Unaccounted transient or permanent changes in sensor performances can compromise the overall quality of datasets obtained with a glider. From the specific perspective of the principal physical variables (temperature and conductivity), the main short-term risk is the deterioration of the quality of salinity data that is primarily due to the fouling of the glider’s conductivity sensor, especially when the glider approaches the sea surface and is deployed in coastal waters. Another potential short-term risk is a sudden shift in the glider’s temperature response caused by the blanketing of the temperature sensor by extraneous material. The long-term risks are the intrinsic drifts of the two sensors as specified by the sensor manufacturer. Given the way a glider operates, it is practically impossible to obtain sufficiently representative temperature and salinity samples in the field for effective comparisons with sensor data. Hence, laboratory testing of the temperature and conductivity sensors is one approach used to estimate the qualitative changes in pre- and postdeployment sensor performance in order to mitigate measurable drift effects on the collected data. This paper presents a systematic procedure that can be used to conduct reproducible laboratory evaluations of the temperature and conductivity sensors on a Teledyne Webb Research (TWR) Slocum glider. It is shown that the data from such tests can give useful information regarding the data quality as a result of fouling of the sensors during missions. Obviously, this kind of information is valuable when the reliability of mission-generated temperature and conductivity data, which are used to determine salinity, have to be assessed.
The Department of Oceanography of the Istituto Nazionale di Oceanografia e di Geofisica Sperimentale (OGS) acquired a Teledyne Webb Research (TWR) Slocum electric glider in 2007 to reinforce its marine observational capabilities. The glider, an autonomous underwater vehicle (AUV), was denominated “Trieste 1,” and is designed to work safely down to a depth of about 200 m. This makes it particularly suited for operations in littoral waters. Driven in a sawtooth vertical profile by variable buoyancy and internal mass displacements, it is capable of covering spatial and temporal grids at very fine resolutions, making it an exceptional network component for subsurface monitoring schemes on a regional scale (Schofield et al. 2009; Jones et al. 2005; Creed et al. 2003, 2004).
The Trieste 1 came equipped with sensors for making measurements of conductivity, temperature, and pressure during deployments (Fig. 1). The inlet of the pressure sensor is located on the glider’s fuselage, a few centimeters above the leading edge of the port-side wing. The conductivity–temperature (C–T) module is constituted by a needle-shaped temperature probe and a conductivity measuring cell incorporating antifoulant devices at both ends. The antifoulant devices help to reduce biological growth inside the cell. The C–T module is not pumped but is attached externally below the same wing in a low-drag configuration that allows water to first pass the temperature probe and then flow through the conductivity cell when the glider is underway. The support of the module and the pressure inlet are roughly in line and equidistant from the port-side wing itself (Fig. 1).
The different sensors and their accessories were supplied by Sea-Bird Electronics, Inc. (SBE), and fitted to the glider by TWR during the construction of the vehicle. The “as calibrated” accuracies of the sensors upon dispatch from SBE are 0.002°C for temperature and 0.005 in equivalent practical salinity for conductivity (~0.001 S m−1), but the effective accuracies in the field are probably to be considered no better than 2 times as much (i.e., 0.004°C and 0.002 S m−1) owing to the shipping and subsequent installation operations (SBE 2010, personal communication).
Trieste 1 generates enormous amounts of temperature and conductivity data during missions (the maximum sampling frequency setting for its C–T module is 1 Hz), and we thought it would be reasonable to expect the relevant sensor performances to vary over the course of one or more consecutive deployments. Indeed, the possibility of this occurring is quite real. A glider needs to surface in order to communicate, download new mission parameters, take positional fixes and upload data. At these times, its sensors are particularly vulnerable to physical damage or fouling events that can cause transient or even permanent sensor modifications. There is, thus, an ever-present risk of deteriorating data quality when working with gliders, especially in coastal waters where the phenomenon of fouling cannot be ignored a priori without jeopardizing the integrity of the collected data.
