• Alkire, M. B., and Coauthors, 2012: Estimates of net community production and export using high-resolution, Lagrangian measurements of O2, NO3 , and POC through the evolution of a spring diatom bloom in the North Atlantic. Deep-Sea Res. I, 64, 157174, doi:10.1016/j.dsr.2012.01.012.

    • Search Google Scholar
    • Export Citation
  • Carlson, J., 2002: Development of an optimized dissolved oxygen sensor for oceanographic profiling. Int. Ocean Syst., 6, 2022.

  • Champenois, W., and Borges A. , 2012: Seasonal and interannual variations of community metabolism rates of a Posidonia oceanica seagrass meadow. Limnol. Oceanogr., 57, 347361.

    • Search Google Scholar
    • Export Citation
  • D'Asaro, E. A., 2003: Performance of autonomous Lagrangian floats. J. Atmos. Oceanic Technol., 20, 896911.

  • D'Asaro, E. A., 2010: Calibration of the dissolved oxygen sensors on float 48 and on the Knorr CTD with Winkler bottle samples. The 2008 North Atlantic Bloom Experiment, Calibration Rep. 3, Archived at the Biological and Chemical Oceanography Data Management Office, 83 pp. [Available online at http://data.bco-dmo.org/NAB08/Oxygen_Calibration-NAB08.pdf.]

  • D'Asaro, E. A., and McNeil C. , 2007: Air–sea gas exchange at extreme wind speeds measured by autonomous oceanographic floats. J. Mar. Syst., 66, 92109.

    • Search Google Scholar
    • Export Citation
  • Demas, J. N., DeGraff B. A. , and Coleman P. , 1999: Oxygen sensors based on luminescence quenching. Anal. Chem., 71, 793A800A.

  • Edwards, B., Murphy D. , Janzen C. , and Larson N. , 2010: Calibration, response, and hysteresis in deep-sea dissolved oxygen measurements. J. Atmos. Oceanic Technol., 27, 920931.

    • Search Google Scholar
    • Export Citation
  • Garcia, H. E., and Gordon L. I. , 1992: Oxygen solubility in seawater: Better fitting equations. Limnol. Oceanogr., 37, 13071312.

  • Garcia, H. E., Locarnini R. A. , Boyer T. P. , Antonov J. I. , Baranova O. K. , Zweng M. M. , and Johnson D. R. , 2010: Dissolved Oxygen, Apparent Oxygen Utilization, and Oxygen Saturation. Vol. 3, World Ocean Atlas 2009, S. Levitus, Ed., NOAA Atlas NESDIS 70, 344 pp.

  • Keeling, R., Körtzinger A. , and Gruber N. , 2010: Ocean deoxygenation in a warming world. Annu. Rev. Mar. Sci., 2, 199229, doi:10.1146/annurev.marine.010908.163855.

    • Search Google Scholar
    • Export Citation
  • Kihm, C., and Körtzinger A. , 2010: Air–sea gas transfer velocity for oxygen derived from float data. J. Geophys. Res., 115, C12003, doi:10.1029/2009JC006077.

    • Search Google Scholar
    • Export Citation
  • Klimant, I., Kühl M. , Glud R. N. , and Holst G. , 1997: Optical measurement of oxygen and temperature in microscale: Strategies and biological applications. Sens. Actuators, 38B, 2937.

    • Search Google Scholar
    • Export Citation
  • Körtzinger, A., Schimanski J. , and Send U. , 2005: High quality oxygen measurements from profiling floats: A promising new technique. J. Atmos. Oceanic Technol., 22, 302308.

    • Search Google Scholar
    • Export Citation
  • Martini, M., Butman B. , and Mickelson M. , 2007: Long-term performance of Aanderaa optodes and Sea-Bird SBE-43 dissolved-oxygen sensors bottom mounted at 32 m in Massachusetts Bay. J. Atmos. Oceanic Technol., 24, 19241935.

    • Search Google Scholar
    • Export Citation
  • McNeil, C., and D'Asaro E. , 2007: Parameterization of air–sea gas fluxes at extreme wind speeds. J. Mar. Syst., 66, 110121.

  • Nicholson, D., Emerson S. , and Erickson C. , 2008: Sub mixed-layer oxygen production determined from Seaglider surveys. Limnol. Oceanogr.,53, 2226–2236.

  • Prakash, S., Balakrishnan Nair T. M. , Udaya Bhaskar T. V. S. , Prakash P. , and Gilbert D. , 2012: Oxycline variability in the central Arabian Sea: An Argo-oxygen study. J. Sea Res., 71, 18, doi:10.1016/j.seares.2012.03.003.

    • Search Google Scholar
    • Export Citation
  • Riser, S., and Johnson K. , 2008: Net production of oxygen in the subtropical ocean. Nature, 451, 323325.

  • Tengberg, A., and Coauthors, 2006: Evaluation of a lifetime-based optode to measure oxygen in aquatic systems. Limnol. Oceanogr. Methods, 4, 717.

    • Search Google Scholar
    • Export Citation
  • Uchida, H., Kawano T. , Kaneko I. , and Fukasawa M. , 2008: In situ calibration of optode-based oxygen sensors. J. Atmos. Oceanic Technol., 25, 22712281.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Puget Sound calibrations of multiple floats. (a) Rack holding five floats was lowered over the port side by the research vessel's crane. (b) SBE-9/11 CTD with rosette was deployed off the fantail and took water samples close to the operating floats. (c) Two packages could be navigated in the vertical to a fraction of 1 m using the vessel's echosounder and were no more than 10 m apart horizontally.

