Sea-ice thickness on the northern Canadian polar shelf: A second look after 40 years

This paper presents a systematic record of multi-year sea-ice thickness on the northern Canadian polar shelf, acquired during the winter of 2009-10. The data were acquired by submerged sonar positioned within Penny Strait where they measured floes drifting south from the notional “last ice area”. Ice was moving over the site until 10 December and fast thereafter. Old ice comprised about half of the 1669-km long survey. The average old-ice thickness within 25-km segments of the survey track was 3-4 m; maximum keels were 12-16 m deep. Floes with high average draft were of two types, one 10 with interspersed low draft intervals and one without. The presence or absence of thin patches apparently distinguished aggregate floes comprised of sub-units of various ages and deformation states from units of more homogeneous age and deformation state. The former were larger and of somewhat lower mean thickness (1-5 km; 3.5-4.5 m) than the latter (400600 m; 6.5-14 m). Calculated ice accretion onto the multi-year ice measured in autumn 2009 was used to seasonally adjust the observations to a date in late winter, when prior data are available. The adjusted mean thickness for all 25-km segments 15 with 4 tenths or more old ice was 3.6 m (sample deviation of 0.4 m), a value indistinguishable within sampling error from values measured in the same area during the 1970s. The recently measured ice-draft distributions were also very similar to those from the 1970s.

more than 6 m at some places near the northern coastlines of Greenland and the Canadian Archipelago. This spatial pattern of ice thickness in the Arctic has been attributed to the combined influences of the long freezing and short thawing seasons at high latitude (Maykut and Untersteiner, 1971) and of pack-ice convergence onto these windward shorelines (Tucker et al., 30 1979); the convergence maintains high ice concentration and enhances the formation of thick ice rubble via material failure under pressure (Kovacs, 1972).
A compact area of multi-year sea ice has persisted on the North American side of the Arctic during the last quarter century while the area covered by multi-year ice in the Arctic Ocean has decreased considerably (e.g. Comiso, 2012;Kwok, 2018).
Moreover the numerical models informing the 5th IPCC Assessment report suggest that the same geographic distribution 35 will continue into the future (Collins et al., 2013;Huard and Tremblay, 2013). This evidence has fostered the concept of a future remnant "last ice area" within and adjacent to the north Greenland and Canadian polar shelves (Pfirman and Tremblay, 2009;Pfirman et al., 2010;Hamilton et al., 2014); this area is thought to be ecologically important as a future refugium for ice-reliant Arctic wildlife from bacteria to megafauna (http://www.wwf.ca/conservation/arctic/lia/ [Nov 2021]).
Although clearly of concern, the notion of a threatened last area of ice, now rapidly thinning, is based on remarkably few 40 data from that area, past or present (Lyon, 1984;Melling, 2002;Lindsay and Schweiger, 2015). In the compilation by Lindsay and Schweiger (2015) of Arctic ice thickness observations acquired since 1975 there are points since 2003 that lie within the north-west part of the area of most persistent ice, but unfortunately this region lacks comparative data from earlier times. Moreover the recent data from this area are best viewed cautiously because they are measures of ice-plus-snow elevation above an infrequently glimpsed sea level, from an Earth satellite with a relatively large footprint (20-70 m 45 diameter); in situ validation has been minimal (Kwok et al., 2020).
