Arctic Ocean glacial history

While there are numerous hypotheses concerning glacial e interglacial environmental and climatic regime shifts in the Arctic Ocean, a holistic view on the Northern Hemisphere ’ s late Quaternary ice-sheet extent and their impact on ocean and sea-ice dynamics remains to be established. Here we aim to provide a step in this direction by presenting an overview of Arctic Ocean glacial history, based on the present state-of-the-art knowledge gained from ﬁ eld work and chronological studies, and with a speci ﬁ c focus on ice-sheet extent and environmental conditions during the Last Glacial Maximum (LGM). The maximum Quaternary extension of ice sheets is discussed and compared to LGM. We bring together recent results from the circum-Arctic continental margins and the deep central basin; extent of ice sheets and ice streams bordering the Arctic Ocean as well as evidence for ice shelves extending into the central deep basin. Discrepancies between new results and published LGM ice-sheet reconstructions in the high Arctic are highlighted and outstanding questions are identi ﬁ ed. Finally, we address the ability to simulate the Arctic Ocean ice sheet complexes and their dynamics, including ice streams and ice shelves, using presently available ice-sheet models. Our review shows that while we are able to ﬁ rmly reject some of the earlier hypotheses formulated to describe Arctic Ocean glacial conditions, we still lack information from key areas to compile the holistic Arctic Ocean glacial history. (cid:1) 2013 The Authors. Published by Elsevier Ltd.


b s t r a c t
While there are numerous hypotheses concerning glacialeinterglacial environmental and climatic regime shifts in the Arctic Ocean, a holistic view on the Northern Hemisphere's late Quaternary ice-sheet extent and their impact on ocean and sea-ice dynamics remains to be established. Here we aim to provide a step in this direction by presenting an overview of Arctic Ocean glacial history, based on the present state-of-the-art knowledge gained from field work and chronological studies, and with a specific focus on ice-sheet extent and environmental conditions during the Last Glacial Maximum (LGM). The maximum Quaternary extension of ice sheets is discussed and compared to LGM. We bring together recent results from the circum-Arctic continental margins and the deep central basin; extent of ice sheets and ice streams bordering the Arctic Ocean as well as evidence for ice shelves extending into the central deep basin. Discrepancies between new results and published LGM ice-sheet reconstructions in the high Arctic are highlighted and outstanding questions are identified. Finally, we address the ability to simulate the Arctic Ocean ice sheet complexes and their dynamics, including ice streams and ice shelves, using presently available ice-sheet models. Our review shows that while we are able to firmly reject some of the earlier hypotheses formulated to describe Arctic Ocean glacial conditions, we still lack information from key areas to compile the holistic Arctic Ocean glacial history.
Ó 2013 The Authors. Published by Elsevier Ltd.

Introduction
The glacial history of the Arctic Ocean involves the build-up and decay of marine-based ice sheets on the continental shelves, the development and disintegration of ice shelves, and significant changes in ocean-circulation regimes and sea-ice cover. None of the four other world ocean areas experienced such dramatic physiographic and environmental changes as the Arctic Ocean through the Quaternary glacialeinterglacial cycles. This has been recognized for nearly a century, but a lack of direct field observations led to several postulated, rather contradictory, hypotheses concerning glacialeinterglacial cycles in the Arctic Ocean (Donn and Ewing, 1966;Broecker, 1975;Hughes et al., 1977). When early hypotheses on the environmental setting of the glacial Arctic were developed, i.e. suggesting on one the hand an extensive ice shelf in the central Arctic Ocean during glacial periods (Mercer, 1970) and on the other hand sea-ice free conditions (Donn and Ewing, 1966), there were few field data available to test them. The modern perennial Arctic Ocean sea-ice cover has delayed data collection, but during the last decade, in particular during the last years with substantially reduced summer sea-ice extent, data collection has increased by an order of magnitude (Polyak and Jakobsson, 2011).
Since the Arctic Palaeoclimate and its Extremes (APEX) program started in 2007 as a continuation of two preceding programs PONAM (The Late Cenozoic Evolution of the Polar North Atlantic Margins, 1988e1994 (Elverhøi et al., 1998a)) and QUEEN (the Quaternary Environment of the Eurasian North, 1996e 2002 (Thiede et al., 2004)) numerous field campaigns to the Arctic Ocean have been carried out (Jakobsson et al., 2010a). These have revealed new insights into the extent of ice sheets on the continental margins bordering the central Arctic Ocean, and how they fed into ice shelves that occupied regions of the central basin. Here we present an overview of present knowledge on the Arctic Ocean glacial history. It is divided into regions where the state-of-the art glacial history is presented based on mapped glaciogenic landforms, sediment stratigraphy and established chronologies (Fig. 1). We address the spatial extent of ice sheets that occupied the northernmost Arctic continental shelves and drained into the Arctic Ocean, including their possible extensions as ice shelves. The central Artic Ocean is treated as a separate region with a focus on how the marine sediment record has documented glacialeinterglacial cycles. We also include recent developments concerning numerical simulations of Arctic ice sheets and glacial paleoceanographic conditions. We focus the overview on the Last Glacial Maximum (LGM) and the maximum extent of Quaternary ice sheets. The overarching question is what do we currently know about the past extent of ice sheets, ice streams and ice shelves, and related oceanographic changes, in the Arctic Ocean during the Quaternary? Which are the most important outstanding questions today? Finally, all data presented in our review allow us to revisit and shed new light on previous hypotheses regarding glacial conditions in the Arctic Ocean.

The glacial Arctic Ocean: hypotheses and theories
In the 1950s and 1960s the driving mechanisms behind the glacial cycles were debated. Following two papers on the topic Donn, 1956, 1958), Donn and Ewing (1966) suggested that a sea-ice free Arctic Ocean was required as a moisture source to build up the northern components of the large Northern Hemisphere ice sheets. Furthermore, they suggested that sea-ice formation in the Arctic Ocean towards the end of glaciation led to ice sheet decay due to an efficient blockade of the moisture required to form precipitation. Their theory implied that changes in Arctic Ocean sea-ice extent were a dominant control on Northern Hemisphere glacialeinterglacial dynamics. The Milankovitch theory advocating an orbital forcing behind glaciale interglacial cycles (Milankovitch, 1920) was at the time beginning to gain ground (Broecker, 1966), but was still being met by scepticism until the 1970s when more deep sea sediment cores became available and dating methods improved (Hays et al., 1976). Although the Arctic Ocean sea-ice extent still is considered a relevant factor for the moisture supply and mass balance of the large Quaternary ice sheets (Colleoni et al., 2009), the overall icee age cycles are believed to be paced by orbital parameters (Imbrie et al., 1992). The first sediment cores from the central Arctic Ocean retrieved from drifting ice stations (Clark, 1971) did not contribute to this original debate due to the irregular preservation of calcium carbonate microfossils that prevented consistent application of key paleoceanographic proxies and hampered the establishment of reliable age models (Backman et al., 2004;Alexanderson et al., 2014).
Another hypothesis that was formulated before much field data were available from the central Arctic Ocean suggested that a vast floating ice shelf covered the deep waters around the North Pole during past glacial maxima. Although previously postulated by Sir William Thomson in 1888 as a likely consequence of a glacial climate, Mercer (1970) was the first to promote this hypothesis based mainly on physiographic analogies between the Arctic Ocean and West Antarctica. Several authors picked up on the ice shelf theory and developed it further (Broecker, 1975;Hughes et al., 1977;Grosswald, 1980;Denton and Hughes, 1981;Hughes, 1999, 2008) (Fig. 2). In its most extreme form, a 1000 m thick ice shelf was hypothesized to have covered the entire Arctic Ocean, even south of the Fram Strait. This massive ice shelf was argued to be a critical stabilizing element, by exerting backpressure, for marine ice sheets grounded on continental shelves and flowing into the Arctic Ocean (Grosswald and Hughes, 1999). The hypothesis of an Arctic Ocean ice shelf was considered extreme by much of the glaciological community, although by the mid-1990s evidence indicating deep grounding of ice, likely derived from ice shelves, began to be documented (Vogt et al., 1994;Jakobsson, 1999;Jakobsson et al., 2001Jakobsson et al., , 2008bPolyak et al., 2001). Subsequently, the mapping of widespread glaciogenic bedforms and the dating of sediment cores retrieved from areas of ice grounding on submarine ridges in the Arctic Ocean, now support the presence of an ice shelf in the Amerasian Basin of the central Arctic Ocean, in particular during Marine Isotope Stage (MIS) 6, w135 ka BP . The development and stability of this ice shelf is supported by a conceptual oceanographic model indicating that the influx of Atlantic water occurred at a much greater depth during glacial periods than today, thus preventing this relatively warm water mass from reaching into the Amerasian Basin where it would cause basal melting of an ice shelf . This pattern of glacial paleo-circulation would in turn mean that ice shelves had a smaller likelihood of developing in the Eurasian Basin of the Arctic Ocean, where they would be exposed to warm inflowing Atlantic water.
Ice shelves may also form through a combination of seaward extension of glaciers and extensive multiyear thickening of landfast sea ice, i.e. sikussak, established along the coast and in fjords (Jeffries, 1992). Bradley and England (2008) postulated that this kind of extremely thick multi-year sea-ice cover developed towards the end of the Last Glacial period as a consequence of a more stagnant Arctic Ocean with a lower relative sea level and a diminished influx of warm Atlantic water. Such pervasive thick sea ice, termed paleocrystic ice, is suggested to have been massively discharged from the Arctic Ocean through the Fram Strait at w11 14 C ka BP (Bradley and England, 2008). The authors propose further that the export of paleocrystic ice at the end of the LGM may have disrupted North Atlantic deep water formation and thereby caused or contributed to the Younger Dryas cold snap. It should be noted that the hypothesis by Bradley and England (2008) is only one of a large number of theories suggested to have caused the Younger Dryas cold event.

