The Argo project intends to continuously monitor temperature and salinity of the upper 2000 m of the global ocean by use of autonomous, vertically profiling floats. They are currently generating the largest oceanographic dataset that ever existed, covering most of the world’s oceans. However, the use of these instruments in the polar oceans is seriously impeded by the presence of sea ice, as floats are hindered from transmitting their profile data or, even more seriously, potentially damaged when ascending to, or being at, the ice-covered sea surface. The authors present a cost neutral ice sensing algorithm (ISA), which alerts for the likely presence of sea ice. In this event, the profile is aborted and no surfacing attempted. To retrospectively track floats that actively remained under the sea ice because of ISA, acoustic tracking via the RAFOS technique was tested in the Weddell Sea. Last but not least, the most recent version of floats features the option of interim storage of profiles that could not be transmitted in real time (iStore). With these three developments, the ice-compatible float system to reliably extend Argo into the Antarctic Ocean in the near future was completed. Additional improvement might include using faster satellite communication links (Iridium or Argos 3) to reduce the float’s risk-prone surface drift.
About 90% of the observed increase of the heat content of the entire climate system between the 1950s and 1990s occurred within the ocean (Levitus et al. 2001). To adequately monitor future changes of the water mass properties, profiles of temperature, salinity, and velocity of the upper 2000-m depth of the world’s oceans are collected by the Argo system (Gould et al. 2004). The project aims to maintain statistically an array of 3000 autonomous profiling floats. It is expected to have at least one float in an area of 3° × 3°. While the float coverage is already moderate or good at low and midlatitudes, float deployments at higher latitudes (>60°) remained marginal.
Since 1999 a total of 92 floats were deployed in the Weddell gyre by the Alfred Wegener Institute. Early deployments (until 2003) focused on the region from 50° to 60°S along the Greenwich meridian. Although the ice coverage in this area is comparatively low, 23 of these 35 autonomous profiling explorer–typed floats (APEX; manufactured by Webb Research Corporation, Falmouth, Massachusetts) became engulfed by the growing sea ice. The others drifted north and left the sea ice zone. From the 23 floats encountering sea ice, 16 floats ceased to transmit data during the first winter within the sea ice–covered area. During the second winter six floats again entered the sea ice zone and only three of them prevailed until spring. None of these floats outlasted the third winter season (Fig. 1). Thus, the probability of these early floats to endure a winter when encountering sea ice is estimated to be approximately 37% (13 of 35). The losses were most likely due to crushing between the ice floes while drifting at the surface or due to damage to the Argos antenna when hitting the bottom of the ice cover during ascend.
How could this situation be improved? Discussions on how to achieve ice compatible, vertically profiling floats used to focus on different concepts of actually detecting sea ice above the float while it was ascending. However, while some people, like Peter Hoeg’s fictitious character Smilla (Hoeg 1993), can indeed develop a formidable “sense of snow and ice,” the cost-effective implementation of such skills into Argo floats forms a serious technical challenge. One option to distinguish between open-ocean and ice-covered conditions would be based on upward-looking sonar (ULS; Harms et al. 2001). During ascend to the sea surface, the float emits pings or chirps and evaluates the signal form of the return echo to estimate the probability of sea ice above. This approach requires extensive modifications in the hardware (including size and weight of the float), control systems, and data processing of the float and noticeable energy consumption in the order of 40 W h for the acoustic emission (about 3% of available energy) and in the same order for the buoyancy adjustment. Another possible method is to let the float knock against the underside of the ice with a tough antenna and hope that it penetrates the ice (Nevala 2005). This technique is a necessity in perennial ice-covered regions like the high Arctic. It utilizes small areas that are ice free or covered with thin ice to full capacity. However, the method depends on fast satellite communication systems (e.g., Iridium or Argos 3) that only now have become reliable. Both techniques provide information on local ice conditions, directly above the float, with little information on the conditions in the wider vicinity of the expected surfacing position. However, sea ice conditions may change rapidly, particularly during the transitional months with partial sea ice coverage. Hence, the detection of ice at a single point just prior to the surface attempt is insufficient; in fact it should cover the area within which ice floes might potentially reach the float while at the surface (about 1 km in diameter) to guarantee that the float will not be enclosed and possibly crushed by ice. To avoid this risk, an alternative approach is feasible in the Antarctic Ocean, which is largely ice free during the summer. There, near-freezing surface waters indicate the onset of sea ice formation, while a warmer surface layer indicates ice-free conditions. This natural ice indicator may be exploited to steer the floats free of ice encounters, as discussed in the next section.
2. The ice sensing algorithm
The basic assumption of the ice sensing algorithm (ISA) is that the probability of the presence of sea ice is related to the temperature in the water column below. However, only the records of the few and sparse early standard Argo floats enduring the austral winter (Fig. 1) provided the first year-round time series of near-surface waters in the Weddell Sea. These revealed that during the cooling season (approximately from March to August) the temperature within the upper 150 m decreased monotonously until it almost reached the freezing point (approximately −1.9°C; see Fig. 1). Although the freezing point is a function of salinity (and pressure), the typical salinity variations in the Weddell Sea have only a small influence on the freezing point. From about 1000 float observations within the Weddell gyre (defined from the Antarctic continent to 57°S and from the Antarctic Peninsula to 30°E) the mean salinity and standard deviation at a 35-m depth were determined as 34.13 and 0.18, respectively. The change of the mean salinity of three standard deviations results in a change of the freezing point of 0.03°C.
