9. History of Batch-to-Batch Comparative Studies of International Association for the Physical Sciences of the Oceans Standard Seawater

In 2010, the Thermodynamic Equation of Seawater 2010 (TEOS-10) was adopted as the standard description for the thermodynamic properties of seawater in oceanography. TEOS-10 replaced the previous standard, the equation of state of seawater 1980 (EOS-80). Millero (2010) has reviewed the history of the equation of the state of seawater, and Pawlowicz et al. (2016) have reviewed the history of oceanic salinity measurements. Since 1902, and through a variety of changes in the definitions of standards and salinity, a certified reference material known as standard seawater (SSW) has played an important role in ensuring the comparability of observed salinity data (Bacon et al., 2007; Culkin & Ridout, 1998; Culkin & Smed, 1979). SSW is created in batches and the most recent batch is numbered P165. Each batch of SSW is created by sampling seawater from a particular region of the North Atlantic, filtering it to remove organic material that could cause problems in long-term storage, carefully measuring a particular characteristic of the seawater, and then bottling and labeling it. Before 1978, the characteristic was chlorinity; since then, the characteristic has been the conductance ratio relative to a specified KCl solution (Chap. 8).

Because of changes to the definition of salinity and to the container used for distributing SSW, the consistency among definitions and the comparability of certified values of SSW between different batches have been examined in a number of studies (e.g., Aoyama et al., 2002; Bacon et al., 2000, 2007; Culkin & Ridout, 1998; Kawano et al., 2000, 2001, 2006; Mantyla, 1980, 1987, 1994; Millero et al., 1977; Park, 1964; Poisson, 1975; Poisson et al., 1978; Takatsuki et al., 1991; Uchida et al., 2020). Although the consistency between batches of SSW has improved in recent decades (Bacon et al., 2007; Kawano et al., 2006; Takatsuki et al., 1991), an inconsistency of about ±0.001 in Practical Salinity (S_P) still exists (Uchida et al., 2020). The magnitude of this inconsistency is significantly greater than the expanded uncertainty of ±0.0004 for the certified values (in S_P) estimated for batches of SSW prepared shortly before 2007 by Bacon et al. (2007). Careful corrections for this inconsistency have made it possible to detect small changes in S_P (−0.0006 ± 0.0001 dec⁻¹) in the deep ocean that might be related to global climate change (Uchida et al., 2020). This means that salinity measurements conducted in the deep ocean for climate studies must be of the highest possible accuracy, as close to the resolution of the salinometers (0.0002 in S_P) as possible.

In this study, we review the history of those studies of batch-to-batch comparisons and present new results of batch-to-batch comparisons of old and recent SSWs. We also fill in the blanks in the long-term batch offset table for batches P113 and P117. We then describe our current effort to establish a high level of comparability between salinity measurements.

Since the 1950s, electrical conductivity meters (salinometers) have been widely used to measure the salinity of seawater (e.g., Park & Burt, 1965a, 1965b). A salinometer is usually calibrated with SSW by adjusting the salinometer to read the conductivity corresponding to the chlorinity of the SSW on the assumption that the relationship between chlorinity and conductivity is the same for all batches of the SSW.

Park (1964) first examined the reliability of SSW as a conductivity standard. Computed salinities from certified chlorinities were compared with salinities estimated from conductance measurements by using an inductive salinometer for batches P15 (prepared in 1937), P18, P24–P26, P29–P32, P35, and P36. Poisson (1975) also conducted measurements of electric conductivity relative to a KCl solution for batches P37, P49, P50, P53, P56, P62, and P64 by using a Jones-type bridge and a cell with bright platinum electrodes. These two studies revealed that the salinity estimated from the conductivity of some batches was higher than that calculated from the certified chlorinity. The Joint Panel on Oceanographic Tables and Standards (JPOTS) recommended that the measurements be repeated by other laboratories, and 26 batches of SSW prepared during the period 1962–1975 were distributed to four laboratories. The results of one laboratory were published independently (Millero et al., 1977), and Poisson et al. (1978) summarized the results of the inter-laboratory, batch-to-batch comparisons using salinometers with the results from the previous studies of Park (1964) and Poisson (1975) for batches P37–P41, P44, P46–P56, P59–P62, P64, and P66–P71. Chlorinity-based salinities (S_Cl) were computed from certified chlorinities (Cl) using the equation S_Cl = 1.80655 Cl. The discrepancy between the S_Cl values and salinities estimated from conductance measurements (mean ± SD was 0.0018 ± 0.0020‰) was considered significant, and it was suggested that variability in the relationship between the chlorinity and conductivity of SSW was the cause of this discrepancy. Although the density, pH, and concentrations of silicate and dissolved organic carbon were also measured for those batches, those measurements did not suggest a clear explanation for the discrepancy. Poisson et al. (1978) concluded that the electrical conductivity as well as chlorinity of each batch of SSW should be certified as soon as possible.

