Comparability of calibration strategies for measuring mercury concentrations in gas emission sources and the atmosphere

A primary mercury gas standard was developed at Van Swinden Laboratory (VSL) to establish an International System of Units (SI)-traceable reference point for mercury concentrations at emission and background levels in the atmosphere. The majority of mercury concentration measurements are currently made traceable to the empirically determined vapour pressure of mercury. The primary mercury gas standard can be used for the accurate and precise calibration of analytical systems used for measuring mercury concentrations in air. It has been especially developed to support measurements related to ambient air monitoring (1– 2 ngm−3), indoor and workplace-related mercury concentration levels according to health standards (from 50 ngm−3 upwards) as well as stationary source emissions (from 1 μgm−3 upwards). The primary mercury gas standard is based on diffusion according to ISO 6154-8. Calibration gas mixtures are obtained by combining calibrated mass flows of nitrogen and air through a generator holding diffusion cells containing elemental mercury. In this paper, we present the results of comparisons between the primary gas standard and mercury calibration methods maintained by NPL (National Physical Laboratory in the United Kingdom), a National Metrology Institute (NMI), and the Jozef Stefan Institute (JSI), a Designated Institute (DI). The calibration methods currently used at NPL and JSI are based on the bell-jar calibration apparatus in combination with the Dumarey equation or a NIST (National Institute of Standards and Technology in the United States) reference material. For the comparisons, mercury was sampled on sorbent traps to obtain transfer standards with levels between 2 and 1000 ng with an expanded uncertainty not exceeding 3 % (k = 2). The comparisons performed show that the results for the primary gas standard and the NIST reference material are comparable, whereas a difference of −8 % exists between results traceable to the primary gas standard and the Dumarey equation.

concentration levels in the environment are key to underpin global efforts to control and reduce the mercury emissions, to meet the obligations of legislation and to protect human health.
The majority of measurements of mercury concentrations are currently traceable to the empirically determined equations describing the saturated vapour pressure of mercury, usually via a bell-jar calibration apparatus (Brown et al., 2008a, Brown 35 et al., 2008b. This apparatus allows a saturated concentration of mercury vapour in air to develop in a confined space in equilibrium with ambient conditions, from which a known amount of mercury can be removed for calibration purposes. Several empirical equations are available, e.g., the Dumarey and the, more recently proposed, Huber equations (Dumarey et al., 1979, Dumarey et al., 1985, Dumarey et al., 2010, Huber et al., 2006. The agreement between these equations is unsatisfactory, with vapour pressures from different equations differing by more than 7 % at 20 ᵒC (Brow et al., 2010. To 40 remove the dependency of mercury concentration measurements from these empirical equations, and to provide stability and comparability, a primary mercury gas standard has been developed by VSL (Van Swinden Laboratory, National Metrology Institute in the Netherlands), to provide metrological traceability to the International System of Units (SI) , de Krom et al., 2020. The SI-traceable primary standard, working according to ISO 6145-8 (ISO 6145-8, 2005), can be used for the calibration of instruments and the certification of mercury gas generators. Furthermore, sorbent traps can be spiked with 45 known amounts of mercury from the primary standard using pumped sampling, according to ISO 16017-1 (ISO 16017-1, 2000). The transfer standards obtained can be used as certified reference materials to calibrate field instruments or for comparisons such as mandatory for calibration and testing laboratories to show their conformity assessment under ISO/IEC 17025 (ISO/IEC 17025, 2017). Such a comparison can only be successful if it allows participants to validate their measurement procedures with respect to SI-traceable standards as required by ISO/IEC 17043 (ISO/IEC 17043:2010). 50 Measurements of mercury concentrations in air are typically carried out using a pump to sample air at a monitoring location, at a known rate for a known time, onto an adsorption tube (Pirrone et al., 2013, Pandey et al., 2011. The trapping of mercury is usually performed with a sorbent tube using gold to form an amalgam and trap the mercury. To increase the surface area available, the gold is often dispersed on silica or a similar support , Brown et all., 2017. Mercury in this adsorption tube is then thermally desorbed and measured. Furthermore, in 2019 a new technical specification was adopted in 55 Europe, which describes the use of gold amalgamation traps for sampling and determination of mercury compounds in flue gas (NVN-CEN/TS 17286, 2019). This technical specification is based on the United States Environmental Protection Agency (US EPA) 30B method (Method 30B, 2017). That method in turn uses carbon sorbent traps for the determination of total gaseous mercury emissions from coal-fired combustion sources. The sorbent traps contain carbon typically activated with iodine or another halogen (Živković et al., 2020). 60 To demonstrate the robustness and comparability of transfer standards obtained with the primary standard, in this paper we present the results of comparisons with current calibration methods maintained, using the bell-jar in combination with the Dumarey equation or the NIST (National Institute of Standards and Technology in the United States) liquid standard reference https://doi.org/10.5194/amt-2020-314 Preprint. Discussion started: 21 September 2020 c Author(s) 2020. CC BY 4.0 License. material (SRM 3133) for calibration of the equipment. For the comparisons, gold sorbent tubes and carbon sorbent tubes were sampled with mercury to obtain transfer standards. For the first comparison, between NPL (The National Physical Laboratory 65 in the United Kingdom) and VSL, gold sorbent tubes have been sampled with mercury amounts between 2 ng and 10 ng.
Calibration of the analysers were performed with injections from the bell-jar in combination with the Dumarey equation.
During the second comparison, between JSI (Jozef Stefan Institute in Slovenia) and VSL, amounts between 10 ng and 1000 ng have been sampled and the NIST SRM 3133 has been used to calibrate the analyser used.

