Analysis of PM 2.5 Speciation Network Carbon Blank Data

Transcription

1 Analysis of PM 2.5 Speciation Network Carbon Blank Data James B. Flanagan, Max R. Peterson, R.K.M. Jayanty, and Ed E. Rickman Research Triangle Institute, Research Triangle Park, North Carolina ABSTRACT This presentation describes results of an analysis of field and laboratory carbon blank data for the PM 2.5 Chemical Speciation Trends Network (STN), which is managed and funded by EPA/OAQPS, and for which Research Triangle Institute is the contract laboratory. Data from over two years of network operation were used for this analysis. PM 2.5 chemical speciation samplers are multi-filter units which include a quartz fiber filter for collecting PM 2.5 aerosol samples. These filters are analyzed for carbon constituents using an adaptation of NIOSH Method 5040, a thermal-optical transmittance method. This method, which relies on thermal evolution of carbon species followed by quantitation by FID, does not identify specific compounds, but reports method-defined aggregate filter loadings of organic carbon (OC), elemental carbon (EC), and total carbon (which is the sum of OC and ED). Almost 1500 laboratory QC filters; 2727 trip and field blank filters; Data span February 2000 through March 2002; Data were managed using Microsoft Access, and were analyzed using both MS Excel and SAS. The data analyses performed for this report resulted in several conclusions that have direct bearing the blank correction issue. The most significant of these are as follows: 1. Laboratory blank values are smaller (by an order of magnitude) than field and trip blank values. Furthermore, any contribution to the overall blank levels is already implicitly included in the field and trip blanks. Therefore, the laboratory blank levels are not necessary for blank-correcting the routine exposure data. 2. Trip Blanks and Field Blanks have very similar statistical characteristics (e.g., mean and standard deviation). Therefore, the Trip and Field Blank data will be combined when calculating the blank correction factors. 3. Average blank levels differ by sampler type. Therefore any blank corrections must be computed using the appropriate sampler-specific blank value. 4. Average Field and Trip Blank masses were found to be between 10 and 30% of average routine (24-hour) levels. Data users should consider the impact of this finding on their data applications. 5. It was also found that the analytical "split time" used to separate the OC and EC fractions by the analyzer's software is usually early (before oxygen is added) for most blanks and other low-level samples. Because of this systematic pre-oxygen split, refractory OC that volatilizes from the filter without forming char can be mistakenly categorized as EC in some blanks. Pre-oxygen splits are much 1

2 less likely in more heavily loaded filters. This finding has implications for how blank data might be used to correct the OC/EC data for 24-hour environmental samples. INTRODUCTION The original question that prompted this investigation was to discover if it is desirable and feasible to blank-correct the OC/EC data, and if so, to identify the significant variables that define how the data should be stratified for blank correction. Network Description The PM 2.5 chemical speciation trends network (STN) consists of approximately 54 monitoring sites dedicated to characterizing major aerosol mass components in urban areas of the United States for discerning long-term trends and providing an accountability mechanism to assess the effectiveness of emission mitigation programs. The initial implementation of the trends network began with the installation of the first 12 sites in November The pilot network operated from February 9, 2000 through the end of July, 2000; at the completion of the preliminary sampler and network evaluation period, additional trends network sampling sites were brought into operation along with many state and local speciation sampling sites. Today, there are over 200 chemical speciation sites in operation, sampling at one-in-six- and one-in-three day intervals. This data analysis reported here considered data from the 54 STN sites, plus most of the samplers operated by state and local agencies. As part of the ongoing QC program for the network, regular trip and field blanks are included among the samples shipped to the field sites. Field blanks are run at a frequency of 10% (one field blank for every 10 scheduled routine samples), and trip blanks are run at a frequency of 3% (one trip blank for every 30 scheduled routine samples). Trip blanks are implemented by mounting filters into the modules, shipping the loaded modules to the field and back to RTI without opening the modules' containers. Field blanks are implemented similarly, except that the modules are removed from their packing and are mounted on the sampler, removed, and repacked. Analytical Instruments All carbon analyses are performed on Sunset Laboratory Inc. Thermal/Optical Carbon Analyzers. Carbon analyzers used in the speciation program and their histories are given in Table 1. 2

