lower-limit of OC. Optical methods define EC to be all visible-light absorbing particulate matter, and relate the EC loading or concentration to the measured light absorption via a calibration coefficient. The quantity reported is frequently called “light absorbing carbon” (LAC) rather than EC. Some thermal methods assume that all carbon evolved below a certain temperature is OC, and higher temperature carbon is EC. It has been suggested by some researchers that these quantities be called “low temperature” and “high-temperature” carbon [Turpin et al., 2000]. Other thermal methods volatilize OC by heating in the absence of oxygen, and combust EC by heating in the presence of oxygen. During oxygen-free OC volatilization, some OC pyrolyzes and becomes EC. Without correction, this would lead to an overestimate of particulate EC. Thermal-optical methods correct for the pyrolytic conversion of OC to EC by monitoring the transmittance or reflectance of the filter during analysis. Pyrolysis reduces the transmittance of light through the filter; as EC is removed the transmittance increases again. Pyrolyzed OC is estimated as the amount of carbon removed in order for the transmittance through the filter to regain its prepyrolysis value. This quantity is added to the OC value to provide total particulate OC and EC.

[4] Sampling artifacts also complicate particulate carbon measurement. A quartz fiber filter is a typical sampling medium for the measurement of OC and EC, because it can be heated to greater than 900°C and thus is suitable for thermal carbon analysis. However, it has a large, initially clean, surface that adsorbs some organic vapors while collecting particles. While the collection efficiency of a quartz filter for organic vapors is small, this positive artifact can lead to errors comparable to particulate OC concentrations [Turpin et al., 1994, 2000]. In addition, Eatough et al. [1993, 1995] has argued that semivolatile organic compounds in the particulate phase are removed by volatilization due to the pressure drop across a quartz fiber filter (negative artifact). Changes in sampling conditions (e.g., temperature, relative humidity, organic vapor concentrations) also result in changes in the gas-phase/adsorbed-phase partitioning of organic compounds, resulting in samples that are more representative of the conditions at the end of the sampling period.

[5] Two general approaches have been used to account for sampling artifacts: (1) estimation of the size of the artifact, and (2) reduction of the size of the artifact. A backup quartz fiber filter is frequently used as a “dynamic blank,” to provide an estimate of the adsorption of organic vapors on a quartz sampling filter. Placement of the quartz backup filter behind a Teflon filter, rather than behind a quartz filter, provides a larger estimate of adsorbed vapor, and this Teflon-quartz backup is believed to be a more accurate adsorption estimate [McDow and Huntzicker, 1990; Hart and Pankow, 1994; Turpin et al., 1994]. Sometimes a denuder is used upstream of the sampling filter to reduce the quantity of adsorbed vapor on the sampling filter. Volatile losses could be enhanced in the reduced vapor-phase environment downstream of a denuder. Therefore adsorbents are sometimes placed downstream of the sampling filter. These materials collect whatever vapors are not removed by the denuder plus whatever particulate organic material volatilizes from the collected particles. Again, samples of particle-free ambient air can be used to investigate the magnitude of organic vapors penetrating through the denuder and adsorbing on the quartz sampling filter and the downstream adsorbents (i.e., remaining positive artifact). A more extensive discussion of organic sampling artifacts is provided by Turpin et al. [2000] and Huebert and Charlson [2000].

[6] A few carbonaceous aerosol intercomparison studies have been conducted previously. For example, the 1986 Carbonaceous Species Methods Comparison Study (CSMCS) sponsored by the California Air Resources Board was carried out to investigate uncertainty associated with the sampling and analysis of carbonaceous aerosol (12 participants) [Countess, 1990; Hering et al., 1990]. The OGI in situ thermal-optical carbon analyzer [Turpin et al., 1990], an Aethalometer [Hansen and McMurry, 1990], and a photoacoutic spectrometer [Adams et al., 1990] were the semicontinuous methods at that study. Total carbon had a pooled coefficient of variation for integrated samples analyzed by different laboratories (e.g., aliquots of ambient PM and source samples analyzed by different analytical methods) of 9.0%, whereas the coefficients of variation for OC and EC were 25.8% and 52.3%, respectively [Countess, 1990].

[7] Since the 1986 intercomparison, a number of time-resolved semicontinuous carbon analysis methods have been developed. This is important because atmospheric processes that affect fine particulate carbon concentrations occur on a time scale of minutes to hours. For example, one-hour OC and EC measurements have been used to estimate the relative contributions of primary (emitted in particulate form) and secondary (formed in the atmosphere) OC to total carbonaceous aerosol [Turpin and Huntzicker, 1995; Lim, 2001]. Such estimates aid the development of effective control strategies for atmospheric fine particulate matter. Short-time resolution could also provide information relevant to human exposure and acute health effects. Continuous measurements with better than 5-min resolution are needed for aircraft studies of the radiative forcing of atmospheric aerosol. In addition, improvements in measurement sensitivity enhance the time-resolution of carbonaceous measurements made in clean marine environments in support of radiative-forcing climate studies.

[8] An intensive field campaign of the Atlanta Supersite Experiment was carried out at a ground-based measurement site on Jefferson Street in midtown Atlanta, GA, during the summer of 1999. Semicontinuous measurements of OC and/or EC were made with 5 instruments as a part of the experiment. This paper describes these methods and their collocated measurements and discusses the sampling and analytical uncertainties associated with semicontinuous carbon measurement.