After considering a new equation of state of seawater, the JPOTS recommended the Practical Salinity Scale 1978 (PSS-78) (UNESCO, 1981), and the relationship between chlorinity and S_Cl was subsequently abandoned in favor of a conductivity–S_P relationship (Millero, 2010). Batches of SSW since P91, labeled with the results of conductance measurements, have all been made in basically the same way by Ocean Scientific International Ltd. (UK).

Because SSW has been considered to have a lifetime of a few years, long-term monitoring of SSW characteristics has been carried out by performing batch-to-batch comparisons during periods of overlapping lifetimes. These comparisons have then been cumulatively “stitched together” to form a table of long-term batch offsets. Mantyla (1980) used a salinometer to conduct batch-to-batch comparisons of batches P35–P37, P39–P47, P49, P51, P52, P54, P56–P59, P61, P63, P65, P66 P69, P70, and P72–P84. Mantyla (1980) then combined these results with the results of 10 other published and unpublished batch-to-batch comparison experiments, including the results from Park (1964), Poisson (1975), Millero et al. (1977), and Poisson et al. (1978), to make a batch offset table for batches P29–P84. Because all of these comparisons were relative to an arbitrarily selected reference batch of SSW, adjustments to a common batch of SSW were usually made based on overlaps of the same batches of SSW in the various comparisons. Because batch P79 was the standard used to determine the KCl concentration for PSS-78, the batch offset table for batches P29–P84 was proposed relative to batch P79.

Mantyla (1987) used a salinometer to continue batch-to-batch comparisons for batches P36, P40–P47, P52, P59, P61, P63, P70, P73, and P76–P102 and updated the batch offset table for batches P29–P102 by combining those results with the results of Mantyla (1980). The offsets were calculated relative to a new reference based on the assumption that the average of the batch offsets for P91–P102 was zero. The rationale was that SSW was certified by the K₁₅ value defined by the PSS-78 rather than by the chlorinity from batch P91, where K₁₅ is the electrical conductivity ratio relative to a KCl solution (32.4356 g kg⁻¹) at a temperature of 15 °C (International Practical Temperature Scale of 1968 [IPTS-68]) and a pressure of one standard atmosphere.

After Mantyla (1987), Takatsuki et al. (1991) used a salinometer to conduct batch-to-batch comparisons for batches P90, P100, P104, P106, P108, P111, and P112. Mantyla (1994) also used a salinometer to make batch-to-batch comparisons for batches P103–P110. Aoyama et al. (2002) used salinometers to conduct interlaboratory batch-to-batch comparisons for batches P110, P112, P114, P116, P118–P124, and P127–P129 and expanded the batch offset table proposed by Mantyla (1987) up to P129 by combining those results with the results from Takatsuki et al. (1991) and Mantyla (1994). Because the results of Aoyama et al. (1998) (non-peer-reviewed letter, International WOCE Newsletter, 32, 5–7) were later published as a peer-reviewed paper (Aoyama et al., 2002), those results are referred to as Aoyama et al. (2002) in this manuscript.

Culkin and Ridout (1998) recalibrated SSW batches P120–P129 relative to newly prepared solutions of KCl and examined the variation with age of their K₁₅ values for a maximum of two years. They also examined the shelf life of SSW stored in a borosilicate glass bottle for batch P123 and confirmed no significant change in the K₁₅ value for three years. Bacon et al. (2000) also examined aging variations of batches P115–P132 by using salinometers to conduct batch-to-batch comparisons. The results were compared with the batch-to-batch offsets proposed by Aoyama et al. (2002).