Information on sea ice is also sparse within the adjacent part of the notional last ice area that overlies the Sverdrup Basinthe geological name for the northern Canadian polar shelf. The extent of ice here and its composition by ice type have been monitored and charted by the Canadian Ice Service since about 1970, to provide guidance for navigation during August and September. The drift of ice through the area has been determined sporadically by following identifiable floes in satellite 50 images and by satellite tracking of beacons on floes that fortuitously entered the area. Knowledge of ice thickness is based solely on bore-hole measurements, weekly single-point values near Canadian weather stations since the 1950s and on survey transects of bore-hole measurements acquired during on-ice seismic surveys during the 1970s. There are in addition a handful of targeted surveys of extreme features. Available data have been reviewed by Melling (2002). More recent publications have been focused on ice kinematics and concentration, including fast ice (Galley et al, 2012;Howell and 55 Brady, 2019), systematic estimates of ice inflow from the Arctic Ocean via feature tracking in satellite images (Alt et al 2006;Kwok 2006;Agnew et al 2008;Wohlleben et al., 2013), decadal fluctuations in multi-year ice concentration (Howell et al 2008(Howell et al , 2010(Howell et al , 2013(Howell et al , 2015Tivy et al 2011) and summertime deterioration of multi-year floes in the area (Dumas et al 2007;Howell et al 2009).  shelf, annotated with the names of its larger islands and waterways. The basin is an area of consistently high ice concentration, with old ice comprising 85-90% of the ice present, except in the area north of Bathurst Island (70%) and in Norwegian Bay (50%) 65 (Melling, 2002). Some of this ice enters from the Arctic Ocean in the north-west during the few months of mobility in late summer and autumn; this ice tends to be highly deformed and extremely thick. A second population develops within the basin from areas of 70 first-year ice that do not melt during the brief summer (Flato and Brown 1996) journey across the basin takes 3-5 years because the ice is land-fast between the High Arctic islands for up to ten months per 80 year. The time taken to cross the basin varies with conditions of ice congestion, wind and temperature in summer.
The average thickness of sea ice tends to decrease towards the south-east. In the 1970s, survey-line means as high as 5-7 m were measured near the north-western margin of the polar shelf and as high as 3-4 m near the exits to the North West Passage (Fig. 2). Melling (2002) proposed that the cause of this progressive thinning with distance from the outer shelf was a small sustained heat flow to the ice from the ocean; this mechanism that has been further documented by Melling et al. 85 (2015).
Future severe ice conditions on the Canadian polar shelf are critically dependent on whether all multi-year ice disappears from adjacent areas of the Arctic Ocean; cut off the supply and the waters between the Queen Elizabeth Islands could be a very different place. It is useful to note that palaeoclimatic indicators reveal that there was heavy ice in 95 this area in summer even during the Holocene Warm Period, 5000 years ago (Dyke et al 1996;Dyke and England 2003), when Arctic climate was 2-3°C warmer than that during the 20th century.
The thickness of multi-year ice presently over the 100 northern polar shelf is not known; there have been no observations since the late 1970s. Seeing the declining ice thickness of the central Arctic (Lindsay and Schweiger, 2015) one might anticipate that multi-year ice over the northern polar shelf has also thinned 105 dramatically. If so, the feasibility of exploiting the proven natural gas reserves of the Sverdrup Basin would be greatly enhanced. Severe ice conditions here in the 1970s were critical in limiting expeditious supply of the exploration activities, and of great concern in any plan to develop the reserves and to ship out produced and liquefied natural gas by tanker. 110 There is also motivation from the perspective of ice dynamics. A range of factors appear influential in the loss of ice from the central Arcticthe pattern of ice circulation, reduced residence time, albedo feed-back, warming ocean, warming atmosphere, poorer recruitment of second-year ice. The seas of the Canadian High Arctic are relatively isolated from the Canada Basin, but they do harbour some of the heaviest multi-year ice in the Arctic. Study of ice here, where gain and loss factors for sea ice may be different, could provide new perspectives on sea ice and climate change. 115 The goal of this study was a timely second assessment of ice within the notional last ice area. Perhaps paradoxically, we chose not to work in the area itself but rather to observe ice as it was leaving via channels along its southern margin. The reason, as common in Arctic field science, was logistical. Access to the interior of the Sverdrup Basin would have been very costly by icebreaker and perhaps not practical even in summer. Use of aircraft in winter would have been cheaper but contingent on the presence of first-year ice suitable for landing ski-planes, which is scarce here. These factors provided a 120 strong motivation for making observations at the most southerly useful locations. We chose Penny Strait for the first year of observation with an eye further west on Byam Martin Channel in future years. Because inter-annual variation was likely to be appreciable, we planned to sustain observations over several years. Drilling or electromagnetic induction radar (Haas et al., 2009) can provide ice thickness transects at good (better than 25 m) 125 spatial resolution but require human presence; these methods are impractical for year-round sea-ice observation at isolated locations. We chose to use ice-profiling sonar (IPS), a submerged up-looking sonar optimally designed for measuring underice topographic profiles. The IPS can operate autonomously for as long as three years (Melling et al., 1995). Moored at a fixed location, it measures the submerged depth of ice (draft) as the difference between the depth of the instrument (from a pressure measurement) and its distance from the ice bottom (from an echo). The IPS is used in combination with Doppler 130 sonar (ADCP) which measures ice-drift velocity; together the instruments provide data from which a detailed topographic transect of multi-year ice floes, ridge keels and leads can be constructed. The spatial separation of survey points is proportional to ice-drift speed, so that when ice becomes immobile the IPS is limited to measuring the slow local growth and decay of ice stalled above it.
