Readme File for Huebert Group C-130 OC/EC Version 5.0 Re-submitted 7 May 2004 May 2004: There are several major changes to our data in this version, which are explained in more detail below. 1. The OC/EC split time is now determined by referencing the laser transmittance to a blank laser transmittance, which eliminates the temperature-dependence and the uncertainties in Sunset's software corrections for it. 2. For most of the FT samples and a few BL samples, only TC is now reported. The efforts we made to apportion TC between EC and OC in Version 4 produced unreasonable mass absorption efficiencies, which seemed to be the result of apportioning TC with inadequate basis. We do include columns of inferred OC and EC based on the TC/EC ratio of other samples, but in Version 5 there are only a few direct FT measurements of EC and OC. 3. We have reworked our backup CIG filter data using a scheme that eliminates low-temp and high-temp peaks which we believe to be artifacts. The result is that now 2/3 of our samples have valid CIG analyses, and half of those are above the MDL so that they have significant values for evaporated (potential negative artifact) carbon. The small loading and resulting small S/N from aircraft samples has made it hard to detect the loss of a few tens of percent of the ambient OC. 4. The same PC enhancement factor of 2.0 was used for both OC and EC, making the TC data more sensible. Previously we had used the slightly higher NSS-derived enhancement factor of 2.2 for OC and 1.8 for EC. 5. We no longer post data for carbonate carbon. After further studying the thermograms, we do not feel this is a reliable way to quantify carbonate. March 2003: The attached data set has been revised as a result of several QA/QC processes. Some of the values have changed considerably, and not always as an across the board percentage. Among other things, the uncertainties are now all propagated and include the Mader uncertainty (from multiple punches from one filter). We have also improved the correction for our particle concentrator. Please see description of revisions and their rationale in Sections 3.3, 3.4, and 5.2. Paragraphs starting "March 2003" describe the changes to this version of our data. Frankly, I (BJH) continue to be concerned about the correctness of the OC/EC split in our thermograms. My concern arises from changes in the light transmission that happen even when no carbon is coming off: they appear to be related to changes in the temperature of the oven, its windows, and the filter punch. Several factors (other than charring and EC loss) clearly modify the light transmission during analysis. Among these may be the volatilization of coating material on the EC, which we believe increases the EC specific absorption substantially. On the other hand, in lightly-loaded samples there is not much charring but the transmission still moves around quite a bit during an analysis. Since most samples have a lot of carbon coming off during the steps where O2 is present (and since often less than half is apportioned to EC), small changes in the transmission due to other factors could cause a large change in the derived amount of EC. These issues would not affect the total carbon determination, but they could have a significant impact on the reported EC and OC concentrations. We continue to work on this issue in the lab and are collecting more EC-dust samples for lab tests. Oct 2002: The attached data set has been revised in accordance with the ideas/suggestions presented and discussed at the ACE-Asia Data Workshop in Boulder, CO, in May, 2002. There are major changes in processing of thermograms and in PC-BOSS sampling efficiencies (which have now been measured in the lab). Title: Aircraft C-130 Aerosol Mass of organic and elemental carbon < 1-2 um (Huebert) Authors: PI: Barry J. Huebert Co-PIs: Byron Blomquist Delbert Eatough Steve Howell Technician who ran samples and processed data: Tim Bertram Dept. of Oceanography 1000 Pope Rd., MSB 407 (only needed for courier deliveries) University of Hawaii Honolulu, HI 96822 USA 1-808-956-6896 phone 1-808-956-9225 fax 1.0 Data Set Overview This data was collected from the NCAR/NSF C-130 during the ACE-Asia program. It consists of samples collected using a PC-BOSS sampler (with a preconcentrator and an organic vapor denuder) and analyzed using a Sunset Labs Model 3 thermal/optical carbon analyzer. We report total organic carbon (OCsum = the sum of OCq and OCvol), organic carbon that collected on a quartz filter (OCq), organic carbon that volatilized from the quartz and was captured by a backup carbon-impregnated glass filter (OCvol), elemental carbon (EC), and carbonate carbon (CC). Each of these is accompanied by an uncertainty. The uncertainties in are from a rigorous propagation of errors analysis. Included in the data set are the start and stop times of each sample, as well as the location and altitude of the C-130 at the start and end of each sampling interval. These locations and altitudes are GPS values from the NCAR-RAF data system. More detail on the location of each leg and the conditions of each flight can be found in the C-130 Flight Reports and C-130 data in the ACE-Asia catalog (http://www.joss.ucar.edu/ace-asia/dm/). A paper describing the data set has been published in JGR: Huebert, B.J., T. Bertram, S.G. Howell, D. Eatough, J. Kline, and B. Blomquist, Measurements of Organic and Elemental Carbon in Asian Outflow During ACE-Asia from the NSF/NCAR C-130, J. Geophys. Res., In press (ACE-Asia Special Section B), doi: 2004jd004700, 2004. The first reported samples are from the first research flight (RF01), flown on 31 March 2001 Japan time. (The takeoff time was just before midnight on 30 March UTC. All times and dates in the data file are UTC.) All research flights have a sample name starting with RF and were flown from Iwakuni MCAS, Japan. The last sample is from a return ferry flight (FF06), on 8 May 2001, between Wake Island and Hawaii. Not all parameters are reported for all flights, largely due to logistical considerations (e.g., we could not bake CIG filters prior to ferry fights and analyze them immediately after). 