Conversion of NOAA atmospheric dry air CH₄ mole fractions to a gravimetrically prepared standard scale

2.1. Standards Preparation

[5] Most of the materials used for preparation of the gravimetric standards were described by Novelli et al. [1991]. Briefly, standards were prepared in 5.9 L aluminum cylinders (Luxfer, Riverside, California). The cylinders were fitted with brass, packless diaphragm valves, and then their internal surfaces were cleaned with a proprietary process (Scott-Marrin, Inc., Riverside, California). Our prior experience with such aluminum cylinders indicated that they are capable of storing CH₄ in dry air mixtures with extremely good stability, especially at near-ambient atmospheric levels of CH₄. Ultrapure air (Scott-Marrin, Inc.) was used as the diluent gas. At the time that the standards were made (1993–1994), we verified that the CH₄ mole fraction in each cylinder of diluent gas was at or below our detection limit of ∼5 nmol mol⁻¹. Since then our detection limit for CH₄ in air has improved to ∼1–2 nmol mol⁻¹, and subsequent analysis indicates that the diluent gas contained ∼5 nmol mol⁻¹ CH₄ (see below). Aliquots from the cylinder of pure methane used to prepare the standards were analyzed for the most likely impurities (N₂, O₂, and H₂O vapor). N₂ and O₂ were below our detection limits of ∼100 ppm, and H₂O vapor was <10 ppm. These levels of impurities would have a negligible affect on calculated CH₄ mole fractions.

[6] The CH₄ in dry air standards were prepared on a vacuum manifold [see Novelli et al., 1991, Figure 1b] using three techniques to insure that the accuracy of the gravimetric scale was not biased by the preparation technique. All standards were prepared from a single parent with a CH₄ in dry air mole fraction of 1.0654%. This parent mixture was prepared with the second technique described below. Its mole fraction was consistent with a series of eight standards with a range of CH₄ mole fractions from 1.5406 to 0.06549% in zero air.

[7] In the first technique a predetermined pressure of the high-concentration CH₄ in air mixture is added directly from its cylinder through the manifold to another 5.9 L cylinder that was previously evacuated to ≤4 Pa and weighed relative to another nearly identical cylinder (to minimize buoyancy effects) on a Sauter platform balance. This cylinder is weighed again on the platform balance to determine the number of moles of CH₄ added. It is reconnected to the manifold for addition of the final pressure of zero air and then weighed a final time. Maximum dilution at each step is about a factor of 100. Eight standards were prepared with three successive dilution steps and one with two steps.

[8] In the second technique a 10 mL stainless steel tube fitted with a valve is evacuated and then filled to a predetermined pressure with a gas mixture from a 5.9 L cylinder containing a high CH₄ concentration. The tube is weighed five times relative to another, similar tube to minimize buoyancy effects on a (Mettler model AE163) balance and then reattached to the manifold. The contents of the tube are expanded into a 5.9 L cylinder that was previously evacuated to ≤4 Pa and weighed. After closing the valve on the tube the remaining high-concentration CH₄ mixture is repeatedly flushed from the manifold into the cylinder with zero air. The 10 mL stainless steel tube is then weighed relative to the “blank” tube five times. The total number of moles of CH₄ and zero air added to the cylinder is determined by the difference in weight of the tube before and after its contents are expanded into the cylinder and the fraction of the high-concentration mixture that is CH₄. The cylinder is then pressurized to its final pressure, usually ∼9.65 MPa, and weighed again to determine the number of moles of zero air added. The CH₄ in dry air mole fraction is calculated as the number of moles of CH₄ divided by the total number of moles (CH₄ plus zero air). This technique is sensitive to the effects of small leaks, particularly when the manifold is being flushed with zero air.

