To address some of the knowledge gaps outlined above, a set of eight buoys carrying a suite of oceanographic sensors was deployed upstream in the Transpolar Drift in October 2019 as part of the MOSAiC Distributed Network (DN). The instruments were installed within ∼ 40 km of the German research icebreaker R/V Polarstern, which was planned to stay anchored to the same ice floe for an entire year and constituted the main MOSAiC ice camp (also referred to as the Central Observatory). Incorporating a diverse collection of autonomous instrumentation, the DN was intended to spatially extend and complement the extensive work program conducted by the different teams in the Central Observatory and to bridge the gap between the spot measurements and larger-scale airborne and satellite observations. The DN consisted of several complex instrument systems distributed to a dozen main sites, complemented by a large number of additional simple GPS buoys in the wider DN. Of these sites, eight were equipped with the buoys presented in this study and were each co-deployed with a handful of other instruments (see below). This distribution, together with the ice drift, covered the mesoscale (of the order of a few tens of kilometers) and, to some extent, the submesoscale (of the order of 1 km). While we were originally aiming for less spacing between the individual buoys to better resolve the submesoscale processes, the challenging ice conditions and absence of sufficiently stable floes required us to change the plan and extend the deployment area to reduce the risk of early failures. The exceptionally dynamic ice pack in winter 2019/2020 led to a failure of several other buoys in the early period of MOSAiC (including three of eight units described here; see Sect. ).

Our focus was on the upper water column and, in particular, the whole upper ocean mixed layer across most of the Eurasian Basin. The primary aim of the expedition was to capture local processes in the strongly stratified region of the Transpolar Drift, which features near-surface waters that are much fresher and colder than those found in the Nansen Basin close to the inflow of water of Atlantic origin. The drift started in the Amundsen Basin, as close to the Siberian shelves as possible, where the Transpolar Drift stream originates. That allowed a continuous drift into late spring without relocating the ice camp and with a chance to recover the buoys. Placing instruments in the upper 100 m was expected to capture processes in the seasonally deepening and subsequently restratifying mixed layer and to also cover part of the upper halocline throughout much of the year.

We required a sufficient number of measurement points in the vertical to give a good indication of the structure of the mixed layer and the upper halocline. This especially aimed at resolving passing eddies or fronts, while minimizing the number of sensor packages and associated costs. Deployment alongside ocean profilers at different sites within the DN, such as the Woods Hole Ice-Tethered Profiler or the Drift-Towing Ocean Profiler DTOP;, ensured that the vertical structure of the upper water column was measured to 1 dbar resolution at least once a day. At the same time, our systems aimed at capturing transient phenomena, such as internal waves or eddies, requiring measurements as frequent as 2 min.

The sensor platform, hereafter referred to as Salinity Ice Tether (SIT), is a floatable buoy built by Pacific Gyre (California, USA) and comprised an oval surface float that housed the main electronics and batteries, connected to a 100 m long inductive modem cable with a 10 kg terminal weight attached to the bottom (Fig. a). Along this tether, five SBE37IMP MicroCAT CTD (conductivity–temperature–depth) instruments built by Sea-Bird Scientific, California, USA, and hereafter referred to as CTDs, were mounted at depths of 10, 20, 50, 75, and 100 m. Following the naming convention of the general AWI (Alfred-Wegener-Institut) buoy program, which was also implemented for all MOSAiC buoys, we identify the individual buoys with an ID consisting of the deployment year (in this case, 2019), followed by a buoy-type-specific letter (in this case, O) and a running number (from 1 to 8). The corresponding buoy IDs are 2019O1 to 2019O8 for the eight SIT systems. Furthermore, we refer to the individual CTDs by the buoy ID and the nominal CTD deployment depth, i.e., 2019O1-10m is the CTD attached to the tether of buoy 2019O1 at a nominal depth of 10 m.