2.0 Instrument Description Air was conveyed into the aircraft by a gradually-curved solid diffuser inlet. It had a double elliptical inlet tip that was flight tested on the pre-PELTI test flights and found to reduce flow separation. The tip inside diameter was 5.38 mm, the I.D. of the curved tube that conveyed the flow into the fuselage was 31.8 mm, its radius of curvature was 61.0 cm, it entered the fuselage at a 65 degree angle (not quite perpendicular but still flowing toward the tail), and the inlet tip was 30.5 cm from the aircraft skin. The cutoff size of this inlet has not been established, but based on the PELTI experiment we expect that it passes particles smaller than a few um with high efficiency. For mineral particles that do not stick to inlet surfaces, it may have a high efficiency for particles many um in diameter. The samples were collected using a PC-BOSS sampler (J. Lewtas, et al., Aerosol Sci. Technol., 34, 9-22, 2001) developed by Delbert Eatough of BYU. Once inside the fuselage, the air was conveyed by a 2.54 cm id tube to a splitter: a small flow (12 lpm - all rates are in actual liters per minute) went to the side filter, while the major flow (135 lpm) passed into a particle preconcentrator (PC, bypass flow of 100 lpm and sample flow of 35 lpm). The PC (Y. Pang, et al, J. Amer. Waste Man. Assoc., 51, 25-36, 2001) is a virtual impactor with a cut size around 0.1 um that was designed to increase the concentration of particles by a factor of about 4 (but our was far from ideal, see Section 5.2). The enriched flow from the preconcentrator passed then into a diffusion denuder (D.J. Eatough, et al., Atmos, Environ., 33, 2835-2844, 1999) containing 15 parallel 4.5 x 58 cm strips of carbon-impregnated glass fiber (CIG) filters separated by 2 mm. This denuder removed most organic vapors, some of which could have caused a positive artifact by adsorbing on our quartz sample filters (after which they would have been indistinguishable from actual organic aerosol). The organic and elemental carbon aerosol samples were collected on baked quartz filters (Pall Gelman 2500QAO, 47 mm, baked 16 hrs at 550C), which was prepared and handled in the same manner as most other ACE-Asia OC/EC filters. (We used a method developed at Cal Tech, as described by Lynn Salmon on October 6, 1994. Personal communication, Rick Flagan) Any organic carbon that evaporated from the quartz filter during sampling (potentially a negative artifact) was captured by a CIG filter (S&S GF3649, 47 mm, baked 10 hours at 325C under nitrogen) immediately behind the quartz. One unique feature of the PC-BOSS sampler is a side filter that pulls off a small flow of sample air at the splitter upstream of the preconcentrator for QA/QC purposes. We exposed a Teflon filter in this side position during one OC/EC sample per flight, to compare with the sulfate from the quartz filter. (Since all SO2 should be denuded by the CIG denuder, the sulfate concentration derived from the quartz sample filter should equal that derived from the Teflon side filter.) During the entire rest of the flight we exposed a single quartz filter in this side sampler (the "Sideall" quartz filter) which thereby collected enough EC to achieve an adequate S/N for comparison with the sum of the EC found on all the remaining quartz sample filters. This allowed us to determine the actual enhancement by the virtual impactor and (as it turns out) to identify the loss of about half the EC and NSS mass in the PC or denuder. This loss was consistent for both sulfate and EC and from flight to flight, so we have corrected the OC and EC for this factor. Thermal mass flowmeters were used to continuously record the flows through the side filter, sample filter, and preconcentrator bypass. These also facilitated the manual adjustment of all flows to design conditions during the first minute or so of each sample. The analysis of samples was done using a Sunset Labs thermal/optical analyzer. It heated the sample to specified temperatures for specified times, converting evolved carbon to CO2 and then CH4 for analysis by FID. The thermal program was the same as that used by other groups in ACE-Asia. It involved heating to four temperatures in a He environment to drive off OC, with the final step at 870 C. After cooling the sample down to 550 C, a He/O2 mixture was introduced and the sample was heated further to drive off elemental carbon (EC). The instrument measures the transmission of laser light through the filter to enable the separation of EC from OC that charred during the first stages of heating. CIG filters were also analyzed with the Sunset Labs instrument: they were heated at 20 C/min to a maximum front oven temperature of 300 C in a He environment. At the highest He-only temperature, CO2 is also driven off from some carbonates (which are common in dust). Therefore, when there was a significant amount of apparent OC at that temperature we treated another aliquot of the sample overnight with HCl (to drive off the carbonate as CO2) and re-analyzed it to determine how much of this carbon was organic vs carbonate. (This procedure became operational as of RF07.) A punch that was supplied with the analyzer was used to take a 1.0 x 1.45 cm section of each 47mm filter for insertion into the analyzer. For most of our flights we masked off a 20 mm x 28.5 mm center portion of the 47 mm filter area, so the particles were concentrated on a smaller surface. This increased the amount of analyte per area of filter, and thus improved our signal to noise (S/N) ratio. We assume that the analyte was evenly distributed across the filter's exposed surface. 