International Comparison of a Hydrocarbon Gas Standard at the

Feb 6, 2014 - Good practice would be to ultimately confirm the low mole fraction region of that calibration curve using a certified reference material...
2 downloads 10 Views 1MB Size
Download PDF
Article pubs.acs.org/ac

International Comparison of a Hydrocarbon Gas Standard at the Picomol per Mol Level George C. Rhoderick,*,† David L. Duewer,† Eric Apel,‡ Annarita Baldan,§ Bradley Hall,∥ Alice Harling,⊥ Detlev Helmig,@ Gwi Suk Heo,# Jacques Hueber,@ Mi Eon Kim,# Yong Doo Kim,# Ben Miller,∥ Steve Montzka,∥ and Daniel Riemer∇ †

Chemical Sciences Division, Materials Measurement Laboratory, National Institute of Standards and Technology, 100 Bureau Drive, MS-8393, Gaithersburg, Maryland 20899-8393, United States ‡ Earth System Laboratory, Atmospheric Chemistry Division, National Center for Atmospheric Research, Boulder, Colorado 80307, United States § Van Swindin Laboratorium, Thijsseweg 11, 2629 JA Delft, NL ∥ Global Monitoring Division, Earth Systems Research Laboratory, National Oceanic and Atmospheric Administration, 325 Broadway, Boulder, Colorado 80305, United States ⊥ Analytical Science Team, National Physical Laboratory, Hampton Road, Teddington, Middlesex, U.K. @ The Institute of Arctic and Alpine Research, University of Colorado, Boulder, Colorado 80309-0450, United States # Korea Research Institute of Standards and Science, 267 Gajeong-ro, Yuseong-gu, Daejeon, Republic of Korea ∇ Division of Marine and Atmospheric Chemistry, Rosenstiel School of Marine and Atmospheric Science, University of Miami, Miami, Florida 33149-1098, United States S Supporting Information *

ABSTRACT: Studies of climate change increasingly recognize the diverse influences of hydrocarbons in the atmosphere, including roles in particulates and ozone formation. Measurements of key nonmethane hydrocarbons (NMHCs) suggest atmospheric mole fractions ranging from low picomoles per mol (ppt) to nanomoles per mol (ppb), depending on location and compound. To accurately establish mole fraction trends and to relate measurement records from many laboratories and researchers, it is essential to have accurate, stable, calibration standards. In February of 2008, the National Institute of Standards and Technology (NIST) developed and reported on picomoles per mol standards containing 18 nonmethane hydrocarbon compounds covering the mole fraction range of 60 picomoles per mol to 230 picomoles per mol. The stability of these gas mixtures was only characterized over a short time period (2 to 3 months). NIST recently prepared a suite of primary standard gas mixtures by gravimetric dilution to ascertain the stability of the 2008 picomoles per mol NMHC standards suite. The data from this recent chromatographic intercomparison of the 2008 to the 2011 suites confirm a much longer stability of almost 5 years for 15 of the 18 hydrocarbons; the double-bonded alkenes of propene, isobutene, and 1-pentene showed instability, in line with previous publications. The agreement between the gravimetric values from preparation and the analytical mole fractions determined from regression illustrate the internal consistency of the suite within ±2 pmol/mol. However, results for several of the compounds reflect stability problems for the three double-bonded hydrocarbons. An international intercomparison on one of the 2008 standards has also been completed. Participants included National Metrology Institutes, United States government laboratories, and academic laboratories. In general, results for this intercomparison agree to within about ±5% with the gravimetric mole fractions of the hydrocarbons.

N

atmospheric composition, trends, source attribution (HC patterns), oxidizing capacity, and OH/O3/NO3/halogen chemistry.3 Benzene is a precursor to particulates, and toluene is important for determining sources of methane (HC

onmethane hydrocarbon compounds (NMHCs), particularly in urban environments, are important precursors and contributors to atmospheric photochemical processes. These processes lead to the formation of particulates and secondary photo-oxidants, such as ozone.1,2 The important species of methane and isoprene largely control background ozone, and ethene, acetylene, 1,3-butadiene, and m/p-xylenes also contribute. Alkanes C2 to C5 are also important to © 2014 American Chemical Society

Received: November 19, 2013 Accepted: February 6, 2014 Published: February 6, 2014 2580