To ensure an operational time of ∼ 1 year, the individual CTDs were set to record data internally at 2 min intervals, independent of the buoys' own sample and transmit intervals. The surface buoy itself recorded the GPS position and surface temperature and carried a submergence sensor. Furthermore, the buoy controller polled all CTDs for an additional measurement according to a preconfigured buoy sampling interval, which could, in principle, be adjusted by sending a reconfiguration command via iridium if necessary. However, throughout our experiment, all buoys were set to take a measurement and immediately transmit the corresponding data at a fixed 10 min interval, which was chosen to ensure an operational time of the surface buoy of at least 1 year. In summary, the internal sampling interval of the CTDs was 2 min, while an additional CTD measurement was obtained and transmitted by the buoy every 10 min together with the corresponding auxiliary (meta)data.

Upon removal of a magnet at the side of the hull after deployment, the buoy was switched on and automatically activated the internal 2 min sampling of the attached CTDs. The buoy controller sampled all of the CTDs 1 min before a scheduled transmission. After reading the auxiliary sensors, the buoy controller sampled the GPS at the scheduled transmit time to obtain a valid GPS position for transmission. The data resulting from this measurement cycle were then compressed into a file and sent via the iridium short-burst data (sbd) protocol to a shore-based server, which decoded the data and provided the data for download. In the standard buoy firmware, pressure is by default rounded to 0.1 dbar before transmission to save on data size and transmission costs. Unfortunately, this could not be changed during operation. Salinity was calculated on the server side based on the transmitted temperature, conductivity, and pressure measurements according to the Practical Salinity Scale 1978 (PSS-78). For the final dataset, however, practical salinity was recalculated from these variables using the MATLAB Gibbs–SeaWater (GSW) Toolbox version 3.05.5 .

The SBE37 MicroCAT CTD itself (either in its normal configuration or inductive modem configuration) is a reliable and accurate sensor that has been used for studies of physical oceanography on various platforms all over the world for decades, particularly for moored operations (e.g., ). The stated initial accuracy for this sensor type is ±0.003 mS cm-1 for conductivity, ±0.002 ∘C for temperature, and ±0.1 % of the full-scale range for pressure. For our instruments, the pressure rating was between 100 and 1000 m due to the limited availability of 100 m pressure sensors at the time of manufacturing. The initial accuracy for a 100 m calibrated pressure sensor (0.1 % of the full-scale range) is thereby ∼ 0.1 dbar. As stated earlier, the buoy firmware only transmitted 10 min pressure records to the first decimal, resulting in a vertical resolution of 0.1 dbar compared to 0.02 to 0.002 dbar (0.002 % full-scale range) of the CTDs themselves. Since the accuracy is of the same order of magnitude, we do not consider the transmission limitation for pressure as being significant in relation to our final dataset. For datasets that include the internally recorded data of the recovered CTDs, the pressure records as sent by the buoy have been interpolated to improve their quality (see Sect. ). The typical sensor stability is rated as 0.003 mS cm-1 and 0.0002 ∘C per month for conductivity and temperature, respectively, and 0.05 % of the full-scale range (pressure) per year.

The buoys were deployed along with a suite of other platforms during the set-up phase of the MOSAiC Distributed Network by the Russian icebreaker Akademik Fedorov in early October 2019 (Fig. b–f). A few days before reaching the deployment destination, the instruments were successfully tested on the deck. Ice floes were selected based on an inspection of high-resolution satellite imagery and helicopter surveys. The ice conditions were challenging for permanent installations since thin ice was predominant in the deployment area . The station work was either done from the ship or by Mi-8 helicopters. The instruments were co-installed on selected ice floes together with SIMBA-type ice mass balance buoys , snow buoys , and DTOPs.

The snow buoy, SIMBA, and SIT buoys were installed close to each other, while the DTOP was installed at a distance of at least 70 m away. After a deployment site with sufficiently stable level ice was chosen, the buoy tether was laid out on the ice, and the CTDs were attached to their designated depth positions along the cable. The instruments were covered with towels for protection during handling on the ice and snow. The cable and instruments were manually lowered into the ocean through a 25 cm diameter hole in the ice, and the surface float was placed on top of the hole. Air temperatures were as low as -15 ∘C. Table  summarizes the deployment information for each instrument.

