Sea Ice Drift and Arch Formation in the Robeson Channel Using Daily Coverage of Sentinel-1 SAR Data During the 2016 – 2017 Freezing Season

Robeson Channel (RC) is a narrow sea water passage between Greenland and Ellesmere Island in the Arctic. It is a pathway of sea ice from the central Arctic and out to the Baffin Bay. This paper uses a set of daily Synthetic Aperture Radar (SAR) images from Sentinel-1A/1B, acquired between September 2016 and April 2017, to study kinematics of 10 individual ice floes as they approach then drift through the RC. Tracking of 39 selected floes was visually performed in the image sequence and their speed was calculated and linked to the reanalysis 10 m wind from ERA5. Results show that drift of ice floes is remarkably slow while in the compact ice regime upstream of the RC unless the floe is surrounded by water or thin ice. In this case the wind has more influence on the drift. On the other hand, ice floe drift is found to be about 4–5 times faster in the open drift regime within the RC and clearly influenced by wind. A linear trend is found between change in wind 15 and change in ice drift speed components, both along the length of the channel. Case studies are presented to reveal the role of wind on ice floe drift in details. The study also addresses the development of the ice arch at the entry of the channel. It started development on 24 January and matured on 1 February 2017. Details of the formation process, using the sequential SAR images, are presented. The arch’s shape continued to adjust by rupturing ice pieces at locations of cracks under the influence of northerly wind (hence the contour keeps displacing northward). The study highlights the advantage of using the 20 high-resolution daily SAR coverage in monitoring aspects of sea ice cover in narrow water passages where the ice cover is highly dynamic. The information will be particularly interesting for possible applications of SAR constellation systems.


Introduction
One of the exit gates for sea ice flux from the Arctic Basin to southern latitudes is through Robeson Channel (RC). It is located between Greenland and Ellesmere Island (Canada), with its northern location around 82°N, 62°W (Fig. 1). It 25 connects the Lincoln Sea (a southern section of the Arctic Ocean) to the Kennedy Channel, which opens to the Kane Basin.
These three water bodies are known as Nares Strait, which opens south to the Baffin Bay. RC is a short and narrow passage (about 80 km in length and 30 km wide) and more than 400 m deep along its axis.
Oceanographic measurements in the RC are not commonly performed. Herlinveaux (1971) found the dominant surface 30 current to be from north to south with an average velocity in April and May increased from about 0.36 km h -1 near the surface to nearly 0.9 km h -1 at 80 m depth. Strong southerly current around 1.08 km h -1 was also measured in the western section of the channel during early spring of 1971 and 1972 with fluctuation of 0.43 (Godin, 1979). When they used two https://doi.org/10.5194/tc-2020-44 Preprint. Discussion started: 4 March 2020 c Author(s) 2020. CC BY 4.0 License. ocean simulations to study the circulation and transport within Nares Strait, Shroyer et al. (2015) found that the mean current structure in south of RC depended on the existing of landfast ice.

35
The sea ice cover in the RC comprises a combination of seasonal (first year ice, FYI) and perennial ice (multi-year ice, MYI) both imported from the Arctic Basin through the Lincoln Sea. The only locally grown ice is found in narrow strips adjacent to the land at the two sides of the channel. Based on earlier results of ice thickness and motion retrieved from optical satellite sensors and reconnaissance flights in the 1970s, Tang et al. (2004) estimated the ice flux crossing the RC to be around 40 40×10 3 km 2 . Using a record of ice displacement retrieved from Radarsat-1 images during , Kwok (2005 found the average annual ice area flux to be 33×10 3 km 2 . Rasmussen et al. (2010)

