Evidence of Abrupt Transitions Between Sea Ice Dynamical Regimes in the East Greenland Marginal Ice Zone

Sea ice modulates the energy exchange between the atmosphere and the ocean through its kinematics. Marginal ice zone (MIZ) dynamics are complex and are not well resolved in routine observations. Here, we investigate sea ice dynamics in the Greenland Sea MIZ using in situ and remote sensing Lagrangian drift datasets. These datasets provide a unique view into ice dynamics spanning spatial scales. We find evidence of tidal currents strongly affecting sub‐daily ice motion. Velocity anomalies show abrupt transitions aligned with gradients in seafloor topography, indicating changes in ocean currents. Remote‐sensed ice floe trajectories derived from moderate resolution satellite imagery provide a view of small‐scale variability across the Greenland continental shelf. Ice floe trajectories reveal a west‐east increasing velocity gradient imposed by the East Greenland Current, with maximum velocities aligned along the continental shelf edge. These results highlight the importance of small scale ocean variability for ice dynamics in the MIZ.


Introduction
The Arctic is warming at over twice the rate of the global average as a result of rising greenhouse gas concentrations (Taylor et al., 2022). A hallmark of Arctic change is a thinning and retreating ice pack (Comiso et al., 2017;Maslanik et al., 2007). Areas of sea ice decline are seeing stronger momentum transfer to the ocean (Polyakov et al., 2020) and intensifying ocean eddies (Manucharyan et al., 2022), amidst other effects (Feldl et al., 2020). Furthermore, previously ice-bound sea routes have rising ship traffic as the ice edge moves northward (Boylan, 2021;Dawson et al., 2018), increasing the potential need for accurate drift forecasts.
Quantifying sea ice dynamics in the marginal ice zone (MIZ) has been an ongoing challenge (Dumont, 2022). In situ sea ice motion measurements are primarily retrieved from moorings and drifting buoys. These instruments are difficult to deploy and generally exhibit a low residence time in MIZ. As a result, there are rarely multiple buoys in the MIZ at any given time. Observations from numerous drift expeditions have demonstrated a tight coupling between ice and ocean variability in the MIZ at small and moderate length and time scales (Leppäranta & Hibler, 1987), including effects of eddies (Morison et al., 1987), boundary currents (Quadfasel et al., 1987), waves (Squire, 2020), inertial oscillations (Gimbert et al., 2012;Lammert et al., 2009), and tidal currents (Heil et al., 2008;Vasulkar et al., 2022). Remote sensing provides an invaluable complement (e.g., Kozlov et al., 2020;Manucharyan et al., 2022), however low data acquisition rates (on the order of days) inherently limit the physical processes being resolved (Kwok, 2010). Sea ice dynamics at sub-daily time scales, for instance, are important for measuring lead opening rates and ridging (Hutchings et al., 2011), key parameters for modeled sea ice growth and air-ocean fluxes, as well as for resolving tidal and inertial motion. Furthermore, drift estimates from microwave radiometers have higher uncertainty during the summer due to melt processes (Tschudi et al., 2020) and due to increased small-scale variability (Yoshizawa et al., 2023 This study aims to characterize marginal ice dynamics in the Eastern Greenland Coast and the Fram Strait region across a broad range of scales and throughout a full season. The complex bathymetry surrounding the Fram Strait and Svalbard is known to result in strong, spatially heterogenous tidal currents (Fer et al., 2015;Padman et al., 1992;Vasulkar et al., 2022) as well as narrow and fast boundary currents (Foldvik et al., 1988;Richter et al., 2018). The spatial heterogeneity of the region, compounded by high interannual variability (Smedsrud et al., 2017), makes it difficult to inter-compare results from prior drift expeditions. As coupled models begin to achieve sufficient resolution to capture small-scale sea ice-ocean interactions (Arbic, 2022), comprehensive observations are needed to constrain models and evaluate model performance, especially in high latitudes where uncertainties in small-scale ocean variability (including tides) result in large inter-model differences (Stammer et al., 2014;Zaron & Elipot, 2021). To this end, we leverage two new observational datasets, each providing Lagrangian measures of ice motion: buoy drift trajectories from the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC, Nicolaus et al. (2022)) and ice floe trajectories obtained from moderate resolution remote sensing imagery via the Ice Floe Tracker algorithm (IFT, (Lopez-Acosta et al., 2019)). The MOSAiC drift trajectories represent the highest density of in situ ice dynamics observations yet collected. Ice floe trajectories from IFT allow investigation of variation in ice motion at high resolution, as the motion of individual ice floes is measured rather than the area-averaged velocity that is provided by traditional remote-sensing velocity products, and provide a step toward filling the summer gap in satellite-based drift observations.