In this paper, we describe a systematic, reproducible procedure for testing the C–T module of a TWR Slocum glider in the laboratory. This procedure can be used to assess the performances of the module’s temperature and conductivity sensors under controlled conditions. More specifically, as we will show, the procedure can assist in evaluating the eventual impact of fouling on the response of the conductivity sensor. Note, however, that our methodology tackles only what could be construed as “accruing” errors that are caused, in one way or the other, by material changes in sensor components over time. It cannot, and does not attempt to, address dynamic errors. Such errors are generated by the alignment geometry of the CTD system and the uneven sampling rates of the different sensors comprising it. The absence of a pump and the discontinuity between the temperature and the conductivity sensors in the C–T measuring circuit are other causes for concern from this point of view. A pump would have ensured a constant response time for the conductivity sensor by maintaining a steady flow of water through it. The pump guarantees that a pair of ducted T and C sensors see the same water parcel, as well as enables the T and C response times to be more closely aligned in time (due to response time matching) and space (due to positional distance and transit time of the water parcel between the sensors), both of which are necessary to effectively eliminate salinity spiking. Another advantage of pumping is the ability to effectively remove the cell thermal mass errors inherent in all of the conductivity measurements made while passing through a thermocline. These errors are on the order of 0.03 in terms of the practical salinity (henceforth referred to as just “salinity”) per 1°C change in temperature. The amplitude of this effect scales as the inverse of the flow through the cell, which is why pumping allows for its correction.
The paper is organized in the following way: Section 2 describes the laboratory setup we use while testing. Section 3 outlines the main steps of the testing procedure. An example of the kind of results that can be obtained with our procedure is furnished in section 4. Section 5 illustrates the cleaning method that is employed in order to evaluate the possible effects of fouling on the conductivity sensor. Section 6 presents a general discussion on the performance of the Slocum glider’s C–T sensors, based on the reported test results for Trieste 1. We close the paper with some concluding remarks in section 7.
2. The laboratory setup
The work was carried out in the Centro di Taratura Oceanografico, the temperature- and humidity-controlled oceanographic calibration laboratory of the OGS. The best solution involved immersing the glider’s C and T sensors in a Guildline model 5010 thermostatic calibration bath (stability ≤ 0.002°C day−1) filled with filtered natural seawater (filter size: 0.22 μm). The glider is positioned on the bath without its wings using appropriate supports and is rotated carefully so that the C–T module is completely submerged (Fig. 2). Plain plastic wrap is wound around the glider and tucked in under the edges of the bath in order to minimize perturbation of the bath environment once a temperature set point is attained. The salinity of the seawater in the bath tends to rise and fall slightly when the set points are, respectively, much higher or much lower than the laboratory temperature setting (usually 21° or 23°C) resulting from the increased evaporation at higher temperatures and some condensation at lower ones. This is one more reason why we determine the salinity of the bath at every temperature set point, as described in the following section.
The Guildline bath draws in water near the surface and pumps it out at the bottom under a false bottom. The resulting circulation pattern maintains the measuring cell of the conductivity sensor properly flushed despite its immobility in the water (Fig. 3).
To further ensure a uniform flow through the conductivity cell, a SBE 5M submersible pump is connected to the rear of the C–T unit, that is, to the free end, away from the temperature sensor (Fig. 4), using an ad hoc exhaust port/spigot assembly (SBE 1992). This pump also serves to minimize the possible trapping of air bubbles within the cell. Once it is attached, the pump is powered on, and then left activated until the end of the testing.