Primary mercury gas standard
The primary gas standard has been developed as an elemental mercury (Hg 0 ) gas generator to establish metrological traceability of mercury concentration measurement results, based upon a gravimetric approach, for ambient air levels as well as higher concentrations , de Krom et al., 2020 .
The working principle of the primary mercury vapour generator is based on diffusion according to ISO 6145-8 (ISO 6145-8, 75 2005). Specially developed diffusion cells filled with Hg 0 are kept under nitrogen in temperature and pressure-controlled diffusion chambers. The generated Hg 0 vapour diffuses from the cell through a capillary. By weighing the diffusion cells at regular time intervals on a high-resolution analytical balance (AX1006 mass comparator with a Mettler AT1005 balance, Mettler, Switzerland) the Hg 0 mass flow rate (diffusion) is determined. Standard Hg 0 gas mixture concentrations are prepared by mixing the Hg 0 vapours in nitrogen with known flows of matrix gas, e.g., purified air. 80 The characterisation of different diffusion cells has been described previously by I. de Krom et al (de Krom et al., 2020). Using diffusion cells with a capillary of 3 mm in diameter, mercury concentrations between 0.1 µg m -3 and 2.1 µg m -3 can be obtained with an expanded uncertainty of 3 %. Diffusion cells with a capillary of 33 mm in diameter can generate mercury concentrations between 5 µg m -3 and 100 µg m -3 with an uncertainty of 1.8 %. In this project both types of diffusion cells have been used to obtain the primary mercury gas standard. 85

Preparation of transfer standards
Transfer standards are prepared via pumped sampling of known volumes of the primary standard for a known time onto sorbent traps according to ISO 16017-1 (ISO 16017-1, 2000), using a specially designed multi-sampling manifold, made up of 6 thermal Mass Flow Controllers (MFCs) operating in sucking mode. Each MFC is connected to a 3-way valve. All valves are https://doi.org/10.5194/amt-2020-314 Preprint. Discussion started: 21 September 2020 c Author(s) 2020. CC BY 4.0 License.
controlled with a timer, that allows them to switch simultaneously. The sampling is done under controlled environmental 90 conditions.
The mercury amount collected onto the sorbent material can be calculated using equation (1).
Hg is the mercury amount on the sorbent material in ng, ̅ (Hg) is the mercury diffusion rate from the diffusion cell(s) in ng min -1 , t is the sampling time in minutes, (sample) is the pumped sampling flow in L min -1 and ( ) is the total gas 95 flow in L min -1 through the generator to obtain the primary mercury gas standard. As an example, the standard measurement uncertainty associated with a mercury amount of 10.5 ng, obtained using three diffusion cells with a capillary of 3 mm in diameter, can be calculated from equation (1) using the law of propagation of uncertainty of the Guide to the expression of Uncertainty in Measurement (GUM) (BIPM, 2008). In Table 1 the uncertainty budget belonging to the mercury amount is shown. The expanded uncertainty is 0.3 ng (k = 2), which is equivalent to a relative expanded uncertainty of 3%. 100