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

Analytical Chemistry

Article

be to ultimately confirm the low mole fraction region of that calibration curve using a certified reference material (CRM). Comparison of these measurements made by different researchers over many years requires use of long-term stable (+10 yr) and accurate calibration standards that deliver the target hydrocarbons at levels which mimic atmospheric content. In 2008, the Gas Metrology Group (GMG) at NIST initiated a research program to determine the feasibility of producing NMHC gas mixtures at the picomoles per mol level.15 A suite of five gravimetric primary standard mixtures (PSMs) containing 18 NMHCs in nitrogen covering a mole fraction range of ≈50 to 230 pmol/mol were developed and contained in aluminum cylinders. That initial study resulted in an internally consistent suite of PSMs. While NIST had previously studied and reported on long-term (10 year) stability of NMHC mixtures at the nanomoles per mol (ppb) level,16 there was little public information on the long-term stability of these picomoles per mol NMHCs gas mixtures contained in aluminum or other types of cylinder materials. Pollman et. al. reported on the preparation and stability of C 2 to C7 hydrocarbons in glass flasks, prepared by the U.S. National Oceanic and Atmospheric Administration (NOAA). Samples in glass flasks showed agreement (within ±5%) of the lighter alkanes compared to a reference sample over a period of 1 year.17 However, the data for two alkenes, ethene and propene, showed increases over that period of time of 25% and 30%, respectively. Results also showed an overestimate of up to 50% for n-hexane over year stability data of samples taken in NOAA network borosilicate glass flasks. One may need to consider that these flasks are at ambient pressure; we urge caution when comparing against compressed gas cylinder mixtures and data as hydrocarbons have been known to increase in aluminum cylinders at near ambient pressures (from unpublished data). This paper reports on the preparation of a fresh suite of NMHC PSMs in aluminum cylinders that were used to assess the stability of the original 2008 NMHC PSM suite and on the results of an international comparison for one of the original 2008 PSMs that included laboratories who prepare their own primary standards and/or make measurements of atmospheric whole air samples.

patterns). Many anthropogenic sources, including automobiles and other transportation vehicles, power stack emissions, and biomass burning, as well as terrestrial and oceanic biogenic processes, emit NMHCs into the atmosphere. These and other considerations result in considerable interest in measuring the levels of NMHCs at ground level and in the upper atmosphere (e.g., the 1982 hydrocarbon study in Sydney, Australia, reported by Nelson and Quigley,4 measurements made from 1984 to 1988 in 69 cities in the United States described by McAllister et al.,5 a 1992 Southern Oxidants study by by Apel et al. in Atlanta, Georgia,6 and a description of the measurement of low molecular mass NMHCs emitted from the mid-Atlantic ocean region by Plass-Dülmer et al.7 Several general overviews on NMHC studies in major cities and the global atmosphere have also been published.8,9 These and other studies find that typical NMHC mole fraction levels range from tens of nanomoles per mol (parts per billion; ppb) in urban atmospheres to well below 10 pmol/mol (parts per trillion; ppt) in remote environments.2 Gas standards containing NMHCs have been key components of these measurement studies. The 1992 Southern Oxidants Study used a National Institute of Standards and Technology (NIST) gas mixture, certified by NIST using their primary standard mixtures (PSMs) of NMHCs, in the range of 5 to 125 nmol/mol per component.6,10,11 In 2003, the World Calibration Centre for Volatile Organic Compounds (WCCVOC), a part of the World Meteorological Organizations (WMO) Global Atmosphere Watch (GAW) program, coordinated the first comprehensive intercomparison among the GAW−VOC community.12 The main focus of the 2003 study was documenting the analytical capabilities of laboratories making measurements of NMHCs. While the WMO/WWCVOC coordinated the 2003 comparison, the National Center for Atmospheric Research (NCAR) in Boulder, Colorado, was asked to prepare gas standards, for comparison, in canisters and to send them to participating laboratories in Brazil, Canada, Czech Republic, Finland, Germany, Ireland, and Slovakia.11 NCAR had previously obtained NIST-certified reference NMHC standards to compare to their scale (agreement was within ±5%) and for traceability to NIST. Each laboratory determined mole fractions for as many as 73 NMHCs in these canisters, using their own calibration technology or other available standards. The reported mole fractions for some NMHCs differed significantly from the reference values among the participants ranging from agreement of 0.1% to over 1000% relative. Excluding one laboratory with consistently poor results, the average relative deviation for all other laboratories was 20%. The agreement for the 21 NMHCs traceable to NIST was 9.5%. Culminating in 2006, the multiyear Accurate Measurements of Hydrocarbons in the Atmosphere (AMOHA) study evaluated gas chromatographic methods used across Europe to determine NMHC mole fractions, finding good agreement for some NMHCs and significant differences for others.13,14 The AMOHA study used low nanomoles per mol calibration standards supplied by the National Physical Laboratory of the United Kingdom. Participants needed to dilute these standards to picomoles per mol levels. While dilution is an established technique for creating a calibration curve in gas analysis, there are potential complications, including establishing the purity of the diluent gas and the extent of adsorption/desorption of the NMHCs throughout the dilution system. Good practice would

■

EXPERIMENTAL SECTION NIST had experience at the picomoles per mol level for key halocarbons18−20 and some VOCs such as benzene. NIST previously developed a series of certified gas standards with NMHCs at mole fractions of ≈5 nmol/mol for federal and state governments and academia.21 Procedures for preparing PSMs and CRMs have been documented and demonstrated with Standard Reference Material (SRM) 1800: Non-Methane Hydrocarbon Compounds in Nitrogen, which contains 15 NMHCs.22,23 The previously developed SRMs contained only a subset, 18, of the EPA PAMS target compounds and the 30 component EU Directive ozone precursor lists. Those SRMS also did not include some important hydrocarbons such as methane, ethene, acetylene, isoprene, 1,3-butadiene, or m/pxylene. NIST had previous experience with these compounds and had determined that acetylene, isoprene, and 1,3-butadiene, while they appear to be stable at micromoles per mol levels, were not stable in Aculife aluminum cylinders at the (5 to 20) nmol/mol level.24,25 Methane was not considered, as it is present at ambient concentrations well above picomoles per mol, and was not relevant for the analytical procedures used in this study. Therefore, these key hydrocarbons were excluded 2581

