OCADSAccess DataDiscrete / Bottle DataGLODAPEvaluation of Inorganic Carbon Quality

Evaluation of Inorganic Carbon Quality

The first task of the GLODAP synthesis project was to assemble a merged data set for each basin. The working data set was assembled at Princeton University (PU) and include all of the DOE survey cruises, all of the NOAA OACES cruises and many international WOCE and JGOFS survey cruises to obtain comprehensive spatial coverage. As the data sets were assembled, consistency was checked by comparing property-property and property vs depth plots for stations that are near (within 50 to 100 km) the intersection of cruise lines (the so-called crossover analysis).

This procedure is the first level of quality control and indicates, but does not eliminate, the possibility of systematic differences between cruises or oceans. The next step is to recommend adjustments to the inorganic carbon data based on a comprehensive check of analytical and data reduction procedures, analysis of crossover, and regional analysis of cruise data. This is necessary to produce a gridded data set of data that is both precise and accurate on a global scale. The quality assurance/quality control (QA/QC) procedure involved a careful examination using the following techniques:

Analytical and Calibration Techniques

  • Total carbon dioxide (TCO2) analysis and calibration. All TCO2 samples that were retained in this synthesis work were analyzed by coulometric titration. The primary differences between the various groups were the sample volume use, the level of automation, and the primary calibration method. On many cruises the coulometer (UIC, Inc.) was coupled to a semi-automated sample analyzer (Johnson and Wallace 1992; Johnson et al. 1985, 1987,1993, 1998). The most common system, a single-operator multiparameter metabolic analyzer (SOMMA), was typically outfitted with a 20- to 30-mL pipette and was calibrated by filling a gas loop with a known volume with pure CO2 gas, then introducing the gas into the carrier gas stream and performing subsequent coulometric titration (Johnson and Wallace 1992; Johnson et al. 1987,1993, 1998). Some systems were calibrated by analyzing sodium carbonate standards. In TCO2 systems that were not coupled with a semi-automated sample analyzer,the sample was typically introduced manually by a pipette or a syringe.
  • Total alkalinity (TALK) analysis and calibration. All shipboard TALK measurements were made by potentiometric titration using a titrator and a potentiometer. TALK was determined either by characterizing a full titration curve (Brewer et al. 1986; Millero et al. 1993; DOE 1994; Ono et al. 1998) or by a single point titration (Perez and Fraga 1987). Analytical differences were in the volume of sample analyzed, the use of either an open or closed titration cell, and the calibration methods. Results were obtained from different curve-fitting techniques such as Gran plots, nonlinear fitting, or single-point analysis.
  • Fugacity of CO2 (fCO2) analysis and calibration. Two different types of instruments were used to measure discrete fCO2 samples. With each, an aliquot of seawater was equilibrated at a constant temperature of either 4 or 20°C with a headspace of known initial CO2 content. Subsequently, the headspace CO2 concentration was determined by nondispersive infrared analyzer (NDIR) or by quantitatively converting the CO2 to CH4 and then analyzing the concentration using a gas chromatograph (GC) with flame ionization detector. The initial fCO2 in the water was determined after correcting for loss (or gain) of CO2 during the equilibration process. This correction can be significant for large initial fCO2 differences between the headspace and the water, and for systems with a large headspace-to-water volume ratio (Chen et al. 1995).
  • pH analysis and calibration. The pH measurements were determined by a spectrophotometric method (Clayton and Byrne 1993), with m-cresol purple as the indicator and either scanning or diode array spectrophotometers, or by using pH electrodes.