The trajectories of the buoys during their 10-month-long drift are presented in Sect. . Buoys 2019O2, 2019O7, and 2019O8 failed within the first 5 weeks after their deployment. Buoy 2019O5 failed in July 2020, most likely due to ice ridging, having survived for 9 months. The other four buoys were recovered from the decaying ice field in the marginal ice zone of Fram Strait in August 2020 by the German and Russian icebreakers R/V Polarstern and Akademik Tryoshnikov.

The processing described above yielded eight individual time series, with one for each buoy. Only three time series (2019O2, 2019O7, and 2019O8) comprise a few weeks, while five (2019O1, 2019O3, 2019O4, 2019O5, and 2019O6) are ∼ 10 months long. For 2019O1, 2019O3, 2019O4, and 2019O6, we supply a merged product that combines buoy and CTD data, while for 2019O2, 2019O5, 2019O7, and 2019O8, only the transmitted buoy records are available.

The measured oceanographic variables are conductivity, in situ temperature, and pressure, and the derived variables are practical salinity and depth. A primary quality flag (1 – good, 2 – modified, 3 – questionable, 4 – bad, and 9 – missing data) is given for each of these five variables. Each measurement has a corresponding timestamp. Only the buoy measurements (indicated by a buoy_flag) initially had a GPS record, and a position was given to the higher-resolving CTD records by linear interpolation. The drift speed was calculated from the difference between GPS positions. Additionally, the buoy measured surface temperature and a submerged Boolean, which indicates whether the buoy was in seawater or not. Refer to Table  for a summary of all variables in each processed dataset. An overview of all individual datasets and the corresponding collections for raw and processed data are given in Table .

The drift trajectories of all eight buoys are shown in Fig. . Their journey alongside the Central Observatory, from the deployment area north of the Laptev Sea through the Transpolar Drift until their recovery in Fram Strait, took about 10 months. The buoys traveled over the Gakkel Ridge from the Amundsen Basin to the Nansen Basin in March/April 2020. The drift accelerated strongly from June onwards upon reaching the Yermak Plateau. While a detailed discussion of the drift pattern is beyond the scope of this paper, it should be noted that the relative positions of the buoys within the array did not change much before finally entering the Fram Strait area (Fig. ).

In the following, we take a closer look at each of the processed datasets shown as time series plots in Figs.  and . The corresponding T-S diagrams are provided as a supplement in the Appendix (Fig. ). As stated earlier, we refer to a specific instrument time series by the buoy name with the nominal instrument depth. For simplicity, we use the buoy name and instrument references as a synonym for the time series itself.

Buoy 2019O1 (Fig. a) was generally performing well, except for some problems at the beginning of its life cycle. 2019O1-10m and 2019O1-20m show a suspiciously low salinity for the first few days after deployment, before adjusting to consistent levels. 2019O1-75m and 2019O1-100m had the same issue at the very beginning (lasting only a few hours), while it took 2019O1-50m a full 2.5 weeks before plausible salinity data were collected. We assume that these instruments were affected by ice formation in the conductivity cells, which is a known issue when instruments are submerged while still being much colder than the seawater. We acknowledge, however, that it is quite unusual that it takes an instrument such a long time to recover, even when installed in seawater at freezing point. An alternative explanation is that bubbles were trapped in the conductivity cell that were eventually able to escape the cell. This is also a known issue, especially with shallow deployments. After this initial adjustment period, 2019O1 continued to collect reasonable data throughout its operational time. The unit was recovered on 5 August 2020.

Buoy 2019O2 (Fig. a) started fine before all instruments were suddenly lifted by 5 m over 1 h during an event on 12 November 2019. They remained in place for 1 d before they were lifted again by 8 m in a similar event. The buoy was lost immediately afterward on 13 November 2019.

Buoy 2019O3 (Fig. b) was fine initially, but 2019O3-50m stopped working after several hours following the deployment and never resumed any data transmission. The other four instruments were performing well throughout their operational time. During an event on 17 February 2020, all remaining CTDs were substantially lifted although there was no particularly strong drift at that time. 2019O3-10m and 2019O3-20m were even lifted to depths as shallow as the ice base, which is unusual. The only explanation is that the buoy tether became entangled with the co-deployed DTOP free floating profiler, which possibly moved closer to the SIT as a result of converging pack ice conditions. The sensors returned to their original depths after ∼ 30 min and continued to measure correctly until the buoy was recovered on 3 August 2020.