  • View in gallery

    Determination of four of the five calibration points for July 2010. Duplicate Winkler samples at each level (black circles) were taken at, above, and below the float package (small yellow triangles) guided by data from the CTD (green dots). Calibration point (inverted magenta-outlined triangles) for the floats based on the Winkler samples was chosen at each level. Accuracies of better than 1 μmol kg−1 were achieved. Float and CTD oxygen values have been offset from their factory calibrations in this figure to best show their relative variations.

  • View in gallery

    Variations in the SBE-43 temperature function. Shown is (a) for six different calibrations of sensor 15 normalized by the value at T = 10°C and labeled by date. Membrane was changed between the third and fourth calibrations and for the last calibration. (b) Absolute value of change in near 25°C as a function of time interval between calibrations for all sensors.

  • View in gallery

    (top) SBE-43 frequency function F(f) for sensors 139, 15, and 156 at multiple calibrations indicated by different colors. (bottom) Deviation of F(f) from the first calibration.

  • View in gallery

    Deviation of optode oxygen from in situ and laboratory calibrations for three optodes all with the same batch of membrane. Optode oxygen was calculated using the factory calibration. Colors indicate date; symbols indicate optode number. Colored lines show the model [Eq. (6)] that approximately fits these data.

  • View in gallery

    Optode bias at 100% saturation as a function of time. Small symbols with text show individual calibration points. Large symbols and text show these points interpolated to float deployment dates. Gray lines show optode model [Eq. (6)] with a spread of ±0.5%. The inset shows details of interpolation for Typhoon Megi and Ocean Station Papa deployments with thick bars spanning the calibration value plus or minus its estimated standard deviation. Colors and symbols distinguish the three different optodes.

  • View in gallery

    (a) Time series of oxygen for float 67, optode 861, measured during Typhoon Megi marked with comparison points (yellow). Factory-calibrated optode (magenta) is offset (red) based on extrapolated Winkler samples. Factory-calibrated SBE-43 (blue) is scaled based on optode offset to form the final SBE-43 calibrated data (black). (b) A least squares fit between the SBE-43–optode difference and the oxygen level at the comparison points gives the SBE-43 offset at 100% saturation.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 469 329 43
PDF Downloads 240 171 12

Calibration and Stability of Oxygen Sensors on Autonomous Floats

View More View Less
  • 1 Applied Physics Laboratory, and School of Oceanography, University of Washington, Seattle, Washington
  • | 2 Applied Physics Laboratory, University of Washington, Seattle, Washington
© Get Permissions
Full access

Abstract

The calibration accuracy and stability of three Aanderaa 3835 optodes and three Sea-Bird Electronics SBE-43 oxygen sensors were evaluated over four years using in situ and laboratory calibrations. The sensors were mostly in storage, being in the ocean for typically only a few weeks per year and operated for only a few days per year. Both sensors measure partial pressure of oxygen, or equivalently saturation at standard pressure; results are expressed in this variable. It is assumed that sensor drift occurs in the oxygen sensitivity of the sensors, not the temperature compensation; this is well justified for the SBE-43 based on multiple calibrations. Neither sensor had significant long-term drift in output when sampling anoxic water. Sensor output at 100% saturation differed from the factory calibrations by up to ±6% (averaging −2.3% ± 3%) for the SBE-43 and up to −12.6% for the optodes. The optode output at 100% saturation is well described by a single decaying exponential with a decay constant of approximately 2 yr and an amplitude of 28%. The mechanism of this drift is unknown but is not primarily due to sensor operation. It may be different from that experienced by sensors continuously deployed in the ocean. Thus, although the optodes in this study did not have a stable calibration, their drift was stable and, once calibrated, allowed optode and SBE-43 pairs mounted on the same autonomous floats to be calibrated to an accuracy of ±0.4% over a 4-yr period.

Corresponding author address: Eric A. D'Asaro, Applied Physics Laboratory, University of Washington, 1013 NE 40th Street, Seattle, WA 98105. E-mail: dasaro@apl.washington.edu

Abstract

The calibration accuracy and stability of three Aanderaa 3835 optodes and three Sea-Bird Electronics SBE-43 oxygen sensors were evaluated over four years using in situ and laboratory calibrations. The sensors were mostly in storage, being in the ocean for typically only a few weeks per year and operated for only a few days per year. Both sensors measure partial pressure of oxygen, or equivalently saturation at standard pressure; results are expressed in this variable. It is assumed that sensor drift occurs in the oxygen sensitivity of the sensors, not the temperature compensation; this is well justified for the SBE-43 based on multiple calibrations. Neither sensor had significant long-term drift in output when sampling anoxic water. Sensor output at 100% saturation differed from the factory calibrations by up to ±6% (averaging −2.3% ± 3%) for the SBE-43 and up to −12.6% for the optodes. The optode output at 100% saturation is well described by a single decaying exponential with a decay constant of approximately 2 yr and an amplitude of 28%. The mechanism of this drift is unknown but is not primarily due to sensor operation. It may be different from that experienced by sensors continuously deployed in the ocean. Thus, although the optodes in this study did not have a stable calibration, their drift was stable and, once calibrated, allowed optode and SBE-43 pairs mounted on the same autonomous floats to be calibrated to an accuracy of ±0.4% over a 4-yr period.