Results of Shipboard Analysis of Certified Reference Materials

Certified Reference Materials (CRMs) were used on many of the cruises as secondary standards for TCO2, with some exceptions during the Pacific Ocean and Atlantic survey (See Table 2 in Lamb et al, 2002). Routine analysis of shipboard CRMs helped verify the accuracy of sample measurements. Certification of the CRM for TCO2 is based on vacuum extraction/manometric analysis of samples in the laboratory of C. D. Keeling at Scripps Institution of Oceanography (SIO). A complete CRMs information can be found in: Information on Batches of CO2 in Seawater Reference Material page. Most groups which routinely ran CRM samples for TCO2 also analyzed the samples for TALK. The CRMs were certified for TALK in July 1996. However, archived CRMs produced prior to 1996 were calibrated as well so that post-cruise adjustments of TALK could be made (See Table 3 in Lamb et al, 2002) CRMs at the time of measurements were not available for the other carbon parameters.

Replicate Samples

Replicate samples were routinely collected and analyzed at sea, thus allowing the analyst to determine the overall precision of the measurement. The imprecision of replication includes the error associated with the collection and handling of the carbon sample, as well as the analytical precision. In addition, replicate samples for TCO2 were collected and stored for analysis ashore at SIO by laboratory of C.D. Keeling (see Guenther, P. R., C. D. Keeling, and G. Emanuele III. 1994b. Oceanic CO2 Measurements for the WOCE Hydrographic Survey in the Pacific Ocean, 1990-1991: Shore Based Analyses. SIO Reference Series, Ref. No. 94-28. University of California, San Diego).

Consistency of Deep Carbon Data at the Locations Where Cruises Cross or Overlap

One approach for evaluating the consistency of the cruises was to compare data where cruises crossed or overlapped. A location was considered a crossover if stations from two cruises were within 1° (~100 km) of each other. If more than one station from a particular cruise fell within that limit, the data were combined for the comparison. For this analysis, only deep-water measurements (>2000 m for the Pacific Ocean, >2500 m for the Indian Ocean, and >3000 m for the Atlantic Ocean) were considered, because CO2 concentration in shallow water can be variable, and the penetration of anthropogenic CO2 can change relationships between the carbon parameters measured at different times. Once the stations were chosen, the data were plotted against potential density referenced to 3000 dB (or 4000 dB in the Atlantic) since water moves primarily along isopycnal surfaces. In order to quantitatively estimate the mean difference between legs, each of the two fitted curves for a restricted deep water density range was evaluated at evenly spaced intervals covering the range of space common to the selected stations from both legs. A mean was taken of the differences, and standard deviation was calculated.

Multiple Linear Regression Analysis

Another approach used to evaluate the data at the crossover locations was a multi-parameter linear regression analyses (MLR). Brewer, et al. (1995) and subsequently others (Wallace 1995; Slansky et al. 1997; Goyet and Davis 1997; Sabine et al. 1999), have shown that both TCO2 and TALK concentrations in deep and bottom waters can be fit well with MLR functions using commonly measured hydrographic quantities for the independent parameters. The geographic extent over which any such function is applicable depends on the number of water masses present, and the uniformity of chemical and biological processes which have affected the carbon species concentration in each water mass.

Isopycnal Analyses

At a few locations in the North Pacific the estimated offsets at the crossovers were not consistent with the offsets from the basinwide MLR analysis. In an attempt to determine whether the limited number of stations analyzed biased on the crossovers, we expanded the crossover analysis to include additional stations along each cruise and/or stations from neighboring cruises. The deep (> 2200 m) station data were averaged at specific potential density (sigma-3) values and fitted with a 2nd-order polynomial function. The average differences and standard deviations were determined from evenly spaced differences along the curves. The range of values observed for a particular cruise at each isopycnal level indicated whether the stations initially used in the crossover analysis were offset from the surrounding stations. Although more assumptions about oceanographic consistency are necessary, the additional stations used in the isopycnal analysis can provide a better estimate of the difference between cruises because more data points are included in the analysis.