[9] The third technique is a variant of the second that does not require reweighing of the sample tube after expansion of high-concentration CH₄ in dry air mixture into the cylinder. This is achieved by including the tube in the multiple flush/expansion steps to flush residual CH₄ from the tube and manifold into the cylinder. It is assumed that essentially all CH₄ from the high-concentration mixture is transferred to the cylinder. Tests with CFC-12 using 12 flush/expansion cycles suggest that 99.981 ± 0.018% of the original high-concentration mixture is transferred to the cylinder. This would introduce a potential error of 0.3 nmol mol⁻¹ at 1750 nmol mol⁻¹ (i.e., background ambient CH₄ abundance), an error that is small compared with other sources of error. Standards prepared with the two tube techniques were single-step dilutions starting with the 1.0654% (mole fraction) mixture.

[10] Prior to analysis the contents of the cylinders were mixed by rolling them on the floor. Cylinders then stood idle for at least 3 days between standard preparation and analysis.

[11] Eighteen CH₄ in dry air standards were prepared, and they are summarized in Table 2 by cylinder identification number, preparation method, CH₄ value determined from the gravimetric method, and ratios of the prepared value to the value determined on the CMDL83 scale. In addition to the 16 standards that define the new scale, two others were prepared at nominal values of 30 and 100 nmol mol⁻¹ to test the analytical techniques near the low end of the scale.

2.2. Analytical Methods and Procedures

[12] The standards prepared with the gravimetric technique were analyzed by a gas chromatograph (GC) with a flame ionization detector (FID). To determine the relationship between our old scale and the gravimetric one, each gravimetrically prepared standard (“grav”) was treated as an unknown and analyzed relative to one of our existing standards, in effect normalizing measurements of the gravs to the CMDL83 CH₄ scale. The analysis sequence alternated between aliquots of standard and grav, always starting and ending with standard so that each aliquot of grav was bracketed by aliquots of standard. Details of the experimental procedures used for CH₄ measurements at CMDL have been described previously [Dlugokencky et al., 1994, 1995]. The gravimetric CH₄ scale was evaluated over the past 10 years on four analytical systems, but most of the analysis presented here is based on our most recent analytical system. Earlier systems, as described by Steele et al. [1987] and Dlugokencky et al. [1994], differed from newer systems in chromatographic scheme, system control and data acquisition, and chromatogram peak integration algorithms. Our current analytical system for standard calibrations is based on a Hewlett-Packard 6890 gas chromatograph with flame ionization detector, multiposition stream selection valves for sample selection, a two-position, six-port valve for injection of gas onto the GC column, and custom software for valve control and chromatogram peak integration. A single, 3.2 mm OD, 3 m long column packed with 80/100 mesh HayeSep Q at 40°C is used for CH₄ separation. Sample loop volume is 5 mL. Carrier gas is 99.9995% N₂, further purified by passing it through a heated metal-oxide catalyst (Trace Analytical CAT-1) and a 50 cm long, 2.1 cm ID trap filled with a 50/50 mixture of 13X (8–12 mesh) and 5A (0.16 cm) molecular sieve pellets. Carrier gas column head pressure was set with one of the GC's electronic pressure controllers to give a flow rate of 36 mL min⁻¹. The flame is fueled by 40 mL min⁻¹ H₂ (99.999%) and supported by 250 mL min⁻¹ 40% O₂ (99.98%) in N₂. The analysis sequence was as follows. The sample loop was flushed with standard or unknown gas at 100 mL min⁻¹ for 30 s, the flow was stopped by switching the stream selection valve to an “off” position, and 15 s were allowed for the sample loop contents to relax to ambient pressure before the loop's contents were injected by the gas sample valve. Sample and standard gases were introduced to the system alternately from a four-port stream selection valve, of which two ports were plugged and used as off positions. Air was injected by actuating a six-port sample valve, and CH₄ was separated with a retention time of ∼56 s (see Figure 1a). At the end of the run, the six-port sample valve was switched to “load” position in preparation for the next aliquot. Total run time for each aliquot analyzed was 2 min 45 s. The most significant difference between this system and older designs was that the single column yielded narrower peaks that were easier to detect and integrate than the old two-column (silica gel precolumn and molecular sieve 5A main column) systems. We insured that results from the single porous polymer column technique agreed with those from the two-column systems, within experimental error. The higher resolution of the HP35900E analog to digital (A/D) converters (24-bit compared with 16-bit for the stand-alone integrators) also contributed to better repeatability.

