SI Traceability and Scales for Underpinning Atmospheric

Metrologia

PAPER Recent citations SI traceability and scales for underpinning - Amount of substance and the mole in the SI atmospheric monitoring of greenhouse gases Bernd Güttler et al

- Advances in reference materials and To cite this article: Paul J Brewer et al 2018 Metrologia 55 S174 measurement techniques for greenhouse gas atmospheric observations Paul J Brewer et al

- News from the BIPM laboratories—2018 Patrizia Tavella et al View the article online for updates and enhancements.

This content was downloaded from IP address 205.156.36.134 on 08/07/2019 at 18:02 IOP

Metrologia

Bureau International des Poids et Mesures Metrologia

Metrologia Metrologia 55 (2018) S174–SS181 https://doi.org/10.1088/1681-7575/aad830

55 SI traceability and scales for underpinning

2018 atmospheric monitoring of greenhouse

MTRGAU Paul J Brewer1,6 , Richard J C Brown1 , Oksana A Tarasova2, Brad Hall3, George C Rhoderick4 and Robert I Wielgosz5 S174 1 National Physical Laboratory, Hampton Road, Teddington, Middlesex TW11 0LW, United Kingdom 2 World Meteorological Organization, 7bis, avenue de la Paix, Case postale 2300, CH-1211 Geneva 2, Switzerland 3 National Oceanic and Atmospheric Administration, 325 Broadway, Mail Stop R.GMD1, Boulder, CO 80305, United States of America P J Brewer et al 4 National Institute of Standards and Technology, 100 Bureau Drive, MS-8393 Gaithersburg, MD 20899-8393, United States of America 5 Printed in the UK Bureau International des Poids et Mesures, Pavillon de Breteuil, F-92312 Sèvres Cedex, France

E-mail: [email protected] MET Received 21 June 2018, revised 2 August 2018 Accepted for publication 6 August 2018 10.1088/1681-7575/aad830 Published 7 September 2018

Abstract Paper Metrological traceability is the property of a measurement result whereby it can be related to a stated reference through a documented unbroken chain of calibrations, each contributing to the measurement uncertainty. The stated reference can be the International System of Units, and 1681-7575 more specifically a realisation of a primary standard with its value and uncertainty expressed in SI units, which offers many benefits such as long term stability and the ability to reproduce the standard within its stated uncertainty at any time. Alternatively, measurements can be 5 made traceable to an empirical scale, an approach that has evolved for practical reasons to meet the needs of certain measurement communities. We explore the benefits and drawbacks of these two measurements systems as they have evolved for greenhouse gas measurements S174 at background levels and outline the advantages of using elements from each to secure a more robust measurement infrastructure. In particular, this work examines the importance S181 of compatibility, the accuracy of measurement results and the benefits of comparisons with independent primary standards. Keywords: metrology, traceability, calibration, SI, scale, atmospheric monitoring

Introduction the measurement uncertainty. The reference in a measurement system, to which measurements are made traceable, can be The goal of implementing metrological traceability within a realised in different ways. Here we consider two approaches measurement system, is to ensure that measurements can be for a measurement system for greenhouse gas amount-of-sub- comparable and compatible anywhere at any time [1]. A met- stance fractions in the atmosphere. The first provides trace- rological traceability chain [1] must be established from the ability to the international system of units (SI) and the second (value of the) measurement result to the value embodied in, or to a defined scale. In particular, we consider the advantages carried by a material, or obtained by a measurement procedure and disadvantages of each approach and how the measure- [1]. Metrological traceability is not just a simple statement ment system may ultimately benefit from an approach that [2]. It must be based on an unbroken chain of quantity values combines elements of each. in a specified material or measurement, each contributing to There are some prominent examples of the use of scales within metrology, for example for temperature measurements 6 Author to whom any correspondence should be addressed.

