The MOSAiC Drift: Ice conditions from space and comparison with previous years

We combine satellite data products to provide a first and general overview of the sea-ice conditions along the MOSAiC drift 25 and a comparison with previous years. We find that the MOSAiC drift was around 25% faster than the climatological mean drift, as a consequence of large-scale low-pressure anomalies prevailing around the Barents-Kara-Laptev Sea region between January and March. In winter (October April), satellite observations show that the sea-ice in the vicinity of the Central Observatory (CO) was rather thin compared to the previous years along the same trajectory. Unlike ice thickness, satellitederived sea-ice concentration, lead frequency, and snow thickness during winter month were close to the long-term mean with 30 little variability. With the onset of spring and decreasing distance to Fram Strait, variability in ice concentration and lead activity increased. In addition, frequency and strength of deformation events (divergence and shear) were higher during summer than during winter. Overall, we find that sea-ice conditions observed close (~ 5 km) to the CO are representative for the wider (50 km and 100 km) surroundings. An exception is the ice thickness: Here we find that sea-ice near the CO (50 km radius) was 4% thinner than sea-ice within a 100 km radius. Moreover, satellite acquisitions indicate that the formation of 35 large melt ponds began earlier on the MOSAiC floe than on neighbouring floes. Christian Katlein, Xiangshan Tian-Kunze, Robert Ricker, Philip Rostosky, Janna Rueckert, Suman Singha, Julia Sokolova https://doi.org/10.5194/tc-2021-80 Preprint. Discussion started: 8 April 2021 c © Author(s) 2021. CC BY 4.0 License.

introduced. Where possible, we compare data in the full resolution of the respective satellite with mean values formed over a 50 and 100 km radius (Fig. 1, buffer).

Lagrangian sea-ice tracking
To investigate whether the 2019/2020 drift was comparable to previous years, we made use of satellite sea-ice motion data to 85 reconstruct the pathways the ship would have taken if the experiment had started in one of the previous 14 years (October 2005(October -2018 instead. The satellite-based sea-ice pathways were determined with a drift analysis system called IceTrack. The system traces sea-ice forward in time using a combination of satellite-derived, low-resolution drift products (Krumpen et al., 2019, Belter et al., 2020. In summary, IceTrack uses a combination of three different ice drift products for the tracking of sea-ice: i) motion estimates based on a combination of scatterometer and radiometer data provided by the Center 90 for Satellite Exploitation and Research (CERSAT, Girard-Ardhuin and Ezraty, 2012, 62.5 x 62.5 km grid spacing), ii) the OSI-405-c motion product from the Ocean and Sea Ice Satellite Application Facility (OSISAF, Lavergne, 2016, 62.5 x 62.5 km grid spacing), and iii) Polar Pathfinder Daily Motion Vectors (v.4) from the National Snow and Ice Data Center (NSIDC, Tschudi et al., 2016, 25 x 25 km grid spacing). The IceTrack algorithm first checks for the availability of CERSAT motion data, since CERSAT provides the most consistent time series of motion vectors starting from 1991 to present and has shown 95 good performance (Rozman et al., 2011;Krumpen et al., 2013). During summer months (June-July) when drift estimates from CERSAT are missing, motion information is bridged with the OSISAF product (2012 to present). Prior 2012, or if no valid OSISAF motion vector is available within the search range, NSIDC data are applied. The reconstruction of "virtual" floes for these 14 years works as follows: Sea-ice is traced forward in time on a daily basis starting on October 4 (1996 -2019) until July 31 (303 days). Tracking is discontinued if ice concentration at a specific location along the trajectory drops below 50%, 100 which the algorithm defines as the position where the ice melted.
To assess the accuracy of this Lagrangian tracking approach, Krumpen et al. (2019) reconstructed the pathways of 56 GPS buoys deployed between 2011 and 2016 in the central Arctic Ocean. The displacement between real and virtual tracks is approximately 36 ± 20 km after 200 days and considered to be in an acceptable range. To number the accuracy of IceTrack in 2019/2020, we reconstruct the drift of the CO and 23 additional DN buoys (Krumpen and Sokolov, 2020). A comparison of 105 Fig. 2a with 2b (Supplement) shows that the reconstructed drift of the CO is in close agreement with the observed drift.
However, when the CO entered Fram Strait (red box), the reconstructed track lags behind the real one. The limited performance of IceTrack in Fram Strait is the result of a general underestimation of drift speeds by low resolution satellite products in this area (Sumata et al., 2014). It becomes particularly evident when looking at the reconstructed drift of the additional 23 DN buoys deployed in the vicinity of the CO (Supplement S.1c). Within the first 200 days, the reconstructed DN trajectories 110 deviate only slightly from real tracks (28 ± 15 km after 200 days), but once the DN reaches Fram Strait (south of 82.5°N, after 250 days), the distance between real and reconstructed pathways is gradually increasing. The comparison of the CO drift with the drift of the previous 14 years is therefore limited to the first 250 days.
ice concentration uncertainties can be significantly larger (up to 25%; Spreen et al., 2008). However, these are uncertainties of individual grid cells and mean biases for the averaged 50 and 100 km radii are lower. In summer or during warm air intrusions 125 sea-ice underestimation due to wettened ice surfaces, ice lenses or higher liquid water content in the snow, or melt ponds might occur. Such a period is observed during MOSAiC from mid-April to May 2020 and discussed below. During that time period, we show for comparison sea-ice concentration from an inverse multi-parameter retrieval using optimal estimation (Scarlat et al., 2020). The spatial resolution of this dataset is approximately 40 km. During and following the warm air intrusion and associated rain on snow it shows more correct ice concentrations but is yet not available for the previous years and thus cannot 130 be used as primary dataset here.