MOSAiC Drift Trajectories
Drift trajectories (n = 108) from sea ice buoys deployed during the MOSAiC Expedition were selected such that trajectories (a) contained at least 30 days of data between 1 May and 1 September 2020, (b) had sampling rates of at least once every two hours, and (c) drifted through the Fram Strait. Buoys were deployed in an array surrounding an instrumented Central Observatory (CO) (Bliss et al., 2022(Bliss et al., , 2023Krumpen & Sokolov, 2020;Nicolaus et al., 2022). Distance between buoys in the MOSAiC Distributed Network (DN, red lines in Figure 1) ranged from less than 1-60 km from the CO on 1 May. Nine additional buoys in the Extended Distributed Network (ExDN, gold trajectories in Figure 1) were between 150 and 530 km away from the CO. Positions were interpolated to a 1-hr grid using cubic splines following de-spiking. Drift velocity was computed using centered differences after projecting buoy positions onto the NSIDC North Polar Stereographic grid. Sea ice concentration (SIC) from AMSR2 (Meier et al., 2018) was interpolated to individual buoy positions to find the last date where the buoy remains in sea ice (SIC >0). Hourly 10-m wind velocity from the ERA5 reanalysis (Hersbach et al., 2020) was interpolated to buoy and ice floe coordinates. Bathymetry data was obtained from the International Bathymetric Chart of the Arctic Ocean version 4.1 (Jakobsson et al., 2020). Observations at the DN are summarized via median and percentiles for robustness against outliers and to avoid making assumptions about the shape of the distribution.

Ice Floe Tracker Algorithm
The Ice Floe Tracker (IFT) algorithm (Lopez-Acosta et al., 2019) was used to identify and track sea ice floes in 250 m resolution optical imagery from the Moderate Resolution Imaging Spectroradiometer (MODIS) spanning the spring-to-summer transition from 2003 to 2020. IFT is a feature-tracking algorithm and can resolve floes with areas of at least 1 km 2 . Corrected-reflectance MODIS imagery was downloaded from NASA Worldview (https:// worldview.earthdata.nasa.gov). Daytime MODIS images from both Aqua and Terra satellites were analyzed and assigned timestamps using the Satellite Overpass Identification Tool (Hatcher et al., 2022). Both Aqua and Terra provide daily observations, with time offsets between the two satellite images of between 20 and 90 min. Floe trajectories were resampled to daily resolution prior to calculating daily displacement, with gaps of at most 1 day in length filled by linear interpolation. In total, drift trajectories (2-22 days) are available from 15,534 floes; median trajectory length is 2 days, and the total number of estimated drift displacements is 42,613.

Frequency Analysis
Rotary spectral analysis was performed using the University of Hawaii PyCurrents Python library (https://currents. soest.hawaii.edu/hg/pycurrents/). Velocities for 20-day trajectory segments were de-trended by removing the centered 35-hr mean, filtered with a Hann window. Resulting spectra were then smoothed with a 3-point boxcar filter. Note that while changing the size of the de-trending window affects the magnitude of the spectra for lower frequencies, it does not alter the position of spectral peaks.