3. The testing procedure
The glider is operated in the “lab mode” (WRC 2005) with its factory calibration settings and the C–T sampling frequency set to the maximum (1 Hz). The Guildline 5010 bath that is filled with the filtered seawater is cycled down from around 30° to 0°C in step changes of circa 5°C in order to provide temperature and conductivity set points where glider C–T readings and comparable reference temperature and conductivity values are acquired. Note that we obtain conductivity set points by varying the temperature of the seawater in the bath; we do not test different salinities in the bath with seawater. At each temperature set point, a reference temperature value is estimated from a set of readings made using a calibrated Rosemount 162CE Standard Platinum Resistance Thermometer (SPRT) coupled with a Hart Scientific 1590 Precision Digital Thermometer (Fig. 5). A reference conductivity value corresponding to the specific reference temperature is obtained by inverting the measured salinity of a water sample collected after the SPRT and glider C–T measurements are completed. The salinity determinations are carried out using a Guildline 8400B Laboratory Salinometer, standardized with International Association for the Physical Sciences of the Oceans (IAPSO) Standard Seawater. Finally, means are computed for the different sets of glider temperatures and conductivities, and the corresponding residuals with respect to the relevant reference values are estimated. The entire operation takes about 8–10 h.
As a rule, the Guildline bath is allowed to settle for at least 1 h at a temperature set point before sampling with the SPRT and the glider C–T module is initiated. Typically, about 5-min worth of data are recorded each time, and the stability of the bath during the sampling interval is monitored by following the SPRT temperature trace on the display of the Precision Digital Thermometer.
All of the elements of the reference measuring systems are maintained to within declared specifications by monitoring their performances regularly, adhering to recommended usage and upkeep practices, and scheduling servicing with a manufacturer immediately when laboratory quality assurance procedures indicate a developing problem. The Rosemount 162CE SPRT is checked, and, when necessary, recalibrated to subrange 11 (0.01°–29.7646°C) of the International Temperature Scale of 1990 (ITS-90) using a Hart Scientific 5901 Triple Point of Water Cell and a Hart Scientific 5943 Gallium Melting Point Cell. The 1590 Precision Digital Thermometer is controlled using an externally calibrated, certified Hart Scientific 5699 SPRT with the same ITS-90 fixed-point cells. This SPRT also serves as an independent temperature reference for comparisons if required. The Guildline 8400B Salinometer is standardized using a bottle of IAPSO P-Series Normal Standard Seawater (salinity = 35) prior to every salinity sample analysis run. Immediately after standardization, a bottle of IAPSO 38H-Series High Salinity Standard Seawater (salinity = 38) is measured to determine the instrument offset and linearity in the characteristic salinity range of the Mediterranean Sea where we operate our glider. The standardization and the offset and linearity checks are always repeated every 24 h during testing. A full-scale linearity check of the salinometer (10 ≤ salinity ≤ 38) is performed at least once every 6 months in-between returns to the factory employing IAPSO P-Series, 10L-Series (salinity = 10), 30L-Series (salinity = 30), and 38H-Series Standard Seawaters.
The combined standard uncertainties associated with the reference temperature and conductivity determinations are 0.0015°C and 0.000 17 S m−1, respectively (Table 1). In the case of temperature, the reported value takes into account the following uncertainty sources: the stability and uniformity of the Guildline 5010 bath, the stability of the SPRT, and the accuracy of the Precision Digital Thermometer (operated with the built-in 100-Ω internal reference resistor and a nominal current of 1 mA), used for making the SPRT resistance measurements. As for conductivity, the stated value comprehends the uncertainty in the measurement of reference temperature and the accuracy of the Guildline Salinometer used for analyzing the salinity samples.
The repeatability associated with the depicted setup is illustrated in Fig. 6 where the variabilities associated with the measured bath temperatures and the readings acquired by the glider’s C–T module across four separate test runs carried out on four different occasions between January 2008 and September 2009 are shown. As can be seen, the Guildline bath was stable to better than 0.0015°C every time. The standard deviations for the relative glider temperature and conductivity measurements were within 0.002°C and 0.0003 S m−1, respectively. There are likely to be some small errors in the glider measurements owing to the fact that the electronics were at room temperature (21° ± 1°C) during the testing, but the magnitude of the variabilities would seem to indicate that they are extremely small for both the temperature and the conductivity sensor. The working of the SBE 5M pump does not have any discernable effect on the stability of the temperature bath over the typical sampling interval of 5 min that we employ.