2. Experiment

[9] The Jefferson Street measurement site was located in a mixed industrial-residential area about 4 km northwest of downtown Atlanta. Five semicontinuous carbon analyzers were installed in air-conditioned trailers. Each instrument sampled ambient PM2.5 through individual sampling inlets located about 2 m above the trailer roof from 3 August until 1 September 1999.

[10] The Rutgers University (RU)/Oregon Graduate Institute (OGI) in situ thermal-optical carbon analyzer (New Brunswick, NJ), Rupprecht and Patashnick (R&P) 5400 ambient carbon particulate monitor (Albany, NY), Radiance Research particle soot absorption photometer (PSAP; Seattle, WA), Aerosol Dynamics, Inc. (ADI) flash vaporization carbon analyzer (Berkeley, CA), and the Magee Scientific AE-16 Aethalometer (Berkeley, CA) were operated by RU, Atmospheric Research and Analysis, Inc. (ARAI), ARAI, ADI, and Harvard School of Public Health (HSPH), respectively. Only two instruments, the RU/OGI in situ thermal-optical carbon analyzer and the R&P 5400 ambient carbon particulate monitor, reported both OC and EC. Measurement cycles ranged from 1 min to 2 hour (e.g., 1 min PSAP; 2 hour RU). Sampling and analytical protocols for each method are described below and summarized in Table 1.

Table 1. Sampling and Analytical Methods for Organic (OC), Elemental (EC), and Total Carbon (TC) Measurements During the Atlanta Supersite Experiment in 1999a
InstrumentOperatorSampling MediumOCECTC
  • a

    NA: not applicable; GFF: glass fiber filter; QFF: quartz fiber filter; ADI: Aerosol Dynamics, Inc.; ARAI: Atmospheric Research and Analysis, Inc.; HSPH: Harvard School of Public Health; R&P: Rupprecht and Patashnick; RU: Rutgers University; OGI: Oregon Graduate Institute.

ADI flash vaporization carbon analyzerADIimpactorflash vaporization in airNANA
Particle soot absorption photometer (Radiance Research)ARAIGFFNAlight absorption (10 m2/g assumed)NA
Aethalometer (Magee Scientific)HSPHQFFNAlight absorption (12.6 m2/g assumed)NA
R&P 5400 ambient carbon particulate monitorARAIimpactor275°C in air750°C in airOC + EC
RU/OGI in situ carbon analyzerRUQFF700°C in He (pyrolysis, adsorption correction)850°C in 2% O2 in HeOC + EC

2.1. Rutgers/OGI in Situ Thermal-Optical Carbon Analyzer

[11] The Rutgers/OGI in situ thermal-optical carbon analyzer measured 1-hour average OC and EC concentrations every other hour. Ambient air was pulled at 16.1 L min−1 through a 2.5 μm cutpoint cyclone. This air stream was split evenly between the “aerosol analytical unit,” that collects particles and some adsorbed vapor on a quartz fiber filter and the “vapor analytical unit” that provides a dynamic blank by pulling particle-free ambient air through a quartz fiber filter (particles are removed first with a Teflon filter). After collection for a prescribed period, air is purged from the unit and analysis begins. The “aerosol unit” quartz fiber sampling filter is heated stepwise to 700°C in helium to volatilize OC and in 2% oxygen in helium to 850°C to combust EC. The transmittance through the quartz fiber filter is monitored with a He-Ne laser throughout the analysis to correct for the pyrolytic conversion of OC to EC. The quartz fiber filter in the “vapor analytical unit” is analyzed in 2% oxygen in helium at 750°C. This dynamic blank provides an estimate of the adsorption of organic vapors on the aerosol unit's sampling filter. It was subtracted from the measured OC in the aerosol unit to obtain the reported particulate OC concentrations. This method is described in detail by Turpin et al. [1990]. An in situ thermal-optical transmittance (TOT) carbon analyzer inspired by this instrument but with only one analytical unit is now manufactured by Sunset Laboratory, Inc. (Forest Grove, OR).

[12] EPA has selected thermal-optical transmittance (TOT) carbon analysis with the NIOSH temperature protocol (top temperature for OC removal in He is 870°C) for laboratory analysis of the national network speciation samples. Another common laboratory carbon analysis protocol is thermal-optical reflectance (TOR) with the IMPROVE temperature protocol (top temperature for OC removal in He is 550°C). The TOR/IMPROVE method has been reported to result in OC concentrations 10–15% lower and EC concentrations roughly a factor of two higher than the TOT/NIOSH method [Chow et al., 2001; G. A. Norris et al., Comparison of particulate organic and elemental carbon measurements made with the IMPROVE and NIOSH method 5040 Protocols, submitted to Aerosol Science and Technology, 2002] (hereinafter referred to as G. A. Norris et al., submitted manuscript, 2002). To what extent these differences are due to the method of pyrolysis correction (transmittance versus reflectance) versus the top temperature for OC volatilization is unclear. The RU/OGI in situ carbon analyzer corrects for pyrolysis by monitoring the transmittance through the filter (TOT) rather than the reflectance (TOR) because of the observation that EC generated during the analysis by pyrolysis of OC appears to deposit on the fibers throughout the filter giving the filter a gray appearance on the back as well as the front [Turpin et al., 1990]. For this reason, TOT is likely to report lower EC concentrations than TOR. The developers of the first (laboratory) thermal-optical carbon analyzers determined that heating the sample to 650°C in helium was sufficient to volatilize OC based on analysis of samples and standards [Huntzicker et al., 1982]. In the Atlanta Supersite Experiment and previous studies, the top temperature used for the RU/OGI in situ carbon analyzer was varied in the field prior to the beginning of the study. Temperatures higher than 700°C did not yield additional non light-absorbing carbon in Atlanta. Therefore this temperature was selected as the top temperature for OC volatilization in the in situ carbon analyzer during the Atlanta Supersite Experiment. OC and EC measurements would differ from those obtained with a top temperature of 550°C (i.e., IMPROVE protocol) if some OC is present that requires temperatures greater than 550°C for removal. Use of an even higher temperature for OC removal (i.e., 870°C) is unlikely to yield more OC but is likely to result in greater pyrolysis and therefore the accuracy of the pyrolysis correction becomes more important. The high-OC volatilization temperature in the NIOSH protocol (870°C) was adopted to facilitate the removal of carbonate carbon, which is important in mining samples.