These instruments were placed on a submerged mooring ( Fig. 3) with the ADCP near the seabed and the IPS at 45-m depth, 135 safe from moving ice. Because there was no suitable geomagnetic reference for ice-drift direction at this locationthe geomagnetic inclination was 88.3° -the section of the mooring beneath the ADCP was torsionally rigid (i.e. no twisting). All ice-drift measurements by the ADCP were made relative to its fixed, but unknown, heading that was later deduced during the processing of data from the directions of measured tidal currents. This method has been used successfully since 1998 for oceanographic moorings 140 in Cardigan Strait, Hell Gate and Nares Strait (Münchow and Melling, 2008).
The mooring was placed in Penny Strait, on the western side that is favoured by multi-year-ice leaving the northern polar shelf; the south-moving stream of 4-9 tenths' pack ice on this side that was our observational target can be 145 seen in the late-summer ice-concentration map from the Canadian Ice Service (Fig. 4).
Ironically these same streams pose challenges for the deployment and recovery of moorings. We were able to deploy our mooring from an icebreaker (CCGS Henry Larsen) within this fast moving heavy ice. 150 However for recovery, the ship needs to reach and hold station at the predetermined location of the mooring, requirements that would have been difficult under the conditions encountered at deployment. We retrieved the mooring in late May 2010, operating from the surface of fast ice accessed by aircraft. Details of the data collection are summarized in Table 1, Table 2  155 and Table 3   ice within the pack. The limiting circumstance, which occurs every winter in this area, is 10 tenths ice held immobile by 185 coastal geometry. With fast ice established after mid-December, the wind could move neither ice nor water. In September and again in November, the direction of prevailing wind was north-west, along strait, moderate in speed in September but weak in November. During October and December, wind blew from the north-east, across the strait, moderate in October and strong in December.

Ice drift with wind and current 190
Sea-ice drift was derived via Doppler analysis of sound scattered by ice back to the ADCP when operating in bottomtracking mode (Melling et al., 1995). Ice tracking was independent of current measurement by the same instrument, except for the determination of direction. The fixed heading of the sonar throughout the deployment (ensured by the torsional rigidity of its mooring) was determined by aligning the ebb and flow directions of the measured tidal current with those predicted by a tidal model (https://www.bio.gc.ca/science/research-recherche/ocean/webtide/index-en.php [Nov 2021]). 195 Sampling error, which dominates uncertainty in ice velocity measured by the Doppler method, was better than 0.5 cm s -1 for a 15-minute average value with our operating configuration.
The progressive vector of ice drift over the mooring, approximating the ice-drift trajectory through the strait, is displayed in  October 6-8. Drift to the south-east occurred during the intervening periods, except that ice was almost motionless during November 5-15. Most of the net south-east displacement occurred with prevailing north-west wind during September 5-28 (251 km) and October 15-27 (374 km).