Following Aoyama et al. (2002), Kawano et al. (2000) conducted batch-to-batch comparisons of batches P70, P88, P94, P114, P116, P119, P121, P123, P124, P127–P129, and P132-P135, and Kawano et al. (2001) used salinometers to conduct batch-to-batch comparisons of batches P123 and P132–140. Kawano et al. (2006) used salinometers to conduct interlaboratory batch-to-batch comparisons of batches P133 and P135–P145 and expanded the batch offset table of Aoyama et al. (2002) up to P145 by including the results of Kawano et al. (2000, 2001). Kawano et al. (2006) also filled in the blanks of the batch offset table for batches P115, P125, P126, P130, and P131 by using results from Culkin and Ridout (1998) and Bacon et al. (2000). Kawano et al. (2006) reanalyzed those results (Table 5 in Kawano et al., 2006) because they were regarded as batch-to-batch comparison experiments and were consistent with the other results of Kawano et al. (2006). Finally, Kawano et al. (2006) proposed a new batch offset table relative to the new reference. The new batch offset table assumed the average of the batch offsets for P130–P145 to be zero because consistency among those batches was much better than the consistency of the older batches. Kawano et al. (2006) demonstrated a reduction of inconsistency at 105 crossover points of the World Ocean Circulation Experiment (WOCE) sections by adding 14 crossover points in the Indian Ocean to the 91 crossover points in the Pacific Ocean and Atlantic Ocean reported by Aoyama et al. (2002). The standard deviation of duplicate measurements of S_P was reduced from 0.0020 to 0.0017 by applying the SSW batch offset correction. Kawano et al. (2006) also demonstrated a reduction of unrealistic S_P changes with time in the deep ocean by applying batch offset corrections to WOCE sections and their repeat hydrographic sections 10 year apart.

Bacon et al. (2007) estimated for the first time the uncertainty of the label value of manufactured SSW. Following Culkin and Ridout (1998), they recalibrated SSW batches P130–P144 with reference to newly prepared solutions of KCl. No significant change was found in the label value outside the expanded uncertainty with a coverage factor of 2 (S_P difference of 0.0004) for as long as 5 year after the SSW batch manufacture.

The container used for SSW was changed from a soda-glass ampoule (up to P139) to a borosilicate glass bottle (from P140) in 2000. The only exceptions were P138 (both ampoule and bottle) and P142 (ampoule). For batch P138, Kawano et al. (2006) evaluated the bottle-type SSW, and Bacon et al. (2007) evaluated the ampoule-type SSW.

Following Kawano et al. (2006), Uchida et al. (2020) used salinometers to conduct batch-to-batch comparisons of batches P138, P141, P142, and P144–P163 and expanded the batch offset table of Kawano et al. (2006) up to P163 with a revision of the offset for P145. Because the Bacon et al. (2007) results suggested little change in the characteristics of SSW between batches, in apparent contradiction to the batch-offset tables being determined by other workers, Uchida et al. (2020) compared batch-to-batch offsets for batches P130–P144 with the corresponding batch-to-batch differences estimated from Bacon et al. (2007) and found no significant change in label S_P outside the expanded uncertainty (S_P difference of 0.0004) in both cases. Thus, it appears that the batches considered by Bacon et al. (2007) were of unusually consistent quality.

In practical terms, the use of batch corrections does have consequences for ocean climate research. Uchida et al. (2020) detected recent freshening (−0.0006 ± 0.0001 g kg⁻¹ dec⁻¹) in the deep North Pacific by applying the batch-to-batch corrections to conductivity–temperature–depth (CTD) data (36 CTD profiles during two recent decades) accumulated at time series station K2 (47°N, 160°E; water depth 5215 m). It was difficult to detect such a slight trend in salinity without applying the batch corrections. Uchida (2019) expanded the batch offset correction table up to P165 following Uchida et al. (2020) (the expanded batch offsets were −0.0005 and −0.0001 in S_P for batches P164 and P165, respectively).

The batch offset values proposed by those previous studies were summarized in three papers (Kawano et al., 2006; Mantyla, 1987; Uchida et al., 2020). Figure 9.1 shows the batch offsets of S_P for all of these studies relative to the new reference proposed by Kawano et al. (2006) and fills in the blanks in the batch offset values for batches P113 and P117 (see Sect. 9.6). The batch-to-batch variation was quite large for batches before P90 (1980), which were calibrated based on chlorinity, but it has been significantly smaller for batches after P91 (1980) which were calibrated based on KCl reference. However, there has been a long-term trend in the offsets, which have decreased by about 0.0025 over the last 40 year. Uchida (2019) has provided the batch offsets for batches P29–P165, and that online dataset will be expanded for the most recent batches.