2.2. R&P Ambient Carbon Particulate Monitor

[13] ARAI operated an R&P 5400 ambient carbon particulate monitor. The monitor pulls ambient air through a 2.5 μm cutpoint cyclone inlet at a flow rate of 16.7 L min−1 and collects particulate matter on a stainless steel impactor plate with a 0.14 μm cutpoint at a flow rate of 16.7 L min−1 [RWTÜV, 1997]. Organic carbon (or low temperature carbon) is measured by heating the collected PM2.5 sample to 275°C in air. This temperature was selected so R&P carbon monitor samples would match the OC/EC split of a batch of concurrently collected ambient samples analyzed using a laboratory thermal-optical method. The sample temperature is then increased to 750°C for subsequent EC (or high-temperature carbon) analysis. Pyrolysis does not occur because oxygen is present at all times during the analysis. However, agreement between thermal-optical methods and thermal methods like this one is likely to depend on the mix of sources/composition of the organic matter. Impactor collection has the advantage that the adsorption of organic vapors on the substrate is expected to be small. However, particle loss due to bounce and low collection efficiencies for small particles could be a problem. The dual collector/analyzer design enables simultaneous collection and analysis. One-hour average OC and EC concentrations measured every hour were reported. The monitor is described extensively by Rupprecht et al. [1995].

2.3. Magee Scientific Aethalometer

[14] HSPH used a Magee Scientific AE-16 Aethalometer to measure particulate EC (also called light-absorbing carbon, LAC, or black carbon, BC). The analytical method is described in detail by Hansen et al. [1984]. PM2.5 is collected on an automatically advancing quartz fiber filter tape at a flow rate of 4 L min−1. The sample filter advanced when the filter optical density exceeded 0.75. The optical absorption at 880 nm wavelength is measured every 5 min through the quartz filter. Conventionally a specific attenuation cross section (σaeth) of 16.6 m2 g−1, recommended by the manufacturer and based on the study of Gundel et al. [1984], has been applied to calculate BC (or LAC, EC) concentrations. Babich et al. [2000] suggested that 12.6 m2 g−1 is a good conversion factor for particulate EC in urban cities based on the correlation between the Aethalometer data and EC values analyzed with a Desert Research Institute (DRI) thermal-optical reflectance (TOR) carbon analyzer using IMPROVE temperature protocol for carbonaceous aerosols in 6 urban cities in the United States [Chow et al., 1993]. A factor of 12.6 m2 g−1 was used to convert absorption data of the Aethalometer to EC concentrations in this study. Hourly average EC concentrations were reported.

2.4. Radiance Research Particle Soot Absorption Photometer

[15] ARAI made continuous EC measurements with a Radiance Research PSAP. The PSAP is based on the integrating plate technique in which the change in optical transmission of a filter caused by particle deposition is related to the optical absorption coefficient using Beer's law and a calibration coefficient [Bond et al., 1999]. This is the same underlying measurement principle as that employed by the Aethalometer. Sample air was drawn through a cyclone with a 2.5 μm cutpoint, then through a glass fiber filter, at a flow rate of 1.26 L min−1. Sample and reference measurements of optical attenuation at 565 nm were then related to EC, assuming a specific absorption coefficient of 10 m2 g−1 [Bond et al., 1999]. The PSAP was equipped with a “dilution” filter (provided by the manufacturer) with a nominal filtered to unfiltered air ratio of 8.1:1. Subsequent testing of the filter revealed a dilution ratio of just under 4.0, which was used by ARAI to correct the final data set. Sample filters were changed out each day or when the filter transmittance dropped below 0.70, whichever came first. Continuous (i.e., 1-min) EC measurements were made and the corresponding hourly average EC concentrations were reported.

2.5. ADI Flash Vaporization Carbon Analyzer

[16] ADI operated a sulfate system to which a NDIR carbon dioxide analyzer was added to measure organic compounds that were directly oxidized to CO2. Particles are humidified, collected by impaction, and analyzed by flash vaporization. The cycle time is 10 min. Ambient particles were sampled at 3.7 L min−1. Particles larger than 2.5 μm were removed by a single-jet impactor, and organic vapors were removed using a 300-channel, 10 mm long activated carbon denuder (Novacarb, Mast Carbon Ltd, Surrey, UK). Downstream of the denuder, 2.7 L min−1 was directed to the sulfate-carbon system (remainder to nitrate system). The sulfate-carbon flow was humidified to 85 ± 5% RH, and humidified particles were collected by impaction on a 4 mm wide 0.08 mm thick nichrome strip or 0.05 mm thick platinum strip (after 25 August, 1600 EST) downstream of a single 0.61 mm (0.024 in) diameter nozzle operated at sonic conditions. The sample collection cell was housed in a ventilated enclosure to provide near–ambient temperature at the point of collection. After the 8-min collection period, the collected sample was analyzed by flash-heating (red hot) into in a 0.8 L min−1 flow of dry, CO2-free air, and evolved vapors were detected by a Li-Cor 6252 NDIR detector and a UV fluorescence SO2 analyzer operated in series. The temperature of the flash is not known.