Previously published reconstructions of ice-sheet extent
The most inclusive set of compilations of Quaternary ice-sheet extent is in Ehlers and Gibbard (2004), which includes papers dealing with the glacial history of the Arctic (Dyke, 2004;Funder et al., 2004;Hjort et al., 2004;Kauman and Manley, 2004;Zazula et al., 2004;Svendsen et al., 2004b) (Fig. 3). The Eurasian icesheet extent over several time slices was compiled by Svendsen et al. (2004a). They concluded that its maximum extent occurred during MIS 6 towards the end of the Saalian glaciation (w140 ka), and included a northward ice extension from Svalbard onto the Yermak Plateau (Fig. 2). The glaciogenic features behind this conclusion were later found to more likely originate from grounding of large ice shelf fragments exiting towards the Fram Strait Jakobsson et al., 2010b). Whether the Barents Sea Ice Sheet ever extended out from Svalbard onto the Yermak Plateau during the Quaternary remains unresolved. Several  . The general circulation of Atlantic water in the present Arctic Ocean is shown with red arrows (modified from Rudels et al., 2012). Sediment cores discussed in Section 4 and shown in Fig. 11aee  authors have pointed out striking differences between the Quaternary ice sheets, in particular between the Eurasian components of LGM (Late Weichselian) and MIS 6 (Late Saalian). While the LGM had a relatively small extension over West Siberia, landforms and glacial deposits suggest that the largest ice sheet during MIS 6 covered much more of this region (Svendsen et al., 2004a). The answer to why the MIS 6 ice sheet generally grew so large has been suggested to reside in the orbital configuration (Colleoni et al., 2011). The incoming solar radiation during MIS 6 springs is 20 W m À2 lower than during the LGM and the springs were also longer. This would allow snow cover to remain longer each season and lower the impact from summer melt.
A noticeable difference from the Dyke (2004) LGM limit of the north-western Laurentide ice sheet (LIS) was proposed by England et al. (2009) who inferred that the entire Banks Island was covered by a predominately cold-based ice sheet (Fig. 3). More recent fieldwork across western Banks Island provides further evidence for the advance of the NW Laurentide Ice Sheet onto the polar continental shelf (Lakeman and England, 2013). Furthermore, the Innuitian Ice Sheet (IIS) coalesced with the LIS to the south and also reached the polar continental shelf during the LGM (England et al., 2006). Although most of the islands were occupied by cold-based ice, at least half a dozen ice streams drained directly into the Arctic Ocean from the LIS and IIS via the deep marine channels of the CAA, supplying significant ice and sediment to the Arctic Ocean during the LGM (Stokes et al., 2005(Stokes et al., , 2006England et al., 2006England et al., , 2009, discussed below. There are not many LGM ice-sheet reconstructions published that include the entire circum-Arctic region and inferred ice extent into the Arctic Ocean. One such early reconstruction is that of Hughes et al. (1977), which contain the hypothesized 1-km thick ice shelf in the Arctic Ocean (Fig. 2). They suggest that all the Arctic Ocean continental shelves were glaciated at LGM including that of the East Siberian, and Chukchi seas (Fig. 2).
The global glacial isostasy model ICE-5G by Peltier (2004) includes ice-sheet extension, thickness, and land topography at time slices from the LGM to present. ICE-5G is based on a theoretical model of Earth's isostatic rebound following glacial unloading as well as input from regional ice margin mapping efforts, like the QUEEN program (Svendsen et al., 2004a) and by Dyke et al. (2002Dyke et al. ( , 2004. ICE-5G has been used to define boundary conditions in many general circulation model (GCM) experiments addressing the LGM climatic and environmental conditions (Abe-Ouchi and Otto-Bliesner, 2009). This means that uncertainties in spatial reconstructions of LGM ice-sheets, as presented in this review, carry forward into the simulation results.

Continental shelves and adjacent bathymetric highs
In this section we describe glaciogenic landforms, stratigraphy and chronology for the regions shown in Fig. 1. We introduce the nomenclature of glaciogenic landforms and their interpretation used in this section in Appendix 1.  Hughes et al. (1977), which includes a 1-km thick ice shelf in the Arctic Ocean. Note that all the Arctic Ocean continental shelves are glaciated including that of the East Siberian, and Chukchi seas. In the minimum version by Hughes et al. (1977), the shallow East Siberian Sea is ice free.

