Interannual variability in Transpolar Drift summer sea ice thickness and potential impact of Atlantiﬁcation

. Changes in Arctic sea ice thickness are the result of complex interactions of the dynamic and variable ice cover with atmosphere and ocean. Most of the sea ice exiting the Arctic Ocean does so through Fram Strait, which is why long-term measurements of ice thickness at the end of the Transpolar Drift provide insight into the integrated signals of thermodynamic and dynamic inﬂuences along the pathways of Arctic sea ice. We present an updated summer (July/August) time series of extensive ice thickness surveys carried out at the end of the Transpolar Drift between 2001 and 2020. Overall, we see a more 5 than 20% thinning of modal ice thickness since 2001. A comparison of this time series with ﬁrst preliminary results from the international Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) shows that the modal summer thickness of the MOSAiC ﬂoe and its wider vicinity are consistent with measurements from previous years at the end of the Transpolar Drift. By combining this unique time series with the Lagrangian sea ice tracking tool, ICETrack, and a simple thermodynamic sea ice growth model, we link the observed interannual ice thickness variability north of Fram Strait 10 to increased drift speeds along the Transpolar Drift and the consequential variations in sea ice age. We also show that the increased inﬂuence of upward-directed ocean heat ﬂux in the eastern marginal ice zones, termed Atlantiﬁcation, is not only responsible for sea ice thinning in and around the Laptev Sea, but also that the induced thickness anomalies persist beyond the Russian shelves and are potentially still measurable at the end of the Transpolar Drift after more than a year. With a tendency towards an even faster Transpolar Drift, winter sea ice growth will have less time to compensate the impact processes, such as 15 Atlantiﬁcation have on sea ice thickness in the eastern marginal ice zone, which will increasingly be felt in other parts of the sea ice covered Arctic.

. a) Map showing all EM-based summer (July/August/September) SIT measurements (dark blue circles) obtained between 2001 and 2020, as well as July/August mean sea ice concentration 25×25 km) for the period from 2000 to 2019 (Lavergne et al., 2019). Enclosed area (red, 80.5-86 • N and 30 • W-20 • E) indicates the selected area of interest (AOI, see Table 1 for an overview of the corresponding expeditions and basic SIT statistics for the selected AOI). The black line (86 -90 • N, 50 • E) shows the transect of Russian sea ice observations. b) Summer (July/August) mean (red circles) and modal (light blue circles) SIT based on EM measurements conducted in the AOI (dark blue circles in a)). Diamond shows modal EM SIT measured on the MOSAiC floe in July 2020 and filled circles indicate mean and modal EM SIT values obtained during the IceBird MOSAiC campaign in September 2020. Vertical lines indicate standard deviations of mean (red) and modal (light blue) SIT values of the individual profiles from each year. c) shows the fractional occurrence of first-year (FYI, white), second-year (SYI, grey), and multi-year ice (MYI, black, ice older than two years) for the individual years. The age classification is based on ICETrack calculations of the number of days the ice particles travelled along their trajectories. a number of years and the area covered during these campaigns (Fig. 1 a)) includes a wide range of different ice types from various sources, a careful analysis is required for the investigation of interannual SIT variability.
For the current study we focus on the SIT measurements from a selected area of interest (AOI, enclosed area in Fig. 1 a)) north of Fram Strait. Sea ice reaching Fram Strait originates from multiple regions of the Arctic, which means long-term 55 observations of SIT in its vicinity provide insight into integrated Arctic-wide thermodynamic and dynamic changes in the sea ice cover (Hansen et al., 2013). While previous studies recorded substantial thinning in Fram Strait in summer (Hansen et al., 2013;Spreen et al., 2020) and across Fram Strait (79 • N) SIT gradients in spring during the first decade of the 21st century (Renner et al., 2014), we focus on the evolution of summer (July/August) SIT further upstream of the Transpolar Drift. With the AEM being towed by a fixed-wing aircraft longer transects and ultimately a greater areal distribution of the measurements 60 are achieved as compared to other in situ observations. The objectives of this study are to extend the summer SIT time series (from 2012 to 2020), first published by , at the end of the Transpolar Drift and investigate the interannual variability in SIT in the selected AOI close to the export gate of Arctic sea ice. We will use the Lagrangian sea ice tracking tool, ICETrack (Krumpen, 2018) to determine the of the observed SIT variability a thermodynamic model is applied along the determined sea ice trajectories to reconstruct the AOI-sampled SIT. In addition we will compare the SIT changes in the AOI to long-term observations gathered during regular Russian cruises from Franz Josef Land to the North Pole. This additional comparison is conducted to discuss whether the observed changes are limited to the AOI or induced during ice formation and transit through the Arctic Ocean. Finally, we will use the unique opportunity to compare the long-term SIT time series to IceBird and GEM measurements conducted within the 70 framework of the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC). At 85 • N and 136 • E the German icebreaker RV Polarstern (Alfred Wegener Institute, 2017) moored to a 2.8 × 3.8 km sized ice floe in October 2019 . After about 9 months of drifting through the Arctic Ocean, RV Polarstern and the MOSAiC floe reached the selected AOI in summer 2020. This allows us to consider the MOSAiC floe in the context of the long-term time series, and determine whether the SIT of the MOSAiC floe in 2020 was exceptional or in agreement with historical observations. 75 2 Data and methods