[13] Each aliquot from the gravimetrically prepared standards (treated here as an unknown) was bracketed by aliquots of standard gas. The standard was one of the 15 existing CMDL primary CH₄ standards mentioned earlier, described in detail by Dlugokencky et al. [1994]. Methane dry air mole fractions were determined for each aliquot of unknown as the ratio of the unknown peak response to the average peak response of the bracketing reference gas aliquots multiplied by the CH₄ mole fraction assigned to the reference gas. Peak heights and areas have been used as the quantitative measure for instrument response [see Dlugokencky et al., 1994], depending on the integration tool used. When we employed stand-alone integrators (as on our early analytical systems), we used peak heights because we obtained a factor of 2–3 better precision than with peak area. Height is a legitimate substitute for area when chromatographic peak shapes are near-Gaussian and retention times do not vary, as in our case. Recent versions of our analytical systems used an HP35900E A/D converter with an integration algorithm running on a Unix-based computer (developed by P. Salameh, Scripps Institution of Oceanography, 1993 [Prinn et al., 2000]); with these systems, height and area response give about the same precision, so we use area. Each determination of the CH₄ mole fraction for a cylinder was based on the average of 20 replicates of unknown. Repeatability (expressed as 1σ) was typically ≤1.5 nmol mol⁻¹ for ambient mole fractions (∼1700 nmol mol⁻¹), which corresponds to a relative precision of <0.1%. Our integration algorithm uses a user-defined “peak threshold” to determine the start and end of peaks on the basis of the rate of change of the chromatogram signal (i.e., the slope). The choice of peak threshold was critical for good system performance. Large thresholds can accommodate a noisy baseline and day-to-day variations in chromatographic shape, but they result in nonlinear response over large ranges of CH₄ mole fraction; small thresholds ensure a linear response, but they can result in poorer repeatability and potential integration of erroneous peaks. Through a combination of laboratory experiments and simulations, we determined that a peak threshold of 100 counts s⁻¹ (see Figure 1b) maintains good repeatability and insures a linear response over the range of CH₄ mole fractions of interest.

[14] Each gravimetrically prepared standard with a CH₄ mole fraction more than ±20% from ambient had a dedicated regulator which was always “conditioned” with the air from the cylinder. Immediately prior to each analysis the regulators of grav and standard gas cylinders were pressurized to cylinder pressure by opening the cylinder valve completely. Then the cylinder valve was closed, and the pressure bled from the regulator through the low-pressure side. This process was repeated four times. Cylinders were stored with the regulators pressurized and the cylinder valves closed.

3.1. Evaluation of Gravimetric Scale

[15] The gravimetric scale was evaluated by comparing each mixture with one of the primary standards in the CMDL83 scale. Through intensive intercomparison of these primary standards with themselves and other primaries we can be certain that their CH₄ mole fractions are not changing over time. We have looked closely at the calibration histories of our primary standards and estimated that our CH₄ mole fraction scale is stable to 0 ± 0.1 (2σ) nmol mol⁻¹ yr⁻¹ [Dlugokencky et al., 1994]. Based on the standard error of the mean for multiple measurements of each cylinder (i.e., σ/√n, where σ is standard deviation, n is the number of determinations, and each determination is the mean of 20 aliquots), the uncertainty with which each grav is tied to the CMDL83 scale is ±0.3 nmol mol⁻¹. In Figure 2a, CH₄ mole fractions determined for the gravimetrically prepared mixtures against the CMDL83 scale during 2003 and 2004 on two GC systems are plotted on the abscissa, and the prepared gravimetric values are plotted on the ordinate. One of the GCs utilized the two-column chromatographic scheme of earlier systems, and the other utilized the newer single-column scheme. Each plotted “point” is actually five overlapping determinations. A first-order polynomial has been fitted to the measurements with a least squares method: χ_grav = (1.0124 ± 0.0007)χ_cmdl − (4.8 ± 1.1), where χ_grav is the mole fraction calculated from the gravimetric technique for each grav and χ_cmdl is the CH₄ mole fraction determined versus the CMDL83 primary scale. Uncertainties are 1σ. Only those gravimetric standards with mole fractions in the nominal range 300–2600 nmol mol⁻¹ were used in the curve fit. The gravimetrically prepared scale therefore gives CH₄ mole fractions that are 1.24% greater (or 21 nmol mol⁻¹ at 1700 nmol mol⁻¹) than the CMDL83 scale. A small difference in slope was observed between the gravs prepared by direct dilution (technique 1) in a cylinder (1.2% greater than CMDL83) compared with those produced with the tube dilution (techniques 2 and 3) methods (1.3% greater than CMDL83). Since none of the techniques is preferred over the others, results from all techniques are used.