Sea-ice thickness
Sea-ice thickness (SIT) along the MOSAiC drift track during the Arctic winter season from October 2019 through April 2020 is analysed using two satellite remote sensing data sets. The first data set is based on radar altimeter data from the CryoSat-2 135 (CS2) mission of the European Space Agency (ESA). We use SIT retrievals generated at the full resolution of the altimeter with an approximate point spacing of 300 m and swath width of 1650 m along the ground-track of the satellite. The method of the SIT retrieval for each radar waveform is based on Ricker et al. (2014) with updates described in Hendricks et al. (2020).
The data set is named the AWI CryoSat-2 sea-ice thickness product version 2.3 and it is accessible through the website meereisportal. de (last access: February 15, 2021). In this study, we specifically use the Level-2 pre-processed (l2p) product, 140 which contains a daily collection of all data points along an orbit with valid freeboard data between October 1, 2019 and April 30, 2020. Subsets of all orbits within a day are generated based on their distance to the noon (UTC) position of the CO. For each subset we compute the mean SIT, the interquartile (IQR) and interdecile (ICR) SIT range, as well as the number of data points in each daily subset. According to the study logic, the search radius for the SIT subsets is chosen as 50 and 100 km and we only use individual orbits that provide at least 50 data points. Hence, we do not show data from a smaller (e.g. 5 km) search 145 radius, as very few orbits were close enough to the CO.
The second data set used for the SIT estimation is the merged CryoSat-2/SMOS (CS2SMOS, version 203) SIT product (Ricker et al., 2017). CS2SMOS provides gridded SIT data at a resolution of 25 km, which is significantly lower than the CS2 l2p data, however the underlying optimal interpolation provides gapless SIT information, also north of the CS2 orbit limit of 88°N.
Each daily updated CS2SMOS SIT field is based on an observation period of 7 days and we use the centre of this period as 150 the reference time to subset SIT data around the CO position at the selected radii. CS2SMOS data are based on CS2 l2p and Soil Moisture and Ocean Salinity (SMOS) SIT data. The SMOS retrieval provides thickness information of thin sea sea-ice, which complements the CS2 l2p data. The data merging uses a background field extending two weeks before and after the observation period, thus the temporal coverage is shorter than that of the CS2 l2p data and ranges from October 18, 2019 to April 12, 2020. In addition, the selection of SIT observations in the CS2SMOS data may vary from the CS2 l2p regional 155 coverage as we use the grid cell centre positions within 50 km and 100 km radius around the CO to compute the daily mean CS2SMOS SIT value. Thus, the number of CS2SMOS SIT observations selected may depend on the position of the CO relative to local grid cell coordinates. However, we do not expect this to cause a selection bias due to the smoothness of the CS2SMOS SIT data and the size of the search area.