EM sea ice thickness measurements
Electromagnetic induction (EM) SIT measurement systems take advantage of contrasting electrical conductivities between sea ice and sea water to determine the distance between the ice-water interface and the EM device (Kovacs and Morey, 1991;Haas et al., 1997). In 2001, measurements were conducted using a Geonics EM31 ground-based EM instrument (GEM). The GEM 80 was pulled over the ice on a sledge and obtained the distance to the ice-water interface (Haas, 2004). GEM measurements included SIT values over melt ponds and pressure ridges, however, open water and thin ice were not adequately represented in the data sets due to the practical limitations of sampling those areas on foot (Haas, 2004). The 2020 GEM measurements on the MOSAiC floe were taken with the Geophex GEM-2, a broadband EM sensor used for advanced thickness observations (Hunkeler et al., 2016). After 2001, SIT was obtained using the airborne EM system (AEM), EM Bird, that was towed by a 85 helicopter (in 2004) and the research aircrafts Polar 5 and 6 (from 2010 onwards). The EM Bird was operated between 12 and 20 m above the ice surface . Following Pfaffling et al. (2007), SIT was calculated as the difference between EM-derived distance to the ice-water interface and laser altimeter-recorded distance between EM device and the airsnow interface. EM measurement accuracy is within 0.1 m to drill-hole measurements over level sea ice, while water inclusions within pressure ridges lead to a general underestimation of ridges by as much as 40 to 50% (Pfaffling et al., 2007;Haas et al., 90 2009).
Thickness measurements using the ground-based and airborne EM methods always represent the total combined sea ice and snow thickness (Haas et al., 1997). Given the study period from mid-July to mid-August and following climatological values of snow depth (Warren et al., 1999;Renner et al., 2014;, we assume a 0.1 m snow or weathered layer thickness, which is negligible for the EM measurements. More snow may still have been present during episodic precipitation 95 events, but likely melted within a few days. In order to ensure comparability of the available EM-based measurements from 2001 to 2020 only data taken between 80.5 to 86 • N and 30 • W to 20 • E (AOI, Fig. 1 a)) were selected for the analysis. Following , the AOI was also selected to be north of Fram Strait to concentrate the analysis on sea ice that was shaped along its pathways through the Arctic rather than by local melt phenomena in Fram Strait. Finally the selected AOI allows for a more reliable analysis of 100 the trajectories of the sampled sea ice since low resolution sea ice motion products used for Lagrangian tracking are highly uncertain in Fram Strait (Krumpen et al., 2019). Expedition logistics and the prevailing weather conditions prevented us from acquiring continuous and overlapping measurements over the full AOI each year. However, following Rabenstein et al. (2010) the lengths of the conducted EM profiles were adequate to consider the data to be sufficiently homogeneous in time and space and representative for the sampled region and time of year. Table 1 provides an overview of all relevant field campaigns, 105 duration, profile lengths, basic statistics and references for the measurements from within the selected AOI.
The analysis of trends and interannual variability of summer SIT in the AOI is based on temporal and spatial averages and the most frequently occurring EM SIT -the mode of the distribution. Modal SIT is a representation for the thickness of thermodynamically grown level ice, while mean SIT includes thermodynamically and dynamically grown sea ice and therefore is an indication for the general variability of SIT (Haas, 2017). Prior to the calculation of summer mean and modal SIT 110 values from all available data points within the predefined AOI ( Fig. 1), SIT values < 0.1 m, including open water values, were excluded to avoid biases due to different fractions of open water areas in the data sets.