The contributions of diurnal and semi-diurnal tides to the sea ice velocity were determined from harmonic analysis as in Pease et al. (1995). The harmonic model has the form where ′ ( ) = − is the position anomaly in polar stereographic coordinates relative to the centered 35-hr mean, t is the time, C, D, A k , and B k are unknowns, and ω k are the frequencies of the O1, K1, N2, M2, and S2 tidal constituents. The model is solved for each Cartesian position component separately. Maximum tidal currents were estimated by calculating velocity using centered differences on the predicted position anomalies then calculating the daily maximum of these velocities. Model fit was evaluated with the coefficient of determination.

Deformation Analysis
The horizontal deformation of sea ice was computed via the Green's Theorem method (Hutchings et al., 2012(Hutchings et al., , 2018. Strain rates were estimated using sets of buoy arrays manually identified from buoy positions on the first of the month (00:00 UTC) and filtered with a 5-km spatial filter. Arrays of three or four buoys were selected to provide Observations with estimated sea ice concentration less than 15% were masked prior to calculating percentiles. Panels (f) and (g) show area-weighted means (solid lines) and standard deviations (shading) of divergence and maximum total shear strain rates. Lower left: Buoy positions and sea ice conditions on July 12th (h), July 26th (i), and August 6th (j). The blue star marks the location of 2020P225, which was located at the Central Observatory. Imagery from MODIS accessed through NASA WorldView. Right: Drift trajectories for buoys within the Distributed Network (red) and additional buoys (gold) overlaid on ocean bathymetry. The light gray trajectory corresponds to buoy 2020P225 as in panels (h)-(j). The dotted portion of the trajectory marks locations where interpolation with the satellite product indicates 0% SIC and MODIS imagery shows that the buoy is next to the ice edge. Colored circles mark the first day of each month.
full, non-overlapping coverage of the sea ice while reducing the occurrence of skinny triangles leading to inaccurate strain rate estimates. As a result, 80, 74, and 70 arrays were produced for May, June, and July, respectively. Hourly strain rates were computed for each individual array during the month in which the array configuration was defined, and a time series was constructed by appending the mean strain rate components (divergence and maximum shear) across all arrays for each month. Strain rates were not computed following breakup of the CO on 31 July 2020 (Nicolaus et al., 2022). The mean divergence rate was computed as an area weighted mean to account for variation in array size across the DN.

Overview of MOSAiC Summer Drift
From May to September 2020, the MOSAiC Distributed Network (DN) transitioned from central Arctic pack ice, through the Fram Strait, and into the East Greenland Current, with most buoys eventually exiting the MIZ into the Greenland Sea ( Figure 1, right panel). We approximate the transition times in the following paragraphs by providing the dates for buoy 2020P225, which was deployed near the Central Observatory; note that the DN had an effective radius of ≈60 km, so crossing an underwater feature such as a shelf edge at a constant speed of 0.2 m/s takes nearly 7 days.
During the initial stage, the array remained in pack ice over the Nansen Basin (May 1st to June 13th, Figure 1). Drift speeds were low (median speed 7 cm/s) except for the second week of May, coinciding with a strong cyclone. This event induced deformation in the array that peaked at the time of maximum wind speed. As the array drifted over the Yermak Plateau (June 13-July 12), drift speed increases up to 14 cm/s. Deformation rates began to increase as the first few buoys moved into a shallower region. The ensemble median drift speed anomaly increased as the array drifted over the plateau boundary displaying an apparent diurnal oscillation. Note that the wind speed was not noticeably different from that of the previous month, suggesting the importance of ocean forcing. Both divergence and shear increased at this time. Divergence, like the sub-daily velocity, showed a diurnal oscillation pattern. The oscillation is evidence of working leads repeatedly opening and closing. A similar feature was noted during the N-ICE2015 campaign, suggesting that this diurnal pattern is a recurring feature of the region (Itkin et al., 2017). Earlier measurements of divergence in ocean currents suggest tides as a mechanism for this feature (Padman et al., 1992). This possibility will be examined in Section 5. The periodic changes in divergence preceded the oscillation in the median drift speed anomaly, hinting at the role of internal ice stresses inducing deformation.