Bild vergrößern

Although SSW has a finite lifetime, unused vials of older batches can sometimes be found in the storage areas of marine institutions, leftover as new batches are received and integrated into standard measurement protocols. In an attempt to check the consistency of the long-term batch correction table (Fig. 9.1), which was created by stitching together batch-to-batch comparisons made over many years, a collection of historical SSW stored from as far back as 1969 was created, and as salinometer (Autosal model 8400B; Guildline Instruments, Ltd., Canada) was used to conduct batch-to-batch comparisons of this collection at the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) at the time of the batch offset determination for P164 in September 2020 (Uchida, 2019). The dates of measurement were 7 September (for P155–P164), 8 September (for P148–P154), 9 September (for P138, P141, P143–P147), 17 September (for P130–P137, P139, P142), and 18 September (P51, P82, P84, P85, P88, P93, P103, P112, P114, P119, P121–P124, P127, P128). At least five bottles of P161 were measured on each day of the measurements to correct for temporal drift of the salinometer. Measured S_P was calibrated against the batch-offset-corrected values by adjusting the mean value for the reference batches P160–P163.

Figure 9.2 shows the differences in S_P from the batch-offset-corrected values. The differences for older SSWs that had been sealed in soda-glass ampoules and stored for more than 20 year were highly variable (maximum S_P difference of 0.08), very much greater than the proposed batch offsets, and they tended to be positive (Fig. 9.2a). Although evaporation of water was unlikely to have affected the salinity of the SSW in the ampoules, we observed suspended matter resembling glass flakes and crystals on the inner wall of the glass. The seawater inside the ampoules as well as the glass container may have undergone chemical changes. Therefore, these results were probably unreliable. However, the oldest SSW, P51 (produced in 1969), seemed to have been kept in a refrigerator for a long time (but recently stored at room temperature for several years), and the average of the differences of the S_P for five ampoules was close to zero (−0.0023), although the variation was large (SD of 0.0167).

However, differences for more recent SSWs stored in borosilicate glass bottles were close to zero (maximum S_P difference of 0.0012) (Fig. 9.2b). With the exception of the oldest batch (P138) among the borosilicate glass bottles and batches P146–P150 (12–15 year-old), the differences of S_P were within ±0.0003, smaller than the expanded uncertainty of SSW S_P of 0.0004 according to Bacon et al. (2007). However, for P146–P150, there was a negative bias (Fig. 9.2b). The history of the batch-to-batch comparisons suggested that the S_P of these batches had drifted with time at a rate of about −0.0005 per decade, albeit with considerable uncertainty (Fig. 9.3) (Uchida, 2019).

Changes in conductivity/density/salinity relationships in real seawater have been linked to changes in the concentrations of seawater constituents that are involved in biogeochemical cycling (Pawlowicz et al., 2011). To determine whether similar factors were related to batch offsets, total alkalinity (TA) and concentrations of dissolved inorganic carbon (DIC), nitrate, silicate, ammonia, and dissolved organic carbon (DOC) were measured for batches P138, P141, P143–P151, P155, and P164 in 2020–2021 (Table 9.1). The focus of the study was the temporal drift of the S_P of the SSW (Fig. 9.3).

Bild vergrößern

Bild vergrößern

Uchida et al. (Chap. 10) have examined changes to the composition of SSW since 2010, including the results reported here. They found that (i) silicate increased with time (5.47 μmol kg⁻¹ year⁻¹), (ii) DIC decreased with time (−13.5 μmol kg⁻¹ year⁻¹), (iii) TA, nitrate was constant, and (iv) DOC and ammonia were constant for batches after P148, although they were highly variable for batches before P147. For batch P145, TA and DIC were much lower than usual (~420 μmol kg⁻¹ and ~300 μmol kg⁻¹, respectively). As Poisson et al. (1978) also found, there was no clear relationship between these compositional changes and the batch offsets of S_P (ΔS_P values in Table 9.1). However, the decrease of S_P (e.g. −0.0013 per 15 year for P147) might have been partly related to the rate of decrease of DIC (−203 μmol kg⁻¹ per 15 year) reported by Uchida et al. (Chap. 10), because a DIC change of 203 μmol kg⁻¹ will change S_P by 0.0014 (Pawlowicz, 2010).