[17] The system was calibrated weekly with aqueous oxalic acid and sulfate standards applied directly to the collection strip. Data were corrected for the average field blank (0.9 μgC m−3) measured by sampling filtered air. The system measured only directly oxidizable carbon. Organic carbon was estimated as twice the oxidizable carbon, based on comparison with filter data from one previous study conducted in the winter of 1999 in California [Dutcher et al., 2000].

2.6. Integrated Impactor Samples

[18] A Micro-Orifice Uniform Deposit Impactor (MOUDI) collected PM2.5 at 30 L min−1 on precleaned aluminum foil disks in seven stages with cutpoints of 1.78, 0.97, 0.56, 0.32, 0.18, 0.098, and 0.056 μm in aerodynamic diameter. A cyclone (Dp = 2.5 μm) upstream removed particles larger than 2.5 μm. A precleaned quartz fiber after filter placed after the final stage collected smaller particles and adsorbed some gases. The MOUDI was operated on a 12-hour schedule. OC and EC analyses of the MOUDI samples were made by TOR using the IMPROVE temperature protocol [Chow et al., 1993]. Pyrolysis correction cannot be made directly for samples collected on aluminum foil. Therefore the OC-EC split of collocated filter samples analyzed by TOR using the IMPROVE protocol was assumed to apply to concurrently collected MOUDI samples. These measurements are reported here for the purpose of comparison with the semicontinuous methods. The sum of carbon concentrations measured on all seven stages (carbon in 0.056–2.5 μm particles) with and without the after filter is compared below with semicontinuous concentrations averaged over the same time periods.

2.7. Integrated Filter Samples

[19] Nine filter-based samplers collected 24-hour average PM2.5 samples for carbon analysis, while two samplers collected 12-hour samples, including the MOUDI described above. A “relative reference” data set for OC, EC, and TC was calculated by averaging across the nine 24-h average samplers on a given day for each species [Solomon et al., 2003a, 2003b]. The “relative reference” data were used in this paper for comparison with the semicontinuous samplers averaged to the same 24-hour periods. The “relative reference” data and integrated sampler collection and analysis protocols are described in detail by Solomon et al. [2003b].

[20] All integrated filter-based samplers collected PM2.5 for carbon analysis on baked quartz-fiber filters at flow rates between 6.7 and 16.7 L min−1. Six of the samplers (Anderson RAAS, MetOne SASS, URG MASS, the FRM, R&P Speciation, and the MOUDI) did not use denuders to remove organic vapors that adsorb on quartz fiber filters. However, the R&P speciation sampler did include a backup quartz fiber filter to estimate and correct for this positive artifact. Three samplers used multichannel denuders with carbon-impregnated filters (the BYU and TVA PC-BOSS and the ARA PCM), and two used XAD-4 coated annular glass denuders (CIT PCM and URG-VAPS) to remove or reduce the adsorption artifact. The two PC-BOSS and the two PCM samplers (denuder-containing samplers) also used either carbon impregnated filter, quartz fiber, or XAD-impregnated quartz fiber backup filters to collect organic compounds that volatilized off of the particles collected on the front filter. The challenges posed by negative and positive organic artifacts are discussed by Turpin et al. [2000]. OC concentrations reported for the integrated samplers in the study database and used in this paper were not corrected for artifacts, nor were corrections made for denuder collection efficiencies.

[21] Integrated samplers were analyzed for OC, EC, and TC (TC = OC + EC) by one of three methods. Most samples were analyzed by TOT using the NIOSH temperature protocol (NIOSH Method 5040) [Birch and Cary, 1996; NIOSH, 1998]. The PCB-TVA and the ARAI PCM samples were analyzed by TOR using the IMPROVE temperature protocol, and the PCB-BYU samples were analyzed by temperature-programmed volatilization [Ellis and Novakov, 1982; Tang et al., 1994]. Recent comparison studies [Chow et al., 2001; G. A. Norris et al., submitted manuscript, 2002] suggest that for typical urban samples OC is usually about 10–15% lower and EC is typically about a factor of two higher by TOR with the IMPROVE protocol than by TOT with the NIOSH protocol. Total carbon agrees to better than 10%.

3. Results and Discussion

3.1. Intercomparison Between Semicontinuous Measurement Methods

[22] Table 2 summarizes participant-reported instrument detection limits for OC, EC, and TC during the Atlanta Supersite Experiment. The minimum detection limit for the semicontinuous instruments ranged from 0.1 to 2.0 μgC m−3. Figure 1 shows 10 min-average PM2.5 OC concentrations for the ADI flash vaporization carbon analyzer and hourly average OC, EC, and TC concentrations for the other semicontinuous samplers. Concentrations agree reasonably well with the exception of a few time periods. Table 3 provides descriptive statistics for TC, OC, and EC concentrations. Median measured OC concentrations were 6.5, 7.8, and 8.4 μgC m−3 for the ADI flash vaporization carbon analyzer, R&P ambient carbon particulate monitor, and RU/OGI in situ thermal-optical carbon analyzer, respectively. Median measured EC concentrations were 1.00, 2.5, 2.0, and 1.9 μgC m−3 for the PSAP, R&P carbon monitor, Aethalometer, and RU/OGI thermal-optical carbon analyzer, respectively.

Figure 1.