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

Analytical Chemistry

Article

the parent to prepare fresh picomoles per mol PSMs, since the concentrations of the NMHCs varied from 50 to 500 nmol/ mol. The older, but recently analyzed, parent PSM ALM032445 was used to provide concentrations in line with the original picomoles per mol PSMs. Table S-2 of the Supporting Information lists the concentrations of the hydrocarbons in ALM032445. Weighing Apparatus. The 5.9 L cylinder was weighed on a Mettler XP26003L top-loading balance having a capacity of 10 kg and 0.001 g sensitivity. This balance consistently results in a standard deviation of the average mass weighings from 0.001 to 0.002 g. The 29.5 L cylinders were weighed on a Mettler SR64001 top-loading floor balance with a 64 kg capacity and 0.1 g sensitivity. A minimum of five independent weighings (tare, cylinder placement, stabilization, mass recording) were made after evacuation, after addition of the NMHC mixture from the parent PSM, and after completion of diluent addition. Preparation of Fresh Picomoles Per Mol Primary Standard Mixtures. Three new “child” PSMs were prepared in August of 2011. Each cylinder was purged twice with diluent N2 (≈ 1.4 MPa). The second purge was analyzed chromatographically to confirm the absence of NMHC contamination. The cylinders were then evacuated, charged with an aliquot of the selected parent mixture, and brought to the desired nominal concentration by addition of diluent from one or two N2 source cylinders as needed to achieve the required pressure. The PSM cylinders were weighed after evacuation, after the addition of the parent standard, and after each addition of N2 diluent. When more than one diluent source was used, the mass of diluent from each source was determined from a single weighing rather than as the average of multiple independent weighings. Table S-3 of the Supporting Information list the concentrations and percent relative uncertainties of the hydrocarbons in each of the new 2011 “child” PSMs. Chromatographic Analysis. An Agilent 6890 gas chromatograph with a flame-ionization detector (GC/FID) was used for all analyses. A 30 m × 0.32 mm J&W GASPRO capillary column was used to achieve baseline separation of the compounds. The column was temperature programmed from 70 to 220 °C at 5 °C/min (no isothermal period) and held 3 min with a helium column flow of 4 mL/min. The FID was operated at 250 °C with a makeup flow of 30 mL/min helium. The standards were prepared for injection onto the column using an Entech 7100 preconcentrator, collecting 1000 mL of sample at a flow of 200 mL/min. The sample was collected in a stainless steel 0.32 cm diameter trap, packed with Tenax and glass beads, at −150 °C, transferred by heating to 150 °C to a precolumn (2.5 cm × 0.8 mm stainless steel) held at −150 °C, and then injected onto the main column via a 0.8 mm stainless steel transfer line heated to 150 °C. The Agilent 6890 Chemstation software was used to integrate the peak areas. Integration in all cases was baseline-to-baseline with the exception of a few cases that required subtraction of an impurity on the shoulder of a main peak. The Chemstation software was used to collect the data and a macro program was used to transfer it to an Excel spreadsheet for further analysis. Sample blanks, the diluent N2, used to prepare the new 2011 suite of standards, were run to confirm the absence of significant carry-over. International Comparison. The comparison discussed in this paper grew out of an earlier comparison conducted in 2008 by the European Association of National Metrology Institutes

from this study. Since ortho-xylene was included in the SRM, any stability results could be safely assumed to apply to the para and meta isomers. The intentions of this study were to determine stability and comparison among other laboratories standards and scales at picomoles per mol levels for some hydrocarbons and not, at this time, a complete list. NIST already had well-characterized standards at the (2 to 15) nmol/mol level. Therefore it was decided to use those standards used to develop the SRM to prepare picomoles per mol level standards for this study. The well-characterized pure hydrocarbons and PSMs used to prepare SRM 1800b, Eighteen Non-Methane Hydrocarbon Compounds in Nitrogen,26 were used to prepare five PSMs having NMHCs at 60 to 230 pmol/mol levels.15 These five PSMs were judged to be sufficient to make a first assessment as to the feasibility of preparing a consistent suite of picomoles per mol standards. In 2011, three new PSMs were prepared to assess the stability of the 2008 suite. Cylinders. New aluminum gas cylinders, 29.5 and 5.9 L internal volume, were obtained commercially from Air Liquide America Specialty Gases (ALASG), formerly Scott Specialty Gases, Plumsteadville, PA, and used in the preparation of the fresh picomoles per mol PSMs. The cylinder manufacturer cleaned the interior of the cylinders with a caustic etch followed by an acid wash procedure. The cylinders were conditioned with the proprietary ACULIFE IV (ALASG) chemical vapor deposition process to passivate the inner walls, which has shown to lessen reactivity. Regulators and Tubing. New two-stage, high-purity, low dead volume regulators were purchased for this project from ALASG, Plumsteadville, PA. The body of these regulators was nickel-plated brass with stainless steel diaphragms, Viton seals and seats, and brass pistons. It was specified that these regulators be processed and cleaned without using products that would contribute NMHC contamination. These regulators were used solely at NIST in their analysis of the mixtures. The stainless steel tubing and traps used in the cryogenic preconcentration system (Entech Instruments, California) used for sample introduction were treated with a fused silica ceramic coating, Silonite (Entech Instruments). Sampling lines from the regulators on the cylinders to the sampling manifold were also treated with Silonite to minimize absorption/ desorption of NMHCs. Diluent Nitrogen. Four cylinders of high-purity diluent nitrogen (N2) gas, obtained from Air Liquide America Specialty Gases, Plumsteadville, PA, were used to prepare the picomoles per mol PSMs. All materials were stated by their manufacturer to have a minimum purity of 99.9995% (excluding argon). The purity analyses were performed as per the original PSM suite development in 2008.16 The hydrocarbon impurities found in these nitrogen cylinders are given in Table S1 of the Supporting Information. Hydrocarbons detected were ethane (≈ 5 pmol/ mol) and propane (≈ 4 pmol/mol). The nitrogen sources used to prepare this current suite of standards were much cleaner than those used to prepare the original suite of 5 PSMs. “Parent” Nanomoles Per Mol Primary Standard Mixture. A parent PSM, cylinder ALM032445, containing the NMHCs at ≈250 nmol/mol was used to prepare the fresh picomoles per mol PSMs. This mixture had a history of longterm (+10 yr) stability, and recent comparison to a freshly prepared 250 nmol/mol PSM (designated PSM2502011) prepared by dilution from pure starting materials confirmed its continued stability (+16 yr). PSM2502011 was not used as 2582