Internal Consistency of Multiple Carbon Measurements

An additional independent approach for evaluating the accuracy of data is the examination of the internal consistency of the CO2 system parameters. The CO2 system parameters in seawater can be characterized by temperature, salinity, phosphate and silicate, and two of the four measured inorganic carbon parameters: TCO2, TALK, fCO2, or pH. Thus, the carbon system is overdetermined on cruises where three or more carbon parameters were measured. By comparing estimates using different pairs of carbon measurements, one can evaluate potential offsets. In addition, examination of internal consistency over several cruises lends confidence to the reliability of the equilibrium constants. The constants of Mehrbach et al. (1973) as a refit by Dickson and Millero (1987) were used for this analysis, along with equilibrium constants for other components (e.g., boric acid dissociation, solubility of CO2, water hydrolysis, and phosphoric and silicic acid dissociation) necessary to characterize the carbonate system in seawater as recommended in Millero (1995). This choice was made based on the analysis of a large data set (15,300 samples) obtained from all the ocean basins (Lee et al. 2000; Millero et al. 2002). For this analysis, TALK was calculated using a combination of either TCO2 and fCO2, or TCO2 and pH [adjusted upward by 0.0047 (DelValls and Dickson 1998) for the Pacific and Indian Ocean but not for the Atlantic analysis].

Final Evaluation of Offsets and Determination of Correction to Be Applied

Based on the available information, an assessment was made of the offsets necessary to make the data sets in a basin mutually consistent . Any cruises that showed consistent offsets are adjusted, and the data are combined into a unified data set that is consistent between cruises. Two important points must be considered when evaluating the various approaches used to examine the data quality of the cruises.

First, most of the approaches assume that the deep ocean does not change over the time-period of the various cruises. Thus, very little variability would be expected in the deep waters (pressure > 2000 dbar) at the crossover points. Second, the various approaches have different strengths and weaknesses and may be more or less reliable in different oceanographic regions. Furthermore, the calculated offsets and associated errors may not be directly comparable. As a result, some level of subjectivity is necessarily a part of the adjustments proposed in this section. We have made every attempt to consider all of the various lines of evidence available. Adjustments were based on a preponderance of evidence and only implemented only when we felt an adjustment was clearly necessary.

Indian Ocean

The Indian Ocean database consists primarily of all the US WOCE and NOAA cruises. Carbon measurements were made as part of the JGOFS Global Survey funded through DOE. Details of the WOCE/JGOFS Indian Ocean CO2 measurement program, including personnel, sampling and measurement protocols and data quality assurance/quality control checks are described in: WOCE Indian Ocean Survey Data Comparison at Crossover Stations section and also in the following publications:

We have also included data from the French INDIGO and CIVA1 cruises (contributed by A. Poisson). A summary of the combined data set can be found in:

The WOCE/DOE and CIVA TCO2 data have an estimated accuracy of ±2 µmol/kg. Proposed adjustments for the INDIGO I, II, and III TCO2 data to bring them in line with the WOCE/DOE data are +10.7, -9.4, and +6.4 µmol/kg, respectively. The WOCE/DOE and CIVA alkalinity data have an estimated accuracy of ±3 µmol/kg. Proposed adjustments for the INDIGO I and II alkalinity data to bring them in line with the WOCE/DOE data are +6.5 and +6.8 µmol/kg, respectively. These results have been published by Sabine et al. 1999. As part of this work we also evaluated the quality of the GEOSECS carbon data. Our study indicates that the GEOSECS TCO2 values are 22.6 µmol/kg high relative to the WOCE/DOE data in the deep Indian Ocean. No adjustment was proposed for the GEOSECS alkalinity data.

Indian Ocean Table 1 gives some details about the individual cruises compiled for this project and measured parameters.