Sea ice pathways and source regions
In order to determine the pathways and source regions of the ice that was sampled in the selected AOI we utilized the Lagrangian ice tracking tool, ICETrack (Krumpen, 2018). The starting points for the backward tracking of AOI-sampled sea ice were Ice parcels were tracked backward in time on a daily basis. Termination criteria for the tracking were either met when the ice reached a coastline or when SIC dropped to 25% or less. When SIC reaches 25% or less ICETrack assumes that ice is formed.
The applied SIC product is provided by the Center for Satellite Exploitation and Research (CERSAT) and is based on 85 GHz SSM/I brightness temperatures, using the ARTIST Sea Ice algorithm (Ezraty et al., 2007). The number of days from the first 125 day of tracking until ice formation provided the age of the sea ice sampled in the AOI. The tracking was based on a weighted approach to determine the most appropriate of the three available low resolution sea ice motion products (Krumpen et al., 2019): (i) motion estimates from scatterometer and radiometer data from CERSAT (Girard-Ardhuin and Ezraty, 2012), (ii) the OSI-405-c motion product produced by the Ocean and Sea Ice Satellite Application Facility (OSISAF) (Lavergne, 2016) and (iii) Polar Pathfinder Daily Motion Vectors from the National Snow and Ice Data Center (NSIDC) (Tschudi et al., 2019).

130
CERSAT was prioritized as it provides the most consistent time series of motion vectors (from 1991 onwards). However, when CERSAT data were missing (especially during summer months), OSISAF data were used. Prior to 2012 or when OSISAF data were not available NSIDC data were utilized (Krumpen et al., 2019). A detailed description of the three motion products is given by Sumata et al. (2014). Beside sea ice trajectories, ICETrack provided information about satellite-derived SIT and sea ice concentration (SIC) as well as atmospheric parameters, like surface air temperature, 10 m wind speed and surface pressure in daily increments along the trajectories (Krumpen, 2018). Due to this comprehensive approach to analyse sea ice along its trajectories through the Arctic and its accuracy (Krumpen et al., 2019), ICETrack has been widely used in previous studies, e.g. Damm et al. (2018); Peeken et al. (2018).

Thermodynamic sea ice model
In order to investigate the driving mechanisms of interannual variability in modal SIT in the AOI, ICETrack was combined 140 with a simple one-dimensional thermodynamic sea ice model developed by Thorndike (1992). Parallel to retrieving SIC, SIT, atmospheric parameters, and sea ice motion from ICETrack, the model calculated daily sea ice growth and melt along the determined sea ice trajectories. Latent heat of melting/freezing, ocean heat flux, and conductive heat loss are balanced to model ice growth at the bottom, ∆H ∆t (Thorndike, 1992;Pfirman et al., 2004): The model used along-track snow depth, H snow , computed daily from the Warren climatology (Warren et al., 1999) as well as the NCEP/NCAR re-analysis sea surface temperature (T surf ) data (Kalnay et al., 1996) that were extracted along the trajectories by ICETrack (Krumpen et al., 2019). T 0 is the temperature at the ice-water interface (-1.9 • C) and k is the thermal conductivity (k ice = 2 Wm −1 K −1 , k snow = 0.33 Wm −1 K −1 ). Latent heat of fusion, L, was constant at 3 · 10 8 Jm −3 . ∆t equals 86400 s for daily increments of ice growth. Ocean heat flux, F , was assumed to be constant at 2 Wm −2 . Based on these input 150 parameters the model computed daily changes in SIT along the trajectories. When melt occurred (negative growth for any given day) the model reduces the thickness by an additional 0.005 m for that day to parametrise surface melt (Thorndike, 1992;Pfirman et al., 2004). Modelled SIT values at the end of each track were used to calculate AOI summer mean thermodynamic SIT for each year. This modelled value provides SIT excluding the snow layer that is inherently included in the EM SIT values. We therefore added the Warren snow depth value from the position and time of the relevant EM SIT measurements 155 to the final model SIT averages for the comparison to EM SIT. Like modal EM SIT, the modelled SIT is a representation of thermodynamically grown level sea ice. Snow depth is an important parameter in modelling sea ice growth, and due to the limitations of the Warren snow depth climatology (Warren et al., 1999), a major source for uncertainty (Merkouriadi et al., 2020) in the modelled SIT values calculated for each trajectory ending in the AOI. Following Laxon et al. (2013) and Ricker et al. (2014) we reduced Warren snow depth by 160 50% over FYI and based on comparisons of the Warren snow depth with snow buoy data also over SYI. This step accounts for the fact that Warren et al. (1999) snow depth is based on observations during a period where Arctic sea ice was dominated by MYI with thicker snow. Root mean square errors of Warren snow depth values (Warren et al., 1999) were utilised to calculated the model SIT equivalent of the snow depth errors and provide an estimate of uncertainty of the modelled SIT as a result of the applied Warren snow climatology.