1681-7575/18/05S174+8$33.00 S174 © 2018 BIPM & IOP Publishing Ltd Printed in the UK Metrologia 55 (2018) S174 P J Brewer et al [3]. The unit of the fundamental physical quantity known as statement will include all sources of uncertainty, up to and thermodynamic temperature (T), is the kelvin (K), defined as including the realisation of the SI units themselves. The the fraction 1/273.16 of T at the triple point of water. Triple- degree of equivalence of measurement standards can then be point-of-water cells provide a convenient realisation of this calculated by the difference in their values, to the extent to definition. For temperatures other than the triple point of which this difference is smaller than the combined uncertain- water, direct measurements of thermodynamic temperature ties of the values of the standards. require a primary thermometer based on a well-understood In the field of gas analysis, reference materials are usu- physical system whose temperature may be derived from ally disseminated in high pressure cylinders using gravimetric measurements of other quantities. In practice, primary ther- methods to assign the reference value. There is a broad con- mometry is difficult and time consuming and not a practical sensus amongst metrologists that ‘verification’ is required to means of disseminating K. As an alternative, the International ensure the ‘primary method’ is behaving as expected [7]. The Temperature Scale provides an internationally accepted recipe implication is that within gas metrology, amount-of-substance for realising temperature in a practical way. The present fraction values can be realised and their uncertainties calcu- scale is the International Temperature Scale of 1990 (ITS- lated with traceability to the mole. This is achieved via mass 90) (Recommendation 5, CI-1989). The values assigned to and purity measurements, tables of atomic weights and the the fixed-point temperatures used in ITS-90 and pre-ITS-90 molar mass constant Mu. These values should then be verified scales were based on the state-of-the-art knowledge of T at by comparison against an independently prepared gas mixture the time of inception of each scale. Estimates of the differ- of similar composition. This process is intended to demon- ences between T and the ITS-90 (T90) are available with strate that errors in preparation, loss of material to cylinder consensus estimates provided for T − T90, for selected meas- walls, and reaction of gaseous components within the mixture, urements from 4.2 K to 1358 K, as well as a recommendation which would compromise the ‘complete understanding’ of the for analytic approximations to T − T90 for the range 0.65 K to method are well understood. It is agreed that this verification 1358 K. The values of T − T90 are small, and would be neg- step provides a good demonstration that the value and uncer- ligible for the vast majority of users. Considerable research tainty have been properly assigned, based on a gravimetric activities have gone into understanding the factors that can value fully traceable to the SI [8]. The general methods for lead to variations in the fixed points (e.g. impurities and iso- the realisation of gas standards with assigned amount-of- topic composition), so that the scales can be realised inde- substance fractions is described within the written standard pendently in National Metrology Institutes (NMIs) around ISO 6142-1:2015 [8]. This methodology has been success- the world achieving comparable results. As will be explained fully used by NMIs to develop standards for greenhouse gases later, this differs from the scale approach developed for green- for different measurement communities from emission levels house gases, where the scale is realised by one unique set of down to background atmospheric levels and below. Standards standards, held by one institute. produced by such methods can routinely be offered with relative standard uncertainties of 0.1%, but for special applica- tions and over particular amount-of-substance fraction ranges Discussion these can be reduced to 0.025% or below. Traceability to the SI for the amount-of-substance fraction of greenhouse gases in air The scale approach The role of a NMI is to provide national standards for mea- An alternative approach is a measurement system based on surement and in many cases manage the national measure- traceability to a ‘scale’, often a family or collection of gas ment system comprising calibration laboratories, certification standards, based on an agreed reference value or reference and accreditation bodies, and legal metrology [4]. It is also method. This system offers superior precision as comparisons to carry out the necessary comparisons with other NMIs, are made to one source and the absolute accuracy of the now through the International Committee for Weights and artefact has limited impact on measurement compatibility, Measures’ Mutual Recognition Arrangement (CIPM-MRA) provided that all measurements are traceable to the same ref- [5], to demonstrate international equivalence of the national erence, and that reference is stable. Within the greenhouse standards. In most cases this is based on their traceability to the gas field (particularly at the global scale), this practice arose SI. This system was adopted at the 10th General Conference partially out of the realisation that gas amount-of-substance on Weights and Measures (CGPM) on which to establish a fraction measurements could be performed with precisions practical system of measurement for international use [6]. that were often better than the uncertainties associated with Today the SI is the fixed reference point for almost all modern gas mixture preparation. In addition, the key quantity to be science and technology. The base units are regarded as dimen- measured was the relative difference in amount-of-substance sionally independent and form a coherent set of derived units. fraction within a measurement network. Within this network It is coherent in the specialist sense of a system whose units the key consideration was the level of compatibility that could are mutually related within the equations of chemistry and be achieved between different measurement sites. A funda- physics, with no numerical or scaling factors other than 1. In mental characteristic of the scale approach is that a scale, addition, the realisation of a standard expressed in SI units, once defined, is intended to serve as a fixed reference over a implies that the accompanying measurement uncertainty specific amount-of-substance fraction range. For greenhouse