First comparison: 2 -10 ng mercury
Amasil sorbent tubes (PS Analytical, UK) with a specific area of 100 m 2 g -1 have been used in this study. The sorbent material 105 used in these tubes is gold-coated silica. The primary standard is obtained using two diffusion cells with a capillary of 3 mm in diameter (diffusion rate of 2.92 ng min -1 ± 0.09 ng min -1 (k = 2)). The sorbent tubes were sampled, during three different rounds (in May, July and September 2017), with 5 different amounts (2 ng, 4 ng, 6 ng, 8 ng and 10 ng) using a sample flow of 0.5000 L min -1 ± 0.0018 L min -1 (k = 2) for all sorbent tubes. During round 1 (May 2017), 5 sorbent tubes per amount were sampled simultaneously, 2 sorbent tubes were analysed by NPL and 3 by VSL. Thereafter, 2 tubes per amount were sampled simultaneously in round 2 (July 2017) and round 3 (September 2017) which were analysed by NPL (Table 2). Both laboratories calibrated the analyser using injections from a bell-jar in combination with the Dumarey equation.  (Brown et al., 2012)). The instrument was calibrated prior to analysis using a gas-tight syringe, making multiple injections of known amounts of mercury 120 vapour from the bell-jar onto the permanent trap of the analyser, across the range of expected sample concentrations. Sampled adsorption tubes were placed in the remote port of the instrument and heated to 900 ᵒC, desorbing the mercury onto a permanent trap. Subsequent heating of this trap then desorbed the mercury onto the detector for final measurement. Samples are desorbed three times to ensure all the mercury has been removed. It is assumed that the third desorption is equal to the blank level of the tube and this response is taken off the response of the first and second desorption, which are then added together to provide a 125 total analytical response. Quality control injections are made in-between samples to ensure that the analyser is not drifting https://doi.org/10.5194/amt-2020-314 Preprint. Discussion started: 21 September 2020 c Author(s) 2020. CC BY 4.0 License. outside a specified range. A conservative estimate of the relative expanded analysis uncertainty is between 8 % and 10 %. The main components of this uncertainty come from the uncertainty in the response from the adsorption tube during desorption, the repeatability of the instrument response and residual instrument drift (Brown et al., 2008c).

130
At VSL a PS Analytical 10.525 Sir Galahad (PS Analytical, UK) is used to analyse the sorbent tubes. For calibration a commercially available bell-jar has been used for comparison between the primary mercury gas standard and the Dumarey equation. The Tekran® Model 2505 (Tekran, USA) mercury vapour calibration unit is based on the Bell-jar principle. Since the saturation vapour pressure of mercury is a function of temperature, the exact volume injected, and temperature of the 135 mercury saturated air need to be set in order for the Bell-jar to determine the mercury injection amount based on the Dumarey equation (Brown et al., 2008b, Dumarey et al., 1979, Dumarey et al., 1985. Injections, by use of a gas-tight syringe, between 100 µL and 800 µL have been used (including a zero-linearity check point) to calibrate the analyser in the range of 1.3 ng to 10.5 ng.