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

Analytical Chemistry

Article

Table 1. List of Participants, Source of Calibration Standards, and Analytical Methods code

laboratory

INSTAAR

Institute of Artic and Alpine Research, University of Colorado, U.S.a Korean Research Institute of Standards and Science, South Korea National Center for Atmospheric Research, Colorado, U.S.a National Institute of Standards and Technology, Maryland, U.S. National Oceanic and Atmospheric Administration, Colorado, U.S.a National Physical Laboratory, U.K. University of Miami, Florida, U.S.a Van Swinden Laboratory, The Netherlands

KRISS NCAR NIST NOAA NPL U-Miami VSL

standards

analytical method

NCAR, NOAAb own, own, own, own,

gravimetricb gravimetricb,d gravimetricb gravimetricb

microadsorbent trap preconcentration followed by GC/ FID GC/FID/preconcentration GC/FID/preconcentration GC/FID/preconcentration GC/MS/preconcentration

own, gravimetricc own, gravimetricb,d own, gravimetricb

GC/FID/preconcentration GC/FID/preconcentration GC/FID/preconcentration

a

Participant measures whole air atmospheric samples. bCylinder standards at nominal 200 pmol/mol. cRelative responses used by trapping a single 4 nmol/mol standard to generate a response factor per milliliter. dNCAR and U-Miami standards are coprepared.

(EURAMET).27,28 In the 2008 EURAMET study, several nonEURAMET NMIs were also invited to participate, including NIST and Korean Research Institute of Standards and Science (KRISS). The 2008 comparison involved the measurement at nominal 2 nmol/mol for 30 hydrocarbons in a gravimetrically prepared standard mixture. The between participant agreement was generally within ±2% relative of the gravimetric values for the 15 hydrocarbons common to the current study. In 2009, NIST initiated an international comparison of one PSM from the original 2008 suite, cylinder AAL072286, at nominal 200 pmol/mol. Seven laboratories worldwide were invited to analyze this PSM. The participating laboratories are listed in Table 1 and include several NMIs, U.S. government funded agencies, and universities. Table 1 lists the laboratories, their source of calibration standards, and analytical methods. Also, those laboratories that perform regular atmospheric measurements of NMHCs are indicated. NMIs typically do not make atmospheric measurements, but much time and effort is put into the preparation of standards and the assessment of uncertainties. NCAR and University of Miami coprepare their own in-house and commercially available standards; INSTAAR uses NOAA and the NCAR/U-Miami standards, and all other laboratories prepare their own gravimetric standards. Table S-4 of the Supporting Information gives the column, preconcentration trapping material used, temperature, and sample volume trapped by each laboratory.

analysis of peak heights. NOAA scientists have since independently determined that propane is more accurately quantified on their GC/MS with peak areas; back calculating, the mixing ratio assigned by NOAA to AAL072286 based on peak areas is 205.2 +/− 6.2 ppt. However, as the paper was in press, we chose not to change the value and figures throughout, using the original value, but indicating here that current NOAA optimization now shows very good agreement. The 2011 NIST values were determined using all 3 new “child” PSMs prepared in August 2011. NIST January 2013 values were determined using only one of the new August 2011 PSMs, AAL073109. Since this PSM was already ∼1.5 years old in January 2013, stability of the alkenes needed to be checked. Data from analysis in August 2011 and January 2013 were used. Peak areas for the alkene were divided by that of the corresponding alkane; propene/propane, isobutene/n-butane, and 1-pentene/n-pentane. The average ratio (n = 4 in each case) for the August 2011 data was compared to that from January 2013. Figure S-1 of the Supporting Information shows that the ratios were not statistically different within the standard deviation for any of the alkene/alkane ratio combinations. Therefore, the original gravimetric concentrations were used to determine concentrations in PSM AAL072286 for the NIST January 2013 data set. (Interesting note: the alkenes in PSM AAL073109 appear to be stable, unlike PSM AAL072286. Further stability studies will be done in the future to continue to track AAL073109.) Stability Evaluation. The agreement between each measured value, x ± U(x), and the NIST gravimetric value and its k = 2 expanded uncertainty, xgrav ± U(xgrav), was calculated as the percent difference