4. Evaluating glider C–T sensor performance
As mentioned earlier in section 1, SBE assigns as-calibrated accuracies of 0.002°C and ~0.001 S m−1 for our kind of glider temperature and conductivity sensors. These are the accuracies that they expect to reproduce if any such sensors were to be returned to them unused for a calibration check. We, therefore, decided to apply these accuracies as evaluation criteria for testing the performances of our glider’s temperature and conductivity sensors.
The results from a test of the temperature and conductivity sensors of Trieste 1 carried out using the setup and procedure described in sections 2 and 3 above are illustrated in Fig. 7. The test was performed after a series of short deployments in coastal waters in 2008.
Healthy SBE temperature sensors drift in offset, whereas healthy SBE conductivity sensors drift in slope (not offset). The behavior of the temperature sensor was found to be acceptable; the temperature showed maximum deviations of less than ±0.002°C, which are well within the manufacturer’s calibration specifications for it over the whole testing range. On the other hand, the conductivity sensor showed a clear trend with observed differences becoming greater as conductivity increased.
Typically, changes in the performance of a sea-going conductivity sensor can be attributed to the following two main causes: sensor drift and fouling (Oka and Ando 2004; Medeot et al. 2008). Permanent systematic offsets are rare in SBE conductivity sensors because their design features tend to ensure that the measuring elements are considerably well protected. However, they are sensitive to fouling (either singular or continuous events), which alters the measuring cell geometry. This modifies the overall response of the sensor by reducing the sensitivity, resulting in possible low readings with respect to actual conductivities. It is difficult to resolve when the fouling effects start to become evident in the data, so a linear drift is often assumed over time. Pre- and postdeployment sensor validations or factory calibrations can allow for such analysis. After postdeployment validation, the cell is cleaned and the validation is repeated, which is used as the next deployment’s predeployment validation. The procedure we employ to clean the measuring cell to evaluate the effect of fouling on conductivity response is described in detail in the next section.
5. Cleaning the conductivity sensor
Once the behavior of the glider’s conductivity sensor is established over the entire range of operation, the bath temperature is stabilized at some convenient set point (usually 20°C). The conductivity cell is cleaned thoroughly using Triton X-100 and bleach solutions, and flushed with deionized water. The testing setup described earlier is replicated. A first set of reference and glider temperatures and conductivities is obtained, and the corresponding residuals are computed. The cell is cleaned again, and readings are taken. Means and residuals are calculated as before. The conductivity residuals obtained after the cleaning operation are compared to those from the preceding set of measurements to see if they are smaller. The entire process is repeated until the conductivity residuals from consecutive sets of measurements show no significant differences. At this point, we assume that all of the fouling that we could reasonably eliminate has been removed. This is not to say that the conductivity cell has returned to its original factory calibration condition.
The design and location of the conductivity cell make it very hard to clean. The fragility of the mounting configuration, the presence and angle of the temperature probe at the front end, and the lack of any kind of support for attaching tubing are all hindrances to effective cleaning of the cell. We clean the conductivity cell using a slightly modified version of the procedure recommended for this operation by the manufacturer (SBE 2008a,b). A syringe filled with the specified cleaning solution is connected to the free end of the glider C–T unit, as shown in Fig. 8. The plunger of the syringe is pressed so that the cleaning solution flushes the conductivity cell. The expelled wash is discarded, the syringe is refilled, and the whole operation is repeated a few times. Note that, every once in a while, it may be necessary to immerge the conductivity sensor completely in a beaker containing a 1000 ppm bleach solution for a couple of minutes in order to obtain a satisfactory result. This will require the separation of the payload bay midhull section, where the C–T unit is located, from the rest of the glider, and very careful handling (Fig. 9). The conductivity cell is rinsed with fresh deionized water for 5–10 min after every cleaning step (Fig. 10).