S175 Metrologia 55 (2018) S174 P J Brewer et al gas analysis, a scale is often defined by a series of gas mix- Perhaps the most widely known example of a scale for tures, prepared gravimetrically or otherwise, with amount-of- atmospheric measurements is for carbon dioxide. In 1958, substance fractions that span the range of scientific interest. Charles Keeling started collecting carbon dioxide samples at The standards are typically value assigned by methods that Mauna Loa in Hawaii [14]. By 1960, he had established that can provide SI traceable values and associated measure- there are strong seasonal variations in carbon dioxide levels ment uncertainties. However, the ensemble of standards is with peak levels reached in the late northern hemisphere then used to define a calibration curve. It is this calibration winter. A reduction in carbon dioxide followed during spring curve which is taken to be the stable reference for the scale. and early summer each year as plant growth increased in the The stability of the scale and small uncertainties that can be land-rich northern hemisphere. In 1961, Keeling produced achieved for subsequent calibrations in the traceability chain, data showing that carbon dioxide levels were rising steadily then permit different laboratories to achieve relative com- in what became known as the ‘Keeling Curve’. Extension of patibility of measurements (e.g. 0.025% for carbon dioxide the measurement network required detection of small spatial in the northern hemisphere). Independent scales are often and temporal gradients with a high degree of certainty. The compared through the exchange of gas mixtures or through emphasis on compatibility of data from different measure- co-located measurements; essentially an informal method ment sites led to the decision to have traceability to only one of ‘validation’ [9]. For example, the World Meteorological set of primary standards. Introduction of other traceability Organisation (WMO) SF6 scale, identified as X2014, is chains would have led to the variance between primary stand- defined by 17 primary standards (SF6 in synthetic zero-grade ards adding to the variance observed between measurement 1 air), over the range 2–20 pmol mol− [10, 11]. These stan- sites. Standards to define the scale are prepared and sustained dards were prepared over a 14 year period and have standard in high-pressure gas cylinders and have been the subject of uncertainties ranging from 0.2% to 1.2%. A second order intensive research to determine their accuracy and stability calibration curve was generated using gas chromatography [15, 16]. with electron capture detection. This calibration curve was Taking the analogy from temperature measurements, the then used to propagate the scale to secondary standards for symbol used to denote SI and scale traceable amount-of- routine use. In this example the primary standards are highly substance fractions, could be differentiated by use of a sub- consistent with a standard deviation of residuals (from the script in the symbol for the quantity, i.e. x versus xCMDL83 and −1 calibration curve) of 0.014 pmol mol (0.15% of 9 pmol xNOAA04. Similarly, the difference between SI and a particular mol−1). The primary standards may be value re-assigned by institute could also be expressed in this way: e.g. x versus correcting for residuals from the calibration curve, but this xNIST. Whilst this notation is not currently used in practice, is simply a matter of convenience, since it is the secondary its adoption would lead to easy recognition that the values standards that are used for scale propagation. were being reported on a different basis. Over time, x is not Using the scale approach, the level of compatibility of a fixed value and can change depending on how the reference measurements is typically better than can be achieved for value in key comparisons is determined. SI traceable values individual primary standards alone. For example, the uncer- are however rarely corrected, or shifted, unlike scale values tainty associated with primary carbon dioxide standards in as it is assumed that the uncertainty covers the ‘true value’. In the ambient range measured by the National Oceanic and addition, key comparisons for the major greenhouse gases are Atmospheric Administration (NOAA), using the manometric being run by sending standards for comparison measurements method is ~0.2 µmol mol−1 (k = 2). However, the reprodu- to one central laboratory (the BIPM in this case), which allows cibility of value assignments made relative to a fixed scale is the reference value for gas mixtures to be calculated from the about a factor of three better using a non-dispersive infra-red largest consistent set of standards. This has had the advan- (NDIR) method [12], and a factor of ten better using a spec- tage of reducing uncertainties of the reference value with the troscopic method [13]. expectation that a future repeat comparison would produce However, maintenance of a scale requires intensive and reference values that agreed with previous comparison results uninterrupted effort generally at one location or one institute, within the stated uncertainties. as drift in the reference artefact would have significant impli- cations. Because a scale is typically defined by the entire col- Achieving continuity in greenhouse gas data series lection of gas standards, removing or adding artefacts to the collection could result in a measureable difference between sets. The premise used when operating with SI traceable mea- This is true even if that difference is smaller than the uncertainty surements is that they will be compatible within their com- at which the standards can be prepared. Thus, scales are identi- bined stated uncertainties, allowing them to be reproduced in fied by name, and included in WMO Global Atmosphere Watch space and time. The resulting measurements are expected to (GAW) programme’s metadata. Changes are documented when be accurate, i.e. encompass the true value within their stated significant at the level important for the scientific application. measurement uncertainty. The correct implementation of the For example, the WMO scale for measuring methane amount- SI traceable approach for primary standards can be checked by of-substance fractions in the atmosphere has changed over time, comparison with other standards of similar metrological hier- the current scale being identified as WMO-X2004A with pre- archy, an approach that has been formalised in the key com- vious scales being identified as WMO-X2004 and CMDL83. parisons of the CIPM MRA. As the technology for producing