3.0 Data Collection and Processing 3.1 Description of data collection Quartz and CIG filters were baked the night before each flight and loaded into two-stage Teflon filter packs just prior to the flight. This loading was done in a glove box to minimize contamination. The process of collecting samples involved the PC-BOSS sampler described above. On each C-130 flight, up to 8 sets of pre-loaded filter packs would be brought along in a cooler, to minimize loss of semi-volatile particles post-collection. During the first sample, a Teflon side filter was run for sulfate comparison. For the remainder of the flight a single quartz side filter was exposed. When ready for sampling, the operator (either Tim Bertram or Jackie Heath) would load a filter pair into the sample location. When a sampling leg began, all the valves would quickly be opened to start flow through the sampler. Over a period of about a minute, the operator would adjust valves to achieve the design flows through the side filter and the particle concentrator. Flow would continue to the end of that sampling leg, which was usually between 20 and 70 minutes, at which time the valves were closed as nearly simultaneously as possible. This filter pack would then be replaced by one for the next sample. The filter packs were always changed with gloved hands and stored in an on-board cooler. The filters from each flight were unloaded from their holders immediately after each flight. Filters were stored in prepared petri dishes consisting of annealed aluminum foil, press fitted into the top and bottom of the plastic petri dishes. All petri-dishes were secured with Teflon tape and stored in a freezer until analysis. After RF04 the analysis was virtually always done the next day. The data from the first 4 flights is a bit more uncertain because we were still setting up the Sunset Labs analyzer and tuning it for optimal performance. Those samples waited several days before being analyzed. 3.2 Description of derived parameters and processing techniques used Mass flow for the sample and side filters was recorded from two Tylan mass flow controllers on both the RAF data system and our own personal computer. From these recorded values, sample volume was calculated. Each controller was calibrated prior to, during, and after the program. The high volume bypass flow was recorded with a Kurz mass flowmeter, calibrated in the same fashion as the Tylan flow controllers. Uncertainties in the flows were estimated at 5%. To account for the concentrating of particles in the virtual impactor, sample volume (calculated) was defined as the sum of bypass and sample (measured) volume. Losses in the concentrator were addressed with the side filter/sample filter intercomparison and later in laboratory calibrations described in Section 5.2. As described in detail in section 3.3, a correction factor of 2.2 was used to account for loss in the particle concentrator and denuder. (March 2003: See Section 3.3 for a change in this factor.) Relative Humidity, temperature and absolute pressure were recorded at numerous points along the sampling system. Each parameter was stored on our personal computer. 3.3 Description of quality control procedures During each flight one field blank was exposed. This was a filter pack just like the others, which was mounted in the PC-BOSS and exposed for just 10 seconds. Thus, its handling and history were identical to the actual samples. The thermogram (evolved C vs time and temp) for each flight's quartz (or CIG) blank was subtracted directly from the thermogram of each sample quartz (or CIG) filter on that flight. Uncertainties were derived by propagating the errors from the flowmeters (5%), instrument detection limit, the project blank variability, the analytical uncertainty, and the uncertainty in the PC-BOSS correction factor (from the difference between the corrections derived using EC and SO4). Oct 2002: The side filter was an important part of the QC effort, since it allowed us to determine the actual enhancement factor resulting from the preconcentrator, the denuder, and the filter masks. This correction factor was derived from sulfate and EC intercomparison of the side and sample filters. IC results of the sulfate intercomparison yielded a consistent enhancement of 2.2 on the side filter, while thermal techniques for EC depicted an enhancement of 1.8. The correction factor we used was 2.2, since the S/N ratio of the sulfate measurements was considerably better than that for EC. May 2004: Both EC and OC were computed using 2.0 as the correction (PC enhancement) factor. Using different values for the two lead to some logical inconsistencies. In fact, the flight-to flight variation in this factor made the 1.8 (from EC) and 2.2 (from NSS) indistinguishable anyway. One major source of uncertainty in the TOT analysis is the assignment of the time at which the laser transmission through the filter has returned to the pre-charring value, so that any carbon released after that time can be assumed to come from EC. We conducted laboratory tests on quartz filter samples collected at Amami Oshima in April of 2003 to assess this uncertainty. We looked carefully at the laser transmission signal and its relationship to temperature of the oven, which introduced a (4% uncertainty in the transmission value. When we applied this transmission uncertainty to a selection of ACE-Asia thermograms, we