Carbonaceous species concentrations (μgC m−3) measured by ADI flash vaporization carbon analyzer (ADI), Radiance Research Particle soot absorption photometer (ARAI-PSAP), R&P 5400 Ambient carbon particulate monitor (ARAI-R&P), Magee Scientific AE-16 Aethalometer (HSPH), and RU/OGI in situ thermal-optical carbon analyzer (RU) in Atlanta, Georgia, 19–21 August 1999. Date labels are at 0:00 EST.

Table 2. Participant-Reported Detection Limits for Organic (OC), Elemental (EC), and Total Carbon (TC) During the Atlanta Supersite Experiment in 1999a
InstrumentDetection Limit, μgC m−3
OCECTC
  • a

    NA: not applicable; ADI: Aerosol Dynamics, Inc.; R&P: Rupprecht and Patashnick; RU: Rutgers University; OGI: Oregon Graduate Institute.

ADI flash vaporization carbon analyzer2.0NANA
Radiance Research PSAPNA0.1NA
Magee Scientific aethalometerNA0.1NA
R&P 5400 ambient carbon particulate monitor0.50.50.5
RU/OGI in situ carbon analyzer0.30.50.4
Table 3. Carbonaceous Species Concentrations (μgC m−3) for Semicontinuous Instruments During the Atlanta Supersite Experiment in 1999a
SpeciesParameterADIPSAPAethalometerR&PRU/OGI
  • a

    Presented are number of measurements (N), mean, median and one standard deviation of the mean for particulate organic (OC), elemental (EC) and total carbon (TC) for the Aerosol Dynamics, Inc. flash vaporization carbon analyzer (ADI), Radiance Research Particle soot absorption photometer (PSAP), Magee Scientific Aethalometer, Rupprecht and Patashnick 5400 Ambient carbon particulate monitor (R&P), and Rutgers University/Oregon Graduate Institute in situ carbon analyzer (RU/OGI). SD: standard deviation of the mean; NA: not applicable.

OC
 N506NANA542269
 mean6.75NANA7.938.61
 median6.49NANA7.828.38
 SD of mean2.95NANA2.512.45
EC
 NNA506696542269
 meanNA1.262.612.802.33
 medianNA0.992.022.471.93
 SD of meanNA0.891.851.521.52
TC
 NNANANA542269
 meanNANANA10.7210.95
 medianNANANA10.3010.25
 SD of meanNANANA3.893.59

[23] Intermethod comparisons are shown in Figures 24 along with the ratio of mean measurements, Deming linear least squares regression equation, 1:1 line, and coefficient of determination (R2). The Deming regression performs a least-squares fit taking into consideration the uncertainties in both x and y variables, whereas the more common linear least squares fit assumes there are no uncertainties in x [Deming, 1943; York, 1966; Cornbleet and Gochman, 1979]. Since uncertainties in x and y are roughly comparable for data intercomparisons, the Deming regression equations were calculated by attributing equal uncertainties to both variables.

Figure 2.

Total carbon (TC) concentrations (μgC m−3) measured with RU/OGI in situ thermal-optical carbon analyzer (RU) and R&P 5400 Ambient carbon particulate monitor (ARAI-R&P) in Atlanta, Georgia, 3 August to 1 September 1999. The solid line represents the Deming least squares linear regression and the dashed line represents the identity line (1:1). Shown also are the number of samples (N), coefficient of determination (R2), and the ratio of mean concentrations.

Figure 3.

Organic carbon (OC) concentrations (μgC m−3) measured in Atlanta, Georgia, 3 August to 1 September 1999. RU: RU/OGI in situ thermal-optical carbon analyzer; ADI: ADI flash vaporization carbon analyzer; ARAI-R&P: R&P 5400 Ambient carbon particulate monitor. The solid line represents the Deming least squares linear regression and the dashed line represents the identity line (1:1). Shown also are the number of samples (N), coefficient of determination (R2), and the ratio of mean concentrations.

Figure 4.

Elemental carbon (EC) concentrations (μgC m−3) measured in Atlanta, Georgia, 3 August to 1 September 1999. ARAI-R&P: R&P 5400 Ambient carbon particulate monitor; HSHP: Magee Aethalometer; RU: RU/OGI in situ thermal-optical carbon analyzer; ARAI-PSAP: Radiance Research PSAP. The solid line represents the Deming least squares linear regression and the dashed line represents the identity line (1:1). Shown also are the number of samples (N), coefficient of determination (R2), and the ratio of mean concentrations.

[24] Only the R&P thermal carbon monitor and RU/OGI in situ thermal-optical carbon analyzer measured particulate OC, EC, and TC. Total carbon measurements agreed well as shown in Figure 2. They were well correlated (R2 = 83%); the ratio of mean RU to mean R&P total carbon was 1.02, and the ratio of the medians was 1.00. The OC-EC split was somewhat different, as seen in Figures 3 and 4. OC measured by the R&P carbon monitor was 8% lower than OC measured with the RU/OGI in situ thermal-optical carbon analyzer, on average, and R&P EC was 20% higher than RU EC, on average. The differences in OC-EC split are undoubtedly a result of the differences in “definitions” for OC and EC employed by the two instruments (i.e., temperature-defined split for R&P instrument, and evolution with/without oxygen for RU instrument) and/or differences in the collection method (i.e., impactor with a cutoff diameter of 0.14 μm for R&P instrument, and filter collection with correction for adsorption for RU instrument), as discussed above. Measurements from the two instruments were moderately well correlated, with coefficients of determination (R2) of 73% and 74% for OC and EC, respectively. Differences were reflected in the intercepts, rather than in the slopes, which were close to one.