Landforms
The submarine glacial landforms west and north of Svalbard, on the Yermak Plateau, and along the Arctic Ocean margin of the Barents and Kara seas, mapped in Fig. 4, can be divided into those formed in subglacial, ice-contact and glaciomarine environments. The landforms, and landform assemblages, observed mainly from swath-bathymetric data and side-scan sonar imagery are used to make inferences about the extent, flow direction and dynamics of the ice sheet or glacier that produced them.
Streamlined sedimentary landforms occur in most major fjords and cross-shelf troughs west and north of Svalbard (Fig. 4) (Ottesen et al., , 2007Hogan et al., 2010aHogan et al., , 2010b. They indicate past ice-stream flow in the troughs, probably during the Late Weichselian glaciation. Fluting and larger streamlined features have also been observed or inferred from bathymetric data farther east in troughs on the Kara Sea margin (Polyak et al., 1997. Streamlined landforms are particularly well developed in Isfjorden and Kongsfjorden and their cross-shelf troughs (Howe et al., 2003;Ottesen et al., 2007). There is also evidence of their presence on the southern Yermak Plateau, northwest of Svalbard. On the Yermak Plateau, they are interpreted as relict Saalian (MIS 6) features, indicative of transient grounding across the plateau of an ice shelf remnant or an armada of megabergs from the Arctic Basin (c.f. section 2.1). An additional alternative could be the signature of the Fig. 3. Ice-sheet extension during the LGM (white semi-transparent) and Quaternary max (red dotted line). The Eurasian ice limit is from Svendsen et al. (2004a) and the Lau-rentideeInnuitianeGreenland is from the digital GIS files provided with the book Quaternary Glaciations e Extent and Chronology by Ehlers and Gibbard (2004). While several smaller revisions have been made to the shown ice limits, some of the more substantial revisions are indicated with cross hatching; purple (Banks Island) and yellow (Yermak Plateau). New results from the cross hatched area (white) of Chukchi Borderland e Arlis Plateau and East Siberian Sea will likely also lead to future major revisions of ice-sheet extent during the late Quaternary. The LGM topography is from ICE-5G (Peltier, 2004). maximum Quaternary extent of relatively thin ice from Svalbard . Grounding-zone sediments and sets of transverse-to-flow ridges on the north-west Svalbard shelf suggest that ice did not reach the Yermak Plateau during the Late Weichselian (Ottesen and Dowdeswell, 2009;Dowdeswell et al., 2010b). Streamlined landforms in troughs extending northwards from Nordaustlandet indicate that fast-ice flow drained the Austfonna ice cap, likely during the last glaciation .
Mega-scale glacial lineations (MSGLs) and other streamlined sedimentary landforms have been mapped in the northern Barents sea, in Erik Eriksenstretet and Olgastretet, east of Nordaustlandet and around Kong Karls Land (Fig. 4) Hogan et al., 2010a), indicating ice flow into Franz Victoria Trough from eastern Svalbard. This flow direction is almost at rightangles to previously inferred ice flow from a major Late Weichselian ice dome in the central Barents Sea (Lambeck, 1995(Lambeck, , 1996Forman et al., 2004). More work is needed to resolve this apparent discrepancy and much may depend on the dating of the streamlined landforms in Olgastretet and Erik Eriksenstretet (Ingólfsson and Landvik, 2013). Landvik et al. (2013) propose a concept of a succession of ice flow styles (maximum, transitional and local flow) over a glacial cycle in western Svalbard, which might explain the seemingly incompatible ice flow directions observed from eastern Svalbard.
The presence of grounding-zone wedges (GZWs) in the western Svalbard cross-shelf troughs and in the Barents-Kara sector of the Arctic Ocean indicates that post-LGM retreat was episodic, with still-stands punctuating periods of more rapid cross-shelf deglaciation . GZWs are often subtle features in a bathymetric sense, and lack of detailed swath bathymetry and sub- Trough-mouth fans are drawn with a "?" seaward of the British Channel and Voronin troughs as they are based solely on the outward bulging of bathymetric contours and the presence of positive free-air gravity anomalies on the continental slope (Minakov et al., 2012); these features have not been confirmed by geophysical/geological surveying. Submarine landforms are mapped primarily from swath-bathymetric, side-scan sonar and acoustic/seismic data. In cases where only single acoustic profiles were used to map landforms, we have used a dashed line to indicate that the orientation of the features is not known. The submarine landform maps were compiled from: Solheim and Kristoffersen (1984); Vogt et al. (1994); Polyak et al. (1997); Vorren and Laberg (1997); Kleiber et al. (2000); Dowdeswell and Elverhøi (2002) bottom profiles may contribute to their apparent absence in the major troughs of the Kara Sea.
Small transverse ridges occur, for example, in the Bellsund Trough and east of Nordaustlandet (Fig. 4). Morainic ridges also occur on shallow banks at the margins of several cross-shelf troughs in the Barents and Kara Seas, for example Kvitøya Trough in eastern Svalbard (Kleiber et al., 2000). At 250 m water depth in the St. Anna Trough, Polyak et al. (1997) describe ridges that are very similar to ridges in the Olgastretet . A series of ridges up to 30 m high and more than 5 km wide have been identified from acoustic profiles in the central part of St. Anna Trough between 79 and 81.5 N (Polyak et al., 1997). These ridges were interpreted as large moraines but have not been mapped in Fig. 4 because their orientations are unknown.
Seismic reflection profiles from the northern Barents and Kara Sea reveal a thin, relatively continuous cover of Quaternary sediments overlying a prominent erosional unconformity, the Upper Regional Unconformity (URU; Solheim and Kristoffersen, 1984;Vorren et al., 1986). Sediment-core and borehole data show that these units comprise a typical glacialedeglacialepost glacial sequence of basal diamicts overlain by glaciomarine and marinehemipelagic sediments (Elverhøi and Solheim, 1983;Solheim and Kristoffersen, 1984;Polyak and Solheim, 1994;Polyak et al., 1997). The pebbly-mud lithofacies of the diamict units is usually less than 10 m thick, is stiff at its base, and varies in colour with changes in the underlying source bedrock. Glaciomarine muds can be massive or laminated, the latter being interpreted as resulting from changes in the melting regime of a nearby ice margin, as in parts of the Franz Victoria trough (Polyak and Solheim, 1994;Lubinski et al., 1996) and Erik Eriksenstretet . The transition to olivegrey marine-hemipelagic muds, which are regionally ubiquitous, is often gradual and occurs alongside a diversification in foraminifera assemblages indicating ameliorated environmental conditions and an advection of Atlantic water to the northern margin of the Barents and Kara seas (Polyak and Solheim, 1994;Lubinski et al., 1996). Shallow sub-bottom profiles show that the glacial landforms described above (streamlined landforms, moraines) were formed at or near the surface of the glaciogenic diamicts (Solheim et al., 1990;Polyak et al., 1997;Hogan et al., 2010a). Multiple radiocarbon ages from sediments just above the diamicts return ages around 13e 14 cal ka BP, and ages of 10e12 cal ka BP for the glaciomarinee marine transition confirming that this stratigraphy relates to the Late Weichselian glaciation of the Barents-Kara Sea (Elverhøi and Solheim, 1983;Lubinski et al., 1996;Polyak et al., 1997).
Beyond the maximum extent of the full-glacial Eurasian Ice Sheet over Svalbard and the Barents Sea, trough-mouth fans (TMFs) are the largest landforms on the continental margin (Fig. 4). TMFs are identified off the major fjord-trough systems west of Svalbard , larger TMFs also occur along the Arctic Ocean margin of the Kara Sea, offshore of the Franz Victoria and St. Anna troughs ( Fig. 4) (Polyak et al., 1997;Kleiber et al., 2000). Additional TMFs may be present offshore of the British Channel, north of Franz Josef Land, and the Voronin Trough to the west of Severnaya Zemlya, on the basis of convex upper-slope bathymetry and free-air gravity anomalies Minakov et al., 2012). A TMF off Hinlopen Trough appears to have been largely removed by the Hinlopen Slide (Vanneste et al., 2006;Winkelmann and Stein, 2007;Batchelor et al., 2011). Glaciogenic debris flows observed on acoustic profiles of some TMFs are often stacked to form major building blocks of the fans (e.g. Laberg and Vorren, 1995;Dowdeswell et al., 1996;King et al., 1998).
Between full-glacial ice streams, the submarine landform assemblage of inter-trough areas is different from that of crossshelf troughs (Ottesen and Dowdeswell, 2009). Whereas the troughs are dominated by flow-parallel streamlined sedimentary landforms, the inter-ice stream landform assemblage is more typically defined by transverse-to-flow ridges (Ottesen and Dowdeswell, 2009). Such a landform assemblage is present in the fjords of north-westernmost Svalbard (Smeerenbergfjorden, Raudfjorden).
Large areas of the northern Barents Sea floor are marked by iceberg keels that form a palimpsest of cross-cutting furrows. On the Yermak Plateau and the shelf north of Svalbard, there are distinctive zones where very deep iceberg keels have ploughed the sea floor to depths of up to about 800 m (Vogt et al., 1994(Vogt et al., , 1995Dowdeswell et al., 2010b); some plough marks indicate multiple keels on a coherent iceberg up to 7 km wide. These very deep and wide plough marks have been interpreted to be produced by either very large free-drifting icebergs, probably from the first stages of deglaciation of the large Franz Victoria and St. Anna trough ice streams, or from groups of large icebergs frozen into huge multiyear sea-ice floes that may have been present in the Arctic Ocean under Quaternary full-glacial conditions (Vogt et al., 1994;Dowdeswell et al., 2010b;Jakobsson et al., 2010b).
Volume estimates of sediments offshore indicate that a 2e 3 km thickness of rock has been removed from central Spitsbergen since the Eocene (Eiken and Austergard, 1987;Vorren et al., 1991), and at least half of this volume was removed during the Quaternary glaciations (Svendsen et al., 1989;Dimakis et al., 1998;Elverhøi et al., 1998b). Glaciations of limited extent over the Svalbarde Barents Sea region probably began in the PlioceneePleistocene, 3.5e2.4 Ma ago (Knies et al., 2009). Glaciations intensified in the Early Pleistocene 2.6e1.0 Ma and glaciers were initially land-based, with evidence of Early Pleistocene glaciofluvial melt-water transport to the surrounding oceans , before the ice sheets reached the shelf break at around 1.6 Ma and ice-rafted debris (IRD) delivery to the adjacent ocean increased (Knies et al., 2009). Based on isostatic modelling, Butt et al. (2002) suggested that the Barents Sea continental shelf region was subaerially exposed at 2.3 Ma, i.e. at the time when glaciers began to spread over larger areas. Sejrup et al. (2005) suggested that extensive shelf glaciations were taking place in the Svalbard region at 1.6e1.3 Ma, and a large scale intensification of Barents Sea glaciations started about 1 Ma with grounded ice reaching the Yermak Plateau (Knies et al., 2009). The erosional surface from this first ice grounding is located well below the MIS 6 streamlined glacial features mapped on the Yermak Plateau O'Regan et al., 2010). At Kongsfjordhallet, northern Kongsfjorden, western Spitsbergen, there is also evidence for a Spitsbergen glaciation that dates back to the Early Pleistocene (2e1 Ma) (Houmark-Nielsen and . Miller (1982) likewise found indications of Earlymiddle Pleistocene (1e0.3 Ma) glaciations along the southern shore of Kongsfjorden, with at least three emergence cycles and two distinct glacial events. After 0.78 Ma, at least eight full-scale glaciations occurred over Svalbard and the Barents Sea, evidenced by trough-mouth fan deposition along the western Barents margin where cross-shelf ice streams reached the shelf edge (Vorren et al., 2011).
The Svalbard terrestrial record of full-scale glaciations is fragmentary and biased towards Late Quaternary glacial events because of the prevailing erosion at times of major ice-sheet expansion. The Late Quaternary glacial record of western Svalbard (Fig. 5) comprises an extensive Saalian (>130 ka BP) and three Weichselian major glaciations Mangerud et al., 1998). These are fingerprinted by tills in the stratigraphy, overlain by coarsening upwards sequences signifying isostatic rebound and marine regressions. The glaciation curve of Mangerud et al. (1998) is still the best approximation for the glacial events also along the margin to the Arctic Ocean (Fig. 5). However, it probably represents the ice-stream systems of the fjords and cross-shelf troughs along western Svalbard, rather than simultaneous oscillations along the entire margin of the SvalbardeBarents Sea ice sheet (Ingólfsson and Landvik, 2013;Landvik et al., 2013).
Three major ice-stream pulses are suggested for the Last Interglacial/Glacial cycle, at approximately 110 ka, 60 ka and 20 ka ago, interacting with the ocean through delivery of sediments to the shelf break and release of icebergs. The 60 ka and 20 ka pulses do both correspond to input of terrigenous material in sediment cores north of Svalbard, whereas no input from the 110 ka pulse has been found (Winkelmann et al., 2008). These periods of high zonal ice flux could have led to repeatedly extensive ice-shelf formation (Fig. 5). The raised beach record of postglacial emergence on Franz Josef Land (Forman et al., 2004) confirms, together with marine geological evidence (Polyak and Solheim, 1994;Lubinski et al., 1996), that the archipelago and fringing shelf areas and cross-shelf troughs were covered by the SvalbardeBarents Sea ice sheet during the LGM . Svendsen et al. (2004a) imply that ice-sheet oscillations over the Last InterglacialeGlacial cycle along the northern margin of the SvalbardeBarents Sea ice sheet largely mirror its western margin oscillations, but stratigraphical data substantiating this are largely lacking. It has been pointed out that there is still insufficient chronological data to evaluate regional variability in the timing of ice-sheet maxima for different sectors of the SvalbardeBarents Sea ice sheet Ingólfsson and Landvik, 2013).
3.1.2.2. Kara Sea and north-western Siberia. Major glaciations in western Siberia and over the Kara Sea shelf started in Middle Pleistocene, during MIS 16 (the Mansi glaciation), and maximum ice-sheet extension at the south-eastern fringe of Eurasian Ice Sheets occurred during the Samarovo glaciation (MIS 8) (Astakhov, 2004(Astakhov, , 2013. These very extensive Middle Pleistocene ice sheets could have covered the eastern Barents Sea and the Kara Sea margins, west and east of Novaya Zemlya, respectively, and across the continental break into the Arctic Ocean. The most complete Middle Pleistocene terrestrial stratigraphy is on October Revolution Island in the Severnaya Zemlya archipelago (Fig. 1). Möller et al. (2006) report five marine units intercalated with tills, all in superposition, and all of Middle Pleistocene age except for the uppermost till/marine cycle. Till bed II is firmly age-constrained into MIS 6 (Saalian/Taz) whereas the two lowermost till beds are suggested to be from MIS 8 and MIS 10 (or possibly MIS 12), respectively. The highest located marine sediments and beacheridge complexes, tied to the Saalian/Taz deglaciation at the MIS 6/5e transition on Severnaya Zemlya and at Cape Chelyuskin on Taymyr Peninsula, reach w140 m a.s.l. , only some 200 km from the shelf break to the north. This implies a Saalian/Taz ice-sheet thickness over the Kara Sea shelf in excess of 3000 m, as suggested from earth rheological modelling (Lambeck et al., 2006). The timing concurs with possible ice-shelf grounding at 1000 m water depths on the Lomonosov Ridge in the central Arctic Ocean Polyak et al., 2001).
Late Pleistocene glacial events over north Siberia and the Kara Sea are reasonably well constrained. Svendsen et al. (2004a) suggest that there were three major ice-advance/-retreat phases during the Weichselian (Fig. 6), but with successively smaller ice volumes (Möller et al., 2011). The maximum expansion of a Kara Sea ice sheet was in the Early Weichselian (MIS 5d-5b), reaching well south of the Byrranga Mountains on the Taymyr Peninsula, merging with the Putorana Plateau ice cap in the SE and advancing onto the northwest-Siberian lowlands, and reaching the northern shelf edge. The Early Weichselian ice retreat was followed by a limited Middle Weichselian (MIS 4) advance to the south. However, in the north ice reached the shelf edge and merged with the SvalbardeBarents Sea ice sheet to the west. The uppermost till/marine cycle on Severnaya Zemlya suggests that this archipelago was ice-covered through MIS 5d to MIS 4 . Cape Chelyuskin in the east was not ice covered , but the northern Siberian coast was impinged by MIS 4 ice on the Taymyr Peninsula (Alexanderson et al., 2001) that extended well into to the northwest-Siberian lowlands (Astakhov and Nazarov, 2010;Astakhov, 2013). The LGM extent of Kara Sea ice sheet is problematic (Svendsen et al., 2004a). Terrestrial evidence of its advance is only at hand on the northernmost coast of Taymyr Peninsula (Alexanderson et al., 2001), while Severnaya Zemlya was largely ice free during LGM (Raab et al., 2003;Möller et al., 2006). However, based on marine sub-bottom profiling over the Kara Sea shelf, Polyak et al. (2008) argue for the presence of a LGM till over parts of it. LGM ice is suggested to have reached the Vilkitsky Strait in the east and flowed across the shelf towards the Voronin Trough that drains into the Arctic Ocean (Fig. 3). Detailed echo-sounding east of Novaya Zemlya, show no indication for a LGM ice sheet immediately west and north-west of Taymyr Peninsula Dittmers et al., 2008). This speaks against a southern connection between the LGM BarentseKara Sea ice sheet and a local ice sheet on Taymyr Peninsula Dittmers et al., 2008).
Although most stratigraphic and glacio-isostatic data suggest repeated build-up of centres of ice mass in the Kara Sea, ice-flow directional data, as interpreted from till fabrics and glaciotectonics in key sections from Severnaya Zemlya , Cape Chelyuskin , Yamal Peninsula (Forman et al., 2002) and Yugorski Peninsula (Lokrantz et al., 2003), suggest ice flow from other dispersal centres. A possible solution to this enigma is suggested by Möller et al. (2006) and Ingólfsson et al. (2008). In their model of ice sheet build-up and decay they argue that islands and highlands in the periphery of the Kara Sea were critical as nucleation areas for the formation of local ice caps that later merged on the shallow shelf area and here grew into full Kara Sea ice-sheet domes.