Pacific Ocean

Between 1991 and 1999, investigators from 15 different laboratories and 4 countries analyzed carbon measurements on twenty-five WOCE/JGOFS/OACES cruises in the Pacific Ocean. The Pacific Ocean database compiled from these measurements consists of over 136,000 unique sample location from these cruises. Carbon measurements were made as part of the DOE-funded JGOFS Global Survey. Details of the WOCE/JGOFS Pacific Ocean CO2 measurement program - including personnel, sampling and measurement protocols, and data quality assurance/quality control checks - are described in Pacific Ocean section and the following publication:

The Pacific Ocean quality assessment required a much greater effort than did the Indian Ocean assessment. In the Indian Ocean the same parameters were measured with exactly the same equipment, the cruises were carried out as one expedition over a 2-year period, and CRMs were used on all WOCE and NOAA legs. The Pacific measurements involved many of the same principal investigators as the Indian Ocean expedition, but each group used different equipment, different parameter combinations were measured, and the measurements were spread over 8 years. During this synthesis work, we compiled data from 26 cruises in the Pacific Ocean, including data from Canadian, Japanese, and Australian cruises. Our assessment of the Pacific TCO2 data indicates that the reported values are accurate to ±3 µmol/kg after recommended adjustments of +4, -7, and -4 µmol/kg for legs P16N, P17N, and P2, respectively. The TALK data are generally good to ±5 µmol/kg after adjustments of +6, -9, -12, +14, and -6 µmol/kg for legs P8S, P17C, P17N, P2, and P31, respectively. We also adjusted all reported spectrophotometric pH values by +0.0047 to improve the internal consistency with the other carbon measurements.

Pacific Ocean Table 1 gives some details about the individual cruises compiled for this project and measured parameters.

Atlantic Ocean

The comprehensive analysis of the quality of the carbon data of the twenty-three cruises shows overall good agreement and high quality. Based on the extensive analysis, we recommend that the TCO2 and TALK of two cruises (A06 and A07) are not considered appropriate for this synthesis. TCO2 values for A23 are not recommended for use because of a large offset and absence of metadata. In addition the TALK values of A01E are significantly different from neighboring cruises such that these are not used either. Of the lines that had repeat occupations, we recommend that the later (repeat) cruises be used as the primary dataset. This is because the data were more consistent with other data, in part, because improved analysis techniques and because they often were closer in time to the other cruises minimizing the effects of anthropogenic and natural variability. No specific adjustments in TCO2 are recommended. Agreement in TCO2 between cruises that was better than ±4 µmol/kg was deemed acceptable. Although cross-over analysis often showed systematic differences greater than the precision of ±2 µmol/kg, the differences were not systematic for each cross-over, or did not show up in the regional multi-linear analysis. Using an agreement in TALK between cruises of better than ±6 µmol/kg as cutoff, adjustments of +14 µmol/kg and -7 µmol/kg, and -6 µmol/kg are suggested for TALK values on A01W A09 and A17, respectively.

The caveats in the analysis and recommendations should be born in mind. The purpose of the exercise is to create a mutually consistent dataset of TALK and TCO2 for the Atlantic Ocean based on data obtained on the WOCE/WHP, JGOFS and NOAA OACES lines. The analysis of consistency is primarily focused on deep water quantities with the assumption that these values are invariant on decadal timescale. In the well-ventilated Atlantic Ocean where large-scale natural changes appear to manifest through much of the water column, this might not always be the best assumption. Moreover, in this analysis we assume that there are not systematic differences with depth. No comparisons in the upper water column were made both because of seasonal variability in the upper ocean and because the anthropogenic perturbations are most noticeable there.

Atlantic Ocean Table 1 gives some details about the individual cruises compiled for this project and measured parameters.

Details of the WOCE/JGOFS Atlantic Ocean CO2 measurement program - including personnel, sampling and measurement protocols, and data quality assurance/quality control checks - are described in Atlantic Ocean section and the following publications:

  • Perez, F.F., and F. Fraga. 1987. A precise and rapid analytical procedure for alkalinity determination. Marine Chemistry 21:169-82.
  • Millero, F.M., D. Pierrot, K. Lee, R. Wanninkhof, R. Feely, C.L. Sabine, R.M. Key, and T. Takahashi. 2002. Dissociation constants for carbonic acid determined from filed measurements. Deep-Sea Research I, 2002.
Last modified: 2021-03-17T18:30:26Z