45
Sea ice drift is influenced by wind forcing, ocean current and internal stresses within the pack ice (caused by interactions between ice floes, which reduce ice momentum). The latter factor is determined by ice types and concentration within the pack. Other minor factors include the Coriolis force and sea surface tilt. The dynamics of the ice motion is based on spatial and temporal scales (McNutt and Overland, 2003), namely individual floe (< 1 km), multiple-floe (2-10 km for up to 2 days), aggregate floe (10-75 km with 1-3 d time scale), pack ice cover (75-300 km at 3-7 d) and sub-basin scale (300-700 km at 50 7 -30 d). The best coupling with wind occurs at the pack ice scale (also called coherent scale). According to this categorization, the only individual and multiple floe scales are observed in SAR images of the RC. Here, the response to wind is usually floe-to-floe bumping, ridging, redistribution and differential floe motion (McNutt and Overland, 2003).
Tracking individual ice floe motion from a sequence of satellite images is potentially feasible if the temporal resolution of 55 the satellite coverage is reasonable (at least daily). An early attempt is reported in Sameleson et al. (2006) for ice in the Nares Strait using the coarse-resolution satellite data (tens of kilometres) from a passive microwave radiometer at 6.5 GHz.
The authors tracked the motion using only 3 to 5 locations of same floe in a sequence of the satellite images. Due to their fine resolution (tens of meters). sequential Synthetic Aperture Radar (SAR) images are the best tool to monitor sea ice kinematics, particularly if available at short-time scale. However, they have a limited spatial coverage. The earliest studies to 60 estimate sea ice displacement using SAR is presented in Hall and Rothrock (1981) and Leberl et al. (1983), using sequential Seasat SAR images. Later, making use of the more frequent coverage of Radarsat-1 in the western Arctic, the Radar Geophysical Processing System (RGPS) was developed and produced gridded ice motion and deformation data, tracked every 3 -6 days from 1998 to 2008 (Kwok and Cunningham, 2002). A more recent ice tracking operational system (also gridded) is described in Demchev et al. (2017) using a series of Sentinel-1 SAR images.
While RC is covered most of the year by influx of ice from the Lincoln Sea, it is possible that the flow of the ice may be blocked in winter at the entrance of the channel by formation of an arch-shaped ice configuration that spans a transect between two land constriction points at Greenland and Ellesmere Islands. The arch usually collapses in early summer, allowing continuation of the ice flux. Kwok et al. (2010) pointed out that no arch was formed in 2007 leading to a major loss 70 of Arctic ice, which was equivalent to about 10 % of the average annual amount of ice discharged through the much wider Fram Strait (400 km versus 30 km width). This signifies the fact that the entire Nares Strait can represent a major route to the Arctic if the ice arch ceases to form in the future due to thinning of Arctic ice. Moore and McNeil (2018) addressed the collapse of this arch in relation to the recent trend of sea ice thinning. In the present data set, the arch formation started on 24 January 2017 and continued till May 2017. The study includes detailed description of the mechanism of arch's formation.

75
The objective of the study is to utilize the daily Sentinel-1A/-1B SAR coverage of the RC area during a full freezing ice season (September 2016 to end of April 2017) to examine two sea ice features in the RC. The first is the drift of individual ice floes in terms of speed and direction. The information is linked to the 10 m wind reanalysis data to explore the wind influence on the ice drift. Further knowledge about how wind and ice drift are related enables the improvement of sea-80 ice/atmosphere dynamic models (Leppäranta, 2011). The second is monitoring the formation of the ice arch at the inlet of the RC during its 10 days of development until maturity. The advantage of using the daily coverage of the fine resolution SAR in retrieving this information in such a narrow channel is expected to instigate further operational applications of SAR constellation systems (e.g. the recent Canadian Radarsat Constellation Mission (RCM)) with their finer temporal resolution. 85 Figure 1. Map of the RC and its surrounding areas. It is located between Greenland and Ellesmere Island (Canada), with its northern location around 82°N, 62°W. It connects the Lincoln Sea to the Kennedy Channel, which opens to the Kane Basin.  Both IW (swath 250 km at spatial resolution 5m×20m) and EW (swath 400 km at median resolution 20m×40m) modes, both with Level-1 Ground Range Detected (GRD) product were used. All images were acquired in HH polarization. Almost daily 95 coverage of the RC area from both satellites were obtained from late September 2016 to the end of April 2017 (total of 361 images). Images were calibrated to backscatter coefficient in decibel then georeferenced. In order to reduce the image size and the speckle, images were resampled to 50×50 m. While the incidence angle of the EW mode varies between 29.1°and 46.0°across the swath, no correction for the variation of the angle was performed since the backscatter was not used quantitatively.