Draft and underside topography of sea ice
The draft of sea ice was derived from IPS' measurements of water pressure and echo time (Melling et al., 1995). Each draft 220 value was assigned a position along the progressive vector calculated by integration of ice velocity. Because the IPS's field of view is about 0.8 m, redundant data were eliminated by re-sampling the ice-draft at 1-m increments along the drift vector.
The resulting quasi-spatial transect of under-ice topography, 1.669 million data points over the 1669 kilometres, is the starting point for scientific interpretation in this study.
Statistics of ice draft for each 25-km segment of the drift path are displayed in Fig. 8. Although sonar measures draft not 225 thickness, axes showing thickness have been added. A 1.13-ratio of thickness to draft has been used for the minimum and mean values, based on the reasonable assumption of near-isostatic balance at scales of 100 m and more (Melling et al, 1993).
This ratio is consistent with densities of 1025 kg m-3 for seawater, 910 kg m-3 for submerged ice and 870 kg m-3 for subaerial ice (Timco and Frederking, 1996).
Because ridges and keels are not locally in isostatic balance, the ratio of thickness to draft is variable on small (3-30 m) 230 scales in deformed ice (Hibler 1972;Melling et al., 1993). Nonetheless the ratio of maximum draft to maximum height is commonly used as a measure of the thickness of ice ridges, even though the ridge crest may not exactly coincide with the keel. In first-year ice, the ratio is quite high; Kovacs (1972) and Sisodiya and Vaudrey (1981) report 4.5 and 5.5 respectively. The ratio is lower for multiyear ice where thermal weathering and 235 consequent consolidation affect keel draft more than sail height (Amundrud et al 2006). Reported ratios of maximum draft to maximum height for multi-year ice features are 3.1-3.3 (Cox, 1972;Kovacs 240 et al, 1973;Kovacs et al, 1975) The distinction between the multi-year and young-ice populations is clear in the distribution of pack-ice volume versus draft (Fig. 9). These are shown for the entire survey track and for the 25-km segments with draft greater than 2.5 m and less than 1.5 m. The distribution for all ice 265 has a peak near zero thickness, a peak near 3-m thickness and a long  tail to beyond 20 m. The curve for segments with high mean draft shows clearly that the peak at 3-m thickness and the long tail are associated with old ice at this time of year, while the curve for low mean draft is clearly linked to young ice that is only lightly deformed.

Evidence for change in multi-year ice of the Canadian High Arctic
As mentioned earlier, the only prior data on sea-ice thickness in this area were acquired during on-ice seismic surveys during the 1970s and discussed by Melling (2002). The surveys were conducted in the late winter (mid-March to mid-May) when the ice was thick enough to support the sled trains used to move 275 personnel, accommodations and equipment around this remote area. There is value in comparing the 1970's data to the present observations. However because the present data were acquired in autumn and the 1970s data in late winter, a comparison is informative only after adjustments that reflect the seasonal cycle of ice growth and melt. The adjustment of ice thickness from late summer to late 280 winter is more tractable than the converse, because the rate of ice thickening by freezing is better understood than its thinning by melting. Moreover, it is reasonable to assume negligible ice thickening via ridging in the area north of Penny Strait because islands limit the build-up of wind stress and inhibit the ice movement needed to build ridges during most of the winter. 285 Ice growth by freezing is driven by the temperature difference between the snow or ice surface and the ocean and limited by the rate at which heat can be conducted through the ice and snow; the rate of growth slows as the ice and snow thicken. We have calculated wintertime ice growth under High Arctic conditions using 15-day averages of climatological data for air temperature and snow depth at Resolute Bay and Eureka to drive a steady state ice-growth model; assumed values of thermal conductivity were 2.03 W/m/K for sea ice and 0.31 W/m/K for snow. Results are shown in Fig. 10     and traceable to the one dimensional heat-budget model of Maykut and Untersteiner (1971). Using this tool, it can be shown that first-year ice in the colder parts of the marine Arctic can survive its first melt season and gradually evolve into multiyear ice if the cold-season accretion of ice via latent heat loss through the top surface exceeds the warm-season ablation caused by absorption of radiative and oceanic heat. Because the latent heat lost from the top of the ice must diffuse through it from the bottom, its flux is inversely proportional to ice thickness whereas heat gains during the thaw season are proportional 350 to surface area not thickness. As the ice grows thicker, the amount of new-ice accretion during winter decreases; a quasiequilibrium maximum thickness is reached when winter's gains match summer's losses. This maximum is about 3 m for an average seasonal cycle of climatic forcing in the High Arctic (Maykut and Untersteiner 1971;Flato and Brown 1996).