S176 Metrologia 55 (2018) S174 P J Brewer et al gas standards improves, it is expected that the uncertainty of the value of the standard will decrease, whilst enveloping the true value and maintaining consistency with previous standards. Within the field of metrology, metrological compatibility of measurement results is defined as [1] the property of a set of measurement results for a specified measurand, such that the absolute value of the difference of any pair of measured quantity values from two different measurement results is smaller than some chosen multiple of the standard measurement uncertainty of that difference. This means that within a metrological framework, the concept of compatibility of two results cannot be disassociated with the uncertainties of these measurement results. Within a network using the scale approach, data quality objectives (DQOs) are set, of which compatibility goals are Figure 1. The number distribution of the differences between the amount-of-substance fraction of methane reference materials used a subset. The most current definition of compatibility within or distributed by NOAA from 1999–2013 at the time of preparation this setting is taken from the report of the 18th WMO/IAEA and a subsequent analysis (after 1–14 years). The distribution Meeting on Carbon Dioxide, Other Greenhouse Gases and contains 759 data pairs. Related Tracers Measurement Techniques (GGMT-2015) [17], in which it is defined as a property of a set of measurement results, such that the absolute value of the difference between any pair of measured values from two different measurement results is within a chosen value which does not have to be the same as the total combined uncertainty. The GGMT-2015 and VIM definitions of compatibility are significantly different, with the GGMT-2015 definition, not comparing differences in results to magnitudes of measurement uncertainties. Currently for the three major greenhouse gases, carbon dioxide, methane and nitrous oxide, the compatibility goals are ± 0.1 µmol mol−1 (± 0.05 µmol mol−1 in the southern hemisphere), ± 2 nmol mol−1 and ± 0.1 nmol mol−1 respectively [17]. These represent the maximum bias that can be tolerated among different measurement sites in order to iden- tify small trends and gradients in the unpolluted background Figure 2. The results of a WMO comparison experiment. The atmosphere. Compatibility goals are application-specific, and differences between the laboratory result and the reference value (assigned by the WMO CCL) at two nominal amount-of-substance can be relaxed when trends and gradients are relatively large fractions (1750 nmol mol−1, grey circles; and 1950 nmol mol−1, (e.g. regional and urban scales). Maintaining a direct link to black circles). Circles are results reported on the WMO scale, the WMO scales and successfully propagating the scales to and other symbols are those reported on independent scales and working laboratory scales is fundamental to the measurement compared to the WMO scale. Bars are the laboratory and the CCL process. Compatibility is typically assessed by comparing reproducibility estimates combined in quadrature. The WMO compatibility goals are shown as horizontal dashed lines. multiple measurements of the same quantity at a single site (e.g. in situ sampling versus grab samples), by comparing mixture was analysed at the time of preparation and at a later co-located measurements at a single site obtained by two date. The average time between analyses varied between 1 and different laboratories, or by ‘same air’ comparisons where 14 years with a mean of 3.9 years. In figure 1, the standard two laboratories measure the same discrete air sample [18]. deviation of the data is 0.46 nmol mol−1, so that 95% of the Further guidance on determining compatibility of measure- differences measured were within 0.92 nmol mol−1, which is ment results within a network are expected to be developed, approximately the level of reproducibility needed at NOAA including further clarification as to whether compatibility (the WMO Central Calibration Laboratory for methane) to goals should be considered as limit values or as 1 sigma values support the WMO compatibility goal. The WMO scale was for a distribution of differences measured, both having been updated in 2004 [20], in which primary standard value assign- previously reported [17, 20]. ments were revised and all prior results retroactively updated The use of the scale approach to achieve compatible meas- to the new scale without loss of continuity. In practice, this urement results in monitoring networks, despite changes in involves propagating the new scale to secondary and tertiary scale, is partially demonstrated in figures 1 and 2. Figure 1 standards by updating instrument calibration curves that relate shows the distribution of differences in the amount-of-sub- primary-to-secondary, and secondary-to-tertiary. In some stance fraction of methane derived from multiple analysis of cases, such as those involving highly linear instruments, the reference materials (modified natural air in aluminium cylin- scale update might be simplified as a factor that can be applied ders) used or distributed by NOAA from 1999–2013. Each to all measurements.