Landforms
The south-western and central Barents Sea is characterized by shallow banks between deeper troughs, of which Bjørnøyrenna is the largest (Fig. 1). The geomorphic imprints of the Barents Sea ice sheet occurring on the banks are very different from those in the troughs. Banks are characterized by the occurrence of large and small morainic ridges formed in relation to the last deglaciation of the Barents Sea (Elverhøi and Solheim, 1983;Andreassen et al., 2013;Bjarnadóttir et al., 2013) and considered to be diagnostic of slow ice retreat (Ottesen and Dowdeswell, 2009). MSGLs and trough-mouth-fans indicate that the troughs have been occupied by fast-flowing ice streams which during glacial maxima reached the western Barents SeaeSvalbard continental shelf break, whereas grounding-zone sediment accumulations indicate that ice-stream retreat occurred in an episodic manner (Solheim et al., 1990;Faleide et al., 1996;Vorren and Laberg, 1997;Ottesen et al., 2005;Andreassen et al., 2008;Dowdeswell et al., 2008;Winsborrow et al., 2010;Bjarnadóttir et al., 2013Bjarnadóttir et al., , 2014. MSGLs have also been identified on several buried surfaces identified in 3-D seismic data, indicating that ice streams were a common feature during earlier glaciations as well (Andreassen et al., 2004(Andreassen et al., , 2007. The distribution of geomorphic features shows that the ice streams experienced large spatial and temporal variations in extent, flow velocity, and basal conditions. 3-D seismic data from southwestern Barents Sea reveal repeated sequences of glacial rafts and mega-blocks overlain by MSGLs, inferred to be the result of basal freeze-on during periods of ice stream slowdown/quiescence followed by reactivation and resumed fast ice-flow (Andreassen et al., 2004;Andreassen and Winsborrow, 2009). Furthermore, sets of landforms ascribed to the last deglaciation of the upper regions of Bjørnøyrenna indicate that the Bjørnøyrenna Ice Stream repeatedly experienced rapid shifts from fast ice flow to stagnation as inferred from the distribution of MSGLs and crevasse-squeeze ridges respectively Bjarnadóttir et al., 2014) (Fig. 4). However, the upstream extent and duration of these events of ice stream stagnation are unknown. Finally, glacial lineations and recessional features grouped into juxtaposed ice stream flow-sets indicate the occurrence of ice-stream flow switching (Winsborrow et al., , 2012, which also suggests local changes in subglacial thermal regime and ice-flow velocities (Fig. 4).
In the upper parts of Bjørnøyrenna, large plough marks of highly uniform orientation occur immediately downstream of selected trough-transverse grounding-zone sediment accumulations. These are interpreted to indicate events of high iceberg discharge, with mega-iceberg release and ice stream break-up, probably associated with ice-shelf disintegration and/or extremely high ice-flow velocities Bjarnadóttir et al., 2014). The plough marks and mapped corrugation ridges in this part of the Barents Sea are similar to those in Pine Island Bay, Western Antarctica . This landform assemblage is interpreted to represent ice shelf break up, and release of megaicebergs from the grounding zone that move along in an armada and under the influence of tidal motion that squeeze out small ridges in their trails; the corrugation ridges .

Stratigraphy and chronology
The southwestern and southern Barents Sea has been subject to repeated shelf-wide glaciations through several glacial cycles during the late PlioceneePleistocene, with glacial erosion on the continental shelf and deposition and subsidence at the continental margins. This is recorded in thick sediment records on the outer south-western continental shelf (Faleide et al., 1996;Solheim et al., 1996;Vorren and Laberg, 1997;Butt et al., 2000). The glacial sediments are separated from older sedimentary bedrock by the erosional Upper Regional Unconformity (URU) (Solheim and Kristoffersen, 1984;Vorren et al., 1986). At the south-western Barents Sea margin the glaciogenic sediments above URU are up to 3e4 km thick while their thickness is significantly less (typically up to a few hundred metres) and more poorly constrained in the central part of the Barents Sea (Vorren et al., 1984;Elverhøi et al., 1989Elverhøi et al., , 1989. Based on a combination of seismic stratigraphy and well information the sedimentary column in the south-western part of the Barents Sea has been divided into three main seismic sequences representing different depositional conditions (GI, GII and GIII) (Faleide et al., 1996;Butt et al., 2000). Knies et al. (2009) suggested that three main phases of glacial development took place in the Barents Sea. During an initial ice sheet build-up phase (3.5e2.4 Ma) ice growth was restricted to the northern Barents Sea and Novaya Zemlya. A phase of continued ice-sheet growth and southwards expansion occurred 2.4e1.0 Ma, followed by a final phase (<1 Ma) during which Barents Sea glaciations were repeatedly characterised by shelf-wide expansion (Knies et al., 2009).
The last time a Barents Sea ice sheet reached the western shelf edge was around 20 cal ka BP (Vorren and Laberg, 1996;Jessen et al., 2010). Due to limited datable material it is hard to constrain the timing of deglaciation stages on the Barents Shelf. However, the distribution of available ages provides a rough estimate. These indicate that initial deglaciation of the Barents Sea coincided with, and was likely triggered by rising global eustatic sea levels Winsborrow et al., 2010). The outer shelf was deglaciated between 18 and 16 cal ka ago (Rokoengen et al., 1977;Vorren et al., 1978;Bischof, 1994;Polyak et al., 1995;Rasmussen et al., 2007;Aagaard-Sørensen et al., 2010;Junttila et al., 2010;Rüther et al., 2012). At roughly the same time (18e 16 cal ka) the Barents Sea/Scandinavian ice sheets are inferred to have reached their maximum extent in north-western Russia (Larsen et al., 1999;Demidov et al., 2006). Recent geomorphic studies of the sea floor support the idea of a delayed maximum in the south-eastern sector of the Barents Sea and further suggest there was a shift in the focus of maximum ice volume and dynamics to the eastern sector of the ice sheet . The ice-free conditions in the west allowed moisture to penetrate farther east, leading to further ice sheet build-up and advance of this sector. For the central Barents Sea less is known about the timing of ice-sheet retreat, although it may be inferred that the shores of Kong Karls Land and Edgeøya were ice-free at w11.2 cal ka (Salvigsen, 1981;Bondevik et al., 1995), by which time the ice had probably retreated from Bjørnøyrenna and Olga Stret.

Laptev Sea, East Siberian Sea and the Arlis Plateau
The glacial history of the East Siberian Sea and adjacent bathymetric highs in the Arctic Ocean is poorly investigated. The international marine research community only has permit-free access to a relatively small sector of the outer continental shelf between Wrangel Island and De Long islands, i.e. between about 170 E and 175 E, north of 74 40 0 N. In addition, before 2007 persistent sea-ice cover during the summer months made geophysical seafloor mapping difficult in the area.

Landforms
The RV Polarstern expedition ARK-XXIII/3 in 2008 carried out seismic investigations and sediment coring between the Chukchi Plateau and the East Siberian Shelf (Fig. 7) (Jokat, 2009;Stein et al., 2010). The data reveal clear evidence that the outermost edge of the East Siberian Sea towards the central Arctic Ocean has characteristics of a formerly glaciated continental margin (Niessen et al., 2013). Glacial lineations found on the flank and top of the Arlis Plateau, located on the southern Mendeleev Ridge, are interpreted as MSGL and/or iceberg plough marks, which have NNEeSSW and NWeSE directions, respectively (Fig. 7). Here the ice-grounding events are possibly related to floating ice masses from the Chukchi Borderland Jakobsson et al., 2010b) and/ or from the East Siberian Shelf (Niessen et al., 2013).
In seismic profiles across the shelf edge of the East Siberian and Chukchi seas, the top of Neogene sedimentary sequences is truncated. This truncation, accompanied by correlative sedimentary wedges on the upper slope, has tentatively been attributed to ice erosion . In several locations along the East Siberian continental margin, this erosion is associated with MSGL mapped by swath bathymetry or wedges of debris-flow deposits visible in high resolution sub-bottom profiles (Niessen et al., 2013). Debris-flow deposits adjacent to eroded areas on the slope are interpreted as being formed by gravitational redeposition of eroded sediments near former ice-grounding lines. In all locations the MSGL and glaciogenic wedges are covered by hemipelagic sediments, which drape several generations of grounding events. The thickness of well-stratified sediments overlying proglacial and subglacial diamicton ranges from 3 m on top of the Arlis Plateau to 20 m on the East Siberian continental slope. Older glacial streamlined lineations are found in deeper water where they survived subsequent grounding events in shallower water, for example, at about 900 m below present sea level on the Arlis Plateau (Fig. 7). In summary, the mapped landforms on the Arlis Plateau and along the East Siberian Sea margin may stem  (Jakobsson et al., 2008b;Niessen et al., 2013;Dove et al., 2014). Inset shows south-western limit of glaciation mapped on the New Siberian islands (Basilyan et al., 2010). Yellow line is the 350 m isobath which correlates well with observed seaward limits (orange) of iceberg-scoured seabed. Colour-coded arrows show inferred ice flows of different provenance: Orange ¼ Laurentide source; White ¼ Chukchi Shelf; Green ¼ East Siberian Shelf; Black ¼ New Siberian Islands/East Siberian Shelf. Dotted arrows are most hypothetical. Cores shown recovered glacial and iceberg-turbated diamictons on the borderland and shelf, respectively (unpublished data; Polyak et al., 2007;Stein et al., 2010). from thick coherent ice shelves or local ice sheets covering the Chukchi and East Siberian shelves during several glacial cycles in the past.
Iceberg plough marks from the final glacial event are mapped at present water depths between 350 m and 100 m in the entire area between the Chukchi Borderland and the East Siberian slope to 170 E (Fig. 7). The pattern is irregular with crosscutting plough marks. These glaciogenic features are similar to the iceberg scouring described from the Chukchi Sea margin and overlain by deglacialeHolocene marine deposits (Polyak et al., 2007;Hill and Driscoll, 2010). More mapping in this part of the Arctic Ocean is required in order to unravel the glacial history of the outer East Siberian Sea margin and adjacent bathymetric highs.

Stratigraphy and chronology
Terrestrial investigations carried out on Wrangel Island include studies of elevated shorelines, glaciogenic landforms and radiometric exposure dating (Gualtieri et al., 2003(Gualtieri et al., , 2005. Dated raised shorelines, ranging in age between 459 to 780 ka and 64 to 73 ka, are interpreted as eustatic in origin rather than being indicative of former isostatic uplift following ice unloading. Ice-marginal features, such as end moraines, or other glacial landforms are absent in the higher mountains of the island. Exposure dating excludes extensive glaciation during the last c. 84.4 ka (Stauch and Gualtieri, 2008). Permafrost records from the southernmost of the New Siberian Islands contain pollen that suggest a cold tundra-steppe vegetation in the area during LGM (Wetterich et al., 2011). Furthermore, permafrost has been preserved at Bol'shoy Lyakhovsky Island for at least 200 ka suggesting an absence of ice sheets over this entire period (Schirrmeister et al., 2002). In the Laptev Sea region, permafrost records also suggest ice-free conditions during the LGM (Boucsein et al., 2002;Hubberten et al., 2004). Taken together, a large ice-free cold tundra-steppe appears to have stretched from the easternmost Taymyr Peninsula in the west to east of Wrangel Island in the east during LGM. These results contradict the hypothesis of a 2 km thick East Siberian Sea ice sheet centred between Wrangel Island and the East Siberian Islands during the LGM (Hughes et al., 1977;Grosswald and Hughes, 2002) (Fig. 2). However, data from the northern part of the East Siberian Islands provide evidence of glacial impact including large bodies of relict glacier ice, deformed and eroded bedrock and Quaternary deposits, and till units with far-travelled erratics (Grosswald, 1989;Basilyan et al., 2010). This data indicates an ice sheet impinging on the north-eastern part of the archipelago from the East Siberian shelf. The age of this event (or the last ice cover) is estimated as late MIS 6 from thorium/uranium dating of mollusc shells (Basilyan et al., 2010). Sediment cores were recovered along a transect from the Chukchi Abyssal Plain across the southern Mendeleev Ridge . These cores are characterized by prominent changes in sediment colour, grain-size, sediment composition, and degree of bioturbation . In all cores, prominent dark brown intervals were found, which together with the pinkewhite layers and microfossil abundance are used for core correlation and age control . A preliminary age model of the upper part of these cores is based on correlation to cores NP-26 and HLY0503-8JPC Darby et al., 2006;Adler et al., 2009;Backman et al., 2009;Polyak et al., 2009). This age model suggests that one core likely encompass MIS 1 to MIS 8, whereas two of them did not reach MIS 6. Furthermore, the age model inferred in the cores located in water depths between 800 and 900 m at southern Mendeleev Ridge implies that an identified diamicton with erosional structures is related to ice-grounding events that occurred before MIS 5a.