[25] A third instrument, the ADI flash vaporization carbon analyzer, measured directly oxidizable organic carbon (Figure 3), which was multiplied by two, as explained above to estimate total organic carbon for comparison with the other methods. The inferred OC measured by the ADI analyzer was 28% and 17% lower, on average, than OC concentrations measured by the RU and R&P instruments, respectively. In addition, the slopes describing the relationship between ADI OC and the other instruments' OC measurements were significantly different from one. ADI has since decoupled the carbon measurement from the sulfate, which has allowed them to add a heated MnO2 bed to convert all vaporized carbonaceous compounds to CO2 prior to analysis with the NDIR detector. This improvement is expected to make the instrument calibration insensitive to the composition of the organic PM, and therefore improve comparability with other methods.

[26] The RU in situ carbon analyzer collected particles on quartz fiber filters and subtracted the adsorption estimate provided by a quartz fiber filter placed behind a Teflon filter collected concurrently. This approach is grounded in comprehensive artifact studies [Turpin et al., 1994]. However, the correction is large; about 30% of collected OC was adsorbed vapor, and this correction contributes to measurement uncertainty. The R&P thermal monitor and ADI flash volatilization carbon analyzer collected particles on an impactor, drastically reducing adsorption. To improve particle collection efficiencies (i.e., reduce bounce), particles in the ADI system are grown by water vapor condensation prior to collection [Stolzenburg and Hering, 2000]. The impactor collection efficiency for ammonium sulfate is 90% at 0.1 μm dry Stokes diameter and 99% above 0.3 μm dry Stokes diameter. Previous studies reported 89 to 99% collection efficiencies for nitrate in 0.2–1 μm ambient particles with the ADI impactor. Because carbonaceous aerosol is less hygroscopic than ammonium sulfate and ammonium nitrate and fresh combustion could yield EC size distributions with smaller mass median diameters, collection efficiencies might be smaller for carbon than for nitrate. The R&P thermal carbon analyzer employs no device to improve collection efficiencies for small particles and no correction factor to account for losses of carbon as a result of bounce and low collection efficiencies for small particles.

[27] Figure 4 compares EC measurements with the four EC-reporting instruments. The best correlation is found between the two optical absorption instruments, the PSAP and the Aethalometer (R2 = 97%). However, these two optical instruments show a poor agreement (mean PSAP/mean Aethalometer = 0.48; slope = 0.54). Differences between these two methods are undoubtedly associated with the choices of calibration factors applied to calculate EC concentrations from light absorption data. The specific attenuation cross section (σaeth) for the Aethalometer was obtained by comparing its light absorption data with EC determined by thermal-optical reflectance (e.g., TOR using IMPROVE protocol: top temperature in He is 550°C) at specific locations. In this way, σaeth accounts for the matrix effect of the filter. Liousse et al. [1993] reported large variations in σaeth in different environments, which they attributed to differences in the microscopic mixing properties of the particles (matrix effect due to particle properties). Good correlations were found between BC (EC) concentrations determined by the Aethalometer and the DRI TOR carbon analyzer in suburban and urban locations [Allen et al., 1999; Babich et al., 2000]. The Allen et al. results were consistent with a specific attenuation cross section of 16.6 m2 g−1; a value of 12.6 m2 g−1 was calculated from the Babich et al. results and used in this study. Agreement with the in situ thermal-optical carbon analyzer in this study (thermal-optical transmittance; top temperature in He is 700°C) would be obtained by using an attenuation cross section of 15.5 m2 g−1.

[28] An attenuation cross section (σ) of 10 m2 g−1 is a widely accepted value for aerosol EC in situ, although this value can also be expected to vary with variations in microscopic mixing characteristics of particles. This value was applied to obtain EC concentrations from absorption measurements through the glass fiber-sampling filter in the PSAP. The effect of the glass-fiber matrix on the attenuation cross section might alter the effective attenuation cross section, as has been demonstrated for attenuation through the quartz fiber filter of the Aethalometer. This could contribute to the lower PSAP EC concentrations. Agreement between the PSAP and in situ thermal-optical carbon analyzer in this study would be obtained by using an attenuation cross section of 6.0 m2 g−1 for the PSAP (glass fiber filter). Interestingly, Dobbins et al. [1994] obtained the same value for the measurement of fresh smoke.

[29] Calibration factors for optical measurements depend on the properties of the aerosol being sampled and the reference methods used in the calibration. Thermal-optical carbon analysis is often used to calibrate optical methods. It should be noted that the method of pyrolysis correction (reflectance or transmittance) and temperature protocol used in the analysis impact the EC values reported by thermal-optical analysis methods [Chow et al., 2001; G. A. Norris et al., submitted manuscript, 2002]. The final temperature for OC removal must be high enough to remove all OC; however, higher temperatures result in more pyrolysis making the accuracy of the pyrolysis correction that much more important. The choice of calibration factors is the main challenge for accurate assessment of EC by optical methods. If such measurements are being made to assess radiative forcing, conversion to mass concentrations might introduce unnecessary uncertainties. Slopes substantially different from one are found when the thermal and thermal-optical instruments are compared with the optical absorption instruments. Nonzero intercepts are found in comparisons of the R&P thermal carbon monitor and the optical absorption instruments. These observations might indicate that, despite the presence of oxygen throughout the analysis, some OC was incorrectly attributed to EC due to pyrolysis during the R&P analysis (nonzero intercept). Pyrolysis of OC in an oxygen-containing atmosphere is much less likely than in an oxygen-free atmosphere. However, some pyrolysis has been observed for a variety of pure organic compounds and natural organic mixtures heated in an oxygen-containing atmosphere (e.g., 3% pyrogallol; 37% humic acid) [Cachier et al., 1989]. It is also possible that the EC absorptivity (absorbance/mass), used to convert optical absorbance to EC mass, could be nonlinear over the range of absorbances measured or could inaccurately describe Atlanta EC. Also, the RU correction for pyrolysis could overestimate the pyrolytic conversion of OC to EC if the pyrolytically generated EC was not removed first and the pyrolytically generated EC had a different absorptivity than the original EC. The RU/OGI in situ carbon analyzer corrects for pyrolysis by monitoring transmittance through the filter rather than reflectance off the collection surface of the filter because of the observation that pyrolysis of OC to EC generates a coating of dark material on fibers throughout the filter, suggesting that the pyrolytically generated EC is present throughout the filter and that its presence can be more accurately indicated by measuring the transmittance through the filter. This also suggests that it is reasonable to assume that pyrolytically generated EC is removed more easily than original particulate EC [Turpin et al., 1990].