From July 12-August 1 the buoy array drifted southward within the East Greenland Channel and experienced widespread positive divergence and increasing shear, stretching the array on roughly a north-south axis (Figure 1 in panel h and i). Prior to this point, the structure of the deployment array had not undergone significant changes. Sea ice concentration decreased markedly, and we see an increase in 12-hourly oscillations relative to the daily median. Wind speeds were lowest during this period, indicating that the increase in drift speeds must have come from either decreased ice stresses or increased ocean forcing. The role of wind forcing will be explored in Section 4.
Deteriorating ice conditions (Figure 1, panel j) led to the decision at the end of July to dismantle the CO and retrieve many of the autonomous sensors (Nicolaus et al., 2022). By August 1st the main floe had broken apart. A portion of the remaining array drifted southeast and appeared to be drawn into a large eddy, while the rest of the array drifted southwest onto East Greenland Shelf (Figure 1 panels j, k). The late summer ice pack comprised distinct floes among patches of open water. The proximity of filaments of sea ice drawn into vortices indicates the presents of mesoscale ocean eddies. A mid-August storm lead to a spike in drift speeds and was followed by enhanced sub-daily, oscillatory variability. By the end of August, all but 12 buoys had drifted into ice-free waters or ceased operation.

Relationship Between Wind and Ice Velocity at Daily Timescales
The majority of daily to monthly sea ice drift variability in the central Arctic can be explained by variability in the wind (e.g., Thorndike and Colony (1982)). Thus motivated, we begin by assuming a simple relationship between ice and wind where U = u + iv is the complex drift speed, α is a transfer coefficient, which we refer to as the drift speed ratio, θ is the turning angle, U a = u a + iv a is the complex wind speed, and ϵ is the residual. The residual ϵ includes effects of ocean currents, sea surface slope, inertia, Coriolis force, and internal ice stresses, as well as error in the choice of α and θ and error in the ERA5 wind field. Oceanic, atmospheric, and internal ice stresses are an order of magnitude larger than the remaining terms, so we focus on these (Leppäranta, 2011). Changes in ϵ thus provide an indicator of changes in the key forcings for ice motion. The median drift speed ratio over the study period is α = 0.021 and the median turning angle is 20°, in line with previous Arctic Ocean-wide estimates (Brunette et al., 2022, e.g.,). Applying these values to Equation 2 results in the estimated wind speed U est = αe −iθ U a depicted in Figure 2, panel a.
The wind model tends to overestimate the drift speed slightly from May to early June consistent with ice stresses acting against the wind stress. Thereafter, the error in drift speed magnitude ϵ s = |U| − |U est | grows, as does variability in U. The closest agreement to the theoretical value comes when wind speeds are high, (e.g., during the storm in early May), suggesting that the ice approaches free drift during these events. Sustained low wind speeds in mid-to-late July coincided with increased magnitude of ϵ s . Following July 15th, α becomes strongly right-skewed and the median value increases; meanwhile, median θ is close to the same but the likelihood of large deviations increases. Given the low sea ice concentration in the region (Figure 1 panels i, j), it is unlikely that the deviations are due to increases in ice stress. Rather, we surmise that ocean forcing is playing a larger role in ice dynamics at daily timescales from mid-July onward.
The number of buoys decreases as the drift proceeds onto the Greenland Shelf. To supplement the investigation of ice drift variability in the shelf region we turn to the Ice Floe Tracker (IFT). The majority of IFT observations are from the East Greenland Shelf between 70 and 80N (Figure 3 panel a). Low drift speeds along the coast are the result of frequently occurring landfast ice. Drift speeds are enhanced along the shelf boundary ( Figure 3, panel  b). While a small gradient in wind speed contributes to the drift speed gradient, the residuals from the drift model are nearly the same magnitude as the overall drift speed, indicating a major role for ocean currents. For most of the region, this area of enhanced drift speed is also the ice edge. Further north, in the East Greenland Channel, we find local maxima in drift speed away from the ice edge along the shelf boundary. Drift direction, too, tends to follow the shelf boundary. On the northwest corner of the shelf, we see some evidence of a re-circulation pattern with northward flow along the coast turning clockwise to join the southward flow, consistent with model results from Richter et al. (2018).