Wind data
Reanalysis of 10 m level wind is available from a few sources. Four sources were examined in this study: (1) the U.S. National Centers for Environmental Prediction, jointly with National Center for Atmospheric Research (NCEP/NCAR) (Kalnay et al., 1996), (2) the joint NCEP with the Department of Energy (NCEP/DOE) (Kanamitsu et al., 2002), (3) the European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-Interim) (Dee et al., 2011), and (4) its 105 successor ERA5 (C3S, 2017). Specifics of each source are presented in Table 1. The difference between the estimated speed from each source and the speed from Alert station is plotted in Fig. 2 for the period from 1 October 2016 to 30 April 2017.
Alert weather station (Canada) is located at (82.52°N, 62.28°W). This location is close enough to the study area ( Fig. 1). Figure 2 reveals the overestimation of the reanalysis wind when the station's wind measurement is < 10 km h -1 . As the wind 110 measured from the station increases, a systematic underestimation of the reanalysis wind is observed. This is particularly true from the two ERA products. When the speed from the Alert station exceeds 30 km h -1 , reanalysis wind from all sources can be severely underestimated by 20-40 km h -1 . Previous studies show that low-resolution global reanalysis of the wind speed and direction have large errors in the narrow channels of the Nares Strait (Dumont et al., 2009). The present data show that the average absolute deviation of the NCEP/NCAR, NCEP/DOE, ERA-Interim and ERA5 wind reanalysis from the 115 measured wind at Alert station is 9.12, 9.74, 9.04 and 8.92 km h -1 over the period from 1 October 2016 to 30 April 30 2017, respectively. Hence, we chose ERA5 data because of its minimum deviation from Alert's data and finer grid spacing. Data are used to explore links with ice floe drift and study the ice arch development. The grid points from ERA5 reanalysis https://doi.org/10.5194/tc-2020-44 Preprint. Discussion started: 4 March 2020 c Author(s) 2020. CC BY 4.0 License.

Ocean current and sea surface height data
Daily and monthly mean maps of ocean current (vertical coverage at 50 levels from -5500 m depth to surface) and sea surface height are components of the GLORY12V1 reanalysis product covering the altimetry era . It is based on 135 the real-time ocean reanalysis product of Copernicus Marine Environment Monitoring Service (CMEMS) (Fernandez and Lellouche, 2018). Sample maps of both parameters are used in this study to compliment the interpretation of the wind influence.

140
Daily coverage of the fine resolution images of Sentinel-1A/1B SAR (50 m) were used to generate tracking of selected sea ice floes and detect the temporal evolution of the ice arch from its commencement on 24 January until it matured on 1 February 2017. A total of thirty-nine floes were tracked between 26 September 2016 and 10 April 2017. Thirty-two floes moved mainly southward, crossing the inlet of RC, and seven moved mainly northward within the RC. Among the seven floes, five moved in the drifted ice regime before and shortly after the formation of the ice arch and two moved in a polynya-like regime after the formation of the arch. Each ice floe was visually identified in a sequence of daily SAR images (between 4 to 28 scenes), then the floe displacement, drift speed, and direction were calculated. The displacement was determined from the subjectively estimated locations of the same floe in two successive daily images using a code to covert latitude/longitude pairs to distance according to World Geodetic System (WGS84) coordinate system.

150
Three sources of error are implied in the estimation of the displacement. The first is the geolocation error of the SAR image.
The Sentinel-1 product specification (Bourbigot et al., 2016) mentions that the absolute pixel location accuracy is less than 7 m for the IW mode but no number is mentioned for the for EW mode, which is used in this study. The second is the assumption of a linear path (as opposed to curvilinear or meandering path) of the floe between two successive days. This assumption had to be employed because the temporal resolution of Sentinel-1A/1B is not finer than one day. The third is the 155 subjective estimate of the centre of the same floe in successive images. This was also estimated to be within a few pixels.
Assuming these errors are independent and normally distributed, the error in the estimated ice drift speed would be roughly 0.2 km d -1 .
The ice floe speed was calculated using the travelled distance as mentioned above and the period between the two image 160 acquisitions, which varied between 16 and 33 hours. To link the floe speed to the wind at any location of the floe, the reanalysis wind data from ERA5 were acquired from the four grid points closest to the floe. This is to avoid inclusion of wind data, which are usually highly variable, from grid points far from the relevant floe location. The 3-hour wind vectors that acted on the given ice floe during its transition during the period of the two satellite passes are produced in the form of polar maps to qualitatively explore their influence on the floe drift. In addition, the 3-hour wind speed from the four grid 165 points around each floe at each location were averaged to quantitatively link it to the drift speed and generate statistics to quantify the wind influence.
The evolution of the ice arch during its formation period from 24 January to 1 February 2017 was manually delineated in each image. The daily displacement of its two end points was calculated using their latitude/longitude coordinates. For 170 investigating the role of the wind on the progress of the arch shape as well as the location and displacement of its terminal points, daily wind data from ERA5 reanalysis were averaged from the 3-hour interval at the coloured grid points from lines 3, 4 and 5 in Fig. 3.

Results
Two  Figure 4 shows also that ice floes never entered any fjord at any side of the channel. In fact, many fjords become filled by locally grown landfast ice early in the freezing season.