Snow depth
Low resolution snow depth along the MOSAiC trajectory is retrieved from the 7 and 19 GHz channels of the AMSR2 microwave radiometer following the method from Rostosky et al. (2018) and is available via https://seaice.unibremen.de/data/amsr2/SnowDepth/ (last access: February 15, 2021). Following Rostosky et al. (2020), uncertainties are based on Monte-Carlo simulations using varying input parameters for a snow and sea-ice (MEMLS, Tonboe et al., 2006) Mech et al. 2020) microwave emission model. Most sea-ice, snow, and atmosphere properties are not known to the satellite snow depth retrieval (only information about the ice type, multi-year or first-year, is provided). Thus, by varying these properties and evaluating the influence on the snow depth retrieval, an estimate of the uncertainty caused by their unknown state can be obtained. The grid size of the snow depth data is 25 km. Snow depth currently can only be retrieved for multi-year ice areas in March and April (see Rostosky et al., 2018). As the MOSAiC ice floe is in an area of predominantly 170 second-year ice, which radiometrically is considered multi-year ice for the snow depth retrieval, also snow depth for MOSAiC is only available for March and April. Here we present snow depth data at 12.5 km radius around the CO in addition to a 50 km and 100 km average. A comparison with previous years is made with snow depth data extracted along the MOSAiC drift from 2005 until 2019.

Lead detection based on optical data
Sea-ice leads, i.e. lead frequencies and lead fractions along the MOSAiC drift track are derived from Moderate-Resolution Imaging Spectroradiometer (MODIS) thermal infrared data and the Collection 6 of MYD/MOD29 ice surface temperatures (Hall and Riggs, 2019). In order to detect whether a lead is present in a certain pixel we employ the local surface temperature anomaly, which is expected to exhibit significant positive deviations when a lead is present during winter (November to April). 180 This general procedure is followed by the application of a fuzzy inference system that assigns individual retrieval uncertainties to each detected lead pixel. Using this approach, we obtain daily categorical lead maps with separate classes for clouds, seaice, leads, and artefacts, with the latter comprising detected leads with an uncertainty exceeding 30%. The full approach and the resulting products are described in Reiser et al. (2020). From this data set we use daily lead data for the months of November Below lead data at 10 km resolution is compared with mean values formed over a 50 and 100 km radius. Note that days with a cloud fraction above 50 % are excluded from the analysis.

Sea-ice deformation from high resolution radar images
In this study, we quantify sea-ice deformation based on sequential Synthetic Aperture Radar (SAR) scenes obtained by ESA's 190 Sentinel-1A/B satellites along the drift track of the CO. Deformation is the consequence of divergence (opening), convergence (closing) and shear (sliding alongside) between ice floes. Regularly gridded sea-ice drift and deformation fields with a spatial resolution of 1.4 km are retrieved following the method described in von Albedyll et al. (2020). More details about the drift algorithm are provided in Thomas et al. (2008), and Hollands and Dierking (2011. As input for the applied algorithm, we use HH-polarized scenes with a spatial resolution of 50 m. Images over the CO are taken during the entire MOSAiC drift,195 except for the period between January 14 and March 15, when the ship was outside the satellite coverage. The temporal resolution is typically one image per day (with few exceptions). Spatial derivates are calculated from the gridded velocity field and used to derive convergence, divergence and shear (see von Albedyll et al. (2020) for details). To quantify deformation in the vicinity of the CO, we average all grid cells located within a 5 km radius around the ship. To compare deformation in the vicinity of the ship with deformation over a larger area (50 km), averages are computed for 61 x 5 km circles arranged within 200 a radius of 50 km around the ship (see illustration in Fig. 3). In this way we avoid biases due to scaling effects. Exceptionally strong deformation events are defined as events with a magnitude exceeding two standard deviations.