Buoy 2019O4 (Fig. c) was fine from the start. Due to an unintentionally high measurement interval of 30 s, 2019O4-75m ran out of power on 20 April 2020 after a period of intermittent failures starting on 11 January 2020. The buoy and all other CTDs continued to work well until recovery on 14 August 2020.

Buoy 2019O5 (Fig. b) showed some moderate problems. 2019O5-75m stopped communicating upon deployment. 2019O5-10m started to exhibit inductive modem communication issues at the end of October 2019. This began with sporadic missing measurements in the buoy's transmission afterward, which subsequently became more frequent throughout November when roughly half of the measurements were blanks. In the second half of November, the issue gradually disappeared, with transmissions returning to normal on 27 November. From late December 2019 onwards, 2019O5-20m and 2019O5-50m started to exhibit similar issues, again first sporadic and then almost continuously (roughly half of the measurements affected). Interestingly, this problem never affected both sensors at the same time, and the missing transmissions were always alternating and sometimes more or less periodic. From mid-January onwards, 2019O5-50m recovered from this problem, and transmissions were completely normal from the end of January 2020 onwards. However, the issue with 2019O5-20m persisted until mid-February 2020, with failures still being somewhat periodic at times, when the blanks slowly started to decrease. In late February, the problem with 2019O5-20m was also gone, when 2019O5-100m started to exhibit blanks, with the same pattern. It was first sporadic, then much more frequently. In late April, they became only sporadic again, with a very small fraction of blanks during the remaining buoy lifetime. The reason for this behavior remains unclear. On 14/15 May 2020, a drastic event took place. 2019O5-50m and 2019O5-100m showed a sudden 30 m pressure decrease on that day, suggesting that the entire buoy was lifted by that amount, presumably in a substantial ridging event. Communication to 2019O5-10m was lost in the process, as it was probably torn off the tether. At the same time, however, 2019O5-75m started to transmit data again, at a current depth of ∼ 45 m (as expected as the event lifted the instrument from a nominal depth of 75 m). 2019O5-20m was only lifted by 16 m, to a depth of 4 m, suggesting that it was pushed down along the tether by the resistance of the ice bottom. Surprisingly, the sensor survived this incident. The buoy kept collecting reasonable data via the four remaining CTDs (2019O5-20m presumably directly from the ice base), and, interestingly, without any further inductive modem issues. There were two more events, on 13 and 16 June 2020, during which three of the four instruments suddenly moved down 1 m each, except for 2019O5-20m, which was then at 3.8 m and presumably stuck in the ice. Another event during which the buoy was lifted by 4 m occurred on 12 July 2020, shortly before the buoy stopped transmitting any more data.

Buoy 2019O6 (Fig. d) was generally performing well but suffered from an issue with the inductive modem link as well. 2019O6-10m had some minor conductivity cell issues in the first few days after deployment. Inductive modem issues with all instruments on the tether became worse from November 2019 onwards but significantly improved again from 13 April 2020 for unknown reasons. These issues resulted in erroneous values being recorded by the buoy (in contrast to the problems with 2019O5, where measurements were just blank), leading to a large number of outliers in the raw dataset. On 3 August 2020, all CTDs exhibited a sudden increase in pressure of 0.5 dbar, probably related to a draining melt pond (although the submergence sensor was never triggered) or a similar effect that caused the surface buoy to slightly drop. The unit was recovered on 13 August 2020.

Similar to 2019O2, buoys 2019O7 and 2019O8 (Fig. c, d) were also fairly short-lived. 2019O7-75m suffered from conductivity cell issues upon deployment and did not recover. All instruments were lifted by 7 m during an event on 19 October 2019. The instruments were at stable depths for a few days before they were lifted again over a few days between 22 and 24 October. 2019O7-10m was lost eventually, and the buoy stopped transmitting entirely on 26 October 2019. There were three instruments on 2019O8 that had issues with the conductivity cell upon deployment, which lasted until the end of its (short) operational time. On 25 October 2019, all instruments were lifted by 10 m over a few hours, with 2019O8-10m being stuck at the ice bottom and pushed down the tether. A few hours after that, the buoy stopped transmitting.