Driving forces of ice floe motion
As mentioned in Sect. 1, ice dynamics at the floe and multiple-floe scales may not be strongly coupled with wind forcing. It is rather triggered by a combination of wind, ocean current, internal stress within the pack ice, Coriolis force and sea surface tilt. While several studies have addressed the influence of wind forcing on the large-scale motion of the pack ice (which are typically coherent), results presented in this section address the influence of the wind, given the associated ice conditions, on ice drift. Isolating wind from other influences is a challenging task, particularly within a closed pack ice such as the one 230 encountered upstream of the RC. An attempt to achieve this task using a modelling approach is presented in Thorndike and Colony (1982) and Kimura and Wakatsuchi (2000). As mentioned above, links between wind and ice floe drift were explored in two ice regimes; upstream and within the RC separately.

235
Synoptic information of the ice regime north of the RC is presented using the six selected scenes shown in Fig. 6. 245 Figure 6 shows that ice which entered the RC follows a path coming around the northern section of Ellesmere Island as shown by the arrow in the image of 1 December. No ice appears to enter of the RC coming along the coast of Greenland.
The pack ice motion is driven by both the strong northerly winds, which is channelled down the atmospheric pressure gradient from the Lincoln Sea to Baffin Bay (Gudmandsen, 2000) and the ocean current. Additionally, it might be driven by    Table 2). Here, once again, the path was ice-free. During the short period between 30 September and 1 October when south-easterly wind blew at near 30 km h -1 , ice drift accelerated to nearly 9 km d -1 . After 1 October the wind abated but the floe drift continued in southeast direction at a moderate speed between 2-6 km d -1 . When wind blew northward again between 4-7 October (exceeded 40 km h -1 during the 285 first 3 days then became <30 km h -1 ), floe drift did not follow the wind action in the first two days as the two floes were surrounded by high ice concentration. Nevertheless, northeasterly drift is observed between 4-5 October, particularly of floe #2 (Fig. 8), following the strong south-easterly wind during the same period (Fig. 9). This case study demonstrates the effective role of wind on ice floe drift when surrounded by thin ice or water along the wind direction.    Fig. 8) during period between acquisitions of two successive Sentinel-1 coverage.

300
The period is shown in the first column.

2) ice drift within RC
Two features make ice drift inside the RC different from the drift upstream of the RC. The first relates to the ice cover and the second is associated with the ocean current. Prior to the formation of the ice arch on 24 January 2017 (see next section), 305 the channel was filled with ice floes transported from the north but the ice concentration was moderate. Such regime is called "drifted ice", a term normally used when ice concentration is less than 60 %. This contrasts the term "pack ice" used when The second feature is about the dominant southward current in the channel (Fig. 7) that reaches speeds between 0.72 and 1.38 km h -1 . This would be powerful enough to influence the floe drift either before or after the arch formation. Although 315 daily and hourly ocean current data in the RC are available from GLORYS12V1 and PSY4 , it was not possible to explore the relative weights of wind and current forcing on ice drift. An ice dynamic model would be more suitable for this purpose. Nevertheless, the influence of the wind is described qualitatively below with few quantitative data.
Within the RC, ice floes advance along the channel's direction (heading mostly southward but occasionally northward as 320 shown in the track of floe #29 in Fig. 4). Such nearly linear path made it easy to explore links between drift and wind speed.
This was performed by considering their speed components along the channel's length. Scatter plots of these two components are presented in Fig. 10. Positive values indicate motion northward and vice versa. It can be seen that southward-blowing wind (i.e. northerly wind) is always associated with southward ice drift. The situation is different when the wind blows northward. In this case, ice floes may remain drifting southward (influenced by the current). However, as 325 wind accelerates, the ice may eventually drift northward. This is shown in the reversed path of floe #29 from 10 to 18 November (Fig. 4) and explained in the discussions below.  In order to assess the impact of surface wind on ice drift along the RC direction while taking into consideration the drift 350 direction (northward or southward), a scatter plot of the daily change of wind speed (Δꀀ㌳) and floe speed (Δ ㌳) (signed data) was generated (Fig. 11). This is calculated as the speed (of wind or ice) in a given day minus the speed in the previous day.   Figure 12 shows a sequence of daily Sentinel-1 images from 14 to 19 November where many ice floes originated from the north of RC are seen. The path the floe marked with the asterisk (floe #29) is linked to the coincident wind vectors in Fig. 13.