S177 Metrologia 55 (2018) S174 P J Brewer et al

Figure 3. Results of the last three international CCQM key comparisons for assessing the analytical and preparative capabilities of NMIs for measuring methane in an air matrix. Bars show expanded uncertainties.

Figure 2 shows the results of a WMO comparison experiment in which a total of 10 gas mixtures were distributed to 35 laboratories over a two year period (2014–2015). The differences between the laboratory result and the reference value (assigned by the WMO CCL) are shown at two nominal amount-of-substance fractions. The error bars are based only on reproducibility estimates and do not include uncertainties associated with scale preparation, as would normally be the case in a key comparison. The aim of the exercise was to assess the potential for compatibility. In this experiment, 73% of all individual results (51 of 70) and 85% of results reporting on the WMO scale (51 of 60), are within 2 nmol mol−1 of the reference value. It is important to note that within the WMO/ GAW network, compatibility refers to atmospheric measurements, and is only partially demonstrated through analysis of gas mixtures. Nevertheless both figures illustrate the excellent consistency within a network that the scale approach provides. The accuracy of the scale approach has been verified by the participation of NOAA in key comparisons under the CIPM MRA as a designated institute of the WMO. Figure 3 shows the results of three comparisons organised by the CCQM Gas Figure 4. Estimate of differences between SI and WMO NOAA (filled squares) and WMO CSIRO (filled circles) scale, SI and NIST Analysis Working Group. The first (CCQM-P41 part 1) [21] (open squares), and SI and NPL (open circles) traceable amount-of- was organised by the Van Swinden Laboratorium (VSL) in substance fractions for carbon dioxide, methane and nitrous oxide 2003 and assessed the analytical capabilities of laboratories ((a)–(c) respectively), based on results in CCQM-P41, CCQM-K52, for 1.8 µmol mol−1 methane and 365 µmol mol−1 carbon CCQM-K82 and CCQM-K68 comparisons. Different WMO scales dioxide (full complement of results not presented here) in an (filled symbols) are shown in panel (b): values at 1814 and 2214 nmol mol−1 are on the WMO-X2004 methane scale developed at air matrix. CCQM-P41 part 2 [22] was coordinated by VSL NOAA based on gravimetrically prepared primary standards, while in the same year and assessed the preparative capabilities of the value at 1801 nmol mol−1 is on the CMDL83 scale, which was laboratories for methane and carbon dioxide (full complement based on two standards obtained from Biospherics (Portland, OR) of results not shown here) in air at the same amount-of-sub- [22]. In panel (a), the WMO scale naming convention for CO2 was stance fractions as part 1. Ten years later, a third comparison not formally adopted at the time of the comparisons shown. Carbon dioxide results are broadly consistent with scale WMO-X2001. coordinated by the BIPM and NIST (CCQM-K82) [23], NOAA results in panel (c) are on the NOAA-2006 scale. which assessed the preparative capabilities of laboratories for the same mixture composition as CCQM-P41, resulted in 2019. Within each of these comparisons, the key comparison standard uncertainties for reference values of 0.7 nmol mol−1. reference value is considered to be the best estimate of the CCQM-K52 [24] was carried out on carbon dioxide standards SI traceable value and uncertainty that can be demonstrated in 2006 and a further comparison, CCQM-K120 was initiated for the amount-of-substance fraction (x) measured. At the in 2017. The CCQM-K68 [25] comparison on nitrous oxide same time the result produced by NOAA or results traceable standards was performed in 2008 and will be repeated in to the NOAA maintained scale, provides an estimate of the