Corresponding author address: Eric A. D'Asaro, Applied Physics Laboratory, University of Washington, 1013 NE 40th Street, Seattle, WA 98105. E-mail: dasaro@apl.washington.edu

1. Introduction

Measurement of oxygen on moorings, floats, gliders, and other autonomous oceanographic platforms is becoming common. For example, several hundred Argo floats have been equipped with oxygen sensors (Körtzinger et al. 2005), each measuring many dozen profiles per year. Detailed observations of air–sea oxygen fluxes (D'Asaro and McNeil 2007; Kihm and Körtzinger 2010), biological productivity (Riser and Johnson 2008; Alkire et al. 2012), and oxygen minimum zones (Prakash et al. 2012) have been made using such systems. However, the absolute calibration of oxygen sensors can severely limit such studies. For example, air–sea oxygen flux depends on the deviation of oxygen concentration from an equilibrium level near 100% saturation (McNeil and D'Asaro 2007). For a typical deviation in oxygen saturation from equilibrium of 1%, a measurement of the flux to an accuracy of 20% requires an absolute accuracy of approximately 0.2% or about 0.5 μmol kg−1 at 15°C. Detection of predicted long-term decreases in oceanic oxygen, typically 0.1 μmol kg−1 yr−1 (Keeling et al. 2010), places similar tight constraints on sensor accuracy. Manufacturers of the two currently available oxygen sensors quote accuracies an order of magnitude larger. Although previous researchers (e.g., Uchida et al. 2008) report that much higher accuracies can be achieved with careful calibration, uncertainties in calibration and long-term stability remain crucial issues. Here, we address these issues, particularly the long-term stability, with a 4-yr-long series of simultaneous in situ calibrations of both sensor types. Although this work was done to calibrate sensors for autonomous float deployments in hurricanes and typhoons, the results have broader applicability.

2. Sensors

a. SBE-43 Clark cell

The SBE-43 oxygen sensor, sold by Sea-Bird Electronics, is a polarographic Clark cell (Carlson 2002; Edwards et al. 2010) that measures the partial pressure of oxygen relative to an internal anoxic standard from the rate of diffusion of oxygen across a membrane separating the sample from the standard. Because this diffusion depletes the oxygen from a thin boundary layer near the membrane, water must be pumped continuously past the membrane to establish an equilibrium. The measurement thus depends on the rate of pumping, which is chosen so that this sensitivity is not large. For applications where power is limited, the pumping rate is reduced, thereby shifting the calibration somewhat. The measurement is also sensitive to physical contamination of the membrane, which changes the rate of diffusion.

The SBE-43 measures the partial pressure (precisely fugacity) of oxygen, or equivalently the oxygen concentration c (μmol kg−1) relative to (μmol kg−1), the saturation concentration at standard pressure (1013.25 mbar = 101.325 kPa) a known function of temperature T and salinity S (Garcia and Gordon 1992). Thus, the sensor measures , which we express as a percent. Our sensors output a frequency f proportional to the diffusion rate of oxygen. Other models of the sensor output a voltage with a similar equation. The calibration equation
e1
is the product of three terms: F(f) = Soc(f + Foffset) is a linear function with the gain Soc, is quadratic in temperature and describes the temperature response, and describes the pressure response. Each sensor is factory calibrated across a matrix of 17 temperature and oxygen values to determine F(f) and . The function D has a minor (0.7%) effect across the range of pressures (0–200 dbar) considered here and Sea-Bird does not change the value of E. Residuals from the calibration are typically 0.4 μmol kg−1.

Each of our SBE-43 sensors was mounted on the bottom of a Lagrangian float (D'Asaro 2003; Alkire et al. 2012), in line with the pumped Sea-Bird temperature and conductivity sensors. This mounting is upside down relative to that used in Argo floats and required that a small hole be drilled in the plumbing to allow air to vent when the float is first submerged. In the measurements described here, the pump was run continuously but at a slower speed than that used in the factory calibrations. Some of the factory calibrations supplied a correction for this difference (roughly 1%–2% in Soc). For uniformity, we used the uncorrected values. We evaluated four different sensors [serial numbers (SNs) 15, 139, 156, and 173]. The membrane of each sensor was replaced because of damage at some point in the 4-yr evaluation, resulting in seven different sensor/membrane combinations over the period.

b. Aanderaa optode

The 3835 optode sold by Aanderaa Instruments measures the partial pressure of oxygen using the fluorescence quenching (Demas et al. 1999; Klimant et al. 1997) of a platinum porphyrin complex embedded in a gas permeable foil exposed to the water. Fluorescence is measured using the phase shift from an alternating current (AC)-modulated blue excitation to the fluoresced red signal. The sensor also measures temperature. It consumes no oxygen and is thus insensitive to the water flow around it.

Under ideal conditions, quenching is described by the Stern–Volmer equation:
e2
where is the phase shift at zero oxygen and K is independent of oxygen. Both K and are functions of temperature. However, packaging of the fluorescing material and inhomogeneity in its optical properties may lead to more complex forms (Demas et al. 1999). The manufacturer has chosen a calibration equation of the form
e3
Oxygen concentration (not partial pressure) is the product of three terms. The function G is a fourth-order polynomial in a corrected phase with coefficients c(T), each of which is a third-order polynomial in temperature. Scorr expresses the salinity variability of solubility from Garcia and Gordon (1992), and Dcorr is a linear pressure correction, again small over our depth range. The 20 coefficients of G are determined from 35 temperature and oxygen calibration points measured on a sample from a large batch of sensing foil. We have been unable to duplicate the manufacturer's fits from the calibration points. Each foil batch is cut into smaller pieces and used in many different optode sensors. Variability between the foil pieces in each sensor and in the phase calibration of each sensor's electronics is determined from a two-point factory calibration that defines a linear function as
e4
giving the corrected phase (d phase) from the measured phase (b phase).
We find this calibration scheme inelegant because it ignores the known physical principles on which the sensor operates and uses numerically unstable polynomial forms. For our foils the calibration function [Eq. (3)] has a minimum as a function of at a value less than 1% below the oxygen zero point and behaves poorly near zero oxygen. When preliminary fits of the form of Eq. (2) to the factory calibration points are compared to the manufacturer's polynomial fits of Eq. (3), the polynomial forms show an increasing deviation from the Stern–Volmer form above 110% saturation, reaching 4% at the last set of calibration points (~135% saturation). Similar large deviations are found at the edges of the calibration domain in all directions. The polynomial forms therefore appear to introduce significant error. However, because this is not the central focus of this paper, we retain the manufacturer's calibrations, but recast them into the form
e5
thereby eliminating Scorr [because it is part of and expressing the calibration in terms of the physical quantity measured.