found that the EC portion of TC could be changed by up to 15%. We also extracted some of the 2003 Amami Oshima samples with water to minimize charring and found that the char (the difference between the extracted and unextracted samples) burned off over the same time period as did native EC. Since the TOT correction scheme implicitly assumes the char burns off before native EC does, the TOT results are sensitive to the implicit assumption that the specific absorption of native EC and char are similar (Yang and Yu, 2002 ). We do not know how to put limits on this assumption, but it is a potentially significant source of error. We also referenced our raw laser transmittance values to the closest blank laser transmission. The difference between these eliminates the temperature dependence of the laser and the over, givng a much more stable signal for choosing the split time. Since our un-masked area was only large enough to take two punches from each sample, we were unable to do multiple analyses on each sample. Mader et al. (2002) demonstrated that the precision of the TOT analyses was poorer than our blank analysis had suggested. They found that for multiple punches from the same filter, relative standard deviations (RSD) were greater than 0.5 for samples with filter loadings less than 0.2 _g C cm-2. We therefore use as the uncertainty of each individual blank or sample analysis the larger of either: a) the program-wide standard deviation of our blank values (which is very conservative, since it includes all the flight-to-flight variability in how each batch of filters was handled, how long exposed samples had to wait for analysis, etc. that would not affect any one flight's sample-minus-blank computation) or b) the apparent analyte times the RSD derived from Mader et al.'s Figure 3. Two distinct sets of uncertainties were considered. The first group includes only those factors that affect the detectability of EC or OC above the blank value. This includes the uncertainty in the _g C cm-2 of analyte on the sample filter and the blank filter, as well as the limit of detection of the analysis method. Since many of our samples were lightly loaded due to the short sampling time on their flight legs, it is the uncertainty in these three factors (relative to the difference between sample and blank analyte) that determines whether we collected enough sample to derive a concentration. The uncertainty of the above-blank analyte includes the blank uncertainty (the larger of the analyte times Mader's RSD or the standard deviation of all blanks,), the sample uncertainty (done in the same way), and twice the LOD of the instrumental method. The error-bars in Figure 5 are these values, to emphasize the certainty of above-blank values. The remaining multiplicative uncertainties, due to flow meter calibrations ((5%), instrument span calibration ((5%), correction for our particle concentrator efficiency ((20%), and the OC-EC split in the analyses ((15%), are propagated into an additional relative uncertainty of 26%. Essentially this span uncertainty means the whole value and uncertainty could move up or down by 26%, without affecting the issue of whether the sample was significantly above the blank. In the case of ratios like TC/EC (derived from the same analysis of one filter), only the (15% split-point uncertainty applies, since the same calibrations and flow apply to both numerator and denominator. The TC value does not depend on a split point; we have FT samples for which the TC value is significant even though there was not enough charring to define an OC/EC split point. CIG filters were also analyzed using the Sunset Labs instrument, employing a 20(C min-1 constant temperature ramp in high purity helium to a maximum oven temperature of 300(C. The CIG filters were pre-baked at 325(C in nitrogen the night before each flight to avoid contamination and stored in a cooler before and after exposure. CIG field blanks ranged from 0.3 to 1.5 _g C cm-2 averaging 0.8 _g C cm-2, while sample CIG filters ranged from 0.3 to 3.9 _g C cm-2, averaging 1.6 _g C cm-2. The volatilized OC ranged from 0 to 2.5 times the OC collected on the quartz filter. Only 11 of 78 CIG samples had a S/N>2, while 34 had a S/N>1. If we assume unit collection of gas phase organics by the diffusion denuder, any collected semivolatile organic aerosols would no longer be in equilibrium with their gas phase so that some would evaporate in an attempt to re-establish the gas/aerosol equilibrium. This would result in a negative artifact on the quartz filter, but the evaporate would be collected by the backup CIG filter. However, if denuder breakthrough occured, the collection of those residual gas phase organics by the CIG would be misinterpreted as volatilized aerosol OC. Mader et al., (2003) concluded that volatile OC was less than 20% during ACE-Asia, but in a reanalysis of our CIG data since the Mader et al. publication, we have found that 15% of our samples volatilized more than 20% of their OC. We were able to detect TC above its detection limit in all but one of the 92 samples. Twenty-four EC and 16 OC samples were either non-detects (value smaller than its uncertainty) or we could not confidently assign a split point to apportion the TC between them. Most samples were 30 to 60 minutes duration, at a nominally-constant altitude. Denuder shedding was visually evident during parts of the first 4 flights; all contaminated samples were removed from the data set. 3.4 Data intercomparisons To test our analytical procedure, we participated in an ACE-Asia inter-laboratory comparison organized by Jamie Schauer of the University of Wisconsin - Madison. He distributed 2 punches each from four high-volume samples and two blanks. We asked that our samples be kept refrigerated until we were set up in Japan before having them shipped to us so we could analyze them while we were analyzing our aircraft samples. The six ACE-Asia research groups and two labs of Sunset Labs agreed on moderate-level OC to within 4%, low-level OC within 13%,and EC within 13%. This suggests that our analytical capabilities are accurate within those bounds. March 2003: However, it is now clear that this comparison is not totally germane to the lightly-loaded samples we often got from the aircraft. Assessing OC/EC split points for samples that had little absorption raises issues that were not dealt with in the analysis of the more heavily-loaded intercomparison samples. 