Characterization of melt pond coverage using optical Sentinel-2 data
To provide a first quantification of the spatial distribution and temporal development of large melt ponds on the MOSAiC floe, 205 we downloaded all available Sentinel-2 (S2, ESA) satellite images (https://scihub.copernicus.eu/dhus/, last access: February 10, 2021) taken over the ship between end of May and July 31, 2020. Prior the of May, the sun elevation was not high enough for passive optical remote sensing. A total of eight completely or partially cloud-free scenes could be identified. For the https://doi.org/10.5194/tc-2021-80 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. detection of melt ponds, we selected five scenes that are temporally equally spaced, namely June 21, July 1, July 7, July 22, and July 27, 2020. Next, the MOSAiC floe was clipped, and a pond index was calculated by means of a normalized spectral 210 index (e.g. Gignac et al., 2017, Watson et al., 2018 using S2 bands 4 (665 nm) and 8 (842 nm) as input. The pond index is used to differentiate between water and ice/snow. Note that only ponds larger than the spatial resolution of the S2 sensor (10 m) can be detected. We therefore assume that the actual pond cover is significantly underestimated, and that the method is only suitable for providing estimates of the timing and relative changes in pond coverage.

Atmospheric conditions and the Transpolar Drift in 2019/2020 225
Large-scale surface pressure and associated anomalies in 10 m wind speed (shown in Fig. 4) determined the course of the MOSAiC drift and its deviation from the long-term average. October, November, and December were characterized by moderate monthly mean circulation anomalies, oriented mostly such that the winds (and thus the drift) were westward rather than northward, thereby "deflecting" the MOSAiC drift from the North Pole. Starting in January, large-scale low-pressure Siberia and in the Kara and Laptev Seas, in particular in February (up to +10°C, not shown). In contrast, temperature anomalies at the MOSAiC floe were rather moderate most of the time (Fig. 6 middle). Moderate warmer-than-average periods occurred in mid-November, late February, mid-April, and late May, whereas colder-than-average periods occurred in early November 240 and early March, with record-cold temperatures around -35°C. Wintertime cold (warm) anomalies were typically associated with high (low) surface pressure anomalies (Fig. 6 bottom). The positive AO months January, February, March, and April were accompanied by low pressure anomalies also at the MOSAiC floe ( Fig. 6 bottom). Stormy conditions were encountered in particular in these months but also in late November/early December (Fig. 6 top). Apart from these exceptions, meteorological conditions at the MOSAiC floe can be considered relatively normal compared to previous years. 245 The ERA5 data along the MOSAiC trajectory in 2019/2020 agree well with co-located ship observations (Fig. 7), in particular regarding surface pressure. Wind speed tends to be slightly underestimated in ERA5, although it should be noted that the comparison with the raw on-board observations has limitations. However, the winter warm bias in ERA5 over Arctic sea ice of the order of 2-3°C (Fig.7) is consistent with previous assessments (e.g., Batrak and Müller 2019), especially taking into account that the ship temperatures are measured at 29 m with typically higher temperatures than the 2 m ERA5 reference 250 https://doi.org/10.5194/tc-2021-80 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. temperature. Given that these differences are likely systematic and thus similar in other years, the anomalies discussed above are likely not strongly affected.