■

RESULTS AND DISCUSSION The nominal 200 pmol/mol NMHC in nitrogen gravimetric standard PSM AAL072286 was prepared at NIST in 2008 and circulated to participating laboratories. Each participant reported their mole fraction, x, and its k = 2 expanded uncertainty estimate, U(x), to NIST. When standard uncertainties, u(x), are associated with a “large” number of degrees of freedom, expanded uncertainties estimated as U(x) = 2u(x) define an interval, x ± U(x), that is expected to include the true value of the measurand with about a 95% level of confidence.29 Each participant was asked to analyze as many NMHC components of the mixture as possible, concentrating on those that they typically prepare in their own reference standards or measure in the atmosphere. In addition to the January 2008 gravimetric preparation values and their expanded uncertainties, in 2011 and again in 2013, NIST evaluated all of the NMHC mixture components against PSMs prepared in August 2011. Table 2 lists all of the reported results. We note here that the value for propane from NOAA was derived in 2011 from an

D% = 100

x − xgrav xgrav

; U (D%) = 100

U (x)2 + U (xgrav)2 xgrav

where the “D%” notation is that used in the ISO/IEC 17043 international standard.30 The percent difference between xgrav and itself, D%grav, is by definition zero; its expanded uncertainty, U(D%grav), is 100U(xgrav)/xgrav. The D% ± U(D%) were analyzed as functions of the analysis date for each of the NMHCs. Alkanes and Aromatics. Figure 1 displays the D% ± U(D%) gravimetric and analytical results for propane, n-butane, and npentane. Figure S-2 of the Supporting Information displays the results for the remaining 12 alkanes and aromatics. For all 15 alkane and aromatic components of the standard, the NIST analytical results at 3.5 and 5.1 years are largely contained within the ± U(D%grav) interval. This demonstrates 2583

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

2584

1.2 1.0 1.0 1.2 1.2 1.2 1.2 1.2 1.4 1.2 1.2 1.2 1.0 1.0 1.0 1.2 1.0 1.0

197.0 204.3 187.8 210.1 198.9 168.4 189.9 194.5 166.9 196.0 193.6 192.4 195.9 194.5 188.1 195.6 187.3 182.9

15.6 12.0 11.0 12.4 21.4 16.8 11.2 11.4 9.8 11.6 21.6 11.2 11.4 11.4 11.0 19.4 11.0 23.6

U(x)c 10 10 10 10 10 8 10 10 8 10 10 20 10 18 18 20

198

U(x)c

191 206 190 213 204 162 187 192 158 188 184 198 178 180 185

x

Aug 2009

VSL

218.4

202.3 178.2 205.2 189.6 145.8 186.0 185.1 157.9 189.1 192.6 179.3 193.0 199.4 167.1

x

33.0

20.0 16.0 16.0 16.8 10.8 13.4 14.8 14.6 15.2 16.4 22.0 15.0 18.0 22.0

U(x)c

Mar 2010

KRISS

10.2 11.8 6.6 10.8 11.6 3.6 5.6 4.8 3.6

1.0

197.1 192.2 194.3 215.8

205.9

U(x)c

207.6 212.7 186.3 207.5 198.2

x

Feb 2011

INSTAAR

195.8

9.4

7.6 11.3

6.4

206.0 197.6 197.9

12.5

U(x)c

234.4d

x

Jun 2011

NOAA

204.2 205.8 185.9 203.3 197.2 153.6 188.7 193.4 154.0 189.2 190.7 189.8 193.4 193.6 190.8 194.1 194.3 188.6

x 3.2 3.6 1.6 2.6 2.6 3.2 2.4 1.8 4.6 2.6 2.5 1.4 1.2 1.2 1.4 1.6 1.6 1.4

U(x)c

Aug 2011

NISTb

212 217 186 214 212 150 200 204 151 200 199 210 204 201 210 200 197 181

x 15 12 22 7 7 8 8 7 10 8 6 9 7 7 27 11 12 24

U(x)c

Jun 2012

NCAR

222 237 194 212 210 153 195 204 147 202 193 211 201 195 183 195 183 194

x 72 56 24 17 12 15 16 24 14 10 8 30 5 5 10 8 8 14

U(x)c

Aug 2012

U-Miami

202.5 206.5 185.9 202.5 196.6 153.4 187.6 193.5 152.5 188.7 189.4 189.6 193.5 192.8 191.0 192.8 192.8 188.4

x

2.0 2.4 2.0 1.4 2.0 1.6 1.2 1.2 2.0 2.4 2.2 1.5 2.4 3.2 2.6 3.7 1.6 5.4

U(x)c

Jan 2013

NISTc

a Gravimetric preparation. bAnalytical values based on standards prepared in August 2011. ck = 2 expanded uncertainty. dThe value for propane from NOAA was derived in 2011 from an analysis of peak heights. NOAA scientists have since independently determined that propane is more accurately quantified with GC/MS with peak areas; the mixing ratio assigned by NOAA to AAL072286 based on peak areas is 205.2 +/− 6.2 ppt.