Each of our optodes was also mounted on the bottom of a Lagrangian float, about 12 cm to the side and approximately 20 cm above the intake of the SBE-43 pump [see D'Asaro (2010) for a picture]. We evaluated three sensors (SNs 1860, 1861, and 1862), all with the same batch of foil (SN 1701). The initial factory calibration was used throughout.

3. Calibration methods

a. Approach

Oxygen measurements on autonomous floats have not usually calibrated in detail because of the difficulty of doing this with sufficient accuracy. Calibrations have sometimes been done with casts separated from the float by many days and hours (Kihm and Körtzinger 2010). During the 2008 North Atlantic Bloom Experiment (NAB08; Alkire et al. 2012) a single float was calibrated to better than 2 μmol kg−1 from six Winkler/CTD casts generally within 1 km and 1 h of the float, a substantial effort. Even with this small separation, many of the Winklers had to be discarded because the shape of the CTD oxygen profiles did not match those measured by the float. For hurricane deployments, it thus seemed unlikely that we could obtain sufficient accuracy from measurements near the time of deployment. We thus chose to make very accurate in situ calibrations in local waters before and after deployment.

b. In situ

Dedicated oxygen calibration efforts were conducted in Puget Sound, Washington, in July 2008, July 2009, December 2009, July 2010, and December 2010 (Table 1; Fig. 1) to support expected hurricane or typhoon deployments each year. In 2009 and 2010 three or more floats, each with a SBE-43 and optode sensor, were attached to a frame and lowered off the side of the research vessel (R/V) Robertson in 100–200 m of water (Fig. 1a). An SBE-9/11 CTD with a 12-bottle, 2-L rosette was lowered a few meters away (Fig. 1b). Both packages were visible in the ship's echosounder (Fig. 1c) and could be navigated to the same depth to within a fraction of 1 m. The floats were positioned at several levels chosen to span a wide range of oxygen concentrations and allowed to equilibrate for 2000–3000 s. At each level duplicate bottles were taken at the level of the floats, a few meters above, and a few meters below, for a total of six samples near each level. A single Winkler sample was taken from each bottle. Winkler analyses were done at Sea-Bird Electronics by the same technician and equipment used to calibrate the SBE-43 sensor in Sea-Bird's calibration facility. Differences between the duplicate Winkler samples averaged 0.2 μmol kg−1.

Table 1.

Puget Sound calibrations, where values in the method column indicate water sampling type (1, lowered Niskin bottle and 2, CTD rosette) and Winker analysis (a, UW technical services; b, C. Stump; and c, Sea-Bird Electronics).

Table 1.
Fig. 1.
Fig. 1.

Puget Sound calibrations of multiple floats. (a) Rack holding five floats was lowered over the port side by the research vessel's crane. (b) SBE-9/11 CTD with rosette was deployed off the fantail and took water samples close to the operating floats. (c) Two packages could be navigated in the vertical to a fraction of 1 m using the vessel's echosounder and were no more than 10 m apart horizontally.

Citation: Journal of Atmospheric and Oceanic Technology 30, 8; 10.1175/JTECH-D-12-00222.1

The Winkler data were used to generate a calibration point for each of the floats at each of the levels (Fig. 2). These were interpolated by eye to the float level, guided by the profiles from the CTD. This resulted in a set of calibration points, each with an accuracy of better than 1 μmol kg−1.

Fig. 2.
Fig. 2.

Determination of four of the five calibration points for July 2010. Duplicate Winkler samples at each level (black circles) were taken at, above, and below the float package (small yellow triangles) guided by data from the CTD (green dots). Calibration point (inverted magenta-outlined triangles) for the floats based on the Winkler samples was chosen at each level. Accuracies of better than 1 μmol kg−1 were achieved. Float and CTD oxygen values have been offset from their factory calibrations in this figure to best show their relative variations.

Citation: Journal of Atmospheric and Oceanic Technology 30, 8; 10.1175/JTECH-D-12-00222.1

Only surface calibrations were used in 2008. Winkler samples were taken within one meter of the sensors using a hand-lowered Niskin bottle from the R/V Miller. Winkler samples for float 53 on 21 July were analyzed by C. Stump and were of high quality with a mean difference of duplicates of 0.2 μmol kg−1. Analyses for floats 50 and 51 on 23 July were analyzed by University of Washington (UW) Technical Services and had much higher average errors (2 μmol kg−1). Of these, four samples with an average difference in duplicates of 0.5 μmol kg−1 were chosen to include in this analysis.

c. Laboratory

Optode calibration points with zero oxygen and some points at 100% saturation were made (Table 2). Zero oxygen calibrations were performed by immersing the float's optode in a saturated sodium sulfite solution. The 100% saturation readings were made with the optode immersed in bubbled water. All readings were made either in a temperature-controlled water bath set at 20°C (when the sensors were detached from the float) or in a large bucket (when the sensors were attached to the float). In both cases, final readings were taken only after the optode's b phase and internal temperature had fully stabilized. Because the reaction of the sodium sulfite solution with oxygen is exothermic, it can take up to one hour for the readings to stabilize during zero oxygen calibrations.

