As mentioned in Section 2.1, the oxygen sensors consist of a Teflon membrane covering a layer of potassium chloride gel. A constant voltage applied across two electrodes results in a current nearly proportional to the activity of oxygen diffusing across the membrane. This current and the temperature of the cell are measured, and dissolved oxygen (OX) is calculated using an algorithm based on that given by Owens and Millard (1985). Six parameters (Boc, Soc, tcor, pcor, t, wT) are fitted to the CTD oxygen current (Oc), oxygen temperature (OT), and Oc time variation (dOc/dt) by the equation:
OX = [Soc (Oc + t(dOc/dt)) + Boc] * OXSAT(T,S) * exp[tcor[T+ wT (OT -T)] + pcor * (P) …… (1)
where OXSAT(T,S) is the saturation concentration of oxygen at temperature, T, and salinity, S; and P is pressure. The parameters (Boc, Soc, tcor, pcor, t, wT) are determined from a nonlinear least-squares fit (based on the Levenberg-Marquardt method [Press et al., 1988]), against check samples taken from bottles during CTD casts. Because the oxygen sensors show considerable hysteresis (upcast oxygen values are very different from the downcast values) the calibration is made using the downcast CTD values of Oc, (OT), P, T, and S at the density levels where the bottles were closed on the upcast.
In order to satisfy WOCE requirements for vertical resolution, 48 water samples are taken with a 24-place rosette on two consecutive casts at station 2 (36 samples from three casts in early cruises when only a 12-place rosette was used). These data are used to calibrate the conductivity and oxygen profiles. Bottle samples from these casts are not interleaved (i.e., the deepest samples are taken on one cast, and the shallowest are taken on another). This does not pose a significant problem for the conductivity calibration, but because the Beckman and sometimes the YSI oxygen sensors used to drift, in some cases significantly, between casts, the oxygen calibration was more problematic. The oxygen calibration is very nonlinear and sensitive to pressure and temperature changes (cf. Owens and Millard, 1985). If one were to calibrate the deep cast using only the samples taken on that cast (usually below 1000 dbar), the shallow oxygen trace could be wildly in error because there would be a few or no bottle data to constrain it near the surface.
Two sets of parameters were obtained regularly per cruise, corresponding to the casts of Station 1 (Kahe Pt.) and Station 2. At Station 2, we calibrated the sensor to the ensemble of bottle data taken during the so-called WOCE deep and shallow casts. These were the only casts where the distribution of bottle samples with pressure and temperature were adequate to determine those coefficients in the calibration fit to the discrete oxygen data. Bottle samples taken from other casts at this station were also used for calibration. The set of parameters obtained for station 2 was used to calculate oxygen for all the CTD casts of this station, although the CTD data from casts without bottle samples were flagged as uncalibrated. A third set of calibration coefficients from samples at Station 8 (HALE-ALOHA) was also obtained for cruises after HOT-79.
Because the SBE 43 sensor does not output oxygen probe internal temperature data (the sensor does an internal temperature compensation), the wT weighting factor of (OT -T) in equation (1) does not need to be fitted.
Calibration problems due to sensor drift or sensor failure were addressed on a cruise by cruise basis, and are detailed in the following section.