4.0 Data Format The file structure is a tab-delimited ASCII file, with a carriage-return to mark the end of each line. The data is in 21 columns with 103 rows. The first row is a listing of the tabulated parameters. All subsequent rows contain the data from a single airborne sample. None of the "sideall" values are included in the table, only denuded regular samples. -8888 indicates that no valid measurement was made. -9999 indicates a situation in which we made a measurement but the sample was contaminated, dropped, or otherwise made unreliable. The columns are: 1. Start Date and Time of the sample, as YYYYMMDDHHMMSS. For the first 11 samples no seconds were recorded, so the number has two fewer digits than in the rest of the file. All times and dates are UTC. 2. End Date and Time of the sample, as YYYYMMDDHHMMSS. For the first 11 samples no seconds were recorded, so the number has two fewer digits than in the rest of the file. Note that some samples were turned off temporarily (usually for rain showers), so there may be an intermediate off and on, which are listed under "Comments." 3. The UH sample number: Flight Number (RF01), approximate start time as HHMM, and an identifier that this sample was a quartz filter for carbon analysis. 4. EC: µg/sm3 of elemental carbon, from the quartz sample filter only. 5. UncEC: the propagated uncertainty in EC 6. EC from TC: TCq/4.5 for those samples in which the OC/EC split point was ambiguous. This is one way to estimate EC in those samples, for which the TC was a good measurement. The average TCq/EC ratio for all other samples was 4.5. 7. OCq: µg/sm3 of organic carbon on the quartz filter only. 8. OCq_Unc: the propagated uncertainty in OCQ 9. OC from TC: TCq*0.778 for those samples in which the OC/EC split point was ambiguous. This is one way to estimate OC in those samples, for which the TC was a good measurement. The average OCq/TCq ratio for all other samples was 0.778. 10. TCq: µg/sm3 of total carbon, from the quartz sample filter only. 11. UncTCq: the propagated uncertainty in TCq. 12. OCv: µg/sm3 of organic carbon that volatilized and was collected by the CIG backup filter. This is would have been a negative artifact, had we not used the CIG backup filters. 13. UncOCv: the propagated uncertainty in OCv. 14. OCsum: µg/sm3 of total organic carbon, which is the sum of carbon on the quartz filter and any that evaporated onto the backup CIG. For flights with no CIG filters, this column contains -8888. The units are micrograms of carbon per standard cubic meter of air. Standard conditions are 1 atmosphere (1013 mbar) and 298 degrees K. No factor has been applied to calculate the total organic aerosol mass (often called OM, including O, H, and N) from the carbon. 15. OCsum_Unc: the propagated uncertainty in OCTot. 16. Start_Alt: The pressure altitude at the beginning of this sample, in meters above sea level (ASL). In some cases (especially on radiation flights, where the legs were generally shorter), a single sample may have integrated over several altitudes. In other cases the start altitude is probably the better measure of the sample altitude, because the time-lag to start all the samplers at the beginning of a leg means that the plane had probably stabilized its altitude by the time the sample flow began. That same time-lag at the end of the leg, however, increases the likelihood that the listed stop time may be after the plane had begun changing altitudes. Note that these altitudes and locations are provided as a guide only, and serious use of position data should be accompanied by looking at the detailed flight data in the C-130 files, which we cannot reproduce here. 17. Stop_Alt: the C-130 pressure altitude at the time the sample was terminated, in meters ASL. 18. Start_Lat: the gps-derived latitude, in decimal degrees (positive north) at the start of the sample. 19. Stop_Lat: the gps-derived latitude, in decimal degrees (positive north) at the end of the sample. 20. Start_Lon: the gps-derived longitude, in decimal degrees (positive east) at the start of the sample. 21. Stop_Lon: the gps-derived longitude, in decimal degrees (positive east) at the end of the sample. 