203.3 206.9 195.9 202.6 197.5 184.0 188.1 193.1 190.2 189.0 189.4 189.3 193.1 192.1 190.1 192.9 194.0 187.4

ethane propane propene i-butane n-butane i-butene i-pentane n-pentane 1-pentene n-hexane n-heptane benzene i-octane n-octane toluene nonane o-xylene decane

x

Sep 2008

U(x)c

Jan 2008

x

NMHC

NPL

NISTa

Table 2. Reported NMHC Values for PSM AAL072286, Picomoles per Mol

Analytical Chemistry Article

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

Analytical Chemistry

Article

Figure 1. Measured values of propane, n-butane, and n-pentane in PSM AAL072286. The horizontal axis plots the time of each analysis in years from the preparation of the PSM. The vertical axis plots the percent mole-fraction difference in the measured values relative to the gravimetric preparative values, D% = 100(x − xgrav)/xgrav. The black horizontal lines bound the 95% level of confidence interval for the preparative values, ± U(D%grav)/xgrav. The ○ symbols and associated vertical bars represent the D% ± U(D%).

Figure 2. Measured values of propene, i-butene, and 1-pentene in PSM AAL072286. Format as in Figure 1, with the addition of empirical models for the decrease in measured mole fraction with time: D% = a(e−bt −1). The solid curves represent the optimized model for each of the alkenes; the dashed lines bound an approximate 95% level of confidence region about the optimum. The values and approximate standard uncertainties for the model parameters are given in the top left of each panel.

that the mole fraction of these NMHCs did not change over the five-year course of the study, the environmental changes intrinsic to long-distance shipping, and the drop in cylinder pressure from the multiple analyses. The large majority of the participant’s D% ± U(D%) intervals include zero, demonstrating the trueness (i.e., absence of bias) of the xgrav for these compounds. Alkenes. Figure 2 displays the D% ± U(D%) for propene, ibutene, and 1-pentene. Unlike the alkane and aromatic components, the mole fractions of these components decreased to an apparent plateau during the first three years after preparation. These results are in contradiction to Pollman et al.,17 which showed increases of alkenes in glass vessels over time. One might assume the aluminum cylinder wall treatment, Aculife IV, leaves a 100% surface application. The authors assume, as a possible explanation for loss of the alkenes, that this is not the case, and that there are active sites on the cylinder walls. (The aluminum cylinder walls, upon inspection, while appearing smooth and actually rough on a microscale, possibly result in active sites from incomplete treatment.) Thus, the authors speculate that these alkenes became bound to incompletely deactivated sites on the cylinder walls. While too few results are available during the period of greatest decrease to confidently estimate the rate of decrease,

the percent difference values for the three alkene components are well-described by the empirical model D%̂ = a(e−bt − 1)

where t is the measurement time in years from the cylinder’s preparation, a is the asymptotic plateau, b is related to the rate of decrease, and the symbol ″″̂ or “hat” indicates a predicted value. “Best fit” models were defined using standard spreadsheet optimization software to iteratively adjust the a and b parameters to minimize ⎛ D̂ − D ⎞2 ∑ ⎜ % %⎟ ⎝ U (D%) ⎠

where the summation is over all of the reported analytical results for each alkene. The NIST results at 5.1 years were not included in the optimization so that every participant in the study would have equal footing in the analysis (i.e., one analytical data point per participant). The model uncertainties were evaluated using a purpose-built spreadsheet macro system implementation of a parametric bootstrap Monte Carlo (PBMC) technique.31 Each of the PBMC analyses involved optimizing several thousand sets of “pseudo-data”, where each reported x was replaced by a 2585

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

Analytical Chemistry

Article

Table 3. Reported and Corrected Alkene Values for PSM AAL072286 i-butene, pmol/mol

propene (pmol/mol)

a

1-pentene (pmol/mol)

lab

date

years

x

U(x)a

x′

U(x′)b

x

U(x)a

x′

U(x′)b

x

U(x)a

x′

U(x′)b

NISTGrav NPL VSL KRISS INSTAAR NIST NCAR UMiami NIST

Jan 2008 Sept 2008 Aug 2009 March 2010 Feb 2011 Aug 2011 June 2012 Aug 2012 Feb 2013

0.0 0.7 1.6 2.2 3.1 3.6 4.5 4.6 5.1

195.9 187.8 190.0 178.2 186.3 185.9 186.0 194.0 185.9

1.0 11 10 20 3.3 1.6 12 24 2.1

195.9 194.5 199.3 187.9 196.2 195.9 196.0 204.0 195.9

− 6.8 8.9 26 6.1 0.57 11 30 3.0

184.0 168.4 162.0 145.8

1.2 17 8.0 16

184.0 182.3 185.7 172.8

− 15 5.7 24

190.2 166.9 158.0 157.9

1.4 9.8 8.0 16

190.2 188.9 190.8 193.1

− 6.5 6.0 16

153.6 150.0 153.0 153.4

2.0 8.0 15 1.6

184.2 181.5 184.5 185.1

0.9 8.0 15 5.4

154.0 151.0 147.0 152.5

4.6 10 14 2.0

191.0 188.2 184.2 189.7

3.1 9.5 16 4.9

Reported k = 2 expanded uncertainty. b95th percentile of the deviations of the parametric bootstrap random draws from the optimized models.