3 Table 1. History of Carbon Analyzers Used in the Speciation Program Analyzer History Original Placed in service in 1998; upgraded to become the Retrofit analyzer on 3/21/2000; used only briefly in the speciation program to do acceptance testing analysis of prefired quartz filters. Because of few data points, data from this analyzer are not included in the data analysis. New (N) Placed in service 2/10/2000; refurbished to become Second analyzer on 6/6/2001. Retrofit (R) Refurbished reincarnation of Original analyzer; placed in service on 3/21/2000. Software upgraded on 6/6/2001. Second (S) Refubished reincarnation of New analyzer; placed in service 6/6/2001. Third (T) Placed in service on 6/6/2001 The split time is used to apportion TC into OC and EC. OC evolved from the filter is calculated based on total FID response up to the split time, and EC evolved off the filter is calculated based on total FID response after the split time (but before the FID response to the internal standard). The split time between OC and EC is determined by Sunset Lab's calculation software based on changes in the transmittance (or, more correctly, light scattering) of the filter during analysis. The transmittance of the filter during analysis of a typical sample (or a sucrose standard) decreases during the non-oxidizing first heat ramp due to the formation of light-absorbing char from thermal decomposition of nonvolatile OC on the filter. When oxygen is added during the oxidizing second heat ramp the char formed during the first (non-oxidizing) heat ramp and elemental carbon present initially on the filter burn off, and the transmittance of the filter increases. The split point between OC and EC is set as the time in the analysis when the transmittance of the filter reaches the value it had at the beginning of the analysis, which is assumed to indicate that the same amount of light-absorbing material has been removed from the filter as was formed during the non-oxidizing first heating ramp. For a sample containing EC, the transmittance of the filter at the end of the analysis is higher than it was at the beginning of the analysis. LABORATORY QC BLANKS There are two kinds of routine QC blanks that are run routinely in the OC/EC laboratory: Filter Lot Acceptance Blanks and Instrument Blanks. Tables 2 and 3 summarize the OC, EC, and TC results for five different instruments. Only the instruments designated R, S, and T are currently in use. 3

4 Table 2. Lot Acceptance Blank Data Summary Analyzer Count OC (µgc/filter) EC (µgc/filter) TC (µgc/filter) Mean Std Dev Mean Std Dev Mean Std Dev N* R* R** S** T** *1/31/2000-6/6/2001 ** 6/7/2001-3/31/2002 Table 3. Instrument Blank Data Summary Analyzer Count OC (µgc/filter) EC (µgc/filter) TC (µgc/filter) Mean Std Dev Mean Std Dev Mean Std Dev N* R* R** S** T** *1/31/2000-6/6/2001 ** 6/7/2001-3/31/2002 Filter Lot Acceptance Blanks - Filters received from the manufacturer are cleaned in an oven at 900 o C for three hours under a slow stream of filtered air. This removes both OC and EC that may have been present. Filters are cooled and removed from the oven under clean nitrogen gas. Two Filters from each cleaning batch of 100 filters are analyzed (2%). If the acceptance criterion of 1 µgc/cm2 (11.76 µgc/filter) is not met, then the lot is rejected and re-cleaned. Figure 1 shows the Filter Lot Acceptance Blanks TC vs. date of analysis. Data for four different analyzers are shown. No consistent differences between analyzers were found over the two year period for which data are reported

