Systematic errors as the cause for an apparent deep water property variability: global analysis of the WOCE and historical hydrographic data

Abstract

High-quality hydrographic sections occupied during the World Ocean Circulation Experiment (WOCE) have allowed the first estimates to be made of property changes in the deep ocean on a decadal time-scale. The magnitude of the property variability on deep isothermal surfaces (below about 2–3°C) was found to be comparable with the magnitude of possible systematic errors in the data (except for a few regions where formation of deep and bottom waters occurs). However, the problem of systematic errors in hydrographic data was only marginally addressed in the literature. We conducted an analysis of property offsets between the cruises of an expansive global hydrographic dataset, including 1314 cruises. Because of significant differences in quality, a distinction is made between the high-quality modern data (1970–1998) and less accurate historical data (since 1920s). Inter-cruise offsets are determined on isothermal surfaces for salinity, oxygen, silicate, nitrate and phosphate within crossover areas. Offsets are decomposed into a systematic part (the difference of respective systematic errors (biases)) and a non-systematic part (a combined effect of temporal and spatial variability). For the reference subset of N=384 high-quality cruises with M=2201 crossover areas (for salinity) a system of M algebraic equations in N unknown cruise biases is obtained and solved by least squares. Accounting for biases allows a drastic reduction of inter-cruise offsets (a factor of 2–4 depending on property and dataset considered), bringing data from different cruises to agreement within WOCE quality requirements. For a composite WOCE/non-WOCE high-quality dataset calculated root-mean-square inter-cruise offsets before and after adjustment (in parentheses) are 3.2*10 −3 (1.34*10 −3) for salinity, 2.498 (0.782) μmol/kg for oxygen, 2.4 (0.95) μmol/kg for silicate, 0.55 (0.26) μmol/kg for nitrate and 0.045 (0.018) μmol/kg for phosphate. Our results demonstrate, that quality requirements for the WOCE Hydrographic Programme have been obviously fulfilled. Biases for historical cruises are calculated relative to the corrected reference data and are found to be on average a factor of 3–6 larger than the modern cruise biases. Calculated property offsets and biases for both high-quality and historical data are found to be in a good agreement with independent offset estimates. Though small differences of IAPSO Standard Seawater salinity from label salinity are reported in the literature, they can not explain observed inter-cruise offsets, and operator/salinometer related errors seem to be the main cause for inter-cruise differences. Systematic biases can provide an explanation for apparent changes in deep water temperature–parameter relationships as an alternative to natural variability. For instance, we argue that apparent variability of the deep water salinity in the Argentine Basin during 1980s as reported by Coles, McCartney, and Olson (J. Geophys. Res. 101, 1996, 8957–8970) is mostly the result of systematic errors in the data used for their analysis. Removing inconsistencies in a composite data set is also important for producing consistent climatological property fields even in the data abundant regions.