The batch offset table, mostly based on measurements in shore laboratories, was also evaluated by conducting batch-to-batch comparisons during routine oceanographic observation cruises by the Japan Meteorological Agency (JMA). When a new batch of SSW is introduced, batch-to-batch comparisons are performed with the previously used one or two batches to check the inter-cruise consistency of S_P measurements (Table 9.2). Because of the small number of batches in each batch-to-batch comparison, the measured batch-to-batch offset difference between two batches was compared with the corresponding batch-to-batch offset difference calculated from the offset table (Fig. 9.4). The measured differences agreed well with the calculated differences from the offset table, except for the comparison in 2017. The offset difference between P159 and P160 in 2017 differed significantly (S_P difference of −0.0011) from the offset difference calculated from the offset table. Because the results for P157 and P159 in 2016 and for P160 and P161 in 2018 were consistent with the offset table, either the P159 or P160 measurements in 2017 could be the source of the discrepancy.

Bild vergrößern

Following the methodology of Saunders (1986), who used the deep ocean as a natural calibration tank, S_P of North Pacific bottom water at the same location (40°N, 165°E) measured on each cruise in 2014, 2016, 2017, 2020, and 2021 were compared to determine the source of the discrepancy (Table 9.3). The standard deviation of the batch-offset-corrected S_P of the bottom water was small (0.0002) and was equal to the resolution of the salinometers. Because the bottom water S_P in 2017 was calibrated with P159, the offset for P160 measured in 2017 must have deviated by about −0.0011 from the batch offset value of Uchida et al. (2020). Because the offset difference between P160 and P161 in 2018 was consistent with the offset difference of Uchida et al. (2020), only the S_P of the bottles for P160 that were obtained on the KS1704 cruise in 2017 may have been low, but for reasons that are unknown.

There are blanks (batches P113 and P117) in the batch offset table for SSW batches P29–P145 reported by Kawano et al. (2006). Batch offsets for batches P113 and P117 were therefore estimated to fill the blanks by using results of batch-to-batch comparisons reported in previous publications.

Joyce et al. (1992) measured conductivity ratios with a salinometer for several ampoules of batches P113 and P114 during an oxygen/salinity comparison cruise in 1991. Measured S_P were 34.9947 (18:25, July 1st), 34.9937 (18:15, July 2nd), and 34.9970 (18:00 July 5th) for batch P113; they were 34.9963 (18:45, June 30th), 34.9967 (20:30, July 1st), 34.9967 (18:00, July 3rd), 34.9967 (19:30, July 3rd), and 34.9967 (18:00, July 6th) for batch P114. Averages of the S_P were 34.9951 and 34.9966 for batches P113 and P114, respectively. The label values were 34.9937 and 34.9945 for the S_P of batches P113 and P114, respectively, and the batch offset of the S_P was 0.0020 for batch P114. The batch offset of the S_P of batch P113 could therefore be estimated to be 0.0013.

Parrilla (2007) measured 8 ampoules of batch P120 after the salinometer was standardized with batch P117 and reported that the label value (S_P of 34.994) was within 0.0002 of the S_P. Therefore, the batch offset for batch P117 can be estimated to be the same as the batch offset for P120 (−0.0009 in S_P).

The estimated batch offset for batch P117 was evaluated in the deep ocean by applying the batch offset correction to the WOCE hydrographic data (available from CCHDO https://​cchdo.​ucsd.​edu/​). Bottle-sampled S_P data calibrated with batches P117, P123, and P126 obtained during cruises 18HU93019_1 (June 1993), 18HU19940524 (May 1994), and 18HU95011_1 (June 1995), respectively, along the WOCE AR7W line across the Labrador Sea were compared. Data obtained at three stations between latitudes of 57.36°N and 58.23°N for each cruise were used. The T–S relationship for waters deeper than 3000 dbar (maximum pressure of 3657 dbar) was estimated by fitting a line for each cruise, and the S_P at a potential temperature of 1.7 °C (average pressure of 3446 dbar) was extracted from the regression line. The extracted S_P were 34.8906, 34.8867, and 34.8869 for 1993, 1994, and 1995, respectively. The agreement between them was improved by application of the batch offset correction (S_P values of 34.8897, 34.8874, and 34.8875 for 1993, 1994, and 1995, respectively).