Temperature sensors generally drift slowly, at a steady rate, and predictably in a pattern that can be described by a simple offset. For our type of glider temperature sensor, a typical drift of 1 mK (0.001°C) over the first 6–12 months, and successive stabilization in the ensuing 5 yr afterward to better than 1 mK, can be expected (SBE 2010, personal communication). That our testing showed no indication of a significant temperature offset outside of the sensor specification is therefore not surprising, seeing that the glider was less than 2 yr old at the time.
Conductivity sensors, on the other hand, drift in slope mainly resulting from changes in the sensitivity as a result of accumulated fouling. Coatings on the inside of the measuring cell reduce the cell volume, causing an underreporting of true sample conductivities. The effect is clearly evident in Fig. 11, which summarizes the results of the tests carried out on the glider’s conductivity sensor following recovery. The conductivity residuals indicate a normal slope drift with respect to the original calibration, exceeding the sensor’s nominal accuracy of 0.001 S m−1 at the 6 S m−1 calibration point and showing a residual difference of about 0.0015 S m−1 there. This would produce a salinity error of around 0.007–0.008.
This change is summarized in Table 2, which shows the slope coefficients we obtained for the sensor before and after the cleaning of the conductivity cell. The coefficients were computed by applying the method described in SBE (2007) using the data from our tests. It can be seen that the slope of the sensor response in the tested range has effectively changed since the original calibration (a slope of 1.000 and an offset of 0). Prior to cleaning, the slope was found to be 1.000 243, which is equivalent to a salinity error of roughly −0.010 at a temperature of 19°C and a conductivity of 5 S m−1 (pressure = 0 dbar). The majority of the fouling likely consists of coatings from consolidated oils and both organic and inorganic materials that tend to accumulate near the sea surface. If the change in slope was related to fouling, then the cleaning of the conductivity cell ought to ameliorate the conductivity sensor’s response. In fact, in Fig. 11, the conductivity residuals from the test performed after the final cleaning cycle show that the calibration status of the sensor has improved, with values that were now within the sensor’s nominal accuracy specification of 0.001 S m−1. The computed slope value now was 1.000 145, corresponding to a salinity discrepancy at 19°C and 5 S m−1 on the order of −0.006, as opposed to the earlier value of −0.010 (Table 2), which is very close to the conductivity sensor’s nominal accuracy specification of 0.005 for salinity.
From the perspective of data quality, what is interesting is that performing the kind of extended laboratory testing we describe before and after a mission can provide data that are useful for generating slope and/or offset corrections for the readings obtained with the C–T module of a glider in the field. This information is invaluable, particularly when it comes to deriving salinities from the temperature and conductivity data acquired during the mission. Regular checking of the conductivity sensor prior to and after a mission is likely to be the single most productive data assurance measure that one could employ when having to validate field measurements of temperature and salinity obtained with a glider. Clearly, one needs access to adequate means and facilities to accomplish this level of calibration validation. However, the test reported here shows how successful and accurate validations can be done and what considerations are necessary in order to make adjustments or corrections to factory calibration values, including levels of accuracy of reference methods and calibration bath stability and precision requirements.
The usual maintenance and control procedures employed for CTD units in the field, like washing between casts and correcting for conductivity drift using salinity samples collected during profiles, are difficult to implement effectively in the case of glider CTD systems because of the way in which the vehicle operates. We have developed a procedure for testing the C–T module of a TWR Slocum glider in the laboratory by adapting conventional high-quality CTD temperature and conductivity calibration methods to take into account the vehicle’s particular design and functional characteristics. It is shown that testing a glider’s C–T module before and after missions can be useful in evaluating, and possibly correcting, the temperature and salinity data acquired during deployments. We strongly recommend this type of testing, especially in missions involving coastal waters or those where the glider remains in close proximity to the sea surface frequently and for relatively extended periods of time.
This paper was greatly improved by the comments from one of the anonymous reviewers. Our thanks go to Mr. F. Brunetti, Mr. A. Bubbi, and Mr. P. Mansutti of the Technological Development (TECDEV) Group at the OGS for their invaluable support and assistance in all the field operations with the glider Trieste 1 referred to in the paper.