Empirical estimates of the distribution of turning angles and drift speed ratios for the 20 years of summer IFT data ( Figure 3) indicate that both quantities are highly variable and depend on the wind speed. The turning angle distribution for the IFT data is strongly right-skewed. The peak in the θ distribution at 24° is consistent with the expected ≈20° turning angle. The thick tail shows that motion to the reverse of the wind direction is a common occurence. The ice nearly always moves southward along the coast, and the wind is mainly aligned along shore, favoring the southwest direction. Under southerly winds, the ice as a whole does not tend to change directions. In most cases, when the turning angle is close to −180°, the wind direction is southerly (not shown). The highest variability in turning angles and drift speed ratios is at low wind speeds. As the wind speeds increase they come to dominate control of ice drift over the ocean currents. For comparison with IFT, we downsample the buoy drift trajectories to 12:00 UTC, the approximate time of the MODIS daytime overpass, and re-calculate velocity from daily displacements. Empirical distributions of θ and α for the period from July 15th onward are shown in Figure 3. This period is when the majority of the MOSAiC array is within the region sampled by IFT. Due to the relatively small sample size, the buoy-derived distribution is less evenly sampled than the IFT distribution, yet we see that the main features are reproduced. At wind speeds lower than 7.5 m/s, we are much more likely to see high variability in both the turning angle and in the drift speed ratio. Drift speed ratios at high wind speeds are higher in the buoy data than in IFT, which may be due to differences in spatial sampling. As seen in Figure 3, panel b, the highest median drift speeds are found along the shelf break and within the East Greenland Channel, where many of the buoys were located.

Sub-Daily Sea Ice Variability, Inertial Oscillations, and Tides
We now quantify the apparent tide-like oscillation seen in Figure 1. We select 20-day segments of buoy trajectories from four distinct bathymetric regions: the Nansen Basin (NB), the Yermak Plateau (YP), East Greenland Channel (GC), and East Greenland Shelf (GS) (Figure 4a-4d). Rotary spectra show distinct characteristics, with strong signals in both semi-diurnal and diurnal frequency bands everywhere except the deep Nansen Basin (Figures 4a and 4e) indicating that tidal currents play an important role in sub-daily sea ice velocity variability.
In the northern hemisphere, inertial oscillations are clockwise (CW), which manifests as higher spectral power in the CW direction than in the counterclockwise (CCW). The peak in the semidiurnal band for the Nansen Basin trajectories is small but exists in both CW and CCW components. This suggests the possibility of tidal effects on ice motion even in pack ice well away from the shelves. We note as a topic for future research that the east-west velocity component displays a regular semi-diurnal oscillation that is not apparent in the north-south velocity component.
The spread of spectral power across the array (indicated by the shading and dotted lines in Figure 4) is smaller in the NB and YP than in the channel and shelf, reflecting both the coherence expected in pack ice and the area sampled. The CCW semi-diurnal peak is narrow and strong in the GC suggesting a clearer influence of semi-diurnal tides. Over the GS, the diurnal band is no longer distinct, while the semi-diurnal CW band remains strong but increases in spread. Since the shelf region includes a wider range of buoy locations, it is possible that interacting tidal waves and varied bottom topography dilute the tidal signal. Sensor failure and sensor retrieval results in a smaller sample size (26 buoys) representing a region nearly twice as large as the region sampled over the YP, further contributing to the spread in the spectral peak.
The harmonic model assumes that hourly velocity anomalies occur at a limited set of tidal frequencies. When the harmonic model performs well, we interpret that the sub-daily sea ice velocity is consistent with tidal forcing. Tidal constituents are typically estimated from measurements of ocean currents or sea surface height, not sea ice motion; we expect that the additional variability due to imperfect momentum transfer between the surface current and the motion of the ice pack will make the estimate of tidal variability more uncertain. It is therefore notable that we find such strong tidal signals in the ice motion. Implied maximum currents of between 0.1 and 0.2 m/s are seen over the shelf, channel, and plateau, consistent with other tidal current speed estimates (Padman & Erofeeva, 2004;Padman et al., 1992;Vasulkar et al., 2022). These speeds are close to the total drift velocity, hence, tidal currents are likely a major component of ice motion in these regions. For the Yermak Plateau, more than 80% of the sub-daily variance is explained by the tidal currents. The strong change in ocean forcing from the Nansen Basin onto the Yermak Plateau implies a sharp gradient in the ice velocity, inducing deformation. This is confirmed in Figure 1, where we see diurnal oscillation in both divergence and maximum shear that coincides with arrival of the MOSAiC array at the edge of Yermak Plateau.