The MOSAiC drift and a comparison to previous years
We  Figure 9. The average ice concentration between October 4, 2019 to July 31, 2020 amounts to 97%. The seasonal evolution is characterized by a substantial temporal variability over the course of the 303-day long drift. This variability is almost independent of the spatial scale used with only minor differences (±0.5 % deviation from mean) between the ice concentration values determined from the 3, 50, and 100 km radius (Fig. 10). 275 Given the high agreement between the different radii, will limit ourselves in the following to the discussion of the time series with the highest resolution (3 km radius, Fig. 9). From October to July the ice concentration along the MOSAiC drift trajectory agrees well with the long-term 2005/2006 to 2019/2020 average. However, during the first half of the drift (October until end of February) the ice concentration was with 99.5% about 1% higher than the long-term average, while during the second half (March until end of July), it was lower than the long-term average and shows higher variability than the first half. 280 Sea-ice concentration variability stayed below 5% until March 2020, when first significant reductions in ice concentration occurred. At this time, the CO was already positioned north of Fram Strait and the distance to the ice edge was gradually decreasing (compare Fig. 11). With the onset of spring in March/April first major drops in ice concentration below 90% occurred. The strong ice concentration reduction down to 75% from mid-April until beginning of May are due to a false satellite ice concentration retrieval. Observations from the ship confirm that the ice concentration stayed higher during that 285 time period. We can see that at that time a warm air intrusion raised temperatures close to 0°C, which was accompanied by a significant increase in wind speed (Fig. 6). Already below 0°C liquid water content in the snow increases and later refreeze after the warming event can cause ice lenses in the snow. On the floe also refreezing rain on snow was observed. These surface processes and additional weather influence by high water vapour and cloud liquid water affect the microwave polarization difference and likely caused the unnatural fluctuation in ice concentration for the ASI algorithm used here. Other ice 290 concentration algorithms for AMSR2 satellite data (e.g. NASA-Team) showed similar effects (not shown). As an alternative we present the ice concentration from an optimal estimation retrieval (Scarlat et al., 2020) during that critical time period in Figure 9, which takes such effects into account and is in better agreement with ship observations. After mid-May the ice https://doi.org/10.5194/tc-2021-80 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. concentration recovers to almost 100%. In July, the floe started to disintegrate and ice concentration dropped to 85% within a radius of 3 km around Polarstern, and below 60% in the 50 and 100 km radii (Fig. 10). 295 We determine the closest distance to the ice edge from sea-ice concentration maps (Fig. 11). At the beginning of the MOSAiC expedition, the distance from the CO to the ice edge was about 320 km. During October the distance gradually increased to 1000 km due to the freeze-up of the Russian marginal seas. Once the MOSAiC CO approached Fram Strait (March 2020) the distance to the ice edge steadily decreased until the ice margin was reached at the end of July 2020. Note that the winter variability in ice edge distance is caused by polynya activity on the Russian shelf seas. 300