Chukchi margin
The first report of glacial impact on the sea-floor in the Chukchi region dates back 50 years ago (Hunkins et al., 1962). A focused investigation of seafloor morphology and related stratigraphy began, however, first with the 1998e1999 SCICEX swath and subbottom survey Edwards and Coakley, 2003). Data collected since indicate a widespread impact of deep-keeled ice from past ice sheets/shelves e especially on the Chukchi Borderland (Chukchi Plateau and Northwind Ridge), and more sparsely at the ChukchieBeaufort shelf margin (Fig. 7) (Jakobsson et al., 2005(Jakobsson et al., , 2008bPolyak et al., 2007;Engels et al., 2008). Based on the initial mapping of glacial lineations, Polyak et al. (2001) inferred two major sources of eroding ice e from the Laurentide ice sheet to the east and from the Chukchi shelf to the south. However, a lack of evidence for large ice sheets on the Chukchi Sea coast and Wrangel Island, at least in the Late Pleistocene (e.g. Brigham-Grette et al., 2001;Gualtieri et al., 2005) casts doubt on the southern ice source. Accordingly, in later studies mapped glaciogenic landforms on the sea floor in this area have been related to ice originating from the Laurentide ice sheet only, with possibly local ice cap(s) on the Chukchi Plateau (Jakobsson et al., 2005(Jakobsson et al., , 2008bPolyak et al., 2007). Exact pathways, provenance, and ages of these glacial advances remained poorly constrained. An improved understanding emerges from data collected in recent years from the northern part of the Chukchi shelf (Chukchi Rise), which characterizes both sea-floor morphology with underlying shallow stratigraphy (Dove et al., 2014) and deeper strata (Coakley et al., 2011;Hegewald and Jokat, 2013). This new data reveals a widespread grounded-ice presence on the Chukchi Rise with ice source located farther south or southwest indicating ice-sheet centres on the Chukchi and/or East Siberian shelf.

Landforms
The most common features of the shallower part of the Chukchi margin are iceberg plough marks. The cut-off depth of scouring is consistently found at water depth of w350 m (Fig. 7). Isolated or sparsely aggregated plough marks may occur at greater depths; they have curved paths and do not form parallel clusters typical for MSGL. At depths above w130 m scours fade away due to sediment infill, because the inner shelf is swept by currents that remove all pre-existing bedforms. Such plough marks were described earlier from the Chukchi Borderland (Hill and Driscoll, 2010) and are also evident in sub-bottom data from about 167 E, 76 40 0 , at 180 m water depth in the East Siberian Sea (Gusev et al., 2012). Thus, it seems reasonable to assume that the intensively ploughed zone in the depth range between 130 and 350 m extends farther west to an unknown extent along the margin of the East Siberian Sea (Fig. 7).
As iceberg scouring eliminates evidence of previous sea-floor processes, glaciogenic landforms such as lineations and morainic ridges are primarily found at depths >350 m. The lower limit of their distribution varies and often coincides with the inflection point in the slope profile. Around the Chukchi Rise and Plateau this limit commonly occurs at depths between 550 and 900 m, whereas at other sites on the Northwind Ridge it can exceed 1000 m below present sea level (Jakobsson et al., 2008b). In comparison to the Chukchi Plateau and Rise, the distribution of glaciogenic landforms on the Northwind Ridge is fragmentary as the ridge consists of isolated, limited highs, often with steep slopes.
Clusters of parallel, linear to slightly curved grooves and ridges are common for the edge of the Chukchi Rise and Plateau and the highs of the Northwind Ridge. Their pattern and dimensions are characteristic of MSGL identified elsewhere on glaciated shelves. In addition to the shelf edge, MSGL occur in a broad bathymetric trough on the eastern side of the Chukchi Rise (Dove et al., 2014), similar to cross-shelf troughs in the Barents and Kara seas (c.f. Sections 3.1 and 3.2). Several sites such as on the northern Northwind Ridge (Jakobsson et al., 2008b), Chukchi Plateau (Mayer et al., 2010), and in the trough on the Chukchi Rise (Dove et al., 2014) feature drumlinized seabed, which is especially helpful for identifying the direction of ice flow.
Streamlined bedforms east and north of the Chukchi Rise, including its north-eastern edge, are predominantly EeW to SEe NW trending (Fig. 7), consistent with the inferred arrival of eroding ice from the north-western sector of the Laurentide ice sheet (Fig. 7) (Jakobsson et al., , 2005(Jakobsson et al., , 2008bPolyak et al., 2007;Engels et al., 2008). The curved path of this ice flow along the Alaskan margin and then NW-wards across the Northwind Ridge is not well understood, but has been proposed to possibly indicate the presence of very thick shelf ice over the Canada Basin .
On the Chukchi Rise, MSGL mostly occur in the bathymetric trough and on the western edge and appear to have SWeNE orientation, same as at some highs on and near the Northwind Ridge (Fig. 7). The provenance of ice that has formed these features is not yet understood. Possible sources are the Chukchi shelf farther south and/or the East Siberian margin (Fig. 7).
Curvilinear to sinuous, symmetric or asymmetric ridges composed of diamict material are common for the shelf edge, especially between 350 and 550 m water depth around the Chukchi Rise (Dove et al., 2014). Similar ridges occur on the Chukchi Plateau, but their distribution is more sparsely constrained by existing data. Ridges are between 100 and 800 m wide and can grade into depositional wedges up to 50 m thick. Multiple ridges can form long bands running along the shelf edge, where sub-parallel ridges are interspaced with hummocky seabed. Broadly grouped, the ridges are interpreted as moraines and/or GZW. At several locations distribution of ridges indicates up-slope retreat(s) of the groundedice mass after the advance phase(s).
Smaller ridges oblique or transverse to MSGL have been mapped at the north-eastern part of the Chukchi Rise at the flanks of the bathymetric trough (Dove et al., 2014). They may have a diverse origin: two sets of linear to sinuous, sub-parallel ridges may be recessional moraines or basal crevasse-fill, whereas a set of multiple, closely and regularly spaced ridges is similar to ribbed moraines. The adherence of these features to the trough, along with MSGL and a thick package of till, provides further evidence of extensive, possibly pulsed ice streaming from the Chukchi Rise eastwards.
Tills are persistent at the shelf edge around the Chukchi Rise, commonly infilling underlying depressions or channels or forming large cross-sectional wedges, sometimes with multiple till units (Dove et al., 2014). Downslope from the till wedges till-like packages occur with characteristic asymmetric cross-section, probably re-deposited from the eroded sites. The slope and adjacent basins also feature multiple debris lobes up to 50 m thick, interstratified with layered hemipelagic sediments. Altogether, a combination of till wedges, redeposited packages, and debris lobes indicates large amounts of sediment delivered by glaciers to the slope throughout the Chukchi margin.

Stratigraphy and chronology
Based on the position of regional erosional unconformity within Plio-Pleistocene deposits on the Chukchi margin , glacial impact on the seabed in this region has a long history. Direct age assessment of its initiation is not yet possible, but sediment-core data throughout the western Arctic Ocean indicate a sharp increase in iceberg-rafted material at the  Stein et al., 2010;Polyak and Jakobsson, 2011).
Cores from the Northwind Ridge (Fig. 7) constrain the age of the last erosional event associated with the Laurentide-sourced ice to the penultimate glaciation, estimated MIS 4 (Polyak et al., 2007). A younger erosional age of the Last Glacial Maximum (MIS 2) has been obtained only for the shelf-proximal area of the Ridge with water depths <450 m. Data on bedforms from the Chukchi Rise indicate that it was the source of this younger ice, consistent with the Holocene age of sediments overlying iceberg-scoured surface on the Rise and adjacent shelf (Polyak et al., 2007;Hill and Driscoll, 2010). We note that these ages constrain only the last ice impact at a given sea floor site. They do not preclude older grounding events, obliterated by later impacts, neither the occurrence of younger ice shelves that were too thin to reach the sea floor.

Beaufort Sea and the northern Canadian Arctic Archipelago
At least seven primary ice streams discharged directly into the Arctic Ocean from the Laurentide and Innuitian ice sheets during the LGM, some reaching or exceeding 1 km in thickness. The LIS advanced northwestward from the Canadian mainland inundating both the Beaufort Sea and adjoining marine channels and islands of the western Canadian Arctic Archipelago (CAA). There, the LIS coalesced with the southern margin of the Innuitian Ice Sheet (IIS) that inundated the Queen Elizabeth Islands (QEI) to the north. The Canadian Shelf of the Beaufort Sea marks the northwestern limit of the LIS, which during maximal phases of late Quaternary ice-sheet growth extended west along the Yukon Coastal Plain, possibly overriding parts of the Alaskan coast farther west.

Landforms
Three large glacial troughs extend across the Beaufort Sea Shelf and southwest CAA; the Mackenzie Trough, Amundsen Gulf and M'Clure Strait (Fig. 8). Furthermore, eastward (up-ice) from these three prominent troughs, satellite and bathymetric data reveal large MSGLs converging into them from many of the islands and intervening channels of the western CAA (Stokes et al., 2005(Stokes et al., , 2006(Stokes et al., , 2009MacLean et al., 2010).
Where M'Clure Strait and Amundsen Gulf intersect the continental shelf break, a pronounced "duck-foot" pattern in the bathymetry is consistent with the presence of classic TMFs (Stokes et al., 2005(Stokes et al., , 2006 (Fig. 8). Sub-bottom profiles crossing the shelf break at the mouth of M'Clure Strait reveals a series of stacked glacial debris flows supporting this interpretation . In Amundsen Gulf, high-resolution multichannel seismic data also indicate a substantial TMF that purportedly records up to nine Pleistocene advances of the LIS to the shelf break (Batchelor et al., 2013). In contrast, Mackenzie Trough lacks a well-defined large TMF. However, west of this trough, lineations are mapped on the sea floor along the Beaufort outer margin in 400e700 m water depths (Fig. 8). These lineations are interpreted to be of glaciogenic origin and to originate from an ice-shelf that flowed along the Alaska-Beaufort margin and possibly extended to and across the Chukchi Borderland (Engels et al., 2008).