165
Another major source of uncertainty of the modelled SIT is the selected ocean heat flux value. However, due to the simplicity of the selected sea ice model (Thorndike, 1992) and the lack of long-term and Arctic-wide data, for this current study the input of a constant ocean heat flux value was required. We followed previous studies (Maykut and Untersteiner, 1971;Pfirman et al., 2004;Peeken et al., 2018;Krumpen et al., 2019) and selected a constant ocean heat flux value of 2 W m −2 . This value was applied to the sea ice growth model along each trajectory from ice formation to the AOI. shipborne television complex (STK). The STK consists of a high resolution telecamera, a computer for camera control and processing, and a GPS recorder. The system records images of overturning sea ice floes in the vicinity of the moving ship as well as GPS time and coordinates. After manual selection of appropriate images the software is able to measure the detailed geometry of single ice blocks from the ice camera feed and retrieve ice and snow thickness data. The STK system provides SIT with an uncertainty of approximately 4% of the thickness of each floe (Frolov et al., 2007;AARI, 2011).The purpose of this 180 system is to provide navigation data for following ships and reliable SIT data for the validation of satellite-and model-derived SIT. Over the last decades the AARI conducted visual and STK observations regularly during summer (June-August) tourist cruises from Franz Josef Land to the North Pole (     Table 1).
To some degree the interannual modal thickness variability can be explained by the variability in source regions of the ice ( Fig. 2), its age and the number of days with surface temperatures below freezing (-1.9 • C, FDs) the ice experienced during the 205 transit through the Arctic Ocean (Fig. 3 a)).  It has to be noted that the varying number of EM surveys each year and the variation in areal coverage within the AOI of the different surveys makes the analysis of SIT trends challenging. However, large-scale and year-to-year overlapping surveys as well as sampling during the same season each year strengthen the assessment that sea ice sampled in the AOI is changing in thickness and age.

Reconstruction of observed SIT using a thermodynamic model 225
To further investigate the processes driving interannual variability of modal SIT we compare observed AOI values with modelled SIT values of thermodynamically grown sea ice. Sea ice growth along the ICETrack sea ice trajectories was calculated using the thermodynamic sea ice model by Thorndike (1992). The modelled SIT values at the end of the trajectories provide the AOI-mean modelled SIT for each year. Ocean heat flux is the main source of bottom melting (Lin and Zhao, 2019); it is a parameter that is widely debated and still being investigated. It is the sum of heat that enters the surface mixed layer from the deep ocean and heat that enters the surface mixed layer through leads and openings in the ice cover (Zhang et al., 2000;McPhee et al., 2003;Perovich et al., 2011;Wang 250 et al., 2016). Multiple existing studies have shown that Arctic ocean heat flux is highly variable in time and space (Maykut, 1982;Maykut and McPhee, 1995;McPhee et al., 2003;Krishfield and Perovich, 2005;Lin and Zhao, 2019). Nevertheless, the assumption of a constant average ocean heat flux value seemed sufficient for thermodynamic sea ice modelling in the past (Peeken et al., 2018;Krumpen et al., 2020) and is confirmed in this study by the agreement between modelled and modal SIT in all years except 2016.