Sea-ice thickness
Both satellite-based sea-ice thickness products show the expected increase in ice thickness between October 2019 and April 2020 ( Fig. 12 and Table 1). Except for the period between February 14 to March 8, when the CO was positioned north of 88°N, the high orbit density of CS2 allows almost continuous daily coverage at 50 and 100 km radius. The monthly mean 305 thickness within a 50 km (100 km) radius around the CO changed from 0.77 m (0.8 m) in October 2019 to 2.40 m (2.51 m) in April, 2020. The sea-ice thickness distribution is characterized with the interquartile range (IQR) as difference between 75% and 25% percentile and the interdecile range (IDR) as difference between 90% and 10% percentile. The increase in sea-ice thickness was accompanied by a similarly increased IQR and IDR, indicating a wider sea-ice thickness distribution as a result of thermodynamic ice growth and deformation of the older ice class and the formation of young ice throughout the winter 310 season. It is notable that the CS2 L2P sea-ice thickness was consistently thinner at the 50 km radius compared to the 100 km radius (Tab. 1, on average 6 cm (4 %) between October and April). Similarly, IQR and IDR were larger for 100 km than for 50 km, however the larger number of data points in the wider search area may also lead to a higher likelihood of diverse seaice conditions. This is in agreement with findings of Krumpen et al. (2020). According to the authors, the MOSAiC DN was set up at a regional thickness minimum. The local minimum is related to the ice age: Sea ice in the DN was formed three weeks 315 later than the surrounding ice. However, the authors report even larger differences in sea-ice thickness of 36% between the DN area and areas further away.
Results from CS2SMOS mirror these findings of thinner ice close to the CO compared to the larger scale, though the difference are smaller (Fig. 12, Table 1). This can be expected, as the primary input to the CS2SMOS analysis in the central Arctic is CS2 data due to its higher sensitivity to thicker ice than SMOS. The main differences to CS2 L2P are therefore the influence 320 of SMOS in the beginning of the winter and the larger degree of smoothing introduced by the optimal interpolation. The monthly mean sea-ice thickness values in Table 1 are therefore mainly consistent with the exception of the October and November 2019. In this period, CS2SMOS was consistently higher by approximately 0.15m with respect to the CS2 L2P data.
We do not expect that the locally lower thicknesses in the DN are well represented in the CS2SMOS thicknesses, since these are influenced by a larger region due to the interpolation method. The CS2 L2P thicknesses instead are effectively point 325 measurements at kilometre scale and are apparently able to pick up the local thickness gradient. The discrepancy between the CS2 L2P and CS2SMOS thicknesses persisted well into November 2019 and became less prominent afterwards. This provides evidence that the local thickness minimum at the MOSAiC DN became less prominent over the winter season, though still at a detectable level as indicated by the consistently but minor differences at radii of 50 and 100 km.
Since CS2 L2P and CS2SMOS are in general consistent over the winter season, we use CS2SMOS data to compare sea-ice 330 conditions during the MOSAiC drift with the past nine winter seasons in the CS2SMOS data record (Fig. 13). The comparison between the years shows a comparably low sea-ice thickness in the 10-year long data record at the location of the MOSAiC experiment, if not the lowest for segments in the earlier part of the drift. The monthly sea-ice thickness during MOSAiC was approximately 0.4 m lower at the beginning of the drift compared to mean monthly CS2SMOS of all previous winters ( Table   1). The differences reduced towards 0.3 m in April, indicating slightly stronger thermodynamic and dynamic ice growth with 335 respect to the average, potentially aided by the thinner sea-ice in the beginning. Again, the sea-ice thickness difference is https://doi.org/10.5194/tc-2021-80 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. stronger for the 50 km radius than the 100 km radius indicating that the MOSAiC experiment took place in a local sea-ice thickness minimum. The snow depth during MOSAiC was a few centimetres lower but overall, quite average from a satellite microwave radiometer perspective. The time series in Figure 14 shows that the snow depth stayed almost constant from beginning of March until mid-April. Only after the warm air intrusion in April (Fig. 6), increased precipitation led to a small increase in snow depth of about 3 cm. After the initial temperature increase snow depth reduced again, i.e., snow compacted. Mind that the microwave radiometer satellite snow depth retrieval does not react immediately to new snow, which has low emissivity and little scattering. 350

Snow depth 340
The added snow usually shows up after the first few days of snow compaction, which is in line with the in-situ observation from the floe, where snow depth only increased after April 20 while the air temperatures already increased about five days earlier.