Figure 3. Measurement characteristics of participants across all NMHCs. The panel to the left summarizes each participants’ analytical measurements as percent differences normalized to the gravimetric value, D% = 100(x − xgrav)/xgrav. The panel to the right summarizes their measurements as normalized to the combined standard uncertainties, ζ = 2(x − xgrav)/√[U(x)2+U(xgrav)2]. The horizontal axes display the expected bias of the normalized measurements as estimated by the median of all measurements reported by each participant. The vertical axes display the expected variability of the biases as estimated by the MADe of the normalized measurements. The concentric semicircles are for visual reference.

random draw from a normal distribution centered at x with standard deviation of U(x)/2. Each set of pseudodata was then analyzed in exactly the same manner as the original data, the values of interest for each optimized model were recorded, and the uncertainty in each parameter of interest estimated from the variability of the recorded values. Each panel of Figure 2 displays the optimized model function for one of the alkene NMHCs and its approximate 95% uncertainty interval estimated from the 2.5 and 97.5 percentiles of PBMC results. The panels also list the optimum values of a and b and their standard uncertainties, where the uncertainties are estimated as the standard deviation of the PBMC values. While the distributions of the PBMC values for these parameters are roughly symmetric in the near region of the optimum values, the tails of the distributions are skewed toward larger values and the 95% level of confidence intervals are quite asymmetric about the optima. Using the optimized a and b parameters, the participantreported values can be corrected to what they “would have been” absent the observed time-dependent decrease within the PSM:

x̂ = x +

xgrav 100

U (x)̂ =

xgrav 100

P95(|D%̂ − D%|)

where P95 is the function “find the 95th percentile of the set of values”. Table 3 lists the reported and corrected values for the three alkenes. Given that the shape of each of the PBMC model functions is driven by the U(x), it is not surprising that the U(x̂) are quite similar in value to the U(x). For the rest of our analyses involving the alkenes, we substitute for the x̂ but retain the U(x) unchanged. Measurement Performance Summaries. In addition to helping evaluate the stability of the various NMHCs and the trueness of the NIST gravimetric values, the international study of PSM AAL072286 was designed to explore the performance characteristics of the measurement process used by the participants for the various NMHCs. To the extent that the participants in this study represent the NMHC measurement community, the performance characteristics of their measurements can be used to estimate the expected bias and variability for that community’s NMHCs measurements. While combining even well-normalized representations of measurements (such as the D% percent relative differences) of different measurands is of doubtful utility when the measurands are chemically dissimilar or the measurements are made under different conditions,32 the measurements in this study are similar in kind and result from similar processes. To avoid undue influence of atypical results, we characterize combined results with robust summary statistics: the median to estimate the expected value of the normalized representations and the “MADe” to estimate their variability. The MADe is the median

D%̂

Approximate uncertainty intervals for the corrected values can be estimated from the distribution of the absolute differences between the pseudovalues and their optimized model: 2586

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

Analytical Chemistry

Article

Figure 4. Measurement characteristics of NMHCs across all participants. Format as in Figure 3, except that the panels summarize the measurement results for each of the NMHCs, as reported by all participants. The horizontal axes display the expected bias of the normalized measurements for the individual NMHCs as estimated by the median, and the vertical axes display the expected variability of those biases as estimated by the MADe.

signaling both that they somewhat variably underestimate their uncertainties. Figure S-3 of the Supporting Information displays D%- and ζbased summaries for the eight participants summarized over the twelve alkane, three aromatic, and three alkene components of the PSM. While the data are limited, there are striking differences in the patterns between the three panels. This suggests that several participants may have small but systematic chemistry-specific biases in their measurement processes. The alkene panels are based on the corrected data previously discussed. The median D% and MADe for five of the seven participants measuring the alkenes are well within the range from −3% to +3%. This suggests that on average, the NMHC measurements made by these five are relatively unbiased, with a 95% level of confidence range of biases among different NMHCs of about 0% + 2% × 1 = 2%. In consideration of all seven participants, median D% are within the range from −6% to +6%, with the MADe estimates as large as about 5%. The median and MADe ζ values for seven participants corrected alkene data, using their reported Ux, are well within the smallest “score 2” guide semicircle and actually in the range from −1 to +1. This suggests that, on average, these participants consistently provide realistic evaluations of the uncertainties in their measurement processes. Figure 4 displays the D%- and ζ-based summaries for the 18 NMHCs, where the two representations are summarized across up to eight participants. The median D% are all in the range from −2% to about 4%, with the median (and MADe) greatest for benzene. This may reflect relative inexperience of several participants with aromatic NMHCs. The somewhat high MADe values for ethane and propane may reflect difficulties in cryogenically trapping very volatile compounds. The median and MADe ζ values for all but five of the NMHCs are within the “score 2” guide semicircle, indicating that most of the time the uncertainties reported for NMHC measurements are appropriate. However, the five exceptions include benzene and toluene, which again suggests that there may be underappreciated measurement issues and/or preparation of the calibrants, with the aromatics. Figure S-4 of the Supporting Information displays the D%- and ζ-based summaries for the 18 NMHCs, breaking them into NMIs and non-NMIs. Comparison to Previous Studies. Previous intercomparisons, NOMHICE,10,11 AMOHA,13 GAW,12 and EURAMET 88627 used either synthetic mixtures traceable to an NMI or gravimetrically prepared standards from an NMI. While no