Second comparison: 10 ng -1000 ng mercury 140
Gold sorbent tubes have been used for the 10 ng, 50 ng, 100 ng and 500 ng samples and carbon traps for the 1000 ng samples.
Commercially available gold-coated silica (Brooks Rand Instruments, US) and in-house prepared gold-coated high-grade corundum sand (mass fraction gold of 9.6 %) have been used in this study to prepare gold traps. Clean quartz tubes (i.d. 5 mm) were filled with gold-coated silica or corundum (20 mm length) and fixed with quartz wool. Carbon traps were prepared by filling clean quartz tubes (i.d. 5 mm) with 20 mm length of iodinated activated carbon (AIC-500 from Apex Instruments, US) 145 and fixed with quartz wool. For this comparison 3 or 4 sorbent tubes were sampled simultaneously at 5 different levels according to Table 3 using the diffusion cells with a capillary of 3 mm or 33 mm in diameter (diffusion rates of 4.23 ng min -1 or 218.5 ng min -1 respectively). The air flow through the system was kept constant at 3.00 L min -1 ± 0.02 L min -1 (k = 2), while other parameters, i.e., the diffusion cells, sample flow and sampling time were changed during the different entries to demonstrate the variability of the primary mercury gas standard set up. 150  (15) The sorbent tubes were analysed directly after sampling by JSI in the VSL laboratories in November 2018. The amount of mercury on the carbon traps was determined using mercury analyser RA-915M with a PYRO-915+ thermal decomposition 155 attachment (Lumex Scientific, St. Petersburg, Russia), that is based on differential Zeeman atomic absorption spectrometry.
Iodinated activated carbon and quartz wool from the carbon trap were quantitatively transferred to a quartz boat and mercury was released from the sample by combustion in the thermal decomposition unit at 700 ᵒC. The system was calibrated by spiking a known amount of NIST SRM 3133 standard solution (1000 ng) to iodinated activated carbon and determining the corresponding signal using the same procedure as for the sample. The signal of procedural blanks (average value 1.9 ng) was 160 subtracted from the corresponding sample (standard) signal.
The amount of mercury on the gold traps was determined using mercury analyser RA-915M with a modification of the PYRO-915+ to allow direct desorption of mercury from the gold trap in the thermal decomposition unit. The system was calibrated by reducing a known amount of NIST SRM 3133 standard solution (10 ng -500 ng) with tin(II) chloride solution, quantitative purging of the obtained Hg 0 gas onto a gold trap (EPA Method 1631, 2002, and its desorption in the thermal decomposition 165 unit (Shulupov et al., 2004). The signal of procedural blank was always within the instrumental noise and could not be subtracted from the corresponding sample (standard) signal.
The uncertainty of the analytical procedure was estimated using the law of propagation of uncertainty (BIPM, 2008). Estimated relative standard uncertainty of the calibration was calculated from the contributions of the NIST SRM 3133 (0.24 %, k = 1), uncertainties of pipettes used for spiking and preparation of NIST SRM 3133 dilutions, and the uncertainty of the volume due 170 to possible temperature changes. The estimated relative standard uncertainty of the calibration ranged from 0.5 % to 0.6 % (k = 1) for 1000 ng and 10 ng spike, respectively. The relative expanded uncertainty of the whole analytical procedure was usually 2.5 % -3 % (k = 2) and included effects such as sampling repeatability as the greatest contribution, and effects due to calibration and recovery. https://doi.org/10.5194/amt-2020-314 Preprint. Discussion started: 21 September 2020 c Author(s) 2020. CC BY 4.0 License.

First comparison: 2 ng -10 ng mercury
For the first comparison Amasil sorbent tubes were sampled in three rounds with amounts between 2 ng and 10 ng of mercury. Two tubes sampled simultaneously per level, as well as blank tubes, were sent to NPL for analysis after each sampling for each of the three rounds. Directly after round 1 three tubes were analysed by VSL for each sampling level. Both VSL and NPL calibrated their analyser with injections from the bell-jar and the mercury amount is calculated using the Dumarey equation. 180 This approach enables determining the difference between the primary mercury gas standard and the Dumarey equation. After calibration of the equipment the mercury amount on the tubes is verified against the Dumarey equation (Figure 1). https://doi.org/10.5194/amt-2020-314 Preprint. Discussion started: 21 September 2020 c Author(s) 2020. CC BY 4.0 License.  1.4 and 1.5, in  the second round 2.1 and 2.2 and in the third round 3.1 and 3.