Leads 355
Lead frequencies and lead fractions are derived for the surroundings of the MOSAiC CO and the reference period using the method described in Reiser et al. (2020). The lead frequency is a temporally integrated quantity indicating how often a lead is This agrees well with consistently high ice concentration values indicated by the sea-ice concentration climatology during the first half of the expedition (Fig. 9). According to the climatology, higher lead frequencies (> 0.15) in winter are only to be expected near the ice edge and in Fram Strait. The lead frequency anomalies for the MOSAiC year 2019/2020 shown in Fig.   15b indicate no significant deviations from the winter mean climatology. On average, anomalies were slightly negative along the MOSAiC drift trajectory and in the sector between 30°W and 120°E, which again agrees well with the observed slightly 365 higher ice concentration values as compared to the long-term mean (Fig. 9).
Regional differences in lead frequencies can be inferred from monthly lead anomaly maps shown in Fig. 15c-h: The monthly maps reveal anomalously high lead frequencies north of Greenland and Ellesmere Island between November and January.
Moreover, the strong positive anomalies in the Barents Sea in January and in the Beaufort Sea in February/March are worth mentioning (compare Dethloff et al. 2021). However, in the proximity of the CO no significant lead anomalies are found 370 between November and February. Only in March, when the CO was crossing a region of East-West oriented leads, slightly higher anomaly values of up to 0.1 are indicated. In April, leads around the MOSAiC CO were more North-South oriented and strengthened as expressed by higher anomaly values of up to 0.2 (Fig. 15h).
A detailed view on the temporal evolution of lead fractions along the MOSAiC drift trajectory on different radii is presented in Figure 16. Lead fraction is shown for the area around the MOSAiC CO with 10 km, 50 km and 100 km radius, respectively, 375 together with the mean and maximum lead fraction for the reference period and one standard deviation. The mean lead fraction for the area around the CO was slightly increasing towards the end of winter for all of the three ranges shown, which confirms the drift into a region with generally higher average lead frequencies starting in March (Fig. 16a). In general, lead dynamics around the MOSAiC CO were typical for the respective region and point in time with only a few temporally limited, but https://doi.org/10.5194/tc-2021-80 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. significant deviations from the mean. However, meaningful conclusions can only be drawn for the periods in which the cloud 380 fraction for the respective radii was below 50%. Note that in Figure 16 days with missing data and higher cloud fractions are indicated by red dots. A maximum in lead activity was observed on March 4 (at all radii). Several smaller events with lead fractions exceeding one standard deviation from the reference period were recorded on December 11 -12, January 19, January 28, February 1, February 4 -8, March 1 -5, March 11, and April 23 -24 (for 50 km radius, Fig. 16a.). It is striking that these events were only conditionally accompanied by a decrease in ice concentration. 385 Figure 17 shows the time series of divergence and shear rates along the MOSAiC drift track at 5 km and 50 km radii as obtained from Sentinel-1 SAR data. Overall, we find that deformation close to the ship (5 km radius), was representative for the deformation acting on the ice cover at larger distances (up to 50 km). Moreover, we find that mean shear and divergence of 8 390 % d -1 and 2 % d -1 along the MOSAiC drift track are in good agreement with deformation rates obtained from a ship-radar North of Svalbard by Oikkonen et al. (2017).

Sea-ice deformation
The variability of divergence and shear showed a seasonal behavior which is linked to the consolidation of the ice pack and in agreement with findings of previous studies (e.g., Itkin et al., 2017, Hutchings et al., 2011. Monthly averages of the time series indicates that deformation was moderate and balanced in convergence and divergence in the consolidation phase between 395 October and November (Fig. 17). Hereafter, divergence and shear temporarily decreased from December to January. In March to May, divergence and shear went back to a moderate level until a sudden increase in June and July was observed when the MOSAiC CO approached the marginal ice zone. Note that monthly averaged divergence correlates reasonably well with intensified lead activity observed by optical satellites (Sect. 3.6). In spring (March, April), the ice experienced more divergent than convergent motion, which again agrees well with intensified lead activity observed in spring (Fig. 15/16). 400 On daily time scales, divergence and shear were characterized by long quiet phases occasionally interrupted by strong deformation events (video supplement). The average temporal spacing between such deformation events was 2.5 weeks.
However, the events were not uniformly distributed in time, as 60 % of the events took place between October and November (grey bars in Fig. 17). The strongest deformation event within the 50 km radius of the CO was observed on April 14-17, 2020.
By that time, a lead of almost 2.5 km width opened up at 25 km distance of the CO (Fig. 3). 405 Figure 18 presents five cloud-free S2 scenes obtained between June 21 and July 27 that provide a first overview of the temporal and spatial evolution of melt ponds on the MOSAiC floe and its extended surroundings. Melt pond coverage is characterized using the pond index described in Section 2.7, where high values indicate water and low values ice/snow. 410