S178 Metrologia 55 (2018) S174 P J Brewer et al

scale value (xscale) and allows the difference between the two (x − xscale) and their combined uncertainty to be calculated, and is depicted in figure 4. This figure demonstrates a number of key issues, which have been or are being addressed: (a) The uncertainties in the plotted differences have histori- cally been large when compared to the magnitude of the WMO-GAW network compatibility goals for data. There are two factors that have contributed to this, notably the magnitude of the uncertainties in the standards themselves, and the way the comparison has been performed. As can be seen from the results of the most recent comparison in 2013 (CCQM-K82) for methane, both of these issues are being addressed [23, 26]. Comparisons are now carried out in a central laboratory (BIPM), allowing the Figure 5. Comparison of NIST (grey circles) and NOAA (black standards to be compared under repeatability/interme- circles) reference values for methane (left y-axis). Error bars diate reproducibility conditions, resulting in the analytical represent expanded uncertainties. Differences are shown with measurement uncertainty component of the comparison triangles (right y-axis) and the black dashed line represents the to be much reduced. Secondly, the uncertainty of the average difference between the reference value from each institute. Reprinted with permission from [26]. Copyright 2015 American standards has been reduced, and in particular the verifi- Chemical Society. cation uncertainty component, a component which may have been too conservative in previous comparisons. It is ence mixtures (mean difference 2.7 nmol mol−1). While expected that the uncertainty for the reference value will there is certainly room for improvement with respect to decrease for the next carbon dioxide and nitrous oxide meeting strict compatibility goals and reducing uncer- comparisons. There is already some evidence for lower tainties, demonstrating consistency over time is a key uncertainties for nitrous oxide. Analysis of 30 cylinders objective for the network. containing natural air by NIST and NOAA [27], showed a mean difference (NIST-NOAA) in the amount-of- Improvements in uncertainties of SI primary standards for substance fraction of nitrous oxide of 0.20 nmol mol−1, − greenhouse gases with a combined uncertainty of 0.42 nmol mol−1 (b) For carbon dioxide and nitrous oxide the comparisons An SI traceable measurement system could be implemented have not detected a bias between the SI and scale amount- with benefits for the greenhouse gas community, if the uncer- of-substance fraction values. However, it was clear that tainties in the measurements standards contributed in a neg- in 2003 a significant bias between the SI and methane ligible way to the uncertainty of measurements performed at amount-of-substance fraction scale value existed, and this measurement sites. In comparing the magnitudes of network was well documented [19] and corrected for in the change compatibility goals and uncertainties in the values of stan- from the CMDL83 and NOAA04 scale for methane. dards (since uncertainties are added in quadrature) this would However, even though a bias existed between CMDL83 mean that a standard uncertainty one quarter of the magnitude and NOAA04, the CMDL83 was propagated with low of a compatibility goal would increase the standard devia- uncertainty, as can be seen by the close agreement of tion of the measurement results by less than 5%, and would CMDL83 as represented by two different laboratories be negligible (indeed it could be argued that standards with (NOAA and CSIRO). A much smaller residual bias uncertainties one third of the size of the compatibility goals between the NOAA04 scale and SI value for methane would have negligible impact on meeting them). This assumes was revealed in the 2013. A reduction in measurement that compatibility goals are 1 sigma for a distribution of dif- uncertainties for methane standards (0.033% compared ferences measured. Returning to the example of the scale for to 0.2% relative) has been reported [27] resulting from carbon dioxide measurement, NMIs would need to demon- improvements in instrumentation for mass measurements strate comparability and achieve measurement uncertainties and determinations of trace methane in balance air. substantially lower than the WMO GAW programme’s com- (c) There is consistency among these NMIs for carbon patibility goals (which translates to a standard uncertainty in dioxide and methane, although uncertainties are high for the value of the primary standard of ~25 nmol mol−1 for the the former. The quantities x-xNIST and x-xNPL are relatively remote northern hemisphere) to ensure that interchangeability consistent between CCQM-P41 and CCQM-K52, and of reference standards would not impact on global measure- between CCQM-P41 and CCQM-K82, as are the rela- ments [15]. They would also need to provide stable reference tionships between the laboratories shown in figures 4(a) standards as long-term trend detection fundamentally relies and (b). In fact, the difference in methane between WMO on the reference not changing with time. Since the creation (NOAA) and NIST revealed in K82, was confirmed in of the WMO/GAW carbon dioxide scale, significant prog- further bi-lateral comparisons. Figure 5 shows the results ress has been made, as NMIs drive towards addressing these of a NIST-NOAA comparison for 30 methane in air refer- requirements. The smallest expanded uncertainties reported