22. Comments. Two types of comments appear here: 1) times that a sample was turned off briefly to avoid rain showers, and 2) notations that the sample has been flagged as not being totally defendable. The two flagged samples showed no visible signs of any contamination, but occurred on flights where we noted contamination on a few other samples. Their rather high EC values raised suspicion, but we could not eliminate the samples since our visual inspection revealed no obvious contamination. Both were collected in modestly polluted air, so it is possible the high values are valid. This is Version 4.0 of the Huebert group OC/EC data. 5.0 Data Remarks 5.1 Revisions 15 April 2002 Changes from first submission of Huebert Group OC-EC data. After a re-evaluation of the Sunset Labs output, we have changed the values for OC (including carbonate if present) for 6 samples in flights 5 and 6. All other values for these samples and others remain unchanged from the first submission. The changed values of OCQ are in the following samples: RF050455 RF050730 RF050842 RF060155 RF060351 RF060437 The measurement of organic and elemental carbon is a very difficult one. There are a number of opportunities for large positive and negative artifacts, and the separation between "elemental" and "organic" carbon is of necessity operationally defined. Our sampler minimizes these artifacts, but it introduces its own set of uncertainties. The positive and negative artifacts are discussed by Huebert and Charlson (Tellus, 52B, 1249-1255, 2000.) The basic problem is that many vapor phase organic species adsorb readily on baked quartz filters, and are subsequently indistinguishable from collected organic aerosols by the coarse analytical method that pyrolyzes all carbonaceous material indiscriminately. This positive artifact would have been 5-6 times as large as the actual ambient OC on at least one of our flights. How much could one trust a non-denuded quartz sample whose "ambient concentration" was derived by subtracting 80% of the analyte? The existing correction methods involving serial quartz filters have been hard to defend, in part because the thermodynamic conditions on the backup quartz are certainly different from those at the first filter in a system where small temperature or RH changes can dramatically shift the aerosol/vapor equilibrium. Yet serial quartz samplers benefit from simplicity of implementation, so they see wide use. We feel that we have reduced the positive artifact dramatically, and perhaps have effectively eliminated it relative to other sources of error. The long denuder, which had been (laboriously!) loaded with perfectly flat strips of CIG filter and kept sealed to avoid expending its capacity to adsorb organic vapors, probably decreased the concentrations of most species by more than 90% from their ambient values. Thus, even if the initial ambient VOC vapor concentration had exceeded the OC aerosol by 10-fold, the remaining (not very sticky) vapor reaching the quartz would essentially equal the aerosol carbon concentration. The maximum positive artifact would then be 100%, but since these compounds didn't adhere to activated carbon in the denuder, it is unlikely that they will adhere with high efficiency to quartz. The likely maximum artifact might well be in the 10-20% range relative to ambient OCTot. At this point, errors from flowmeters and blanks limit precision. The negative artifact results from the evaporation of ambient aerosol mass when the equilibrium with its vapor phase is disturbed by the sampler and the filter. Acceleration of previously-stationary air to aircraft speeds to bring it into an aircraft unavoidably warms the sample air by 7 C or more, depending on airspeed. The pressure drop across quartz filters under commonly-used conditions is measured in tenths of an atmosphere - a very significant change for volatile species equilibria. The depletion of the vapor phase by adsorption on the filter's front face may cause some of the solid phase to evaporate after collection. Diurnal or altitude-caused changes in ambient temperature may move some species back and forth between phases, evaporating more as the day warms up and condensing it back at nighttime. Thus long samples (multiple hours or altitudes), are likely to at some time experience the evaporation of a previously-collected carbonaceous substance. In a serial-quartz sampler, some of this evaporate might collect on the second filter, and be treated as a positive artifact. In principle we have eliminated the negative artifact by sampling behind the quartz with a CIG filter. It should collect any vapors from evaporation and hold them for analysis later. However, these are messy filters to work with, since they can shed bits of elemental and organic carbon onto the analytical filters. (They are not what one wants to have in a lab, when trying to reduce the EC blank on quartz filters!) The physical integrity of these CIG filters can be damaged by heating them to too high a temperature (we used 325 C), so they cannot be cleaned as rigorously as the quartz (which we heat to 550 C). This means that the analyses of OC collected on CIG can also not go to too high a temperature, or it begins to drive off material that has been on the filter since its manufacture. The upper temperature limit may be a minor issue for quantifying the negative artifact, however, since the species that evaporated from the quartz filter probably volatilize from the CIG filter at a low temperature. However, it means that the blank subtraction process is inexact because residual material was not perfectly removed from the CIG filter prior to its use, causing the blanks to be variable enough to limit the precision of the analysis. Thus the OCvol amount is more variable and less sensitive than OCq in part because one can't clean the CIG perfectly. We also found that the CIG filters had to be baked the night before their use and analyzed within a day of their exposure to avoid the degradation of the CIG blanks. One of the reasons there is no OCvol data until RF05 is that it took us that long to identify and resolve the high variability of the CIG blanks. The storage conditions under which quartz filters kept very nicely caused the CIG filter blanks to become increasingly large and variable. The short sampling time available on an aircraft prevented us from using the most common treatment for variable blanks: collecting a longer sample. The other serious concern about the CIG backup filters is that they may be subject to a positive artifact: any VOC that was not collected by the denuder could get picked up at the slightly different conditions of the backup CIG filter. If the denuder efficiency was limited simply by the ability of big molecules to diffuse to the surface and be collected, then these molecules might well be collected by the CIG filter. If, however, some VOC passed through the denuder simply because that substance doesn't adhere readily to activated carbon, then this substance should also not be collected by the backup CIG filter and no positive artifact would result. In either case, we believe the fractional error this causes in OCsum is likely to be small. The Sunset Labs analyzer reduces but does not eliminate the ambiguity about elemental vs organic carbon. Chemically, elemental carbon would be either diamond or graphite. Soot, a collection of small carbon balls resulting from incomplete combustion, may not be pure carbon. Indeed, the charring of fuels during incomplete combustion can produce tarry substances that are mostly but not exclusively carbon. The cutoff point between impure elemental carbon and heavily charred organic carbon is defined operationally rather than in stoichiometric terms. Our operational definition is to define "EC" as any carbon that evolves during our final heating step, after the O2 is added. The exception to this is that some OC chars during the heating in He and thus does not volatilize. The Sunset Labs instrument measures the transmittance through the filter to quantify this charring, and does not assign carbon in the last heating step to "EC" until the transmittance is back to the value at the start of the analytical cycle. In this way, it should minimize the positive charring artifact that could plague other approaches to measuring EC. The potential errors due to these positive and negative artifacts are certainly reduced and perhaps largely eliminated by the PC-BOSS sampler. The major factor that contributes to the uncertainty of our PC-BOSS sampler is the 2x correction for loss of particles in the denuder. One of the wonderful aspects of the PC-BOSS sampler is its side filter, which allows one to test passing efficiencies in the field. It has no preconcentrator (so its sensitivity is not as good for short samples) and it has no denuder (so it will experience positive VOC adsorption artifacts), but it allows one to test the functioning of these two devices. The measured EC and sulfate efficiencies of 50% are identical to the experience of Eatough using the PC-BOSS sampler on the ground in numerous field programs. It could be due to poor virtual impactor efficiency, losses in the denuder, or both. The fact that both sulfate and EC show the same losses to within 10-20% gives us confidence that this factor is robust. One cannot help feel uneasy, however, when applying a factor-of-two correction. Finally, during the first four research flights we saw (steadily decreasing) evidence of visible specks of carbon from the denuder on our quartz filters. They were strikingly different from the even-colored deposit that came from sampling. We believe this to be material that had been dislodged from the CIG strips in the denuder during shipping and equipment integration. When we first subjected the denuder to the vibration and high air flow of sampling this blew any loose carbon back onto the filters. Fortunately, the specks disappeared completely by RF05. However, we are left in a quandry about a few samples from RF01 and RF02, which have been "flagged" in the Comment column. Several samples on each flight were eliminated because of visible specks. The retained samples did not have visible specks (which would have allowed us to discard them immediately), but they had quite high values of EC or OC relative to the rest of the data set. These high values made us suspicious, but the fact that these samples were collected in fairly polluted air raises the possibility that they really are legitimate concentrations. We elected to leave them in the data set, but with a flag for users that indicates we cannot unilaterally defend them since some samples were clearly contaminated on those flights. 5.2 Revisions 21 October 2002 After discussion at the Boulder workshop with Brian Mader and others, it became apparent that we had reported EC values that had loadings well below the actual detection limit of the Sunset system. Although these sample values were much higher than our blanks and appeared to be significant, a statistical analysis of the reproducibility of small-loading samples revealed that the Sunset instrument was incapable of reproducibly detecting mass loadings smaller than 0.25-0.5 ug C/cm2 (see Brian Mader's manuscript for the actual data). In addition it was apparent that the Sunset instrument has difficulty in setting an appropriate split point for samples with small mass loadings. This is due in part to a minimal change in transmittance usually induced by charring of the OC sample. Apparent changes in transmission associated with the thermal instability of the photodiode and LED resulted in premature split points on many samples. It was determined that samples which were within the statistical limit of the variance in laser transmittance would be set to a default split point at the point of oxygen injection into the front oven. In accordance to these suggestions we reviewed each sample independently to access the position of its split point and whether the mass loading was sufficient for analysis. Due to our short sampling durations (20-60 min) we have numerous samples with insufficient loadings for reproducible analysis. These samples are represented as zero concentration (with an uncertainty that is the DL for that sample) in our data set. In addition, all