Melt pond distribution
One of the most striking features is that at the time when the first cloud-free scene (July 21) was taken, large melt ponds had already developed on the MOSAiC floe. The earlier start of melt pond formation on the MOSAiC floe as compared to the extended surrounding is likely related to the surface topography. Compared to the surrounding floes, the MOSAiC floe was characterized by heavily deformed areas which may have favoured early accumulation of large meltwater ponds. Another possible reason for the early onset of melting may be the high quantity of sediments that were trapped in the ice (Krumpen et 415 al. 2020). The high sediment content temporarily reduced the surface albedo of the floe, which may have favoured early melt of ice.
Within the next ten days, the proportion of large melt ponds on the MOSAiC floe increased considerably, and large ponds also began to form on the neighbouring floes. On July 7, while the total amount of melt ponds was still increasing, a few large melt ponds began to drain. In the final scene (July 27), taken almost three weeks after the draining began and just before the floe 420 was abandoned, large melt ponds had mostly split into smaller ponds and had partially disappeared as a result of several drainage events that were observed in field between July 1 and 27. The (absolute) quantification of melt pond fraction is https://doi.org/10.5194/tc-2021-80 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. limited, as the typical size of the melt ponds observed on the ground were equal to or smaller than the pixel size of the S2 image.

Conclusion
Below we summarize the ice conditions along the drift of the MOSAiC floe and the extended surroundings and compare them to previous years. The analysis is based on satellite data products commonly used for the scientific analysis of sea-ice in the Arctic Ocean. An overview of the atmospheric and sea-ice conditions observed along track is given in Fig. 19.
 A comparison of the MOSAiC trajectory with reconstructed satellite-based pathways for the past 14 years indicates 430 that the drift during the first 250 days of the expedition was around 25% faster than the climatological mean drift.
Deviations from a long-term average drift path are to a large extent the consequence of prevailing large-scale lowpressure anomalies, which resulted in an intensification of the Transpolar Drift between January and March 2020.
 CS2 and CS2SMOS data records show that the mean thickness of sea-ice around the CO (50 km radius) evolved from 0.77 m in October 2019 to 2.40 m in April 2020. Sea-ice near the CO (50 km radius) was thereby 4% thinner as 435 compared to surrounding sea-ice (100 km radius). According to Krumpen et al. (2020), the negative anomaly is due to the younger ice age, as the ice around the CO was formed in a different region and later in the year than the surrounding ice. A comparison with CS2SMOS records from the past nine winters shows that the ice around the MOSAiC CO was comparatively thin (partially the thinnest). In October it was 0.4 m and in April 0.3 m below the nine year average. 440  Unlike the ice thickness, the snow thickness does not differ much from the long-term mean. Data from satellite-based microwave radiometers indicate an average March/April snow depth of 22 cm (25 km radius). This is 3 cm lower than the long-term mean for the years 2006 to 2019.
 From the start of the expedition until April, the average ice concentration within the 50 km radius of the CO was slightly higher (1%) than the long-term mean with low variability. Significant changes in the ice cover occurred only 445 in April and May, as a result of a positive temperature anomaly. Another significant drop in ice concentration took place at the end of the first expedition phase when the floe approached the ice edge (July).
 An analysis of winter (October -April) lead frequencies inferred from MODIS thermal infrared data indicates no significant deviation in lead activity from the mean climatology (2005/2006 -2018/2019). At most, a slight negative deviation from the winter mean is discernible, which agrees well with the positive anomaly in ice concentration 450 between October and April. It is interesting to note that with increasing variability in ice concentration from March onwards, lead activity increased.  A deformation time series derived from Sentinel-1 data gives first insights into divergence and shear events along the MOSAiC drift. Overall, we find that sea-ice deformation on the 5 km radius including the MOSAiC CO was representative for the wider (50 km radius) surroundings. Deformation rates were lower during winter, and higher 455 during summer, which is in agreement with observations from previous studies. The dominance of divergence during spring and summer agrees well with the observed higher lead fractions.     timeseries are smoothed with a 5-day running mean. During spring warm air intrusions with rain on snow caused a significant temporary reduction of the sea-ice concentration (dashed black line). We therefore show in addition an alternative sea-ice concentration data set during that time with uncertainty estimates (red and shaded red; not available for the climatology, see main text).