absolute deviation of a set of values from their median multiplied by a scale factor so that it estimates the standard deviation.33 The median and MADe summaries of D% have chemically interpretable units of relative percent that can be meaningfully evaluated against trueness and precision requirements. However, the D% representations do not address whether expanded uncertainties have been correctly estimated. This lack can be addressed with median and MADe summaries of zetascores xi − xgrav ξi = 2

( ) +( U (xi) 2

U (xgrav) 2

2

)

where bias is normalized to the combined uncertainty of the bias.30 While these ζ-based summaries only provide statistically interpretable units of standard uncertainty, they can be evaluated against normal-distribution expectations: it is very unlikely that |ζ| much greater than 2 will be observed when the U(x) have been appropriately estimated. (To be consistent with one data set per participating laboratory, only the August 2011 NIST data set is included in any ζ and D% calculations.) Figure 3 displays the robust summary statistics for D% and ζ for the eight participants, where the two normalized representations are summarized across up to 18 NMHCs. The median D% for the four NMI participants is within the range from −2% to +2%, with MADe estimates as large as about 4%. This suggests that on average the NMHC measurements made by NMIs are relatively unbiased (0% average bias), with a 95% level of confidence range of biases among different NMHCs of about 8% (0% bias + 2σU95% × 4% MADe). The measurements reported by the four non-NMI participants appear to be systematically slightly positively biased by about 4% but with MADe estimates that are about the same as those for the NMIs. This could arise if the calibrants used by these participants delivered slightly less than the expected values of the analytes. In any case, on average, the NMHC measurements reported by the non-NMIs can be expected with about a 95% level of confidence to be within about 12% of the true value. The median and MADe ζ values for five participants, including all of the NMIs, are within the smallest “score 2” guide semicircle. This suggests that, on average, these participants consistently provide realistic evaluations of the uncertainties in their measurement processes. The median and MADe ζ values for the other participants are greater than 2, 2587

dx.doi.org/10.1021/ac403761u | Anal. Chem. 2014, 86, 2580−2589

Analytical Chemistry

Article

NMIs participated directly in NOMHICE, AMOHA, and GAW comparisons at the nominal (1 to 15) nmol/mol level, they provided standards for the studies or the comparison mixture was traceable to an NMI. We can still compare results from those studies and this current study at nominal 200 pmol/mol. The EURAMET 886 comparison at nominal (1 to 10) nmol/ mol involved mostly NMIs. In the NOMHICE Task 3 study at (1−25) nmol/mol,10 the average difference from the reference value for each hydrocarbon, considering the 11 common to this current work, averaged across all participants was less than 10%. However, the standard deviations of those averages ranged from 10% to 60%. The reported AMOHA study involved 3 different comparisons.13 In the first intercomparison, there were 4 hydrocarbons common to this current study, where the largest percentage differences of individual laboratory results were ethane (24%) and benzene (50.2%). The averaged differences were 1.4% (ethane), 0.36% (propane), 0.32% (n-butane), and 3.4% (benzene). In the second experiment, the majority of participants agreed within 10% of the reference value for common hydrocarbons to this current study, with 5% of the total reported values considering outliers and ranging from 30% to 183% (n-pentane). The differences observed in the third comparison were generally similar to the second, with some improvements. Outliers improved by about 1% with the average amount of the outliers falling from 64% (2nd) to 45% (3rd). Removal of outliers in the third comparison resulted in 2% to 10% mean absolute differences between participants. Data from the GAW12 comparison showed that of the reported values, 43% were within 5% and 69% were within 10% of the reference value for common hydrocarbons to this current work. Most of the outliers ranged from (11 to 80)% with four extreme values well over 100% different from the reference value. Results of the EURAMET 886 comparison27 showed agreement among the participants within 2% for 78% of the reported values and within 5% for 85% for common hydrocarbons to this current study. One laboratory in the study was consistently different from the reference value for all reported hydrocarbons from 58% to 66%. Considering only the NMIs in this current study, the common hydrocarbons (disregarding alkenes) show agreement with the reference values to within 5% for 90% of the reported values, with the largest outliers being about 12% for toluene and 13% for oxylene. This shows very good consistency between nanomol per mol and picomoles per mol studies for the NMIs. Assessing all participant data in this current study, the largest difference between reported values compared to the reference value was 14.6% for propane and 9.2% ethane, with the aromatics largest differences being ≤ (12 to 14)%. All other hydrocarbons showed differences of