Table 2.

Optode laboratory calibrations, where Z = zero oxygen and S = 100% saturation calibration.

Table 2.

4. Accuracy and stability of sensors

a. SBE-43

Linear calibrations in oxygen of the SBE-43 sensors to the Winkler calibrations for the July and December tests in Puget Sound typically differ by 3%–4% near 100% saturation; the predeployment and postdeployment calibrations yield similar results. However, these calibrations span temperatures of 8°–13°C and oxygen saturations of 75%–110% in summer and 75%–100% in winter. Tropical cyclone oxygen values span similar ranges but with warmer temperatures, 22°–30°C. Depending on how the differences between the factory and our calibrations are apportioned between temperature and oxygen components of the calibration, different calibrations are obtained. It is thus important to understand which components of the sensor calibration are changing. We analyze 12 SBE-43 factory calibrations for the sensors used on our three floats.

The temperature function varies by 2%–3% from 10° to 30°C (Fig. 3a). Changes in increase with time (Fig. 3b), with a typical rate of 0.3% per year for temperatures near 25°C. For a typical 6-month interval between calibrations, this can account for only approximately 0.15% of the change in calibration, much less than was observed. Thus, calibration changes for the SBE-43 sensors are primarily due to changes in the linear frequency function F(f). Variations in the slope of F(f), rather than its offset near zero oxygen, dominate the calibration changes (Fig. 4). We will thus use the limited in situ calibration data to vary the slope only.

Fig. 3.
Fig. 3.

Variations in the SBE-43 temperature function. Shown is (a) for six different calibrations of sensor 15 normalized by the value at T = 10°C and labeled by date. Membrane was changed between the third and fourth calibrations and for the last calibration. (b) Absolute value of change in near 25°C as a function of time interval between calibrations for all sensors.

Citation: Journal of Atmospheric and Oceanic Technology 30, 8; 10.1175/JTECH-D-12-00222.1

Fig. 4.
Fig. 4.

(top) SBE-43 frequency function F(f) for sensors 139, 15, and 156 at multiple calibrations indicated by different colors. (bottom) Deviation of F(f) from the first calibration.

Citation: Journal of Atmospheric and Oceanic Technology 30, 8; 10.1175/JTECH-D-12-00222.1

b. Aanderaa optode

We find large (15%) drifts in the optode, but have insufficient data to assess whether the drift occurs in the temperature [Scorr in Eq. (5)] or phase (G) components of the calibration. We assume, as for the SBE-43, that all drift is within the phase component.

Figure 5 shows the deviations of the optodes from our calibration points. Although there are insufficient data to define the entire functional form, some features are clear. There is very little (<1%) change near zero oxygen. The output of the optodes is low compared to the calibrations, with the magnitude of this effect increasing nonlinearly with oxygen and with time. All three optodes behave similarly. The following model captures these features:
e6
where the deviation of the optode-measured oxygen from , the calibration (Winkler) points, increases with oxygen concentration as described by an amplitude Δ and a half-saturation H, and increases with time t (years) starting from a reference time t0 with a time scale (years). Model parameters are fit to the data by minimizing the sum of the squares of deviation of all points excluding the points near , a level at which we believe the optode polynomial calibration is inaccurate, and excluding points near because the functional form is zero by construction. After fitting, the rms deviation is 0.2%, comparable to the accuracy of the calibration points.
Fig. 5.
Fig. 5.

Deviation of optode oxygen from in situ and laboratory calibrations for three optodes all with the same batch of membrane. Optode oxygen was calculated using the factory calibration. Colors indicate date; symbols indicate optode number. Colored lines show the model [Eq. (6)] that approximately fits these data.

Citation: Journal of Atmospheric and Oceanic Technology 30, 8; 10.1175/JTECH-D-12-00222.1

The primary goal of this analysis is to calibrate the optodes near 100% saturation. Accordingly, the model [Eq. (6)] is used to extrapolate the optode reading at each calibration point to 100% saturation, that is,
e7
equivalent, in Fig. 5, to moving along the constant time line through each data point to The resulting time series of optode calibration points at 100% saturation (Fig. 6) show the nearly exponential decrease in the sensitivity of all three optodes, summarized in Eq. (6) by an exponential decay with a time scale of 1.94 yr and a total magnitude of 29%. All but 4 of the 69 data points (6%) fall within ±0.7% of this model.
Fig. 6.
Fig. 6.

Optode bias at 100% saturation as a function of time. Small symbols with text show individual calibration points. Large symbols and text show these points interpolated to float deployment dates. Gray lines show optode model [Eq. (6)] with a spread of ±0.5%. The inset shows details of interpolation for Typhoon Megi and Ocean Station Papa deployments with thick bars spanning the calibration value plus or minus its estimated standard deviation. Colors and symbols distinguish the three different optodes.