Stratigraphy and chronology
Based on the abundance of terrestrially dated sites and documented glacial landforms in northwestern Canada and the western CAA, the advance of the northwest LIS is now recognized to be pervasive, extending to the polar continental shelf during the LGM (England et al., 2006Lakeman and England, 2013). As a result of the LGM ice cover, information on pre-MIS 2 glacial activity for much of the northwest LIS remains undocumented (Kleman et al., 2010). Furthermore, there are still no direct chronological constraints on the limit of the LIS offshore on the polar continental shelf (Dyke and Prest, 1987).
The existence of active ice streams in M'Clure Strait and Amundsen Gulf during the LGM presents a significant update on previous glacial reconstructions that portrayed ice shelves occupying these channels north and south of Banks Island (Dyke and Prest, 1987;Dyke et al., 2002;Dyke, 2004) (Fig. 8). Furthermore, earlier interpretations also showed multiple till sheets on Melville and Banks islands, most of which were interpreted to record undisturbed pre-Late Wisconsinan glaciations (Vincent, 1982(Vincent, , 1983. However, recent field mapping and widespread dating of icetransported shells on both islands indicate that these till sheets all date to the LGM . Indeed, most of the previously mapped multiple till sheets on Banks Island (Vincent, 1982(Vincent, , 1983 are in fact weathered bedrock with only scattered, far-travelled Laurentide erratics deposited during MIS 2 . Most recently, the Laurentide ice retreat has been mapped from the polar continental shelf successively eastward across Banks Island to Prince of Wales Strait, during the Late Wisconsinan England, 2012, 2013). Farther north, on southern Melville island, AMS radiocarbon dates on individual shell fragments were obtained from the Dundas and Bolduc tills  and range from 24 to 49 ka BP . Although these till sheets were originally assigned to two separate Laurentide glaciations, they are now assigned to the LGM. Important implications of these findings are that the northwest LIS had to have advanced across Viscount Melville Sound onto Melville Island after w24 ka BP requiring at least partially ice-free conditions in the Sound during MIS 3 . Granite glacial erratics deposited by the LIS on southern Melville Island occur up at least 235 m above present sea level, recording a minimum thickness of 635 m for the LGM M'Clure Strait ice stream . Furthermore U/Pb zircon dating of some far-travelled erratics collected across the western CAA demonstrate for the first time that their provenance can be assigned to orogenic belts with unique radiometric ages located within the mainland Canadian Shield up to 1000 km to the southeast (Doornbos et al., 2009). These recently identified Shield erratics also would have been en route to the Arctic Ocean via M'Clure Strait and adjoining channels within the central CAA during the LGM (MacLean et al., 2010;Pie nkowski et al., 2012Pie nkowski et al., , 2013. The LIS appears to have reached the Mackenzie delta region on at least two occasions during the Last Glacial cycle. On the mainland, these events are delineated by the Toker Point Stade, and the younger Sitidgi Lake Stade (Fig. 8). Early work based on radiocarbon dates obtained from wood fragments from within the Toker Point till and marine shells from overlying sediments, suggested that this advance was Early Wisconsinan (MIS 5e4) (Murton et al., 2007). As a result, there was a long-standing view that a limited northwesterly advance of the LIS during the LGM (Sitidgi Stade) followed a more extensive early Wisconsinan glaciation (Toker Point Stade/Buckland Glaciation) (MIS 5e4) (Rampton, 1982;Beget, 1987;Vincent and Prest, 1987;Dyke et al., 2002) (Fig. 8). However, more recent luminescence dating of aeolian dune sand on the Tuktoyaktuk Penninsula, now constrain the age of the Toker Point Stade to a short advance between w22 and 16 ka BP, requiring that the Sitidgi Stade represents a deglacial re-advance of the Mackenzie ice stream (Murton et al., 2007). Offshore work on the Canadian Beaufort shelf, provides evidence that the fluvial and aeolian sands underlying the Toker Point till extend out towards the shelf break. This implies a period of prolonged subaerial exposure during the Last Glacial cycle, and possibly, for much of the eastern Canadian Beaufort shelf, throughout the LGM. During this time, thick permafrost developed on the exposed shelf (Mackay, 1959;Murton et al., 2010).
Seismic reflection profiles from the Mackenzie Trough were first used to suggest that the basal unconformity was formed in the Early Wisonsinan (60e120 ka), with evidence of a second ice advance in the Late Wisconsin (Blasco et al., 1990). A recent study based on an extensive set of industry seismic data has largely confirmed these findings (Batchelor et al., 2013). The presence of a buried GZW and two buried lateral moraines indicate a second ice advance out to the shelf edge that followed the initial excavation of the Mackenzie Trough (Batchelor et al., 2013). However, the absence of a distinct TMF in front of the Mackenzie Trough, despite the large available sediment supply, likely indicates that it was only in the late Quaternary that the LIS started to reach this sector of northwestern Canada.
There is evidence that the Mackenzie ice stream may have travelled west along the Yukon and even along the Alaskan coastal plan during pre-LGM glaciations. Herschel Island, immediately west of the Mackenzie Trough, is an ice-thrust moraine resulting from the advance of the LIS. The position of the island coincides with the westernmost moraines on the adjacent Yukon Coastal Plan, both marking the limit of LIS (Mackay, 1959;Rampton, 1982.). Massive ice exposed on Herschel Island shows signs of glaciotectonic deformation and an isotopic signature that resembles buried glacier ice (Fritz et al., 2012). Furthermore, they report radiocarbon-dated plant detritus as young as 16.2 cal ka BP within the glaciotectonized sediments of Herschel Island. If correct, this date would require that the maximum limit of the LIS along the Yukon Coastal Plain and southern Beaufort Sea occurred during late MIS 2. This LGM age is also coeval with the maximum advance of the LIS from the Mackenzie Valley westward into the Richardson Mountains and onto the Peel Plateau to the south. It also corresponds to a much more expansive advance of the LIS across the western CAA during the LGM Lakeman and England, 2013). England et al. (2006) provide a summary of three decades of glacial geology conducted across the Queen Elizabeth Islands (QEI) that constitute the northern half of the CAA. This synthesis reconstructs the configuration, primary divides and ice streams comprising the IIS. The last full development of IIS during MIS 2 is reconstructed, when it advanced from margins similar to present after 19 ka BP, inundating all of the marine channels of the QEI and flowing offshore to an unmapped margin on or beyond the polar continental shelf. The IIS coalesced with the Greenland Ice Sheet throughout Nares Strait and with the LIS throughout Parry Channel. Several prominent marine gateways occupied by these coalescent ice sheets fed ice streams with significant thicknesses towards the Arctic Ocean during MIS 2 documented by the uppermost moraines and erratics left by ice margins that subsequently thinned to widespread deglacial deltas dating to the early Holocene. In northern Nares Strait, Greenland erratics and lateral meltwater channels on Ellesmere Island document an upper ice surface at least 800 m asl, and together with water depths of >600 m in adjacent Hall Basin, require an ice thickness of at least 1400 m locally. How much of this thickness was maintained farther north, where the Nares Strait trunk glacier (likely an ice stream) would have thinned as it crossed the continental shelf of the Lincoln Sea, is currently undetermined. However, at least one deep trough (>600 m) runs northward from northern Ellesmere Island crossing the shelf. Immediately inland, erratics deposited by northward flowing Innuitian ice reach w900 m asl, again recording potential ice delivery (>1 km thick) from this locality to the Arctic Ocean during the LGM (Lemmen and England, 1992).
Other regional ice streams from the IIS flowed northward to the Arctic Ocean through Nansen Sound (>500 m deep), that was occupied by coalescent ice from adjacent Axel Heiberg and Ellesmere islands during the LGM (Bednarski, 1998). A third prominent ice stream flowed northwestward from southern Ellesmere Island via Eureka Sound and Massey Channel, transporting granite erratics >600 km to the Arctic Ocean (Lamoureux and England, 2000;Ó Cofaigh et al., 2000;Atkinson and England, 2004;England et al., 2004). A fourth ice stream from the IIS likely advanced westward through the Prince Gustolf Adolf Sea, possibly buttressed by the northwest LIS that coalesced with the IIS along the eastern coast of Melville Island ).

Landforms
The Northern Greenland continental margin bordering the Arctic Ocean is sparsely investigated due to its remoteness and hard sea-ice conditions. There are yet no marine geophysical surveys of the shelf carried out that reveal whether or not informative glaciogenic landforms exist. Instead, investigations of the northernmost coastal region are the main source of information concerning how far ice extended from Greenland towards the Arctic Ocean during glacial maxima. Along the Peary Land coast, between the ocean and the North Greenland mountain range, there is a 10e 15 km wide coastal plain with landforms and sediments recording glacial and marine events back to LGM. For two summers the "LongTerm Project" studied the Quaternary geology here with the main aim to investigate the relation between glaciation and Arctic Ocean sea ice. The results from this work have provided a substantial database on glacial and environmental events since LGM (Nørgaard-Pedersen et al., 2008;Larsen et al., 2010;Möller et al., 2010;Funder et al., 2011b;Olsen et al., 2012).
Geological mapping of glacial erratics on the coastal plain initially led Koch (1923) to suggest that a local ice cap, the North Cap, developed over the mountain range during LGM, which to the south and southeast merged with the Greenland inland ice sheet. Later fieldworks confirmed this view (Dawes, 1986;Funder, 1989). Intriguingly, both erratics and striations showed that instead of heading directly towards the ocean, the outlet glaciers on the plain were deflected and moved south-eastwards along the coast, and some erratics had travelled long distances from the west. Recent field work added extensive new evidence such as damming of valleys, till fabric, and observations on "epishelf" lakes, dammed between the deflected glaciers and the mountains. The deflection of the glaciers was explained as a result of buttressing by massive and more or less stagnant "palaeocrystic" sea ice in this part of the Arctic Ocean (Bradley and England, 2008), which forced the glaciers to run alongside the coast Möller et al., 2010;Funder et al., 2011b) (Fig. 9). In this scenario the outlet glaciers on the coastal plain would be minor contributors to an ice shelf fed by ice streams in Nares Strait and the Greenland fjords to the west of the plain (Kelly and Bennike, 1992;England et al., 2006) (Fig. 9). These new results indicate more vigorous LGM glacier dynamics especially in the Nares Strait region and the Greenland fjords to the west while a more passive regime prevailed over northernmost Greenland, probably because of precipitationestarvation, as at present where this area contains some of the largest contiguous ice-free areas in Greenland. Recently, in support of this, IBCAO Version 3.0 with a relatively dense amount of spot soundings, shows a glacial trough-like bottom morphology at the northern entrance to Nares Strait that extends directly to the inferred LGM ice limit. These data have been contributed by the Canadian Hydrographic Service , and it seems reasonable that this ice stream could feed an ice shelf. Similarly, at the eastern side of the North Cap, at the Fram Strait coast, an outlet from the Greenland ice sheet carved troughs during LGM outside the mouth of Independence Fjord, and fed shelf-ice that extended to the midouter shelf (Nørgaard-Pedersen et al., 2008;Funder et al., 2011b) (Fig. 9).
About 80 km northeast of Kap Morris Jesup (Peary Land) is the submarine Morris Jesup Rise that begins its northward extension from the Greenland shelf towards the Arctic Ocean (Fig. 9). The northernmost tip of this bathymetric high was mapped with a subbottom profiler (Fütterer, 1992) revealing a glacially scoured seabed extending to approximately 1000 m present water depth (Spielhagen et al., 2004). This area of Morris Jesup Rise was later mapped by a multibeam and chirp sub-bottom profiler (Jakobsson et al., 2008a). Large iceberg scours were found crossing the crest from approximately west to east at a present water depth of 1045 m   (Fig. 9). The part of Morris Jesup Rise surveyed by Oden extends at the shallower end to a water depth of 940 m and contains no signs of grounding by a coherent ice shelf .
During LOMROG 2007 some parts of the flat topped southeastern Lomonosov Ridge was mapped. Sea-ice conditions hampered the geophysical mapping severely, but the few acquired chirp sonar profiles did show that the ridge crest had been subjected to extensive ice grounding at depths shallower than 785 m present water depth. The character of the mapped ice grounding led Jakobsson et al. (2010b) to propose that a large coherent ice shelf extended out to this location from the nearby Ellesmere Is-landeGreenland continental shelf (Fig. 9). Prior to LOMROG 2007, Kristoffersen and Mikkelsen (2006) and Nørgaard-Pedersen et al. (2007) found, from seismic and sediment-core data retrieved from the GreenIce drift station over the south-westernmost part of the Lomonosov Ridge, evidence of ridge top erosion at a depth of about 500e600 m. They suggested that erosion might have been caused by deep draft icebergs or an ice shelf tongue extending at least 100 km north from the shelf edge north of Ellesmere Island.