3.2. CH₄ Impurity in Diluent Gas

[16] In Figure 3a, values assigned to each gravimetric standard are plotted against peak responses for each grav normalized to a cylinder of dry natural air (cylinder ID CC64034 with CH₄ = 1785.0 nmol mol⁻¹ on the CMDL83 scale). Each plotted point consists of four determinations. The intercept, −5.5 ± 1.1 nmol mol⁻¹, is within the uncertainty of that found in Figure 2a. The small intercepts in Figures 2a and 3a (∼−5 nmol mol⁻¹) are significant. At the time the gravs were prepared, all cylinders of zero air were checked for CH₄ contamination, but our detection limit (∼5 nmol mol⁻¹) was not good enough to quantify small CH₄ impurities in zero air. Subsequent improvements to our chromatography, data acquisition hardware, and integration software have allowed lower detection limits, on order of 1–2 nmol mol⁻¹. A tank of Scott-Marrin ultrapure air was analyzed for CH₄ on our newest analytical system, and 4 ± 1 nmol mol⁻¹ was obtained from analysis of multiple aliquots. We also modeled our current CH₄ chromatograms and scaled them from ambient CH₄ mole fractions down to 5 nmol mol⁻¹ to determine if there was integration bias for small peaks. The chromatograms were modeled as exponentially modified Gaussian peaks (i.e., Gaussian peaks with tailing) with random noise comparable in magnitude to real peaks (∼10 μV, peak to peak). The average and standard deviation of 20 chromatograms, integrated with the same parameters used to obtain 4 nmol mol⁻¹ in recent cylinders of zero air, were 4.8 ± 0.7 nmol mol⁻¹. In an independent test, one of us (R. C. Myers) used a commercially available optical technique to determine the CH₄ mole fraction in seven cylinders of Scott-Marrin ultrapure air. One of the cylinders was below the detection limit for CH₄, and the other six ranged between 3 and 6 nmol mol⁻¹. None of these recent tests involved samples from the batches of zero air used to prepare the gravimetric standards; these are no longer available. It is possible that since our gravs were prepared from different batches of zero air, they contain varying contributions of CH₄ from the diluent gas. There is a hint of this in the ratios compiled in Table 2 (and the circles plotted in Figure 2b); standards prepared with the manifold and tube flush method (T2) were made months later than those made with the direct dilution and manifold-only flush techniques, and their ratios, prepared value to value determined on the CMDL83 scale, are high. In fact, uncorrected ratios calculated for standards prepared with the T2 method are comparable to corrected ratios for the other techniques, suggesting that for these standards, no correction for CH₄ in the zero gas is necessary. Errors resulting from the T2 technique from CH₄ left in the manifold and tube would result in ratios that are low, opposite of what was observed, and tests with CFC-12 suggest that these potential errors are small. The evidence for varying amounts of CH₄ in the zero gas is weak and affects only four standards near the ambient portion of the range, so we use the average correction of 5 nmol mol⁻¹ determined by the least squares fit in Figure 2a.