Citation: Journal of Atmospheric and Oceanic Technology 30, 8; 10.1175/JTECH-D-12-00222.1

5. Application of calibrations to sensors

The SBE-43 and optode for each float of each deployment were calibrated in three steps. The 2008 calibrations were evaluated at the time of float deployment in Hurricane Gustav; the 2010 calibrations at the deployment time in Typhoon Megi and at a later deployment at Ocean Station Papa. No deployments were made in 2009. The analysis starts with factory calibrated data, which we denote as and for the SBE-43 and optode, respectively.

a. Determine optode offset at 100% saturation and correct optode at 100% saturation

For each float in each deployment, each of the nearest 100% in situ calibration points was extrapolated in time to the deployment time using the exponential function in Eq. (6). For example (Fig. 6, inset), optode 861 (blue) on float 67, deployed in October 2010, was calibrated using four Winkler calibration points from July 2010 and three Winkler points from December 2010. Each was extrapolated in time (thin blue lines) to the deployment time and their mean was taken as the optode offset (=−11.8% for 861). The standard error of this estimate was taken as the standard deviation of these points divided by the square root of 7. Floats 66 and 68 (optodes 860 and 862, red and black, respectively) have fewer calibration points but are processed in the same way. An “offset optode” dataset , valid at 100% saturation, was computed as .

b. Transfer optode calibration to the SBE-43 and correct SBE-43 at all concentrations

The SBE-43 and optode were compared at a set of hand-chosen points, excluding times of rapid oxygen change when the slow optode response prevents accurate comparison, and, for some floats, excluding the first few hours of deployment during which the optode does not properly equilibrate. Figure 7a shows an example for the 2010 deployment of float 67. From its initial deployment to about yearday 289.6 and between yearday 291.25 and the end of the mission the float executed a slow vertical profile, stopping at selected isopycnals for several hours and producing the stairlike oxygen time series. Between these, the float remained in the mixed layer. Calibration points (yellow) were chosen on each segment of the second profile and in the mixed layer. However, no points were chosen during the first profile, when the optode readings were high and noisy.

Fig. 7.
Fig. 7.

(a) Time series of oxygen for float 67, optode 861, measured during Typhoon Megi marked with comparison points (yellow). Factory-calibrated optode (magenta) is offset (red) based on extrapolated Winkler samples. Factory-calibrated SBE-43 (blue) is scaled based on optode offset to form the final SBE-43 calibrated data (black). (b) A least squares fit between the SBE-43–optode difference and the oxygen level at the comparison points gives the SBE-43 offset at 100% saturation.

Citation: Journal of Atmospheric and Oceanic Technology 30, 8; 10.1175/JTECH-D-12-00222.1

The offset between the offset optode and the factory-calibrated SBE-43 at 100% saturation was determined from a linear fit between and at the comparison points (Fig. 7b), yielding for example, for float 67. The gain of the SBE-43 was adjusted to match this offset:
e8
yielding a calibrated SBE-43 data time series valid at all concentrations. Because the SBE-43 has higher precision and a more certain temperature calibration than the optode, is the primary oxygen dataset resulting from this analysis.

c. Recalibrate the optode at all concentrations

The form of the optode calibration shift is not well determined (see section 4b) at saturations different from 100%. A calibrated optode time series, , valid over the range of our measurements, was calculated from the linear fit of to at the comparison points. The optode is a secondary dataset and the analysis is straightforward, so no examples are shown. The calibrated optodes deviate by up to ±1% from the calibrated SBE-43 sensors at the comparison points.

6. Results

a. Calibrations

Table 3 shows the results of the calibration for each year evaluated at the deployment time for each float. For 2009, with no deployment, the July calibration time was chosen. As expected (Fig. 6), the optode error increased with time. The SBE-43 is −2.3% lower on average, comparable to the typical change in sensitivity because of the slower pumping speed in the float than during the factory calibration. The standard deviation of approximately 3.3% around this is higher than the manufacturer's specification for initial calibration accuracy of 2%, perhaps reflecting the sum of our calibration uncertainty of approximately 0.4%, and some additional drift between the factory calibration and ours. The calibration accuracy is greater for the Ocean Station Papa deployment because the longer records (~12 days) at nearly 100% saturation make the step described in section 5b more accurate. Floats 66 and 67 were deployed ahead of Typhoon Megi within a few kilometers of each other, yet differed in prestorm oxygen levels by 0.6%, within the statistical error. Accordingly, they were each shifted by 0.3% so that they agreed.

Table 3.

Calibration results. For October 2009, calibration was done in July 2009. Calibration for other dates was done at time of float deployment. Offset (%) of factory calibration is from interpolated Winkler samples with 100% saturation at 1013.25 mbar. Offset errors are two standard deviations. SBE-43 bias is the mean of all SBE-43 offsets plus or minus two standard deviations of the mean. SBE-43 accuracy is the mean of SBE-43 offset uncertainties excluding Ocean Station Papa.

Table 3.

b. Sanity checks

For Hurricane Gustav, the three calibrated floats measured a prestorm oxygen level of 100.5%–101.5%, within the range of the 2009 World Ocean Atlas (Garcia et al. 2010) value of 102.5% ± 2%. The deployments were 100 km south of the mouth of the Mississippi River, in a region of potentially strong vertical and horizontal gradients. For Typhoon Megi, the three calibrated floats measured a prestorm oxygen level of close to 100%. This compares favorably with the 2009 World Ocean Atlas at the deployment site (100% ± 2%), with the mixed layer oxygen (99.3% ± 0.4%) measured by a SBE-43 on a CTD cast taken 180 km southwest of the float deployment location from the R/V Revelle, and with the output of a calibrated optode (99.3% ± 0.95%) plumbed into the seawater system while transiting 200 km west of the float deployment site just before the storm's passage. The shipboard optode was matched to the on-station CTD cast data with an offset of 1.8% ± 0.7% near 100% saturation, and independently checked on board to have a zero reading of <0.5%.