Stratigraphy and chronology
Combining the evidence from land with IRD, detrital Fe, and sediment-accumulation rates in nearby ocean cores Larsen et al., (2010) suggest that the proposed ice shelf adjacent to the coastal plain began to build up as early as 30 cal ka BP. Furthermore, they suggest that both here and farther south at the Fram Strait the ice extent on the shelf culminated at 25e20 cal ka BP. After this, retreat began, and shortly before 10 cal ka BP the presently ice free land on the plain had been exposed and glacier margins had reached their present location Möller et al., 2010;Olsen et al., 2012). However, up until c. 8.5 cal ka BP abundant icebergs floated along the coast and deposited large boulders. This shows both that glaciers were calving in the fjords, and that the sea ice was allowed to float freely at least for some time of the year and not locked in palaeocrystic ice or fast ice as at present (Funder et al., 2011a). To the south-east, at the mouth of Nares Strait, deglaciation followed the same pattern, and when the Nares Strait ice stream finally disappeared, an important connection between the Arctic and the North Atlantic Oceans was established allowing the development of modern ocean circulation in Baffin Bay and the Labrador Sea (England 1999;England et al., 2006;Jennings et al., 2011). Larsen et al. (2010) suggested that the final disintegration of the marine ice sheet component was caused by higher summer temperatures and eustatic sea-level rise, but also by inflow of warm Atlantic water through Fram Strait, providing a shallower halocline, and a switch of glacialeinterglacial paleoceanographic circulation regimes. This is in agreement with results from other parts of the Arctic Ocean and its borderland (Bradley and England, 2008;Jakobsson et al., 2010b;Jennings et al., 2011). The possibility that the Arctic Ocean operates with different glacialeinterglacial circulation regimes is further discussed below. The deep ice scouring on the Morris Jesup Rise and the ice grounding on southern Lomonosov Ridge is dated to MIS 6 . This point to that ice shelves most likely were reoccurring phenomena at the Quaternary glacial maxima.

Landforms
The first discoveries of ice groundings in the deep (>800 m below present sea level) Arctic Ocean were made by Vogt et al. (1994) who mapped both lineations and individual plough marks on the Yermak Plateau. This was followed by the discovery of extensive ice grounding on the central Lomonosov Ridge down to 1000 m present water depth (Jakobsson, 1999;Polyak et al., 2001) ( Figs. 1 and 10). Later, the bathymetric highs adjacent to the Arctic Ocean continental margins have been partly mapped during several different icebreaker expeditions. Few mapped areas shallower than 900e1000 m are without plough marks from deep drafting icebergs or more coherent ice masses. More detailed mapping of this part of the ridge crest is needed to further interpret the glaciogenic landforms in the context of the glacial history of the Arctic Ocean.

Stratigraphy and chronology
Modern and Holocene sedimentation rates in the Arctic decrease significantly with distance from the continental margins and oceanic gateways. This largely reflects shelf-to-basin transport processes, inputs from major rivers, and commonly much higher sedimentation rates in the marginal ice zone. In the central Arctic, the two major transport processes for sediments include melt-out from sea ice, and the lateral transport of fine-grained material in ocean currents. As a result, while sedimentation rates of 10's to 100's cm/ka are found along the continental slopes, radiocarbon dating of central Arctic sediments reveal rates of rarely not more than a few cm/ka (Backman et al., 2004). The lowest sedimentation rates occur in regions with the most persistent sea-ice conditions, including the central Amerasian Basin (Levitan and Stein, 2008;Polyak et al., 2009;Stein et al., 2010). This precludes studies of millennial-scale climate variations with a few exceptions where sediment focussing has taken place (Hanslik et al., 2010).
During MIS 2 (14e29 ka; Lisiecki and Raymo, 2005) sedimentation rates in the central Arctic were significantly reduced compared to the Holocene. This reduction has been interpreted largely as a sign of more severe ice conditions and reduced bioproductivity (Darby et al., 1997;Nørgaard-Pedersen et al., 1998;Poore et al., 1999). One of the most striking observations from the Western Arctic is the apparent cessation in LGM sedimentation on the Mendeleev Ridge, where a hiatus between 13.7 and 19.7 14 C yr BP occurs in 8 radiocarbon dated sediment cores (Fig. 11). This break in sedimentation has been previously associated with the development of coherent and thick perennial sea ice in the region during full-glacial conditions, or the development of paleocrystic sea ice, and/or an ice shelf .
In the Eurasian sector of the central Arctic, there is a less distinct break in LGM sedimentation (Fig. 11). For example, on the circumpolar regions of the Lomonosov Ridge, high resolution 14 C dating of Neogloboquadrina pachyderma from some cores reveal continued but diminished sedimentation rates during the LGM, while in other records there is evidence for a correlative LGM hiatus (Hanslik et al., 2010) (Fig. 11). On the Morris Jessup Rise, Lomonosov Ridge, North of Greenland and across the central Eurasian Basin, sedimentation also appears to have continued during the LGM at a reduced rate (Fig. 11). These patterns contrast with sedimentation rates in Fram Strait, Yermak Plateau and at sites close to glaciated continental margins, like north-eastern Greenland and the BarentseKara shelf, where relatively high sedimentation rates persisted (Nørgaard-Pedersen et al., 2003).
Overall, in the Eurasian Basin, clear gradients towards higher sedimentation rates, fluxes of planktic foraminfera and surface water d 18 O, are interpreted as evidence for a less extensive LGM sea-ice cover and the presence of Atlantic water masses (Nørgaard-Pedersen et al., 2003). Relatively high biologic productivity is also reported for LGM sediments along the Northern Barents Sea margin, where upwelling of Atlantic water from katabatic winds coming off the Svalbard/Barents ice sheet may have generated a persistent coastal polyna along the Barents margin to at least the position of the Saint Anna Trough (Knies et al., 1999) (for location see Fig. 1). More recent reconstructions of LGM sea ice, based on phytoplankton derived biomarkers, indicate that perennial sea ice extended south of 81 N in the Fram Strait and over the Yermak Plateau between 29e27.5 ka, and 23.5e17 ka (Müller et al., 2009), and also across the Southern Lomonosov Ridge near the Laptev shelf between 30 and w15 ka (Stein and Fahl, 2013). Fig. 10. Sub-bottom profile across the shallow crest of the Lomonosov Ridge acquired from USS Hawkbill 1999 (Edwards and Coakley, 2003). The ridge crest has been extensively eroded by deep drafting ice.
The development of less severe sea-ice in Fram Strait began w17 cal ka BP, a transition attributed to enhanced inflow of Atlantic water to the Arctic, and the subsequent or coincident retreat of the Svalbard/Barents Sea Ice sheet (Müller et al., 2009). However, perennial sea ice over southern Yermak Plateau between 23.5 and 17 cal ka BP, also corresponds to peak abundances of benthic foraminifera associated with Atlantic water inflow (Wollenburg et al., 2004), suggesting a subsurface inflow of Atlantic water to the glacial Arctic; something that will be further addressed below in Section 5.
The low sedimentation rates that dominate much of the central Arctic during the LGM do not characterize the entire glacial period. Although the temporal resolution of most central Arctic cores remains either too low or too poorly resolved to capture millennial scale changes, evidence for abrupt environmental changes during the Last Glacial cycle (MIS 4e1) are captured by recently published stacked records of ostracode abundance and species composition (Poirier et al., 2012), bottom water temperature reconstructions , and preliminary studies on sea ice biomarkers from Southern Lomonosov Ridge/Laptev shelf (Stein and Fahl, 2013). Similarly, sudden inputs of iceberg-derived IRD to the western Fram Strait during MIS 3 and 2, are linked through the provenance of Fe-oxide grains to calving events of the Laurentide ice sheet (Darby et al., 2002). Similar events are recorded in cores with lower temporal resolution from western and central Arctic (Darby and Zimmerman, 2008). Improved constraints on the timing of these events requires dedicated drilling in more marginal settings of the Arctic where high sedimentation rates have so far prevented the acquisition of sedimentary records capturing the entire Last Glacial cycle.
Beyond the range of 14 C dating, age models for central Arctic sediments become difficult to establish (Backman et al., 2004;Alexanderson et al., 2014). However, across much of the central Arctic, cyclic patterns in sediment composition indicate distinct modes of glacial/stadial and interglacial/interstadial sedimentation (O'Regan et al., 2008;Sellén et al., 2010). Interglacial/interstadial sediments are largely bioturbated fine-grained muds (i.e. dominated by silt and clay), with moderate to low amounts of sea-ice transported coarse-grained material (<10% wt). They generally contain benthic and planktic foraminifer, ostracods and calcareous nannofossils that are indications of moderate productivity under seasonally variable sea-ice. These sediments are often dark brown in colour, owing to an enrichment of manganese hydroxides, derived from riverine and shelf sediments delivered to the Arctic (März et al., 2011;Löwemark et al., 2012). By contrast, glacial and stadial sediments are often beige to grey in colour, are largely devoid of microfossils and signs of bioturbation, and contain higher amounts of coarse-grained material (e.g. Jakobsson et al., 2000;O'Regan et al., 2008;Polyak et al., 2009).
Even in the central Arctic, elevated amounts of coarse-grained material can occur in thick diamictons (decimetres to >1 m thick), as found on the Lomonosov Ridge (Svindland and Vorren, 2002) during MIS 6, 5b, 5/4 and 4/3 Spielhagen et al., 2004). These diamictons are related through timing and mineral composition to the extension of the Barentse Kara ice sheet to the shelf edge, where presumably armadas of large tabular icebergs were shed during the Last 2 Glacial cycles Kristoffersen et al., 2004;Spielhagen et al., 2004). However, there is no indication that such facies were deposited on the Lomonosov Ridge prior to MIS 6, even when looking at the longer 20e30 m Pliocene through Quaternary section recovered during the Arctic Coring Expedition (ACEX) . The oldest recognized diamicton on the central Lomonosov Ridge occurs during MIS 6, and corresponds to the earliest dated period of glacial erosion on the ridge crest, likely caused by the scouring of deep drafted and coherent masses of ice (Jakobsson, 1999;Jakobsson et al., 2010b;O'Regan et al., 2010).
Based on current studies, correlative thick late Quaternary diamictons are not clearly recognized in sediments from the Amerasian Basin. While coarser-grained beige and laminated grey coloured sediments still dominate glacial sediments, in the Amerasian Basin there appears to be an overall increase in the coarse fraction content in the middle part of the Pleistocene. These coarse grained units are often elevated in dolomites, which point towards a provenance in the Canadian Arctic Archipelago (Phillips and Grantz, 2001;Darby et al., 2006;Polyak et al., 2009;Stein et al., 2010). For most of the Quaternary, direct correlations between these events and the growth and decay of the LIS remain difficult because of the poor age control in the marine sediment records, the generally low sedimentation rates in these distal settings, and in contrast to the Eurasian ice sheets, a lack of calibrated ages for the timing of Laurentide ice sheet advances to the shelf edge. On top of this, there remains considerable uncertainty about the timing, nature and extent of Quaternary glaciations in the Bering and East Siberian seas.