Although we believe that the batch offsets and long-term drifts summarized here are reliable estimates of the variability in SSW, these batch offsets are not the only possible source of systematic errors in salinity measurements. Gouretski and Jancke (2001) have reported inter-cruise offsets of salinity for a global hydrographic dataset. They have estimated the offsets for historical cruises to be on average 3–6 times the modern cruise offsets. Such large offsets could not be explained by SSW batch offsets. Purkey and Johnson (2013) have also reported inter-cruise offsets of S_P for WOCE cruises and their reoccupation cruises. Although the application of SSW batch offset corrections improved the agreement among occupations of each section, it did not eliminate the need for ad-hoc offsets to reduce inter-cruise S_P biases. Purkey et al. (2019) have closely examined inter-cruise offsets in S_P for the WOCE cruises and their reoccupation cruises in the South Pacific Ocean and have found that the difference in S_P between cruises occupied after 2000 has been less than 0.0001. They therefore applied an additional ad hoc S_P offset, ranging from −0.0037 to 0.0001, to most of the WOCE occupations in the 1990s. This purely empirical result is consistent with our analysis in suggesting that the change of SSW from soda glass ampoules to borosilicate glass bottles in 2000 improved the stability of the S_P of SSW.

Mantyla (1994) accounted for salinity offsets due to (i) different types of sampling bottles (epoxy-lined Nansen bottles or PVC plastic Niskin bottles), (ii) time lags between water sampling and S_P determination, (iii) batch offsets of SSW, (iv) salinometer response shifts, and (v) differences between up (water sampling) and down (CTD profiling) cast values because of hysteresis in pressure, temperature, or conductivity sensors. Gouretski and Jancke (2001) have suggested an additional possible source of systematic errors in salinity: (vi) the order of samples for salinity drawn from sampling bottles. Specifically, condensation of water may occur on the bottle interior of the headspace, according to the ambient air conditions, if salinity sampling is conducted later (head space becomes large) and is delayed for a long time.

Although there are several possible sources of systematic errors in salinity as described above, only the following possible sources related to SSW are discussed here: (i) within-batch differences, (ii) variations of the chlorinity–conductivity relationship, (iii) initial offsets of label values, and (iv) variation of the conductivity of SSW with age.

Aoyama et al. (2002) have examined within-batch differences in terms of the standard deviation of the S_P of several ampoules from the same batch of SSW (12 batches among P64–P128). They found that the standard deviations of S_P were 0.0001–0.0004 during the first four years after production. For recent SSWs in borosilicate glass bottles, the standard deviations of S_P measured within the 3 year shelf life were also 0.0001–0.0004 for 13 batches from P152 to P165 (Table 9.2). The effect of the within-batch differences on an estimation of the batch-to-batch offsets is therefore expected to be smaller than the typical batch offsets if SSWs are measured within the 3 year shelf life.

The main cause of the batch-to-batch differences among batches up to P90 might be the variation in the relationship between chlorinity and conductivity because the old batches were calibrated based on chlorinity (Poisson et al., 1978). Although the definition of salinity was changed to the definition in PSS-78 for batches after P91, the SSW labels showed both the conductivity ratio and chlorinity until batch P113. Millero and Huang (2009) have compared the S_P calculated from the conductivity ratio with the salinity calculated from chlorinity (S_Cl) by using the values shown on the labels. In this study, the batch-offset-corrected S_P was compared with the S_Cl (Fig. 9.5). The systematic bias (−0.0017 in terms of S_P) of the differences was greatly reduced to −0.0003 by applying the batch offset corrections. However, the variability of the comparison was not greatly reduced. This remaining variability (SD of 0.0019) may then be explained by the variation of the chlorinity–conductivity relationship, because the magnitude of the variation is the same as the magnitude of the batch offset variations for batches earlier than P90 using the chlorinity reference (SD of 0.0018) (Fig. 9.1).

Bild vergrößern

For the initial offsets of label values of SSW, Kawano et al. (2005) examined supplier and lot dependencies of the electrical conductivity of KCl solutions due to impurities in the KCl reagents. They found that this effect on the uncertainty of SSW label S_P values was about 0.0012. However, Bacon et al. (2007) have examined the effect of impurities in the KCl and have found that it is smaller than a S_P difference of 0.0001, although the batch offsets for the period examined by Bacon et al. (2007) were small (S_P SD of 0.0003) (Uchida et al., 2020). Bacon et al. (2007) have estimated the overall expanded S_P uncertainty for the label value of SSW to be 0.0004 and have suggested that the largest uncertainties contributing to the overall uncertainty are (i) measurement of the KCl solution conductivity ratio, (ii) solvent conductivity in the KCl solutions (effectively CO₂ saturation uncertainty), and (iii) measurement of the new SSW conductivity ratio. In fact, Bacon et al. (2007) have reported that the recalibrated S_P values of SSW in reference to KCl solutions vary by ±0.0004 for batches P141 and P144 that were measured within a month. Because the K₁₅ value of the SSW is rounded to five significant digits, resolution of S_P calculated from the certified value is 0.0004, although the resolution of the salinometer model Autosal 8400B corresponds to an S_P difference of 0.0002. The expanded uncertainty of the SSW estimated by Bacon et al. (2007) should be increased slightly to a S_P difference of 0.0005 to take into consideration the uncertainty due to the resolution (0.0002/√3). An initial offset equal to an S_P uncertainty of ±0.0005 may therefore exist for the label value.