In addition to a general data plausibility and consistency check among individual (independent) CTDs installed on the same tether as performed during the data processing, there are a few other methods to assess the general data reliability and quality. These will be elaborated on in this section.

Here we present a prominent example of the type of features observable with the present instrumentation and overall approach – a mesoscale eddy that passed through the buoy array in February 2020. During that time, the MOSAiC observatory drifted towards the northwest at an average velocity of about 0.13 ms-1. The number of loops was considerably less than during the previous November and December (Fig. a). The buoys drifted along a pronounced surface salinity gradient from less saline water in the southeast to more saline in the northwest (Fig. ). The lateral gradients of salinity and temperature are evident from both the time series along each buoy track and in the observations between different buoys.

Although our measurements are confined to discrete depths, the CTD records at 20 and 50 m can still indicate the minimum mixed-layer depth if they show equal temperature and salinity. Thereby, we infer from the records of 2019O4 and 2019O6 that the depth of the upper ocean mixed layer (ML) increased to >50 m from about mid-February onwards. This is in agreement with less frequent observations by an ocean profiler .

Changes in water properties with time at 20 and 100 m depths are shown in the temperature–salinity diagram in Fig. b. The mixed-layer salinity increased with a corresponding decrease in temperature along the freezing line during February. At the same time, salinity and temperature at 100 m decreased. An increase in mixed-layer salinity could be partly explained by a combination of the upward mixing of deep salty water from below and salt rejection during ice formation from above, both forced by two February storms with wind speeds up to 16 ms-1 and surface air temperatures of around -15 ∘C. However, close-by, near-daily measurements of the vertical mixing properties with a dedicated microstructure probe at the central observatory did not show significant mixing underneath the halocline (). Further mixed-layer deepening occurred into early April (Fig. ) when the ML depth reached 100 m. Between February and April, the drift path of the buoys was about 340 km, equivalent to a 270 km straight line distance. They drifted across the front associated with the transition from the Transpolar Drift to the warm and salty inflow of water of Atlantic origin through Fram Strait. These types of upper-ocean salinity gradients generally provide a high potential to form eddies.

During 7–10 February, rapid changes in the water properties were registered by 2019O4, 2019O6, and, to a lesser extent, 2019O1. 2019O6 (blue triangle and lines in Fig. ) and 2019O4 (red triangle and lines in Fig. ) registered these changes 4 and 1 d in advance of R/V Polarstern reaching the same location, respectively. The evolution of the observed salinity values shows a tendency for saltier waters appearing at depths shallower than ∼ 50 m and fresher waters deeper than 70 m. For example, the salinity decreased by 0.22 and 0.17 at 100 and 75 m, respectively, and increased by 0.5 at 50 m depth for buoy 2019O6. Temperature evolution is less conclusive since most measurements were within the range of the cold halocline layer. This feature indicates the presence of an eddy with the approximate geographical locations registered by the buoys as shown by two short lines (beginning and end) on the buoy tracks (Fig. a) and colored shaded areas in Fig. c and d, with buoy temperatures and salinities at four depths of 20, 50, 75, and 100 m. The CTDs at 10 m depth show mostly the same water properties as those at 20 m. Line colors correspond to the triangle colors of the buoys in Fig. a. This event was also observed by the current measurements obtained by the shipboard acoustic Doppler current profiler (ADCP). It shows the presence of the eddy between 11 and 13 February (Sandra Tippenhauer, personal communication, 2021), and the location is denoted in Fig. a (black dots on the dashed line). Most likely, the ship drifted close to the center of the eddy. It exhibited a rather symmetric current structure, with the minimum speed in the center and currents on the sides in opposite directions with speeds up to 20 ms-1. The estimated diameter of the eddy based on the buoy drift was about 30 km. A detailed description of the eddy properties is beyond the scope of this paper and is the subject of ongoing work.