[30] The precision with which carbonaceous aerosol is measured can be depicted by the intersampler precision, expressed as a coefficient of variation (percentage). The pooled CV was calculated from the pooled standard deviations (SD) of concentrations measured concurrently across samplers and the grand mean concentration (X):

equation image
equation image
equation image

where σi = standard deviation of concentrations measured across samplers collected during period i, xi = concentration during period i, ni = number of data points used to calculate σi, and p = total number of periods (number of data points used to calculate X). Because the overall measurement uncertainty increases as measurements approach detection limits, intersampler variability can be expected to increase with decreasing carbonaceous aerosol concentration. Figure 5 depicts the intersampler coefficient of variation (CV) as a function of the carbonaceous aerosol concentration for semicontinuous OC, EC and TC. For Figure 5, all measurements within a single defined concentration range were included in a pooled CV calculation. The intersampler precision reaches a relatively constant value for TC, OC, and EC concentrations greater than 8, 5, and 2.5 μgC m−3, respectively. Intersampler variability increases substantially at lower concentrations, suggesting that reliable measurement at lower concentrations would require longer collection times than utilized in this study. About 14% of OC and 74% of EC measurements made with semicontinuous measurements at the Atlanta Supersite Experiment are below 5 μgC m−3 and 2.5 μgC m−3, respectively.

Figure 5.

Intersampler precision for semicontinuous monitors, expressed as a pooled coefficient of variation (CV), by carbon concentration in Atlanta, Georgia, 3 August to 1 September 1999. N denotes the number of samples in a concentration range. OC: organic carbon; EC: elemental carbon.

3.2. Comparison With Time-Integrated and MOUDI Measurements

[31] Semicontinuous carbon concentrations, averaged over all semicontinuous methods and over the 24-hour periods corresponding to the collection times of the collocated integrated samples, are shown with the relative reference integrated filter measurements and MOUDI measurements in Figure 6. Correlations and ratios of means comparing each semicontinuous measurement and the relative reference values are shown in Table 4. MOUDI concentrations both with after filter EC and with and without after filter OC are shown. Fine-particle MOUDI OC and EC concentrations calculated without the after filter could be slightly underestimated because of under-collection of very small particles (i.e., Dp < 0.056 μm), whereas including the after filter is expected to substantially overestimate OC by including gases adsorbed on the after filter [Zhang and McMurry, 1987]. Comparisons of semicontinuous instruments with MOUDI TC are more meaningful than comparisons with MOUDI OC and EC because MOUDI OC and EC concentrations were estimated from MOUDI TC concentrations by assuming all impactor substrates have the same OC-EC split as concurrently collected filter samples.

Figure 6.

Organic (OC), elemental (EC), and total carbon (TC) concentrations from semicontinuous and time-integrated measurements in Atlanta, Georgia, 3 August to 1 September 1999. Shown are averages from semicontinuous, integrated filter samplers, and integrated MOUDI samples with and without the after filter (AF). Error bars in solid and dotted lines describe variability (1σ) between semicontinuous and time-integrated measurements, respectively.

Table 4. Correlations Between Semicontinuous Measurements and Relative Reference Time-Integrated Values During the Atlanta Supersite Experiment in 1999a
SpeciesParameterADIPSAPAethalometerR&PRU/OGI
  • a

    Presented are the coefficient of determination (R2), number of data (N), and corresponding ratio of mean of semi-continuous measurement (Con) to integrated measurement (Int) for particulate organic (OC), elemental (EC) and total carbon (TC). Aerosol Dynamics, Inc. flash vaporization carbon analyzer (ADI), Radiance Research Particle soot absorption photometer (PSAP), Magee Scientific Aethalometer, Rupprecht and Patashnick 5400 Ambient carbon particulate monitor (R&P), and the Rutgers University/Oregon Graduate Institute in situ carbon analyzer (RU/OGI). NA: not applicable.