S179 Metrologia 55 (2018) S174 P J Brewer et al for carbon dioxide are now approaching the magnitudes of the with much more frequent comparisons for carbon dioxide WMO/GAW compatibility goals, however this work is still planned in the future. someway from meeting requirements [28]. Similarities between underpinning measurements with a The benefits of moving to SI traceability would mean that scale or the SI relate to the operational aspects of dissemi- measurements are consistent with measurements made in nating any sort of artefact-based traceability. Amongst the other areas of science and technology taking into account their NMI community, certain elements of the scale approach are measurement uncertainties and can be combined easily with used within each laboratory to provide a more robust infra- the measurements in more complex calculations, for instance structure, such as preparation of a suite of secondary standards in climate modelling. which are compared to a primary standard from a key com- Aspects of both approaches have been adopted by NMIs parison. The distinction between the two approaches is that and laboratories that propagate scales. For example, NOAA SI traceability in gas analysis involves a level of comparison has made efforts to provide uncertainty estimates following and validation of the primary standards held at each institute accepted methods used by NMIs [29] and has participated in which is repeatedly checked against other NMIs and are equal Key Comparisons on behalf of the WMO. In concert, some in terms of metrological hierarchy. This is not the case for de NMIs have adopted aspects of the scale approach, such as facto primary scales such as NOAA CH4 X2004A as they are using a fixed set of SI traceable primary standards anchored provided by one institute, so validation is performed in-house to performance in a key comparison to define‘ scale’. Those and comparison against other institutes does not impact the primary standards are in turn used to value assign sec- reference value. ondary (or working) standards that remain within the NMI, Uncertainty in SI traceable primary standards is being to propagate that ‘scale’. This is different from the general reduced, and for methane these are expected to be lower in approach used by NMIs in value assigning secondary refer- magnitude than network compatibility goals. On-going com- ence materials using primary standards. In the general case, parisons among NMIs and laboratories that develop primary secondary reference materials are typically available to the reference standards, both formally and informally, will help to public sectors. In these cases, NMIs can offer batch assigned document relationships among reference materials used in gas (or certified) reference materials where the level of uncer- analysis and provide additional information on the stability of tainty is not as demanding or required. However, where very those materials. tight uncertainties are required, for instance to meet WMO compatibility goals, individually analysed reference mat Acknowledgments erials with a higher degree of compatibility are offered [30]. These efforts, together, should provide improved support This work was funded, in part, by the United Kingdom for gas analysis to meet the requirements for atmospheric National Measurement System. BH acknowledges support monitoring. from NOAA’s Atmospheric Chemistry, Carbon Cycle, and Climate program. Conclusions ORCID iDs Excellent compatibility can be achieved within a monitoring network through traceability to a unique scale. While there Paul J Brewer https://orcid.org/0000-0002-7446-417X may be numerical differences between greenhouse gas mea- Richard J C Brown https://orcid.org/0000-0001-6106-0996 surements traceable to scales and SI traceable standards, these are expected to be covered by the measurement uncertainty of References this quantity. Key comparisons have allowed the accuracy of some scales (e.g. those used within WMO/GAW) to be monitored. [1] BIPM, IEC, ILAC IFCC, IUPAC, ISO, OIML and IUPAP 2012 The International Vocabulary of Metrology—Basic Previous results demonstrate that there is a level of consist- and General Concepts and Associated Terms (VIM) ency in the NMI results over time and in the WMO results, (Sèvres: Joint Committee for Guides in Metrology) p 200 even when the scale is represented by a different laboratory [2] De Bièvre P 2010 Accreditation and Quality Assurance that is not maintaining the primary standard (e.g. results from (Berlin: Springer) vol 15 pp 267–8 CSIRO). NOAA now has calibration and measurement capa- [3] Moldover M R, Tew W L and Yoon H W 2016 Nat. Phys. 12 7 [4] 2000 Proceedings of the International School of Physics bilities listed in the BIPM key comparison database and pro- ‘Enrico Fermi’ (Varenna: Italian Physical Society) vides reference values with both an expanded uncertainty (to [5] The International Committee for Weights and Measurements comply with ISO 17025 and 17034) and a reproducibility esti- (CIPM) 1999 www.bipm.org/utils/en/pdf/mra_2003.pdf mate. This provides a useful link between their primary stand- (revised technical supplement 2003) (BIPM) ards and the SI and an on-going framework for assessing drift [6] Comptes Rendus de la 10e CGPM (1954) 1956 p.78 [7] Brown R J C, Brewer P J, Harris P J, Davidson S and van der in these measurements systems over time. Key comparisons Veen A M H 2017 Metrologia 54 L11–8 for the major greenhouse gases are planned every eight years [8] ISO 6142-1:2015 2015 Gas analysis—Preparation of or so, allowing on-going monitoring of the accuracy of scales, Calibration Gas Mixtures—Part 1: Gravimetric Method