During the cooling season, the vertical temperature variations are small, that is, on the order of 0.01°C. Near-freezing point temperatures prevail until December, when warming starts in particular within the upper 50 m (Fig. 1). Thus, the upper 50 m of the missing profiles (white areas in Fig. 2) consist of water with temperatures below −1.79°C. Equipped with this information, we sought an algorithm capable of detecting the likelihood of the presence of sea ice with the following requirements.
The modifications to the existing float design and firmware should be small.
The energy requirement should be small to avoid a reduction of the float’s lifetime.
The ice detection should be robust.
The interannual variability of the ice coverage should be accounted for.
The loss of data resulting from the algorithm should be as small as possible.
The first constraint prohibited the use of additional sensors like ULS or contact sensors. Use of the salinity sensor was, at the time, inappropriate as not all floats featured these sensors and their stability was not clearly established. While most current salinity sensors operate flawlessly, some experience drifts and jumps, which prohibit the use of the salinity sensor in the mission-critical ice detection. Simple season-dependent timing of the surface events (possibly depending on the previous location) is inadequate due to the loss of all surface data and due to regions of perennial ice cover and the variability of the sea ice in both space and time. For instance, the float labeled 56 in Fig. 1 [World Meteorological Organization (WMO) ID 7900080] resurfaced in June 2006 after it was under sea ice for 14 months, whereas float 57 (WMO ID 7900081), deployed nearby at a distance of about 60 km, resurfaced 6 months earlier (Fig. 1). Therefore, the use of the temperature sensor was the most suitable option to comply with all mentioned requirements. Based on the temporal evolution of the near-surface temperature profile, the first ISA abort condition was defined as follows:
1) Abort if the median of the temperature of the seven near-surface temperature measurements between 20- and 50-m depth is less than −1.79°C. The velocity of the ascent is about 0.08 m s−1; hence the upper boundary of 20 m gives up to 4 min for the float to cease ascending before it reaches the underside of the sea ice. The median, rather than the mean, temperature was used to facilitate calculation and to remove eventual outliers.
To reduce the risk of damage in case this condition fails to detect the presence of sea ice, a second safeguard condition was implemented:
2) Abort if a (water) pressure of less than 4 dB is not reached after a period of twice the expected ascent time to the surface at an assumed ascent speed of 0.08 m s−1 and the actual temperature is below −1.79°C.
To record the number of aborts, ISA supplies an ice detection number with each abort, which is transmitted with the regular status message.
Since December 2001, all Alfred Wegener Institute (AWI) floats were equipped with ISA and the area of float deployments was moved to the region south of 60°S where heavy ice conditions occur during extended periods of time. Of the 54 ISA-equipped floats 46 were in contact with the sea ice (Fig. 1). From these 46 floats, 36 endured the first winter season. These floats were expected to deliver a total number of about 2000 profiles until April 2006, of which 907 profiles were not received (missing profiles). For 772 (85%) of these missing profiles the ice detecting count was activated as reported by the floats’ internal counter of aborted profiles. All floats of the 2005 and earlier batches that survived the first winter (12 floats) survived the second winter season as well (Fig. 1). Hence, the probability of ISA-equipped floats to endure a winter is about 80% (48 of 58), that is, more than twice as high as for non-ISA floats.
Further improvements of ISA are forthcoming. Surfacing within partly ice-covered areas occurs more likely at the end of the ice season when ice still exists on top of a warmer surface layer than in the beginning of ice formation in autumn (Fig. 1). Hence, the version of ISA described here performs better during the cooling period than during the warming period. This is due to annual variations of the near-surface temperature structure in correlation with the ice cycle. As a result of convection, the cold layer of the near-surface waters in autumn and winter is relatively homogenous before ice formation starts, while the warming of this layer in spring creates stratification, and not all ice floes have disappeared when the warming starts (Fig. 2). To overcome this restriction, a “retarded” response has been implemented in this (2006/07) season’s batch: once activated, ISA will now need to detect “surfacing conditions” (i.e., the lack of “abort conditions”) for two ascent cycles consecutively before giving the float permission to completely ascend to the surface on the second cycle.
3. Profiling and tracking under the ice
With some areas being ice covered for significant periods of time, substantial numbers of profiles will be aborted and thus not transmitted by the current standard Argo float, although these profiles had been measured by the float. Hence, it is desirable to save these data until they can safely be transmitted at a later date.