OCR2 (N)0.48 (12)NANA0.78 (12)0.52 (13)
 Con/Int0.86NANA0.951.09
ECR2 (N)NA0.13 (12)0.51 (15)0.23 (12)0.52 (13)
 Con/IntNA1.172.442.532.19
TCR2 (N)NANANA0.71 (12)0.59 (13)
 Con/IntNANANA1.171.27

[32] Including only days with no missing data from any of the samplers, mean 24-hour averaged TC concentrations were 10.9, 9.0, 4.0, and 7.2 μgC m−3 (9 days) and mean 24-hour averaged OC concentrations were 7.8, 7.7, 3.8, and 6.8 μgC m−3 (13 days) for semicontinuous, integrated, MOUDI without the after filter, and MOUDI with the after filter, respectively. Mean 24-hour average EC concentrations for days without any missing data were 2.3, 1.0, and 1.4 μgC m−3 for semicontinuous, integrated, and MOUDI EC with after filter, respectively (n = 13 days). The intermeasurement variability of the integrated relative reference filter measurements was substantially higher than that for semicontinuous measurements. Semicontinuous and integrated Relative Reference TC measurements had pooled coefficients of variation of 7% (2 instruments) and 25% (9 instruments), respectively. OC measurements had pooled coefficients of variation of 13% (semicontinuous, 3 instruments) and 28% (relative reference, 9 instruments). EC measurements had pooled coefficients of variation of 26% (semicontinuous, 4 instruments) and 66% (relative reference, 9 instruments).

[33] The variability in particulate TC measured by the relative reference samplers was considerably greater (25%) than the variability typically reported due to differences in analytical methods alone (10–15%). However, comparison with the coefficients of variation from the CSMCS laboratory intercomparison suggests that differences in analytical methods, particularly differences affecting the OC-EC split, could explain most of the variability in reported particulate OC and EC concentrations in this study and the variability in reported semicontinuous particulate TC. Additional uncertainties in relative reference sampler TC concentrations probably result mostly from differences in organic aerosol sampling artifacts. The relative reference samplers handled OC sampling artifacts in a broader variety of ways than the semicontinuous samplers did. Each semicontinuous sampler measuring OC either minimized the positive adsorption artifact by collecting the particulate matter on an impactor substrate that adsorbs little or no organic vapors or corrected for the adsorption of organic vapors on the quartz fiber sampling filter by subtracting a concurrently collected dynamic blank (quartz fiber filter sampling particle-free ambient air). The relative reference samplers included both samplers where a single quartz filter (undenuded) was used to measure particulate carbon without correction for the adsorption of organic vapors on the filter (positive artifact) and samplers where the filter was preceded by a denuder that removed or reduced adsorption and potentially could induce volatilization of collected particulate organics by exposing them to a reduced vapor phase (negative artifact). Solomon et al. [2003b] reported that the former type of sampler measured OC concentrations 5–35% higher and the later measured OC concentrations 20–35% lower than the relative reference sampler average.

[34] In addition to the treatment of sampling artifacts, greater coefficients of variation for the relative reference samplers than for the semicontinuous samplers could also be explained by the facts that (1) the relative reference sampler coefficient of variation calculation includes more samplers (n is greater) and (2) the short sampling times of the semicontinuous samplers minimize variations in conditions over which sampling occurs. Integrated samplers, in contrast to time-resolved samplers, expose collected particles and collection substrates to wide changes in conditions (e.g., vapor concentrations, temperature, and relative humidity) over the course of the day. Such changes are likely to induce adsorption and volatilization (i.e., with heating and cooling or changes in vapor concentrations) to bring the collected materials into equilibrium with the surrounding vapor. Variability due to analytical methods is less likely to explain the greater variability in relative reference samples since TOT, temperature-defined methods, and either TOR or methods calibrated to TOR were used for analysis of both relative reference and semicontinuous samples.

[35] Differences between impactor and filter samples have been observed previously. During CSMCS, Hering et al. [1990] observed that impactor-based systems obtained 29–41% lower OC concentrations than adsorption-corrected filter systems. The Atlanta Supersite Experiment shows similar differences between carbon measured by the MOUDI and filter-based samplers. Filter-impactor differences could result from an undercorrection for adsorption on the quartz fiber filters, and/or losses due to bounce and low collection efficiencies for small particles (< 0.1 μm) in impactor systems [Stein et al., 1994]. If volatile losses were greater in impactors than for filter collection, such a result would also be found; however, a good thermodynamic argument has been made suggesting that the driving force for volatilization is smaller for impactor-collected particles than for filter-collected particles [Zhang and McMurry, 1987].

4. Conclusion

[36] Measurement of carbonaceous particulate matter is subject to larger uncertainties than other major aerosol species. This is because carbonaceous aerosol is comprised of hundreds of compounds including semivolatile organic compounds that partition between the gas and particle phases [Huebert and Charlson, 2000; Andrews et al., 2000; Turpin et al., 2000; Turpin and Lim, 2001]. Automated, real-time and semicontinuous instruments minimize changes in atmospheric conditions during sampling that can lead to both positive and negative artifacts. They also provide the time resolution needed to study atmospheric processes. This paper shows good intersampler agreement between semicontinuous samplers (7, 13, and 26% for TC, OC, and EC, respectively). Distinctive analysis protocols are an important contributor to intermethod differences in OC and EC, and the method of carbon analysis must be considered if data from diverse sources are merged for scientific study. This study exposes the need for standards to reduce uncertainties in the current OC-EC measurements, and demonstrates the successful operation of automated semicontinuous carbon analyzers.

Acknowledgments

[37] These authors would like to enthusiastically thank the leadership and many sponsors of the Atlanta Supersite Experiment, including William Chameides, the U.S. Environmental Protection Agency, and the Electric Power Research Institute. This publication was funded in part by the Electric Power Research Institute. The United States Environmental Protection Agency through its Office of Research and Development partially funded and collaborated in the research described here under assistance agreement CR824849 to Georgia Institute of Technology. It has been subject to Agency review and approved for publication.

Ancillary