2.
The blank tubes were not sampled with mercury but have been returned to NPL to check for contamination during transport 190 and storage. The recovery of all the blank tubes analysed during the three rounds is below 0.1 ng. The repeatability and (over three rounds, with exception of the 6 ng samples) is 3 %. For the 6 ng samples the reproducibility standard deviation is 6.7 %. The reproducibility standard deviation of the verification results is equal to the uncertainty of the mercury amount 195 (Table 1)

Second comparison: 10 ng -1000 ng mercury
During the second comparison 3 or 4 sorbent tubes were sampled simultaneously at VSL and analysed directly at VSL. For 205 the verification the average recovery of the 3 or 4 sorbent tubes is reported (Figure 2). During the verification the analysers are calibrated with the NIST liquid reference material (SRM 3133). For Entry 1 -5 (10 ng, 50 ng and 100 ng (Table 3)) the diffusion cells with a capillary of 3 mm in diameter have been used. The diffusion cells with a capillary of 33 mm in diameter have been used for sampling the other Entries (Entry 6 -14; 50 ng, 100 ng, 500 ng and 1000 ng (Table 3)). As such, for sampling the sorbent tubes with 50 ng and 100 ng, both types of diffusion cells have been used. 210 https://doi.org/10.5194/amt-2020-314 Preprint. Discussion started: 21 September 2020 c Author(s) 2020. CC BY 4.0 License. The repeatability and reproducibility standard deviation of the 10 ng and 500 ng samples have been calculated according to 215 ISO 5725-2 (ISO 5725-2, 2019). Repeatability standard deviations of 0.8 % for both levels have been obtained and the reproducibility standard deviation is 1.3 % and 0.9 % for the 10 ng and 500 ng samples, respectively. For the samples obtained using the diffusion cells with a capillary of 3 mm in diameter an average difference, between the reference values and the verification results, of +3.6 % has been obtained, except for Entry 4. In that case the reference value and verification result do not overlap, with a difference of +9.6 %. For the samples obtained using the diffusion cells with a capillary of 33 220 mm in diameter the difference between the reference values and the verification results is very small, just +0.1 %. The average difference between all the reference values and verification results is +1.3 % with a standard deviation of 3 %. This implies that the primary mercury gas standard and the NIST SRM 3133 are comparable within 1.3 % in contrast to the output of the Dumarey equation used in the first comparison, which are comparable within 8 %. These results are in agreement with earlier research which demonstrated that measured values based on the Dumarey equation are 7 % below those based on the 225 Huber equation .

Conclusion
The work described in this paper describes the comparison between mercury calibration methods maintained by two National Metrology Institutes (NMIs) and a Designated Institute (DI), using the bell-jar in combination with the Dumarey equation or NIST SRM 3133 for the calibration of the equipment. The results of the comparisons show that robust mercury transfer 230 standards can be obtained via sampling of the primary mercury gas standard with a reproducibility standard deviation 3 %, which is equal to the uncertainty of the mercury amount sampled onto the sorbent materials. Transfer standards containing gold or carbon as sorbent material can be prepared with levels between 2 ng and 1000 ng with a relative expanded uncertainty 3 %.
Based on the method used for the calibration of the analyser a difference of -8 % has been obtained between the primary 235 mercury gas standard and the Dumarey equation and a difference of +1.3 % when using NIST SRM 3133 to calibrate the analysers. This implies that the use of the primary mercury gas standard and NIST SRM 3133 leads to comparable measurement results, whereas a difference of approximately 8 % exists between measurement results based on these two versus the output of the Dumarey equation. Based upon this work it should be emphasized that measurement results based on the Dumarey equation have a negative bias of approximately 8 %. 240 The transfer standards obtained with the primary mercury gas standard proved to be useful to establish metrological traceability to the SI units and can be used for the calibration of mercury equipment used in the field, e.g. analysers, bell-jars and gas generators. Furthermore, such transfer standards can be used to benchmark the results of laboratories involved in mercury measurements, e.g. by proficiency tests. The obtained results enable traceable mercury measurement results in emission sources and the atmosphere. These measurements are of fundamental importance to reduce the mercury burden on the 245 environment, to comply with related regulation and protect human health.