of the split points were analyzed and adjusted as described above. We also revisited the SVOC data from our CIG back-up filters. This data has been drastically modified after consultation with Delbert Eatough. The CIG samples were analyzed for carbon that was evolved between 200-300 C. Although there is still some possible error that may be associated with denuder break-through, we are confident that the data presented is an improvement over the previous dataset which calculated SVOC as all of the carbon evolved from the filter between ambient and 300 C. The C evolved below 200 C we believe to be from VOCs picked up during handling or from denuder breakthrough, rather than evaporated OC aerosols. Finally, we re-examined the efficiency of the particle concentrator (PC) in our PC-BOSS sampler, because the measured efficiency for EC and NSS (about 2.2) was so much less than the factor of 4 the design suggests. We generated monodisperse particles of oleic acid and rhodamine dye and measured their passing efficiency. We were alarmed to find that a large fraction of the particles impacted on the edge of the PC virtual impactor receiving slot, apparently due to a misalignment in the machining process. The maximum efficiency (61%) was found for 0.7 µm particles, dropping off to as little as 15% for 3 µm and 27% for 0.1 µm particles. We then used OPC-derived size distributions and our lab-measured efficiency kernels to calculate the PC efficiency for individual samples (with the help of Steve Maria). This analysis assumes that the actual EC and OC distributions were similar to the OPC (effective light scattering) distributions, which is debatable. However, the range of the corrections was only about 10%, so the average correction derived from the side filter has been tweaked only slightly with this analysis. The uncertainties in this third release are estimates, rather than being rigorously derived from propagation of errors analysis. The re-analysis of the thermograms started with an assumption that any sample with less than 0.25 ug C/cm2 of filter was below detection limit, so zero was assigned as its concentration. Samples with between 0.25 and 0.5 ug C/cm2 were examined individually and assigned either zero or a value. The uncertainties in this process are not straightforward and will be sorted out around the end of 2002. Until then we have used uncertainties that are a blend of the Version 1 uncertainties (propagated using flow, blank, and DL uncertainties) and the use of a 0.375 ug C/cm2 DL. 5.3 Revisions 28 March 2003 Version 4.0 of our data includes rigorously propagated errors, which include Brian Mader's observation that the relative standard deviation of multiple analyses on punches from the same filter (done on ACE-Asia samples) became large when the filter loading was less than 0.5 ug C/cm2. We have taken this as a kind of analytical method-dependent lower limit of detection, and propagate it along with the flowmeter, blank, and other analytical uncertainties. A much smaller number of OCV and CC numbers were significant, so the low-S/N values have been removed from the dataset. One sample now has no significant concentrations, but we left its row in the data set so that users could simply paste the new set over the old version in programs that used the earlier data. We have also re-examined our PC correction based on these uncertainties. Some of the side/sample filter ratios were not significant, so we based the PC correction factor solely on 5 EC ratios that had signal to noise ratios greater than 2. That dropped the correction factor down to 1.76, and is responsible for most of the change in values reported here. We continue to work the question of the impact that coating materials have on EC specific absorption. If the coating have as large an effect as our results suggest, they call the thermal/optical approach itself into question: by the time EC is burned off, all its sulfate and OC coatings should be gone, reducing its absorption and increasing light transmission through the filter. Correcting for this effect would decrease our EC amounts and increase the resulting EC specific absorption. Much more lab work is needed on this question. 5.4 Revisions May 2004 These have all been described above. We believe that the version 5.0 data set is far more defendable than previous versions, for all the reasons outlined above. Software compatibility The file OC/EC-C130Huebert21Oct02.txt is tab-delimited ACSII. It should import easily into spreadsheets or other programs. While it can be opened by conventional word processors, the long lines may cause a wrap that confuses the columns. (For the Mac, the application BBEdit Lite is able to handle long lines and huge files very quickly for various types of editing and viewing.) In Microsoft Office 98 Excel for the Mac click "File," "Open," highlight the file, and click "Open." Select "Delimited" when the Text Import Wizard window appears. Choose "Tab" delimiters, which should separate the columns nicely, then "Next." Depending on your use of the data, you may or may not wish to change the column formats prior to clicking "Finish." In Igor Pro, select "Data," "Load Waves," and "Load Delimited Waves." Then browse to the file, highlight it, and click "Open." The window that comes up should offer the proper wave names from the first row. Click "Make Table" to have the loaded waves appear as a table. Highlight wave names 0, 1, and 2, and set their format to "Text," to avoid having the data/time numbers changed to scientific exponential notation. Finally, click "Load" to finish the process. Note that you may wish to replace -8888 and -9999 with something like NaN before trying to do line fits, correlations, etc.