3.3. FID Linearity

[17] The linearity of the combined FID and gas-handling system can be tested with the gravimetrically prepared scale if it is assumed that the gravimetric scale is itself linear (i.e., the scale is without bias related to CH₄ abundance). We feel that with the analysis in section 3.2, this is a safe assumption. Nonlinearities in the detector and gas-handling system would reveal themselves as curvature in Figure 3a, but because the range in CH₄ mole fraction is large, small curvature would be difficult to observe. A more sensitive method for identifying detector nonlinearities is shown in Figure 3b. Here the normalized peak response factor (i.e., (grav area/CC64034 area)/grav mole fraction) is plotted versus normalized response (grav area/CC64034 area). Diamonds are the case that assumes no CH₄ in the zero air, and circles show the case where 5 nmol mol⁻¹ has been added to each grav value. For small relative peak responses (i.e., for small CH₄ mole fractions) the diamonds deviate substantially from constant normalized response/nmol mol⁻¹ CH₄. The nonconstant response/nmol mol⁻¹ indicates either a nonlinear response to the assigned CH₄ mole fractions or an error in the assigned mole fractions. As discussed above, there is strong evidence that the zero air was contaminated with ∼5 nmol mol⁻¹ CH₄. When this 5 nmol mol⁻¹ is added to the assigned value for each grav, the normalized response per mole remains somewhat constant across the entire scale. This indicates that within the experimental uncertainty, a 5 nmol mol⁻¹ correction to the gravs is consistent with a linear FID response.

3.4. Stability of the Gravimetrically Prepared Mixtures

[18] Changing trace gas composition over time in cylinders is common for species such as CO [Novelli et al., 1991]. More than 20 years of experience working with CH₄ standards contained in high-pressure, passivated aluminum cylinders has shown that so long as the air is dry, the CH₄ mole fraction in the cylinder is constant over time. Measurements of the gravimetrically prepared standards against the CMDL scale were plotted as a function of time (1994–2004) and fitted using a weighted least squares technique. Over the nominal range 300–2600 nmol mol⁻¹, slopes ranged from −0.2 to 0.1 nmol mol⁻¹ yr⁻¹, but none were significant at the 95% confidence level.

3.5. Comparison With Other CH₄ Standard Scales

[19] The gravimetrically prepared scale (NOAA04) has not yet been compared directly with scales other than the CMDL83 scale. However, since the CMDL83 scale has been compared directly and indirectly with other scales, we can indirectly compare the gravimetric scale with those used in other measurement programs.

[20] A relatively small difference exists between CSIRO94 and CMDL83 scales. The ratio CSIRO94/CMDL83 is 1.00021 ± 0.00010, or 0.4 nmol mol⁻¹ at 1700 nmol mol⁻¹ [Cunnold et al., 2002]. This difference is in excellent agreement with the mean difference (±1σ) of 0.3 ± 2.2 nmol mol⁻¹ determined by the CSIRO94/CMDL83 flask air intercomparison from 1992 to 2003 (Masarie et al. [2001] and subsequent updates). CSIRO now reports its CH₄ measurements on a scale that was gravimetrically prepared by Nippon Sanso Company and is maintained by Tohoku University (TU). On the basis of the relationship between the TU and CSIRO94 scales reported by Cunnold et al. [2002], TU/CSIRO94 = 1.0119, we calculate that the relationship between TU/CMDL83 is 1.0121. This can be compared with NOAA04/CMDL83 = 1.0124, suggesting a difference between the TU and NOAA04 scales on the order of 0.5 nmol mol⁻¹ at ambient values.