370
There is a remarkable displacement of the floe between 14 and 15 November at a calculated drift speed of 27.0 km d -1 . The wind speed was between 10 and 20 km h -1 and varied over a wide range of angles (north to south as shown in Fig. 13).

375
Apparently, this drift was not as much influenced by the wind as it was by the impact of the incoming floes in such a high ice https://doi.org/10.5194/tc-2020-44 Preprint. Discussion started: 4 March 2020 c Author(s) 2020. CC BY 4.0 License. concentration regime. Between 15 and 16 November, relatively southerly wind blew at speed between 20 and 37 km h -1 but its effect was neutralized, once again, by forces from the incoming ice flux. The entire set of floes appear to drift eastward.
Recall that SSH has a gradient (climatically speaking) that matches this drift direction (Fig. 7). Between 16 and 18 November the same persistent southerly wind continues but it overcame the other effects. The entire set of floes drifted 380 northeast following the wind. The speed of floes #29, indicated in the relevant segments, was highest between 16 and 17 November (11.77 km d -1 ). Between 18 and 19 November there was no wind but the momentum continued to drift along same direction at slower speed.

Case study 3: An ice floe drifted in the polynya within the RC
The track of ice floe #38, which broke off from landfast ice at the Greenland side and drifted north then south in the polynya 395 regime in the RC, is shown in Fig. 14. The track covered the period from 8 to 22 February 2017 (after the arch formation).
The daily wind vector maps associated with selected floe location are presented in Fig. 15. Between 10 and 11 February southerly wind dominated though never exceeded 20 km h -1 . The floe drift (around 12 km d -1 ) matched the wind direction.
Between 13 and 17 February the southerly wind accelerated, reaching 40 km h -1 then 50 km h -1 . This left an impact on the observed floe track that extended northward with floe speed reaching 11.8 km d -1 , 32.2 km d -1 and 20.4 km d -1 on 15, 16 and 400 17 February, respectively. The speed was significantly reduced to 3.7 km d -1 on 18 February as the floe approached the ice arch (a natural barrier). After that day, the wind became northerly and the floe changed its direction of motion to advance southward. It is interesting to note the high floe speed of 43.0 km d -1 between 20 and 21 February and the remarkably highest speed of 99.1 km d -1 between 21 and 22 February. The latter was triggered by the highest wind encountered in this study, gusting to 50 km h -1 . However, it is important to recall that the surface current drives ice motion in the same direction. This case study demonstrates that wind would be the prime driving force of floe motion in a regime of thin ice and combined with the current might set the drift of anomalous speed.

Formation of the ice arch
The ice arch phenomenon is usually associated with the formation of a polynya downstream of the arch. Sometimes it becomes a necessary condition for polynya formation. It is well known that polynya can be driven by wind action that removes newly formed ice (latent heat polynya), and/or warm upwelling ocean water that melts the ice as soon as it is 420 formed (sensible heat polynya) (Smith et al., 1990). However, if continuous advection of ice from a nearby source keeps feeding the area that would-be a polynya, then polynya can only be formed if a natural barrier is developed to block the advection. This obstacle would be an ice arch; a mechanically strong formation that can withstand the massive dynamic load of the advected sea ice. Obviously, this factor is irrelevant to coastal polynyas as they are backed by land (which are more common in the Antarctic region). In the case of the RC, ice arch is usually formed at the inlet of the channel, blocking the ice 425 flux from the Lincoln Sea into the RC. It may collapse a few weeks after formation or persist as late as mid-August (Samelson et al., 2006). The arch observed in the present data set started development on 24 January matured on 1 February and collapsed in May 2017. The mechanism of arch formation is described below. After its initial formation, chunks of ice continue to detach from the arch's contour under action of northerly wind. This alters the arch's shape and the location of its https://doi.org/10.5194/tc-2020-44 Preprint. Discussion started: 4 March 2020 c Author(s) 2020. CC BY 4.0 License.
terminal points on the two constriction points at the Greenland and Ellesmere Island sides. The sequence of the development is revealed in the set of Sentinel-1 images in Fig. 16.
The white dashed line that appear in some panels in the figure represents the arch's contour of the following day. For example, the dashed line in the image of 24 January represents the arch's contour that appears in the image of 25 January and so on. No line is presented if the contour remains unchanged in the following day (e.g. the cases of 25 and 26 January).