It is important to note that the drift speed increased from 0.04 to 0.2 ms-1 between 7 and 14 February. Thereby, the time interval during which the eddy was observed by different buoys decreased, while the drifting distance (or eddy diameter) remained similar. At the same time, 2019O3 and 2019O5 do not show significant changes at a depth of 100 m. Both buoys were located to the northeast of the ship. Likely, they were too far away from the core of the eddy to register a measurable signal.

All three buoys that registered the eddy encountered similar anomalies in temperature and salinity. The salinity decreased by 0.22 and 0.17 at 100 and 75 m, respectively, and increased by 0.5 at 50 m depth, with a corresponding temperature decrease of 0.3, 0.12, and 0.05 ∘C for buoy 2019O6. We also calculated the bulk buoyancy frequency, N2=-gρ0dρ(z)dz, using measurements between 50 and 100 m. The result shows that N decreased during the passing of the eddy from ∼ 8 to 5.7 cph for 2019O6, from ∼ 7.2 to 5.1 cph for 201904, and from ∼ 6 to 4 cph for 2019O1. If we use the simplest way to calculate the first-mode baroclinic Rossby radius, using LR≡NHπf0 with a constant N=5 cph and H=4500 m, then LR is around 14 km (corresponding to a diameter of 28 km) and close to the estimated diameter of the eddy. The distributed nature of the buoys allows a fully synoptic assessment at different points across the array, which could not have been achieved by just one buoy.

The example in Sect.  shows how our data can be used to study mesoscale features in the upper Arctic Ocean. The wider scope of science questions that could be studied with the data is broad – the seasonality could be explored further, as could the variability along the pathways of the Arctic Transpolar Drift (e.g., ). In particular, the buoy data fill the winter gap common to manual in situ observations that are often carried out during seasonally limited ship surveys. The high temporal resolution allows for studying passing transient and submesoscale phenomena, as the buoys acted as a quasi-stationary platform, depending on drift speed. On the other hand, during times of faster drift, the systems measured the upper ocean quasi-synoptically over submesoscales.

This allows us to observe small-scale spatiotemporal variability in temperature and salinity that may be overlooked by traditional platforms. Measurements along a fixed isobath on a scale of hours correspond to a spatial resolution of ∼ 10 km (e.g., like those collected from ships or ITPs) and normally resolve mesoscale features. This may be illustrated by subsampling our buoy data at an interval of 60 and 360 min (Fig. ), revealing a near-surface frontal structure in the T-S diagram with warmer (colder) and fresher (saltier) water at either side of the front. Note that the nominal capacity of our buoys allows <10 min (<∼1 km) resolution data; thus, the buoys further capture phenomena hidden in this frontal structure. For example, there appears to be a density compensation process with warmer, saline waters compensating those with lower temperature and salinity (see black arrows in Fig. a). Another nearby buoy sees the same large-scale front but may additionally observe a local thermohaline intrusion within the front (see black arrow in Fig. b).

The significance of submesoscale circulation has been shown in numerous studies, highlighting the associated strong vertical velocity , restratification processes , conversion of potential to eddy kinetic energy , and feedback between ice and ocean currents around filaments in the MIZ . These physical processes can significantly influence biogeochemical conditions and the ecosystem . Yet, the submesoscale is largely limited to regional models or idealized process simulations (e.g., ) and largely out of reach for state-of-the-art numerical ocean general circulation models, thus necessitating appropriate parameterizations (e.g., ).

Analyzing our dataset further could make use of horizontal wavenumber spectra that can give insights into the length scales of variability induced by submesoscale restratification and internal waves . Furthermore, submesoscale eddies and filaments between mesoscale features can be studied with this piecewise quasi-synoptic dataset.

The scientific results of these potential studies, and the data, could serve to validate numerical models of the ice–ocean system and the fully coupled Earth system. The observations can also serve as input to numerical models, either to initialize or force a simulation or to provide input for assimilation. Finally, the GPS data recorded by our buoys were also used to create the official drift trajectories for several of the main sites of the DN .