255
The studies by Polyakov et al. (2017Polyakov et al. ( , 2020 showed Atlantification is considered to be a positive feedback mechanism (Polyakov et al., 2020) and was mainly observed at the 260 inflow gates of AW into the Arctic Ocean in the Barents Sea (Smedsrud et al., 2013) and north of Svalbard (Ivanov et al., 2012;Onarheim et al., 2014). However, based on mooring and buoy data, Polyakov et al. (2017Polyakov et al. ( , 2020 showed that Atlantification is progressing eastward, impacting ocean stratification and sea ice growth even in the main regions of Arctic sea ice formation in and around the Laptev Sea.  the 2015 winter period. Satellite-derived mean SIT from the ESA Climate Change Initiative Phase 2 (CCI-2) climate data record confirms that a negative SIT anomaly existed in the Laptev Sea during this Atlantification event in 2015 (Fig. 4 b)). This anomaly is likely a result of multiple factors, including increased upward ocean heat fluxes due to Atlantification, but also the 275 observed increase in ice export from the Russian shelves (Krumpen et al., 2013;Itkin and Krumpen, 2017) and ocean heat as a result of increased solar energy input during ice free periods (Perovich et al., 2011;Wang et al., 2016).
In order to quantify the impact of increased upward ocean heat flux on sea ice in the Laptev Sea and in the AOI we adjusted the thermodynamic model to provide a constantly higher ocean heat flux value along the 2016 sea ice trajectories during the Atlantification event observed by Polyakov et al. (2017) in winter 2015 (Fig. 3 c)). Conservative estimates of ocean heat flux 280 during the Atlantification event between January and May 2015 (shaded red area) vary between averages of 4 Wm −2 from moorings closer to the Laptev Sea shelf (Polyakov et al., 2020) and averages of 8 Wm −2 estimated for the moorings further north (Polyakov et al., 2017). Based on these estimates we adjusted the model to provide both values along the red parts of the sea ice trajectories (Fig. 4 a)), which resulted in a mean reduction of SIT of 0.06 (4 Wm −2 run) and 0.20 m (8 Wm −2 run) at the end of the winter (May 2015, Fig. 4 c)). These values are significantly lower than the 0.4 m reduced sea ice growth suggested 285 by Polyakov et al. (2017), which indicates that our adjusted heat fluxes might still be too low. However, the model confirms that increased ocean heat flux reduces sea ice growth in the Laptev Sea. In the AOI, modelled SIT reduced by 0.03 (4 Wm −2 run) and 0.10 m (8 Wm −2 run) in July 2016 (Fig. 4 c)). Although the adjusted model runs show that SIT anomalies induced and surface melt processes altering SIT are likely responsible for a considerable fraction of the observed difference between modelled and observed SIT in the AOI in 2016 as well. Both snow depth (Merkouriadi et al., 2017(Merkouriadi et al., , 2020 and surface melt 305 (Perovich et al., 2014) can vary substantially from year to year and also from the assumptions applied for the presented model.
Especially a thick snowpack, like it has been observed during the 2015 N-ICE campaign Merkouriadi et al., 2017) could significantly limit sea ice bottom growth through thermal insulation (Merkouriadi et al., 2020). However, with limited observations of long-term snow depth and surface melt processes along the investigated sea ice trajectories, we resorted to available climatologies and parametrisations. 310 We argue that the analysis of the 2016 IceBird SIT data allow for a first estimation of the impact of Atlantification on sea ice, its thickness and the time-scale on which these signals can persist. However, it has to be noted that the upward AW heat fluxes vary in strength from year to year, but Atlantification is a process that continuously increases in the eastern marginal ice zone. In fact, Polyakov et al. (2020) showed that the influence of AW heat flux on the ice in the Laptev Sea showed a dramatic increase during the last decade. The upward-directed AW heat fluxes in the Laptev Sea showed an increase during 315 winter periods between 2007/2008 to 2018, that resulted in a more than two-fold reduction of winter ice growth in the last decade (Polyakov et al., 2020).
The example of the 2016 minimum in modal SIT in the AOI is a first indication that the increasing influence of Atlantification potentially persists far beyond the eastern Arctic shelf regions due to its preconditioning effect on SIT. However, to further confirm this discovery it is vital to build continuous long-term SIT time series in the Laptev Sea as well as in the vicinity of the of sea ice, as well as the persistence of SIT anomalies due to oceanic influences.