S180 Metrologia 55 (2018) S174 P J Brewer et al

for Class I Mixtures (International Organisation for Techniques (GGMT-2017) (Dubendorf, 27–31 August 2017) Standardization) (World Meteorological Organisation) in preparation [9] Hall B D et al 2014 Atmos. Meas. Tech. 7 469–90 [19] 2014 17th WMO/IAEA Meeting on Carbon Dioxide, Other [10] Hall B D, Dutton G S, Mondeel D J, Nance J D, Rigby M, Greenhouse Gases and Related Tracers Measurement Butler J H, Moore F L, Hurst D F and Elkins J W 2011 Techniques (GGMT-2013) (Beijing, China, 10–13 June 2013) Atmos. Meas. Tech. 4 2441–51 (World Meteorological Organisation) GAW Report No. 213 [11] Lim J S, Lee J, Moon D, Kim J, Lee J and Hall B 2017 Anal. [20] Dlugokencky E J, Myers R C, Lang P M, Masarie K A, Chem. 89 12068–75 Crotwell A M, Thoning K W, Hall B D, Elkins J W and [12] 2011 WMO/TD-No. 1553 GAW Report ed W A Brand Steele L P 2005 J. Geophys. Res. 110 D18306 (Geneva, Switzerland: WMO) No. 194 pp 101–8 [21] van der Veen A M H 2007 Metrologia 44 08002 [13] Tans P P, Crotwell A M and Thoning K W 2017 Atmos. Meas. [22] van der Veen A M H 2007 Metrologia 44 08003 Tech. 10 2669–85 [23] Flores E 2015 Metrologia 52 08001 [14] Kahele R 2007 Behind the Inconvenient Truth (Hana Hou: [24] Wessel R M et al 2008 Metrologia 45 08011 Pacific Travelogue Inc) pp 10–5 [25] Lee J et al 2011 Metrologia 48 08004 [15] Keeling C D, Guether P R and Moss D J 1986 Scripps [26] Flores E, Rhoderick G C, Viallon J, Moussay P, Choteau T, Reference Gas Calibration System for Carbon Dioxide in Gameson L, Guenther F R and Wielgosz R I 2015 Anal. Air Standards (Publication 42, Environmental Pollution Chem. 87 3272–9 Monitoring and Research Programme) (Geneva: World [27] Rhoderick G C 2013 Anal. Bioanal. Chem. 405 369 Meteorological Organisation) [28] Brewer P J, Brown R J C, Miller M N, Doval Minarro M, [16] Zhao C and Tans P P 2006 J. Geophys. Res. 111 D08S09 Murugan A, Milton M J T and Rhoderick G C 2014 Anal. [17] 2016 18th WMO/IAEA Meeting on Carbon Dioxide, Other Chem. 86 1887 Greenhouse Gases and Related Tracers Measurement [29] JCGM 100:2008 1995 Evaluation of Measurement Data— Techniques (GGMT-2015) (La Jolla, CA, 13–17 September Guide to the Expression of Uncertainty in Measurement 2015) (World Meteorological Organisation) GAW Report (Sèvres: Joint Committee for Guides in Metrology) No. 229 [30] Rhoderick G C, Kitzis D R, Kelley M E, Miller W R, [18] 2017 19th WMO/IAEA Meeting on Carbon Dioxide, Other Hall B D, Dlugokencky E J, Tans P P, Possolo A and Greenhouse Gases and Related Tracers Measurement Carney J 2016 Anal. Chem. 88 3376–85

S181