5 Figure 1. Lot Acceptance Test Blanks vs. Date for Total Carbon LOT ACCEPTANCE TEST BLANKS Total Carbon, All Analyzers, January 31, 2000, through March 31, Total Carbon (µg/filter) Criterion: Acceptance Testing Blank must be < 1 µgc/cm 2 or < µgc/filter New Retrofit Second Third Poly. (New) Poly. (Retrofit) Poly. (Second) Poly. (Third) /30/00 02/29/00 03/30/00 04/29/00 05/29/00 06/28/00 07/28/00 08/27/00 09/26/00 10/26/00 11/25/00 12/25/00 01/24/01 02/23/01 03/25/01 04/24/01 05/24/01 06/23/01 07/23/01 08/22/01 09/21/01 10/21/01 11/20/01 12/20/01 01/19/02 02/18/02 03/20/02 Date Instrument Blanks - Instrument blanks are run daily as a QC check. To run an instrument blank, a previously-analyzed filter is immediately reanalyzed without removing it from the instrument. Because the instrument blank filter is never exposed to ambient air or other contamination sources, instrument blanks are typically cleaner than Lot Acceptance blanks. Figure 2 shows the Instrument Blanks TC vs. date of analysis. Figure 2. Instrument Blanks vs. Date for Total Carbon OC/EC Instrument Blanks Total Carbon, All Analyzers, January 31, 2000, through March 31, /30/00 02/29/00 03/30/00 04/29/00 05/29/00 06/28/00 07/28/00 08/27/00 09/26/00 10/26/00 11/25/00 12/25/00 01/24/01 02/23/01 03/25/01 04/24/01 05/24/01 06/23/01 07/23/01 08/22/01 09/21/01 10/21/01 11/20/01 12/20/01 01/19/02 02/18/02 03/20/02 Total Carbon (µg/filter) Criterion: Acceptance Testing Blank must be < 1 µgc/cm 2 or < µgc/filter New Retrofit Second Third Poly. (New) Poly. (Retrofit) Poly. (Second) Poly. (Third) Date 5

6 TRIP AND FIELD BLANKS Any blank correction factors to be applied to the OC/EC data must be calculated based on the trip and field blank data because lab acceptance blanks and instrument blanks are lower than trip and field blanks by almost an order of magnitude. A survey of the data set's characteristics was conducted in order to identify the important independent variables governing OC and EC blank levels and to determine if blank correction was warranted. Most of the potential independent variables were identified in the original Work Assignment, and included sampler type, geographic location (latitude and longitude), season, correlations with ambient meteorological conditions, correlations with ambient pollutant concentrations, etc. Trip and Field blanks differ from the laboratory QC blanks discussed above because they are mounted into sampling modules, shipped from the lab to the field sites, and returned. Most of the handling procedures for Trip and Field Blanks are identical: the only difference is that Field Blanks are removed from their shipping containers and are mounted (still inside their modules) onto the sampler for a few moments. Table 4 summarizes the characteristics of the four different sampler modules that are being used in the STN program. Although the samplers have been designed to include an XAD-coated denuder to remove gaseous carbon (i.e., gas-phase VOCs and SVOCs) from the sampled air, these denuders have not been used during the study. Table 4. Sampler Module Characteristics. Sampler Type Sample Volume, m 3 per 24 hours Module Materials Module Internal Volume, cm3 Size-selective Inlet MetOne SASS 9.6 Metal 230 external Andersen RAAS 10.4 PTFE-coated metal 31.4 external URG MASS 24 Plastic 32.3 external R&P Uncoated Metal 456 internal impactor shipped with vacuum grease The R&P 2300 module differs from the other sampling modules in one important respect: a greased impactor plate is included inside the R&P 2300 module during shipping. The particle-sizing devices (impactor, cyclone, etc.) for the other sampler types are external to the filter modules. Dow-Corning High Vacuum Silicone Grease is used to fill the impactor well. During module assembly, the grease is smoothed to a flat, planar surface. 6

7 For each sampler type, the Trip and Field Blank levels are very similar. This is not unexpected, since the procedures are so similar. Table 5 shows the comparison of field and trip blanks. Statistically significant differences in average blank levels are found between different sampler types. As Table 5 illustrates, the URG MASS TC levels appear to be about half the levels of the other sampler types. This is probably due to the different materials and sizes shown in Table 4. Table 5. Comparison of Trip and Field Blank Averages. Sampler Type EC, µgc/filter OC, µgc/filter TC, µgc/filter Field Trip Field Trip Field Trip MetOne SASS Andersen RAAS URG MASS R and P Trends (2300) Table 6 compares the average TC blank levels against the average TC levels for 24-hour exposures. Blank backgrounds for three of the four sampler types are high, ranging between 23.9 to 33.9 percent of the average ambient level. For the fourth sampler type, the URG MASS, the ratio of average blank to average Routine was less than 10%. OC appears to comprise about 90% of the blank mass for all the analyzer types except the R&P For the latter sampler, EC comprised approximately 25% of the total blank. We believe that this is an artifact caused by traces of the silicone impactor grease masquerading as EC. Table 6. Blank TC as a percentage of Routine 24-hour exposures. Sampler Type Avg Routine TC Avg Blank TC Blank/Routine, percent MetOne SASS % Andersen RAAS % URG MASS % R&P % This phenomenon is illustrated further in Table 7, which is similar to Table 6, except that it compares Routine vs. Blank EC levels. For three of the samplers, the EC blank is a smaller proportion of Routine level than it was for TC. However, blank EC for the R&P 2300 is over 50% of the corresponding Routine average value. To investigate this phenomenon, the RTI laboratory analyzed some standard samples of the vacuum grease and compared the thermograms against thermograms generated for blank samples from the R&P Similar peaks at the highest temperature region of both thermograms 7