Interpretation of sea ice surveys from the MOSAiC year
The continuation of the IceBird SIT time series in the MOSAiC year 2020 was aggravated by the COVID-19 pandemic which only allowed for survey flights over the AOI from Longyearbyen (Svalbard) and after the usual sampling period from mid-July to mid-August (Table 1). Mean and modal SIT were obtained over the AOI in September 2020 and are shown in Fig. 1 b). The 335 pathway analysis (Fig. 2) confirms the trend, that ice reaching the AOI in summer is increasingly dominated by SYI (Fig. 1 c)).
Although the modal SIT is similar to the 2016 value, it has to be noted that measurements were conducted considerably later in the melt season, which makes a direct comparison difficult and shows that summer melt has a considerable impact on SIT in the AOI.
Due to the late IceBird MOSAiC campaign in 2020 and the ensuing limitations for the comparability to the existing IceBird 340 time series we turn to the only other available SIT data set that was obtained in the AOI during the relevant period between mid-July and mid-August of 2020 -GEM SIT measurements from the MOSAiC floe. Compared to the areal coverage achievable with the AEM, GEM SIT values provide point measurements that are only partly representative for a larger area. Nevertheless, these floe-scale measurements provide the means for an important first estimation of whether the MOSAiC floe is thicker or thinner compared to the sea ice that was sampled in the AOI in the years prior to 2020.
The MOSAiC Central Observatory (CO) and the ice in its immediate vicinity (radius of approximately 40 km), which accommodated the Distributed Network (DN) of various autonomous measurement stations (Krumpen and Sokolov, 2020), was formed during a polynya event north of the New Siberian Islands (Fig. 2) in early December 2018 .
The ice originated in shelf waters less than 10 m deep, drifted eastward along the shallow shelf and ultimately reached deeper waters in February 2019. By the time the German icebreaker RV Polarstern (Alfred Wegener Institute, 2017) moored to the 350 floe in October 2019 (begin of the drift at 85 • N and 136 • E, see Fig. S3) the CO and DN regions (DNR) were surrounded by thicker residual ice that was formed in early November 2018 . Due to the comparably fast drift along the Transpolar Drift (the floe was only about 1.65 years old when it was sampled in the AOI, Fig. 3 a)) the DNR reached the southernmost border of our selected AOI already in the second half of July 2020 ( Fig. 2 and S2). Along its trajectory through the Arctic Ocean DNR SIT was continuously measured using ground-based and airborne EM devices. Unfortunately, technical 355 problems and unfavourable weather conditions limited the availability of SIT measurements covering larger areas in the vicinity of the floe in the second half of July 2020. However, regular GEM measurements were conducted on the remainders of the CO. The GEM thickness results shown here are based on the rapid-release quickview thickness data, have undergone initial quality control, and have been calibrated against manual observations. In order to ensure the best possible comparability to the IceBird SIT time series, we only consider GEM measurements that were conducted while the floe was in the AOI and 360 during the sampling period of the previous measurements. The resulting preliminary AOI SIT values are based on a total of four surveys obtained between July 16 and July 21. Although GEM measurements were only conducted on the central, more stable part of the MOSAiC floe, large-scale AEM measurements conducted over the DNR in April 2020 indicate that modal and mean SIT values measured on the extended MOSAiC floe were representative for the DNR (Fig. S3 and S4). Additional AEM measurements conducted in the beginning of July 2020 confirm that the modal SIT derived from the GEM surveys is in 365 fact reliable and representative for the wider area ( Fig. S3 and S5).
On the basis of AEM surveys conducted over the DNR in April and early July 2020 and the already existing IceBird time series we argue that the modal SIT of 1.71 m measured on the MOSAiC floe is not just representative for the wider area around the floe but also in line with measurements from previous years ( Fig. 1 b)). The modal thickness of the MOSAiC floe is within one standard deviation of the long-term average over all modal SIT values derived for the AOI. This agreement indicates that 370 the MOSAiC floe and its wider vicinity are not exceptional in terms of modal thickness compared to the long-term time series.
The comparison of the MOSAiC floe modal SIT with SIT values reconstructed by the thermodynamic model from Thorndike (1992) confirms that the MOSAiC floe was not exceptionally thin. In fact, Fig. 3 b) shows an underestimation of modal SIT by the model. Despite the fact that when the MOSAiC floe reached the AOI it was of a similar age as the ice in 2016, modal SIT was considerably thicker. This indicates that the MOSAiC floe might have been less impacted by oceanic heat, among other 375 factors, than the ice that reached the AOI in 2016.
Nevertheless, it is important to note that these results are preliminary. Detailed studies of the ice thickness development of the MOSAiC floe along its drift path through the Arctic and its surroundings are the basis for future studies.  Fig. 1 a)). Measurement uncertainty is given by the black bars. Light blue markers indicate the time series of modal AOI EM SIT (Fig. 1 b)).