435
The difference between the visible arch and the dotted line in the image of any given day identifies the ice piece that was chopped by the wind on that day. The 3-hour wind vectors during the period between two daily overpasses of Sentinel-1, obtained from ERA5 reanalysis, are presented in Fig. 17. Data were obtained from the grid points located on lines 3, 4 and 5 in Fig. 3 and shown in the same color as appear in that figure.

440
After seven days of persistent southerly wind, northwesterly wind returned for a few hours between 22 and 23 January. A wide rupture of ice cover, not arch-shaped, is observed in the 23 January image (Fig. 16). Between 23 and 24 January, the dominant northerly wind (about 30 km h -1 ) was enough to cause many cracks in the ice cover and introduce the first visible contour of the arch (image of 24 January). On 25 January, the cracked ice drifted south and another piece of ice was detached from the arch. On that day light southerly wind occurred (<20 km h -1 ) but varied over the entire angular range.

445
Hence the ice drift must have been triggered by another cause (likely sea current). The same wind prevailed until 28 January and no change in the arch was observed. With the continuation of the same wind on 29 January, numerous cracks appeared on that day. When strong northerly wind (reaching 30 km h -1 ) blew between the two satellite overpasses on 29-30 January, the cracked ice was pushed further south, leaving a well-defined arch shape seen in the image of 30 January. A major displacement of the arch's end point at the Ellesmere Island's side (13.88 km) is observed. The arch shape continued to be 450 adjusted on 31 January and 1 February in response to the same strong northerly wind. Note the two large pieces of ice that detached in these two days (Fig. 16). After 1 February, the arch remained unchanged, regardless of the wind speed and direction, until it collapsed on 11 May. The only exception was the breakup of a large piece, defined as floe #39, on 5 March (Fig. 18). This piece broke while the wind was dominantly southerly (between 15 km h -1 and 30 km h -1 ). This is not consistent with the aforementioned scenario of the modulation of the arch shape under the action of northerly wind. However, 455 it should be noted that the rest of ice cover in the image of 5 March (Fig. 18) appears to shift north following the southerly wind as indicated by the arrow.
The arch legs (an engineering term that refers to the end parts of the arch) are observed to be perpendicular to the land contour (see images of 1 and 2 February in Fig. 16). As all the forces in an arch geometry are transferred as a compression 460 forces, the perpendicular ending of the arch shape provides a robust way to transfer the loads directly to the rock base at both sides. This prevents the arch from collapsing (Karnovsky, 2012). Without this configuration, the end point of the arch may   Once an ice floe crosses the entry to the RC and becomes released from the stresses engendered by the surrounding ice it 500 starts to accelerate. While inside the channel, the floe drift speed varies between 15 km d -1 and 45 km d -1 , following the channel's direction (mostly southward but sometimes northward, depending on the wind direction). This nearly linear motion made it easier to explore links between wind and floe drift when components along the channel's length are considered. Regression analysis between these components confirms the increasing influence of the wind on ice floe drift https://doi.org/10.5194/tc-2020-44 Preprint. Discussion started: 4 March 2020 c Author(s) 2020. CC BY 4.0 License.
when the ice cover features thin sheet and water. Change of wind vector is found to be linearly related to the corresponding 505 change in ice floe drift within the RC with less variability in the data after the formation of the ice arch.
Monitoring the ice arch formation revealed its development over a 9 day period from 24 January to 1 February 2017. During this period the arch's shape and its terminal points continued to adjust as northerly wind keeps chopping pieces of the ice cover upstream of the arch at locations of fractures near the arch's contour. Southerly wind has no role on this process as it 510 closes gaps that potentially lead to ice detachment. The process continues until the pack ice upstream of the arch becomes fully consolidated and the arch takes on a mechanically strong dome shape.
This study demonstrates the possibility of generating a non-gridded ice drift vector by tracking the drift of individual ice floes in daily SAR images. Such images are recently available, yet on a limited basis, from Sentinel-1 system, and more so 515 from the recently launched Radarsat Constellation Mission (RCM) (a three-spacecraft fleet). More availability of data from constellation systems is expected in the future from a few national and commercial agencies. The challenge in developing this product remains is developing an automated tracker of individual floes, considering their deformation, rotation, breakup and amalgamation while drifting. 520 Data availability. Sentinel-1 data is available free of charge from the Copernicus Open Access Hub (https://scihub.copernicus.eu/dhus/#/home, European Space Agency, (last access: 2 January 2020). All reanalysis data are publicly available. ERA5 hourly data is available from https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5- Author contribution. MES prepared the manuscript, designed the experiments and performed data analysis. ZHW collected and processed the data. TTL contributed to performance of data analysis and supported writing and editing.