Calculations show that increasing the maximum above 3 m by varying climatic forcing within observed limits is difficult, but that increases in snow depth and ocean heat flux can readily decrease it. 355 Paradoxically a large fraction of the perennial ice mapped for the 1970s by Bourke and Garrett (1987) was appreciably thicker than the 3-m thermodynamic limit. Even then it was obvious that this much thicker ice must have formed by a different process. Narrow zones of much thicker ice (ridges) formed via the piling of fracturing of level ice are ubiquitous in pack ice (Wadhams and Horne 1980). Indeed ridged ice ten or more times thicker than the fragments that comprise it forms within a few hours during wind storms (Amundrud et al. 2004). Because subsequent thermal deterioration of such thick ice 360 amounts to a few metres per year, remnants of a 30-m ridge may persist for a decade or more (Amundrud et al. 2006).
However because ridge remnants are separated and relatively narrow (15-50 m) they cannot form the extensive areas of multi-year ice above 3-m thickness mapped by submarine. However, the stamukhi zone forming the interface between fast and mobile ice is a plausible source for the large floes of thick multi-year ice, past and present. Here cyclic offshore/onshore and shearing movements of the pack driven by storms 365 facilitate the creation of young ice in flaw leads and its subsequent compression into broad expanses of very thick ice rubble near the grounding line (cf. Kovacs and Mellor 1974). The patterns of wind and ice circulation generate the highest ice pressure and rubble-forming potential along the North American side of the Arctic margin (Thorndike and Colony 1982;Colony and Thorndike 1984), particularly against northern Greenland and the western Canadian Arctic Archipelago. This inexorable compression against the coast is evident (Fig. 14) in the drift-pattern of ice revealed by satellite-tracking (IABP, 370 2020), which displays large along-shore movements in both directions that amount to relatively little net displacement alongshore, interspersed with smaller perpendicular movements that are shoreward on average. properties similar to those of past decades. We suggest that thick-ice genesis as stamukhi explains, despite changing climate, the similarity between our 2009 ice thickness data and those from the 1970s in the Canadian High Arctic.  Figure 15 illustrates how averaging alters the probability density of ice draft: the likelihood of both thin and thick ice decreases as the averaging scale increases while the likelihood of near-average ice increases; the multi-year-ice mode becomes more prominent and shifts to higher draft because the ice-draft histogram is positively skewed. These differences are important to bear in mind when interpreting data from low resolution sensors 415 carried by aircraft and satellites.
The Rayleigh criterion was applied to this smoothed ice-bottom relief to detect expanses of ice that were unusually thick on a 250-m scale; the reference level for draft, -2 m, enabled identification of the thickest parts of thick floes; peaks in draft were counted as separate features if intervening values dropped towards this reference level by more than half the peak value. This is an unorthodox use of the Rayleigh criterion, otherwise long established as a tool for identifying ridge keels in well 420 resolved ice-bottom topography (e.g., Wadhams and Horne, 1980). The boundaries of floes containing these features were taken to be those locations where the 250-m averaged draft crossed 1.5-m depth. The length of the measured chord on each floe was taken to be the length of the survey path between these cross-overs; the mean draft of each floe was the average of all draft values in the measured (not smoothed) data between these points.  Figure 16 displays a segment of the smoothed ice-bottom relief, the "keels" identified within that segment and the edges of 425 the floes in which they were embedded. These so-called "keels" are actually broad expanses of thicker ice, and despite their similarity in a vertically exaggerated view, they not associated with individual pressure ridges (Fig. 17). Figure 18 displays the distribution along the survey path of floes potentially hazardous to navigation. The hazard is represented by an index of severity equal to the area of the floe's cross-section associated with ice thicker than 4 m.