Inertial oscillations are difficult to differentiate from semi-diurnal tidal variability at high latitudes. Not only are individual semi-diurnal tidal components very close to the inertial period, but tidally generated waves can become inertially trapped. Confidence that the semi-diurnal cycles can be attributed to tides comes from the relatively long 20-day time window used for estimating tidal constituents, and the presence of strong peaks in the CCW band of the rotary spectra. The presence of inertial oscillations in addition to the tidal variability is indicated by the strong CW peaks in the rotary spectra as well as the timing of increases in sub-daily velocity anomalies following brief periods of strong winds, such as occurred on August 15th.

Discussion and Conclusion
The MOSAiC ice drift observations capture a broad range of summer ice dynamics in a historically undersampled region. The spatial coverage in our study allows us to document the imprint of differences in ocean forcing on sea ice dynamics across distinct bathymetric regions with unprecedented fidelity. Ice drift has a strong stochastic component due to the interaction of highly variable wind and ocean forcing. The particular path taken by the MOSAiC observatory resulted in a month-long residence over the tidally active Yermak Plateau. Gradients in velocity due to the abrupt transition between the basin and tidally active plateau impose strain on the ice, enhancing deformation. Subsequent low wind speeds, decreasing ice concentration, and the proximity of boundary currents resulted in ocean currents dominating the drift through the Fram Strait. Following transit through the Fram Strait, intermittent high drift speed events result from strong wind events, and both tides and inertial oscillations affect the subdaily drift velocities. Low ice concentration late in the summer resulted in the MOSAiC observatory being more sensitive to changes in atmosphere and ocean forcing, unconstrained by internal ice stresses.
Our results emphasize the importance of small-scale, sub-daily motion for ice dynamics (Heil et al., 2008;Itkin et al., 2017). We show that a substantial fraction of observed sea ice velocity at hourly timescales occurs at tidal and inertial frequencies. Characteristics of this variability are distinct within regions defined by bathymetry. Uncertainty in bathymetry combined with sparsely located tide gauges means that tides are poorly constrained throughout the shallow Arctic marginal seas. Models of sea ice drift that fail to take mesoscale ocean variability and tides into account will systematically underestimate drift variability and deformation. The range of dynamical regimes captured during the MOSAiC drift provides a challenging test case for next-generation, tide-resolving coupled sea ice models.
Motion at daily timescales in the Fram Strait summer marginal ice zone bears a spatially heterogeneous imprint of ocean forcing. As ice concentration decreases and ice moves closer to the ice edge, the drift speed ratio increases, and the ice drift becomes less spatially coherent. The wide range of observed turning angles and drift speeds indicate an important role for ocean variability. As we showed through the IFT drift statistics, such variability is not limited to the MOSAiC observational period for the Greenland Shelf region, but is a typical feature of this highly dynamic region. 2722) and the ONR Multidisciplinary University Research Initiatives Program (N00014-23-1-2014). Drift track data used in this manuscript was produced as part of the international Multidisciplinary drifting Observatory for the Study of the Arctic Climate (MOSAiC) with the tag MOSAiC20192020 and the Project_ID: AWI_PS122_00. The authors acknowledge the data collection and processing efforts of the MOSAiC science and logistics teams and of all those who contributed to MOSAiC and made this endeavor possible (Nixdorf et al., 2021). Finally, the authors also acknowledge the use of imagery from NASA's Worldview Snapshots application (https://wvs. earthdata.nasa.gov/), part of NASA'S Earth Observing System Data and Information System (EOSDIS).