[21] Four cylinders of dry natural air, three in 1990 and one in 1992, were exchanged between CMDL and Meteorological Service of Canada (MSC), Downsview, Ontario, Canada [Worthy et al., 1994]. They contained CH₄ mole fractions ranging from 1670.8 to 1793.6 nmol mol⁻¹ on the CMDL83 scale. The average ratio of mole fractions determined (±1σ), MSC/CMDL, was 1.0143 ± 0.0006. Since 1994, 12 additional cylinders of natural air have been compared with CH₄ mole fractions ranging from 1640 to 2173 nmol mol⁻¹. The average ratio of these 12 comparisons is 1.0158 ± 0.0004, so CH₄ mole fractions on the MSC scale are currently 26.9 nmol mol⁻¹ greater than those determined by CMDL at 1700 nmol mol⁻¹. Maintenance of these standard scales by each laboratory has allowed ongoing intercomparisons of measurements from discrete samples collected for CMDL at Alert, Northwest Territory, Australia (82°27′N, 62°31′W). The mean difference in CH₄ mole fraction (±1σ) from March, 1999 through March, 2004, MSC – CMDL83 (after converting MSC measurements to the CMDL83 scale using the ratio 1.0158) is 0.4 ± 1.8 nmol mol⁻¹. Using the scale ratios MSC/CMDL83 and NOAA04/CMDL83, a difference of 0.3% (∼6 nmol mol⁻¹) exists between the MSC and NOAA04 scales.

[22] In an exchange of one cylinder of dry natural air in 1992 between CMDL and the Meteorological Research Institute, Geochemical Research Laboratory (MRI/GRL), Tsukuba, Japan, the ratio MRI/CMDL83 = 1.0133 was obtained [Matsueda et al., 2004]. The inferred ratio with our new scale is 1.0009, or a difference of 1.5 nmol mol⁻¹ at ambient CH₄ abundance. Methane measurements from a common sampling site are not available to make an indirect comparison of the standard gas scales based on measurements for these two labs.

[23] We have previously made qualitative indirect intercomparisons of CMDL monthly mean zonally averaged CH₄ [Dlugokencky et al., 1994] from high southern latitude sites (30–90°S) with measurements from Syowa Station, Antarctica [Aoki et al., 1992], and from Cape Point, South Africa [Brunke et al., 1990]. Such a comparison is useful because at high southern latitudes, CH₄ is well mixed. Since the measurement program at Cape Point and the CMDL measurement program both use standard scales based on standards from Biospherics, agreement between these data sets was excellent. Once our data are converted to the NOAA04 scale, our zonal averages will be greater than the Cape Point measurements by ∼1.2%. The measurements by Aoki et al. [1992] at Syowa Station were on the Tohoku University scale; the mean relative difference in monthly means was ∼1.4%, comparable to the differences with CSIRO mentioned above.

[24] Two cylinders with nominal CH₄ mole fractions of 1800 and 1660 nmol mol⁻¹ have been exchanged with the National Institute for Standards and Technology (Gaithersburg, Maryland); results are summarized in Rhoderick and Dorko [2004, Table 9]. Although NOAA measurements of air in these cylinders were made in 1998, we converted the CH₄ determinations on the CMDL83 scale to the NOAA04 scale. Relative differences between the NOAA04 scale and the NIST CH₄ scale are 0.1% and 0.3%, where NOAA04 values are lower.

3.6. Implications for Atmospheric Methane

[25] While an accurate quantification of atmospheric CH₄ is important for atmospheric chemistry and climate, the small increase in CH₄ abundance (1.2%) that results from conversion of our measurements to this gravimetrically prepared scale will have a negligible impact. As an example, the CH₄ time series from South Pole is plotted in Figure 4. Methane values are ∼20 nmol mol⁻¹ greater on the new scale. The increase in CH₄ annual means at South Pole from 1983 to 2003 increases from 146.6 to 148.7 nmol mol⁻¹ after converting to the new scale. Quantities derived directly from measurements (e.g., global burden, trends, seasonal cycle amplitudes, and spatial gradients) are the least uncertain parameters in our knowledge of the global CH₄ budget. Other important parameters, such as the CH₄ atmospheric lifetime, still have larger uncertainties than those derived directly from observations. It still remains important that all laboratories report CH₄ measurements to international data archives (such as the WMO's World Data Center for Greenhouse Gases in Japan) on a common scale, and it is hoped that labs contributing CH₄ data to GAW will either convert to this new scale or use interlaboratory intercomparison experiments to establish a well-defined conversion factor between their scale and the WMO CH₄ mole fraction scale.