Although thick ice and large floes both present hazard, the index does not distinguish between short transects of very thick 430 ice and long transects of thinner ice. Table 4 compares the 10 floes with the highest severity and the 10 thickest floes. The two groups of floes differ by an order of magnitude in chord length (high index floes being larger) and by 50% in mean thickness (high index floes being thinner), but do not differ significantly in 1-m and 250-m scale maximum thickness. The scatter-plots in Fig. 19 showing all 177 floes suggest that the maximum 250-m averaged thickness of each floe had a weak tendency to increase with the size of the floe and a strong tendency to increase with its mean thickness. Within the full 435 sample of 177 floes (plot not shown), the mean thickness of floes did not vary systematically with floe size.
The data herein provide an opportunity to estimate how frequently floes such as those carefully measured by Johnston (2011Johnston ( , 2019 were likely to be encountered in the Canadian High Arctic in the late 2000s. Autonomous sonar is impartial in measuring all floes that pass overhead, whereas bore-hole surveys are selective. Johnston sought suitable floes during reconnaissance flights by helicopter and selected the most rugged for systematic measurement via drilling. She worked in the 440

Conclusions
Ice-profiling sonar installed in Penny Strait during the 2009-10 winter have provided modern ice-thickness data from the northern Canadian polar shelf, for comparison with prior bore-hole surveys during the 1970s. The two data sets provide a perspective on change within the notional last ice area of the marine Arctic. 470 The ice pack in Penny Strait was a mix of multi-year floes from further north and young ice formed locally. The latter occupied about 60% of the pack ice surveyed because the densely packed south-drifting multi-year floes spread apart as they  The observations in 2009 were acquired at the end of the thaw season, whereas comparative data from the 1970s were acquired late in the freezing season. To allow inter-comparison, the modern ice-draft data were converted to thickness and then seasonally adjusted for wintertime ice accretion. The resulting mean estimated thickness for multi-year in May 2010 was 3.6 m (σ=0.4 m). The average thickness from ten bore-hole lines measured between 1972 and 1980 within 100 km to the northwest of Penny Strait was also 3.6 m (σ=0.6 m). Moreover, the 2010 distribution of ice volume as function of thickness 480 was not obviously different from that in the 1970s in terms of their modes and roll-offs with increasing thickness.
The apparent lack of change in thick multi-year ice near Penny Strait after 40 years of Arctic warming suggests examining its genesis not its wastage. We propose that the formation of Arctic multi-year ice floes thicker than about 3 m has been and continues to be driven by pack-ice pressure rather than by surface thermal energy balance. The build-up of pressure sufficient to build large ridges occurs within a few hundred kilometres of coastlines during strong onshore wind storms. The 485 occurrence of such weather is not obviously reduced by rising Arctic air temperature.
Where such winds prevail, specifically off northern Greenland and the Canadian Archipelago, ridges accumulate into wide bands of thick ice rubblestamukhi zones. Ice rubble piled to a thickness of tens of metres takes many years to deteriorate in the cold High Arctic climate. Meanwhile it evolves into the very thick old ice still found along this coast and for a few hundred kilometres in down-drift directions across the Canadian polar shelf and into the Beaufort and Greenland Seas. 490 The northern Canadian polar shelf continues to present serious ice hazards to navigation. 177 floes of interest were identified within the 739 km of ice having draft greater than 0.95 m. Their danger to navigation was assessed using 4 metrics: chord length, maximum 250-m-average thickness, maximum thickness and the cross-sectional area of ice thicker than 4 m. The 20 most dangerous floes fell into two groups based on these metrics, those with the most ice thicker than 4 m and those with the highest mean thickness. The first tended to have modest mean thickness (3.4-m group average) and large chord length (12.5-495 km group average); the second had much smaller chord length (1.3-km group average) and, obviously, high mean thickness