Comparison to Russian shipborne SIT observations
Due to the position of the selected AOI just north of the Fram Strait the presented SIT time series provides the possibility to 380 investigate interannual variability of the time-integrated signal of Arctic-wide SIT changes. Nevertheless, the selected AOI is a highly variable, and small excerpt of the Arctic Ocean. Additionally, the presented time series is interrupted and still too short to provide insight into SIT changes on climatological scales. For example, the transition from a MYI-dominated towards a FYI/SYI-dominated Transpolar Drift (Kwok, 2018) that was accompanied by a drastic reduction in Arctic SIT and sea ice volume (Kwok, 2018;Spreen et al., 2020)  during the same season as the IceBird measurements and upstream of the AOI one would expect those values to be thicker 395 than the downstream measurements. However, we also attribute the lower estimates of SIT from the Russian observations to the inherent differences between the observation techniques. While visual and STK SIT observations are largely dependent on the ships route and the avoidance of thicker ice patches for faster navigation through the ice, AEM measurements provide SIT distributions on larger spatial scales. We would therefore assume a bias towards, on average, thinner sea ice for the Russian observations compared to AEM measurements. Nevertheless, the Russian observations provide a much longer time series than 400 the AOI measurements, which allows us to confirm general changes in the overall SIT regime in the central Arctic (Kwok, 2018). While the AOI time series indicates further thinning of sea ice between 2010 and 2020, Russian observations show no trend at all.
The length of the Russian observational time series and its ability to show the previously observed regime shift in SIT indicates how valuable consistent long-term time series are. However, the deviations from the AOI measurements also show its 405 limitations and the importance of joint observations for a better understanding of differences and ultimately a better basis for the interpretation of past, present and future ship-based observations from Russian sea ice experts.

Conclusions
This study provides an important extension of the long-term EM-derived summer SIT time series at the end of the Transpolar Drift presented by . We combine these large-scale summer SIT measurements conducted within the 410 framework of the IceBird and previous campaigns with Lagrangian ice tracking and a reconstructed SIT time series from a thermodynamic sea ice growth model (Thorndike, 1992). With this comprehensive approach we explain the observed interannual SIT variability within our selected area of interest (AOI, 80.5-86 • N and 30 • W-20 • E) and investigate the driving mechanisms and source regions of this variability. Based on preliminary results from SIT measurements gathered during the MOSAiC drift experiment, we also put the MOSAiC floe into a historical context in terms of its thickness.