The most recent generation of Argo-compatible Navigating European Marine Observer (NEMO) floats manufactured by Optimare Sensorsysteme AG (Bremerhaven, Germany) are furnished with a CompactFlash card slot and thus a sufficiently sized internal memory to facilitate the interim storage (iStore) of ISA-aborted profiles. The data of the aborted profiles are transmitted during the subsequent summer season (February–March), when ice coverage—and hence risk of damage—is minimal, even when extended surface periods are needed to transmit the larger data volume. Although this procedure is inconsistent with the Argo goal of real-time data, it does deliver valuable information about the undisturbed temperature and salinity distributions under sea ice on seasonal time scales. During Polarstern expedition ANT XXII/3 (21 January–6 April 2005) six iStore-equipped NEMO floats were deployed. Five of those floats delivered a total of 91 profiles collected below the sea ice. The remaining float has reemerged after the summer season (in June 2006) and thus did not deliver the interim-stored profiles yet, which will be transmitted during the 2007 summer season. The available data are currently being processed and will be presented in a forthcoming publication.
To optimally utilize interim-stored profiles, their location (under the sea ice) must be known to an acceptable level of accuracy. Use of travel time measurements of frequency-modulated underwater sound signals (Rossby et al. 1986) allows retrospective tracking of floats by means of RAFOS technology with an accuracy of a few kilometers. Large distances (>1000 km) between the RAFOS signal’s source (a moored sound source) and the receiver (float) are achieved at low- to midlatitudes due to the existence of the Sound Fixing and Ranging (SOFAR) channel. At high latitudes the near-surface temperature is constant or decreasing toward the surface. Thus, a much narrower, shallow sound channel (surface duct) is developed (Jensen et al. 1994). The possible detection range is highly variable due to an unknown influence of sound reflections at the sea ice surface on the coherence of the RAFOS signal (Lee and Gobat 2005).
To estimate the detection range in the Weddell Sea on the basis of field data, in December 2002 three RAFOS sound sources were moored at about an 850-m depth emitting a sound signal of an 80-s 1.6-Hz frequency-modulated sweep near 260 Hz. Furthermore, nine RAFOS floats, ballasted for a 750-m depth, were deployed around Maud Rise. The floats were programmed to surface by the end of February 2004, when about 80% of the Southern Ocean is ice free. Nevertheless, data from only five of these nine floats were retrieved by satellite telemetry (Fig. 3) and tracked with the Matlab-based ARTOA 2 software package (available online at http://www.whoi.edu/science/PO/rafos/).
The correlation between the received and the nominal sound signal, expressed in terms of correlation heights (values 0–7), is a proxy for the quality of the respective reception. The correlation height depends strongly on the distance between the float and the sound source and, in Antarctica, on the season. In austral summer we found large correlation heights for distances up to 700 km, which decreased to about 300 km in winter (Fig. 4). The causes for lesser correlation height during winter are still under investigation, but the difference in stratification of the upper water column, that is, a decrease of the sound velocity from 50-m depth to the surface in summer (Fig. 5) and interference with sea ice are the most likely causes. However, the continuous motion of the float implies continuous times of arrival (TOAs) time series of the sound signals at the float. By virtue of this requirement, TOAs may frequently be identified as trustworthy and used for subsurface tracking, regardless of the quality of the associated correlation height. In other words, poor signals with poor correlation heights can be used if they line up with signals of higher correlation heights (Fig. 4). Our data show that a persistent detection range off 500 km is achieved this way and that thus an effective use of the RAFOS system is possible throughout the year. Assuming a detection range of 500 km requires the installation of a set of about 15 moored RAFOS sound sources to cover the Weddell gyre, that is, to 30°E.
Use of RAFOS subsurface tracking provides additional benefits. Floats have been and are used to infer middepth oceanic circulation patterns, which are estimated for Argo floats from the surfacing positions of two consecutive profiles. The large time lag between the surface positions, generally 10 days, allows only a rather coarse estimate of the circulation pattern (Fig. 6). In regions of significant eddy activity, such as to the southwest of Maud Rise, much of the mesoscale velocity signal becomes aliased. The RAFOS technique allows the determination of these mesoscale current fields as the RAFOS positions are obtained every 24-h interval (Fig. 6).
The three-step process of developing ice-compatible floats is almost completed. The endurance of ice sensing algorithm–equipped floats is about 80%, that is, more than twice as high as for standard Argo floats. Further improvements of the ice compatibility ISA are expected due to an increase of the “abort temperature” and the implementation of a retarded surfacing response. The interim storage and delayed transmission of data from aborted profiles has been successfully tested. The useful range of the RAFOS signal within the Weddell Sea was determined to be at least 500 km year-round, permitting its practical utilization. Hence, with these modifications, profiling Argo floats may be used successfully in seasonally ice covered areas, particularly in the context of climatological and process-oriented ocean research.
We thank Webb Research Corporation and Optimare Sensorsysteme AG for their cooperation. Don Dorson of Bathy Systems suggested the use of the median, rather than the mean, temperature when implementing ISA in the WRC-APEX firmware. R. Schlitzer, AWI, provided the Ocean Data View software (Schlitzer 2006) and L. Kaleschke, University of Bremen, made available the sea ice distribution data. This work was supported by the Bundesminister für Bildung und Forschung (Grant Fkz 03F0367B) and the MERSEA Grant.
Corresponding author address: Dr. Olaf Klatt, Alfred Wegener Institute for Polar and Marine Research, Bussestrasse 24, D-27570 Bremerhaven, Germany. Email: firstname.lastname@example.org