8 were found, strongly implying that the impactor grease was being interpreted by the instruments' software as EC. The software artifact was that, because the grease volatilized without forming char, the "split time" was coming before all the OC had vaporized. The pre-oxygen split seems to occur mostly with blanks and very low-level exposed filters. With higher-level filter loadings containing OC that forms char, the split time comes later in the analysis, eliminating this misidentification of hightemperature OC that does not form char as EC. Table 7. Average EC as a percentage of Routine 24-hour EC averages. Sampler Type Avg Routine EC Avg Blank EC Blank/Routine, percent MetOne SASS % Andersen RAAS % URG MASS % R&P % It should be noted that low-level samples from other types of analyzers also occasionally show a similar artifact, and that the blank EC levels calculated for all analyzer types are probably high. It is impractical to recompute the OC and EC splits for all the blank and low-level samples analyzed during the first two years of the program, because of the huge volume of raw data that would have to be reprocessed. A simpler way to handle blank data is to assume that TC is essentially all OC since exposure of blank filters to EC contamination is unlikely. Figure 3 shows a typical thermogram, showing post-oxygen split and pre-oxygen split. FID response, laser transmission, and temperature are plotted vs. time. The FID signal to the left of the split is integrated as OC, and everything to the right of the split is accounted as EC. TC is independent of the OC/EC split, since it is simply the sum of OC and EC. Note that when the split comes before the addition of oxygen, some of the last OC peak is erroneously counted as EC. The anomalous findings shown above for EC on the R&P 2300 blanks led to discovery of an artifact in the analyzer software. However, other sampler types should not be considered immune to this type of problem. For example, a spot of "char" outside the laser beam area will be subject to this error. It seems likely that a portion of the blank EC levels for all sampler types may be due to misaccounting of OC. 8

9 CONCLUSIONS The data analyses performed for this report resulted in several additional observaations and conclusions. The most significant of these are as follows: 1. Quartz filter blank carbon loadings appear to be composed primarily of volatilizable carbon compounds that do not char and that cause no change in transmittance of the filter as they volatilize. The OC/EC split time for blanks appears to be determined primarily by slight changes in the laser transmittance (or light-scattering) of the filter due to the heating and cooling of the oven and not to any contaminant on the filter. As a result, the OC and EC loadings calculated for blank filters actually represent an artificial division of the TC loading; however, the values may be useful for blank corrections. 2. Laboratory blank values are smaller (by an order of magnitude) than field and trip blank values. The laboratory components of error are already present in the field and trip blank data, and need not be included separately in the blank correction. 3. Trip blanks and field blanks for each sampler type have very similar statistical characteristics (mean and standard deviation). No consistent, positive bias attributable to contamination during field handling could be identified. 9

10 4. The average blank levels differ according to sampler type. Therefore blank corrections should be computed using only data from the respective sampler types. ACKNOWLEDGMENT A portion of the work reported here was performed under Work Assignment 4-05 of EPA Contract 68-D , Jim Homolya, Work Assignment Manager. The following RTI personnel contributed to this report: Mel Richards, Linda Andrews, Jessie Deal, and Mike Riggs. REFERENCE "Analysis of Speciation Network Carbon Blank Data, DRAFT REPORT." Flanagan, James B., and Max R. Peterson, RTI International, September 3,