415
The analysis of pathways and sea ice origin with the Lagrangian ice tracking tool ICETrack reveals that approximately 65% of the ice sampled in the AOI originated in the Laptev Sea. Sea ice reaching the end of the Transpolar Drift is thinning, which has also been shown by Hansen et al. (2013); Renner et al. (2014) et al. (2017, 2020) observed strong Atlantification between January and May 2015. Increased upward-directed ocean heat flux reduced ice growth in the Laptev Sea during this period. Based on the analysis with the thermodynamic sea ice growth model and the AOI EM SIT time series we are able to show how persistent in time and space the impact of Atlantification on Arctic sea ice potentially is. It seems that, due to the fast drift across the Arctic Ocean, winter ice growth was not able to compensate 430 the low initial ice thickness after the Atlantification event. With a tendency towards even faster ice drift along the Transpolar Drift in the future, the impact of Atlantification on sea ice in the eastern marginal ice zone will increasingly be felt in other parts of the sea ice covered Arctic. The presented model analyses also revealed that the assumption of a constant and also our adjusted upward-directed ocean heat fluxes along the sea ice trajectories are insufficient to fully explain the observed modal SIT minimum in 2016. However, it is evident that the influence of oceanic heat, both upward-directed from AW at depth 435 (Polyakov et al., 2017(Polyakov et al., , 2020  SIT measurement approach. Russian shipborne SIT measurements show significant differences to EM-based measurements, but their regularity and spatial consistency enable the depiction of regime shifts in SIT that are hardly resolved by the presented EM SIT time series. Obtaining SIT distributions over large areas and developing and continuing long-term SIT time series will provide unique input data for modelling efforts, and ultimately will improve predictions of Arctic sea ice and its thickness in the future. Continuing regular IceBird measurement campaigns in the vicinity of Fram Strait and combining the results with re-450 liable models, ice tracking tools, and additional up-but also downstream SIT data sets, like the Russian shipborne observations and the Fram Strait ULS time series (Hansen et al., 2013;Renner et al., 2014;Spreen et al., 2020), will prove indispensable for monitoring the complex and radical change of sea ice in the Transpolar Drift system and on an Arctic-wide scale.
All MOSAiC-related data are archived in the MOSAiC Central Storage (MCS) and will be available on PANGAEA after finalisation of 455 the respective datasets according to the MOSAiC data policy.

465
Author contributions. HJB and TK analysed the sea ice thickness data and conducted the backward-tracking and model simulations for the sea ice sampled north of Fram Strait. HJB prepared the manuscript. All authors contributed to the discussion and provided input during the writing process. HJB, TK, and CH participated in the different expeditions during which the EM sea ice thickness data were gathered. AH was vital for the planning and organisation of multiple IceBird campaigns including the one in 2020. IP provided ocean heat flux estimates, calculations for the model analysis, and valuable input to the manuscript. LvA, GB, IR, MW and SH planned and conducted the GEM/EM 470 surveys during Leg 4 of MOSAiC and processed the preliminary SIT data. RR provided preliminary airborne EM SIT data from Leg 3 of MOSAiC. TA, SF and SS gathered and analysed the observational data from the Russian cruises to the North Pole.
Competing interests. All authors declare that they have no conflict of interest.
Acknowledgements. This project was carried out in the framework of the BMBF-funded Russian-German cooperation QUARCCS (grant: 03F0777A).

475
2020 AEM and GEM data used in this manuscript were produced as part of the international Multidisciplinary drifting Observatory for the Study of the Arctic Climate (MOSAiC20192020) during IceBird MOSAiC summer (P6-222_IceBird_MOSAiC_2020) and the RV Polarstern Legs 3 (AWI_PS122_03) and 4 (AWI_PS122_04).
The processing of visual and STK data by T. A, Alekseeva, S. V. Frolov and S.S. Serovetnikov was funded by the Russian Foundation for Basic Research (RFBR) according to the research project number 18-05-60048.
We want to thank the AWI logistics department, the crews of the research aircrafts Polar 5 and 6, the crews of Station North in Greenland, the pilots and crews on RV Polarstern (Alfred Wegener Institute, 2017) and the observers on the Russian icebreakers for their tireless efforts during the various expeditions. This unique collection of data from different expeditions would not exist without you! Finally, thank you to the two anonymous reviewers for their valuable and constructive comments and suggestions during the review process and to editor Dr. John Yackel and the team at The Cryosphere for their support and processing of our manuscript. Table 1. Summary of used research platforms, sampling periods, profile lengths, and basic sea ice thickness statstics for the selected area of interest (see Fig.1 Table 2. Summary of used ships, number of cruises, measurement periods, observation types, and mean sea ice thicknesses (SIT) and uncertainty (Frolov et al., 2007;AARI, 2011) observed during Russian cruises along the 86 -90 • N transect (black line Fig.1 a)) Year