The values recalibrated with reference to KCl solutions for batches P120–P129 have been reported in Table 1 of Culkin and Ridout (1998). The corresponding values for batches P130–P144 that were reported in Table 5 of Bacon et al. (2007) were reanalyzed (Tables 9.4 and 9.5) using the batch offset estimation by Uchida et al. (2020) by assuming that the results obtained on the same measurement day used the same KCl solution. Offsets for the KCl solutions were adjusted so that overlapping data between measurement days were consistent with each other and so that the mean of the average offsets for P120–P144 matched the mean of the corresponding batch offsets (Uchida, 2019). Note that the relative variability of the average offsets was independent of the batch offset table. The adjustment values for each measurement day varied from −0.0003 to 0.0007, and the offset values for each batch also varied from −0.0003 to 0.0006. The average offsets agreed well with the batch offsets (Fig. 9.6). The standard deviation of the S_P differences between the estimated offsets and the batch offsets (Uchida, 2019) was 0.0003. These results suggested that the batch offsets were mostly explained by initial offsets of the label values for batches after P91.

Bild vergrößern

The average of the estimated initial offset of S_P values calibrated in reference to KCl solutions for batches P130–P144 (Bacon et al., 2007) was close to zero (mean ± SD is 0.0001 ± 0.0002), although the S_P offset for batches P120–P129 (Culkin & Ridout, 1998) was slightly biased (mean ± SD is 0.0004 ± 0.0002). The average of the recent batch S_P offsets for P130–P165 was also consistently close to zero (mean ± SD was −0.0001 ± 0.0005) (Fig. 9.1). Therefore, considering the fact that the systematic bias between the S_P calculated from the conductivity ratio and the salinity calculated from chlorinity by using the values shown on the labels for batches P91–P113 was greatly reduced by applying the batch offset corrections (Fig. 9.5), the arbitrarily selected reference for the batch offset table (the batch offsets for P130–P145 was assumed to be zero) would seem appropriate.

In addition to an initial offset, there may also be an aging component (as found for batches P146–P150, Fig. 9.3). Changes in properties of SSW over time, which can occur during storage due to reaction with the glass of the container and microbial activity, may contribute to conductivity changes (Poisson et al., 1978; Mantyla, 1980; also see Sect. 9.4). Changes in properties may have been large for the SSWs sealed in soda-glass ampoules because the quality of the soda glass of the ampoules was lower than that of the borosilicate glass of the bottles used for more recent SSWs, and the air space was larger in the ampoules than in the bottles. The larger air space allowed more “sloshing” inside the ampoules with any motion (Bacon et al., 2007). In fact, the S_P values of SSWs stored in ampoules increased greatly during long-term storage (Fig. 9.2a). A temporal drift and shift of S_P sometimes occurred after the initial calibration, even for SSWs stored in bottles, although the S_P values were relatively stable over time compared to the S_P values of SSWs stored in ampoules.

The S_P values for batches P146–P150 gradually decreased with time, as shown in Sect. 9.3. Batch P145, for example, was calibrated in July 2004 and was measured at JAMSTEC, Japan in March 2005 and at the Woods Hole Oceanographic Institution (WHOI), U.S.A. in April 2005 (see Kawano et al., 2006). The S_P offsets from the label value for P145 were estimated to be 0.0002 and 0.0001 based on the JAMSTEC and WHOI results, respectively. However, the S_P offset of about −0.001 that occurred within the 3 year shelf life after these measurements persisted for 15 year: −0.0010 in 2006, −0.0011 in 2007, −0.0009 in 2009, −0.0015 in 2010, and −0.0011 in 2020 for the measurements at JAMSTEC and −0.0011 in 2011 for the measurement at JMA (see Uchida, 2019; Uchida et al., 2020). In the case of batch P160, only the S_P values of bottles used on the KS1704 cruise in 2017 may have decreased by about −0.001, as described in Sect. 9.5. These results suggest that the S_P values of SSWs are usually stable within their 3 year shelf life but may tend to increase if they are stored in ampoules (batches before P139) and decrease with time for at least some SSWs stored in bottles (batches after P140).