7. Other sensor issues

The analyses here, and in a comparison of an optode and a SBE-43 mounted on a float during NAB08 (D'Asaro 2010; Alkire et al. 2012), identified several other issues with both sensors. During the first day after deployment, the optode data are often noisy and read high by a few percent (e.g., Fig. 7a). This does not occur at the surface and might be because of bubbles of air, trapped behind the optical film, which are compressed at depth and diffuse outward through the film. As has been noted by many others (e.g., Nicholson et al. 2008), the optode's long time response can lead to large hysteresis on profiling instruments. Because the sensor is not pumped, this time can be significantly longer than that specified by the manufacturer (~40 s) depending on how well it is flushed. D'Asaro (2010) reports an optimal value of 153 s on a slowly profiling float, perhaps because of poor flushing around the sensor. Although the optode does not consume oxygen, biofouling on or near the optode can cause reduced readings in low-flow environments. D'Asaro (2010) reports deviations of up to 13 μmol kg−1 compared to an SBE-43 apparently because of this effect.

The SBE-43 must be pumped to make accurate measurements and thus consumes considerably more energy than the optode (e.g., Martini et al. 2007). One strategy to reduce energy is to pump for long enough to bring the sensor nearly to equilibrium before each sample, typically for 20–40 s. However, a complete equilibrium can take much longer than this to achieve [see Edwards (2010) for a detailed analysis of equilibration time], so that the measured value depends on the oxygen level in the cell before the pumping starts. When the pumping stops, a new equilibrium oxygen level is established in the cell, with the sensor consuming oxygen and residual flushing restoring it. This equilibrium value, and thus the measured oxygen on the next pumping cycle, depends on the local oceanographic conditions. D'Asaro (2010) reports that an SBE-43 operated in this way read up to 3 μmol kg−1 high compared to a nearby optode because of the action of surface waves enhancing the sensor flushing near the surface.

In our experience, the optode is more reliable than the SBE-43. All four of the SBE-43 sensors used in our study had their membranes replaced once. Only one of the optodes failed. The SBE-43 appears to have a higher precision than the optode on time scales from 30 s to several hours. We have used it to compute vertical oxygen flux by covariance (D'Asaro and McNeil 2007). Similar attempts using optodes mounted on the same floats have yielded only noise.

8. Summary and discussion

The major surprise in this study was the long-term (multiyear) predictable drift of the optodes during storage, which runs contrary to other reports (e.g., Tengberg et al. 2006; Champenois and Borges 2012) of high stability. Aanderaa Instruments (A. Tengberg 2012, personal communication) reports drifts of similar magnitudes attributed to photobleaching of the optode by the sensor's blue light-emitting diode (LED) during the first approximately 105 samples and now “burns in” new optodes to limit this effect. Our sensors were not burned in, but their drift does not appear to be related to sampling. The observed drift was remarkably stable and nearly independent of the details of how the optodes were stored and/or operated, which varied between sensors in different years. In most years, between deployments the optodes remained attached to the bottom of the floats; the floats were stored in the corner of a windowless basement room with fluorescent lights, which were on during most working days. Typically, they were shipped by truck to Keesler Air Force Base in Mississippi in early August inside of wooden shipping boxes cushioned by closed-cell foam. These boxes were stored in an aircraft hanger until deployment in late August or early September. After a few weeks in the ocean, they were recovered, any biofouling was removed, and they were put back in boxes and shipped to Seattle, Washington. In some years, the optodes were coated with glycerin. In particular, between calibrations in July and December 2009, the optodes were not run at all and remained at Keesler in boxes through October. Nevertheless, the same amount of drift was observed. We speculate that the drift may be a result of some environmental factor during storage—for example, atmospheric ozone—but cannot identify it. It is possible that the observed decay does not occur for optodes in the ocean, perhaps because of the absence of this factor.

The factory calibrations of both sensors were insufficient to obtain accuracies in excess of 1%, as needed for most estimates of air–sea gas transfer. Because the errors of the two different sensors are different, combining both types on the same platform led to a more robust measurement and allowed diagnosis of errors in both. This, combined with repeated, high-accuracy, in situ calibrations were the key elements allowing the required accuracies and precisions to be obtained, both here and during NAB08 (D'Asaro 2010; Alkire et al. 2012).

Accurate interpretation of limited calibrations requires calibration equations that match well the physics of the sensors. Because both sensors measure partial pressure, or equivalently percent saturation at standard pressure, it is best to analyze calibration data for these sensors in terms of this quantity. Our analyses provide additional guidance: drift of the SBE-43 occurs primarily in the sensor gain [Soc in Eq. (1)]; the zero point (Foffset × Soc) and the temperature compensation [] are more stable. Drift in the optode is highly predictable and is much larger at 100% saturation than in anoxic conditions. The environmental factors causing this drift are unknown; its complete functional form is also unknown, but it is nonlinear in oxygen and a decaying exponential in time. We anticipate that using a Stern–Volmer form [Eq. (2)] to calibrate the optode, rather than a polynomial, would lead to more insight and plan to reanalyze these data using this approach.