A lack of traceability of the results of S_P measurement in the International System of Units (SI) is a fundamental problem with the current definition and technology of measuring S_P in seawater (Pawlowicz et al., 2016; Seitz et al., 2011). In contrast to fundamental physical phenomena, measurement standards such as SSW that are prepared by humans are inevitably subject to variations over time (Seitz et al., 2011). At present, it is possible to certify the electrical conductivity of SSW with traceability to the SI within an uncertainty of about 0.02% at the highest level (Seitz et al., 2010, 2019) and the corresponding S_P uncertainty of 0.008. Unfortunately, this level of uncertainty is too large to use in climate studies in the deep ocean (Uchida et al., 2020). Although the batch corrections for SSW are efficient in practice for establishing comparability for S_P measurements, errors in the offset estimation by chaining the batch-to-batch comparison experiments can accumulate. Pawlowicz et al. (2016) have therefore recommended the development of a parallel system of measuring the density of SSW to establish traceability of Absolute Salinity because density measurement results are traceable to the SI at nearly the precision required, and the density uncertainty will not increase with time. An expanded uncertainty with a coverage factor of 2 in seawater density measurement is 1.4 ppm at the highest level (Kayukawa & Uchida, 2021), which is equivalent to 0.002 g kg⁻¹ in Absolute Salinity.

A practical way to establish a high level of international comparability of S_P measurements (e.g., <0.001) would be to have a large amount of secondary reference seawater that was more robust and stable than SSW to estimate batch-to-batch differences of the SSWs due to initial offsets and changes in SSW conductivity ratios over time. Such a secondary reference seawater would require long-term stability in terms of not only S_P (based on conductivity) but also Absolute Salinity (based on density). Howell et al. (2010) have detected inconsistency (0.0013) of the S_P values of SSWs based on their in-house standard seawater when they began using SSW P151 instead of using P150 for the calibration of their salinometer during a period of a few months. Such in-house standard seawater is inappropriate for long-term use, although these inconsistencies could be completely explained by batch offsets (Uchida et al., 2020). The Chinese Primary Standard Seawater for S_P measurements in China is distributed in borosilicate glass bottles (Li et al., 2016) similar to the way IAPSO SSW is distributed and therefore will likely have the same stability problems as SSW.

Multiparametric Standard Seawater (MSSW) (KANSO TECHNOS Co., Ltd., Japan) might be an alternative candidate for such reference seawater, because MSSW is expected to be more stable than SSW with respect to both S_P and density (or Absolute Salinity). MSSW is created from North Pacific Intermediate Water collected from a depth of 400 m in Suruga Bay, south of Shizuoka, Japan, and stored in 500 mL aluminum beverage bottles with plastic inner caps with high gas and water vapor impermeability and is produced by a method similar to that used for the Reference Material for Nutrients in Seawater (RMNS) (KANSO TECHNOS Co., Ltd.), with no air space in the bottle. The use of aluminum bottles and plastic inner caps is meant to avoid the long-term increase in silicate seen in borosilicate glass bottles (Budéus, 2018; Uchida et al. 2011; Chap. 10) as well as other chemical changes (for more details, see Chap. 12).

Uchida et al. (2020) have reported the stability of the S_P of MSSW (lot Pre16) over seven years. They used 8 batches (SSW P153–P162 except for P158 and P160) of IAPSO SSW for the calibration of their salinometers. The standard deviation of the repeated S_P measurements for MSSW was reduced from 0.00036 to 0.00026 by applying the batch offset correction for IAPSO SSW. An ultra-high-resolution (~0.00013 g kg⁻¹ in Absolute Salinity) density sensor based on measurements of refractive indexes by the interference method (Uchida et al., 2019) will contribute substantially toward establishing the traceability of Absolute Salinity measurements to the SI, and it will be suitably calibrated with MSSW. We have started the evaluation of Practical and Absolute Salinities for IAPSO SSWs using MSSW (330 bottles of lot Pre20 produced in January 2022), hopefully for use over the next two decades in laboratories and on research vessels beginning in 2022.