Acknowledgments

This work was supported by the National Science Foundation under Grants OCE 0549887 and 0834340, and benefitted from logistics supplied as part of the Office of Naval Research ITOP program. Floats were air deployed by the Air Force Reserve Command 53rd Weather Reconnaissance Squadron “Hurricane Hunters.” The engineers and staff of the Applied Physics Laboratory provided essential support, as did the officers and crew of the R/V Revelle, R/V Cape Hatteras, and R/V Robertson.

REFERENCES

  • Alkire, M. B., and Coauthors, 2012: Estimates of net community production and export using high-resolution, Lagrangian measurements of O2, NO3 , and POC through the evolution of a spring diatom bloom in the North Atlantic. Deep-Sea Res. I, 64, 157174, doi:10.1016/j.dsr.2012.01.012.

    • Search Google Scholar
    • Export Citation
  • Carlson, J., 2002: Development of an optimized dissolved oxygen sensor for oceanographic profiling. Int. Ocean Syst., 6, 2022.

  • Champenois, W., and Borges A. , 2012: Seasonal and interannual variations of community metabolism rates of a Posidonia oceanica seagrass meadow. Limnol. Oceanogr., 57, 347361.

    • Search Google Scholar
    • Export Citation
  • D'Asaro, E. A., 2003: Performance of autonomous Lagrangian floats. J. Atmos. Oceanic Technol., 20, 896911.

  • D'Asaro, E. A., 2010: Calibration of the dissolved oxygen sensors on float 48 and on the Knorr CTD with Winkler bottle samples. The 2008 North Atlantic Bloom Experiment, Calibration Rep. 3, Archived at the Biological and Chemical Oceanography Data Management Office, 83 pp. [Available online at http://data.bco-dmo.org/NAB08/Oxygen_Calibration-NAB08.pdf.]

  • D'Asaro, E. A., and McNeil C. , 2007: Air–sea gas exchange at extreme wind speeds measured by autonomous oceanographic floats. J. Mar. Syst., 66, 92109.

    • Search Google Scholar
    • Export Citation
  • Demas, J. N., DeGraff B. A. , and Coleman P. , 1999: Oxygen sensors based on luminescence quenching. Anal. Chem., 71, 793A800A.

  • Edwards, B., Murphy D. , Janzen C. , and Larson N. , 2010: Calibration, response, and hysteresis in deep-sea dissolved oxygen measurements. J. Atmos. Oceanic Technol., 27, 920931.

    • Search Google Scholar
    • Export Citation
  • Garcia, H. E., and Gordon L. I. , 1992: Oxygen solubility in seawater: Better fitting equations. Limnol. Oceanogr., 37, 13071312.

  • Garcia, H. E., Locarnini R. A. , Boyer T. P. , Antonov J. I. , Baranova O. K. , Zweng M. M. , and Johnson D. R. , 2010: Dissolved Oxygen, Apparent Oxygen Utilization, and Oxygen Saturation. Vol. 3, World Ocean Atlas 2009, S. Levitus, Ed., NOAA Atlas NESDIS 70, 344 pp.

  • Keeling, R., Körtzinger A. , and Gruber N. , 2010: Ocean deoxygenation in a warming world. Annu. Rev. Mar. Sci., 2, 199229, doi:10.1146/annurev.marine.010908.163855.

    • Search Google Scholar
    • Export Citation
  • Kihm, C., and Körtzinger A. , 2010: Air–sea gas transfer velocity for oxygen derived from float data. J. Geophys. Res., 115, C12003, doi:10.1029/2009JC006077.

    • Search Google Scholar
    • Export Citation
  • Klimant, I., Kühl M. , Glud R. N. , and Holst G. , 1997: Optical measurement of oxygen and temperature in microscale: Strategies and biological applications. Sens. Actuators, 38B, 2937.

    • Search Google Scholar
    • Export Citation
  • Körtzinger, A., Schimanski J. , and Send U. , 2005: High quality oxygen measurements from profiling floats: A promising new technique. J. Atmos. Oceanic Technol., 22, 302308.

    • Search Google Scholar
    • Export Citation
  • Martini, M., Butman B. , and Mickelson M. , 2007: Long-term performance of Aanderaa optodes and Sea-Bird SBE-43 dissolved-oxygen sensors bottom mounted at 32 m in Massachusetts Bay. J. Atmos. Oceanic Technol., 24, 19241935.

    • Search Google Scholar
    • Export Citation
  • McNeil, C., and D'Asaro E. , 2007: Parameterization of air–sea gas fluxes at extreme wind speeds. J. Mar. Syst., 66, 110121.

  • Nicholson, D., Emerson S. , and Erickson C. , 2008: Sub mixed-layer oxygen production determined from Seaglider surveys. Limnol. Oceanogr.,53, 2226–2236.

  • Prakash, S., Balakrishnan Nair T. M. , Udaya Bhaskar T. V. S. , Prakash P. , and Gilbert D. , 2012: Oxycline variability in the central Arabian Sea: An Argo-oxygen study. J. Sea Res., 71, 18, doi:10.1016/j.seares.2012.03.003.

    • Search Google Scholar
    • Export Citation
  • Riser, S., and Johnson K. , 2008: Net production of oxygen in the subtropical ocean. Nature, 451, 323325.

  • Tengberg, A., and Coauthors, 2006: Evaluation of a lifetime-based optode to measure oxygen in aquatic systems. Limnol. Oceanogr. Methods, 4, 717.

    • Search Google Scholar
    • Export Citation
  • Uchida, H., Kawano T. , Kaneko I. , and Fukasawa M. , 2008: In situ calibration of optode-based oxygen sensors. J. Atmos. Oceanic Technol., 25, 22712281.

    • Search Google Scholar
    • Export Citation
Save