Simulation of glacialeinterglacial ocean circulation
During full-glacial conditions with >100 m lower sea level (Lambeck et al., 2002), the Arctic Ocean, including Greenland, Iceland and Norwegian seas, differed significantly from its present day configuration. The closing of the Barents Sea, the Bering Strait and the channels through the Canadian Archipelago, together with the subareal exposure of the shallow ChukchieEast SiberianeLaptev Sea region, represents the most dramatic physiographic reorganization. During peak glacial times, Fram Strait was the single oceanic gateway to the Arctic Ocean ( Figs. 1 and 3). Furthermore, freshwater input to the glacial Arctic Ocean was reduced during glacial conditions, due the termination of low salinity Pacific waters entering through the Bering Strait and an overall reduction of the hydrological cycle (Lohmann and Lorenz, 2000). Different surface wind conditions in the glacial world (Shin et al., 2003;Stärz et al., 2012) may also yield altered circulation in the Arctic Ocean (Stigebrandt, 1985) and deep water formation in the North Atlantic appears to have been reduced (Broecker, 1997;Shin et al., 2003).
Given these pronounced changes, it likely that the Arctic Ocean circulation has a glacial mode, distinctly different from present-day. Based on theoretical considerations and paleoceanographic data, a number of circulation schemes have been proposed. These range from an essentially ice-free Arctic Ocean (Donn and Ewing, 1966;Olausson and Jonasson, 1969) to one that is completely covered by thick sea ice (Weyl, 1968;Bradley and England, 2008). On the basis of the reduced freshwater input, Olausson and Jonasson (1969) hypothesized that the low salinity surface layer was absent in the glacial Arctic Ocean. They envisioned a weakly stratified Arctic Ocean, which was kept ice free through heat transport associated with Atlantic water circulation. A rough estimate, using an annual-mean heat loss of 100 W m À2 from an open-water area of approximately 7$10 12 m 2 , suggests that an oceanic heat transport of about 0.7 PW across the GreenlandeScotland Ridge is required to keep the glacial Arctic sea-ice free. This is about three times greater than the present-day northward oceanic heat flux (Hansen and Østerhus, 2000), indicating that the completely ice-free scenario of Olausson and Jonasson (1969) is less plausible. Indeed, there are paleoceanographic data indicating that the low-salinity surface water, which is presently confined to the Arctic Ocean and along the east coast of Greenland, frequently extended south eastward over the Greenland and Norwegian seas (Dokken and Jansen, 1999;Bauch et al., 2001). The low-salinity surface layer stabilized the water column in the Nordic Seas, allowing an expansion of the sea-ice cover. As a result the deep water sources north of the GreenlandeScotland Ridge shifted from predominantly openocean convection in the central Nordic Seas to brine-related dense water formation in the near-coast shelf areas (Dokken and Jansen, 1999;Bauch et al., 2001;Haley et al., 2008). However, the oceanographic conditions in the northern North Atlantic and the Arctic Ocean appears to have varied significantly within a glacial period, exhibiting pronounced shifts between warmer interstadials and colder stadials (Stigebrandt, 1985;Ganopolski et al., 1998;Dokken and Jansen, 1999).
During glacial times, warmer Atlantic water masses appears to have penetrated northward over the GreenlandeScotland Ridge and into the NorwegianeGreenland seas and Arctic Ocean (e.g. Hebbeln et al., 1994;Bauch et al., 2001;Cronin et al., 2012). The subsurface warming was particularly pronounced during MIS 3, when the central Arctic Basin from 1000 to 2500 m was occupied by water masses that could have been as much as 2e4 C warmer than today . Simultaneously, it appears that temperatures between the surface and 500 m water depth were close to the freezing point, a depth interval occupied today by warm Atlantic Waters. Though the vertical resolution of existing data is limited, this seems to suggest that the cold halocline deepened, resulting in a downward displacement of the glacial intermediate layer of warm Arctic Atlantic Water. A downward displacement of   Table 1). Calibrated ages are presented with AE2 standard deviations, and were calculated using the CALIB 6.0 program with the Marine09 reservoir correction curve. For consistency, an additional regional DR of 300 years was applied to all samples. Figures highlight the generally low sedimentation rates in the central Arctic (AeC) when compared to marginal sites (D, F). A reduction in sedimentation rates is seen across the central Arctic Ocean during MIS 2 (AeC), but it is only in the Amerasian Basin that an apparent widespread hiatus exists. Increasing the DR to 1000 years for samples with uncalibrated ages >10 ka (Hanslik et al., 2010), does not significantly affect the distribution of ages, or the duration of the inferred hiatus (A; hatched bars in histogram). This period of reduced sedimentation appears to correspond to the sea-level lowstand beginning 28e30 ka and lasting until w12e14 ka when Pacific inflow through the Bering Strait resumed (CALIB 6.0.1 Reservoir Correction program. Available online at: http://calib.qub.ac.uk/calib/). the warm Arctic Atlantic Water is also suggested to be the premise for the development of extensive marine shelf-ice complexes in the Arctic Ocean during MIS 6 and presumably also during the LGM (Fig. 12).
To what extent does climate and ocean-circulation modelling provide additional information and constraints on the glacial Arctic Ocean circulation? Most climate-model based efforts have targeted the LGM and model-data comparisons have mainly concerned the Atlantic Ocean and the Southern Ocean (Otto-Bliesner et al., 2007). Moreover, even high-resolution regional ocean-circulation models have difficulties in reproducing the present-day circulation of Arctic and Atlantic waters (Holloway et al., 2007). Recently Stärz et al. (2012) adopted a regional (w30 Â 30 km resolution) North AtlanticeArctic Ocean circulation model to study LGM conditions. Results from a global climate model simulation by Shin et al. (2003) provide boundary conditions at the southern model boundary as well as initial conditions. Under their standard LGM forcing, the Atlantic water entering the Arctic Ocean via Fram Strait has temperatures close to the freezing point. Furthermore, the main halocline shoals to about 100 m in the LGM experiment, compared to 200 m in the present-day control experiment. At first sight this result appears to contradict evidence for the presence of a deep glacial halocline, which Jakobsson et al. (2010b) and Cronin et al. (2012) proposed based on conceptual modelling and paleooceanographic data (Fig. 12). However, the LGM simulation of Stärz et al. (2012) shows that, in the Fram Strait, the outflowing layer from the Arctic extends down to about 1500 m and that the Atlantic inflow has its centre displaced towards the bottom. This is in broad agreement with the glacial Arctic Ocean stratification suggested by Jakobsson et al. (2010b) and Cronin et al. (2012), although the Arctic Atlantic Water temperature in the LGM simulation of Stärz et al. (2012) appears to be somewhat too cold to match the available Mg/Ca temperature proxies from Arctic Ocean ostracodes.
There are also theoretical ideas that can illuminate the glacial circulation of the Arctic Ocean. A more estuarine circulation in the glacial Arctic Ocean, characterized by an extensive low salinity surface layer, could emerge if the northward heat flux across the GreenlandeScotland Ridge was reduced sufficiently (Stigebrandt, 1985;Spall, 2012). If an estuarine circulation mode prevailed, then the generally decreased freshwater input to the Arctic Ocean is expected to cause the upper-ocean low-salinity layer to extend deeper; i.e. a deeper cold halocline (Nilsson and Walin, 2010;Jakobsson et al., 2010b).
The closing of the Bering Strait and the Barents Sea would have impacted ocean circulation patterns both regionally and globally (Hu et al., 2010(Hu et al., , 2012. Stigebrandt (1984) proposed that the salinity difference between the North Atlantic and the North Pacific is presently regulated by the Bering Strait flow. This idea has inspired further speculations on how a closing of the Bering Strait gateway could affect the Atlantic thermohaline circulation (De Boer and Nof, 2004). Circulation modelling suggests that a closing of the Barents Sea may reduce the flow of Atlantic water through the Arctic (Aksenov et al., 2011). Another key factor is the winds over the Arctic Ocean that generate a current flow, which is important for transporting the Atlantic water through the Arctic Ocean (Nøst and Isachsen, 2003). In the glacial state, both the wind forcing and the geometry of the closed depth contours change. Evidently, this would change Atlantic water circulation in the Arctic Ocean.
In summary, our understanding of the glacial Arctic Ocean circulation has been significantly advanced by data-based analyses during the last decade. However, the data also pose some theoretical challenges, including the possibility of a very deep cold halocline and oceaneice-sheet interactions in the glacial Arctic Ocean. It is foreseen, however, that the picture emerging from recent reconstructions together with theoretical and numerical modelling will help to further improve our knowledge of the glacial Arctic Ocean circulation.

Simulation of ice sheets and ice shelf complexes
Prior to the 1970s, only few numerical models simulating ice sheets and ice shelf shelves were available. Stability of ice shelf complexes was mainly discussed for the West Antarctic Ice Sheet in the context of the sparse database of field observations available and/or mathematical analyses of the first theoretical descriptions of ice shelf and ice sheet dynamics (Weertman, 1974;Hughes, 1975;Thomas, 1979). Both Mercer (1970) and Hughes et al. (1977) transferred ideas and concepts developed for Antarctic configurations to the Arctic when they proposed extensive Arctic Ocean ice shelves. Numerical simulations of hypothesized Arctic Ocean ice shelves were, however, impossible at this time. Early numerical ice shelf models, e.g. by Thomas and Bentley (1978), were not treating ice dynamics in the transition zone between an ice shelf and an ice sheet correctly. At present, ice shelf dynamics are being incorporate into advanced ice sheet models, although it still presents a great challenge. To be able to outline how this is done, a brief summary of numerical ice sheet models is provided here.
Based on numerical glaciological concepts from the pioneering work by Weertman (1974); Hughes et al. (1977) reconstructed a huge "late-Würm Arctic Ice Sheet". However, until roughly a decade ago, numerical ice models only captured the behaviour of the slowly moving (tens of meters/year) homogeneous interior parts of land-based ice sheets. These "zero order Shallow Ice Approximation (SIA)" models regard ice sheets as essentially flat in order to simplify the equations governing their dynamics (Hutter, 1983;Kirchner et al., 2011). Due to the shallowness-assumption, SIA models work well in the interior of an ice sheet, but fail in regions with highly variable ice dynamics, e.g. fast-flowing ice streams and outlet glaciers (velocities of up to 1000 m/yr) and when approaching the grounding-line zone. Furthermore, ungrounded floating ice including ice shelves and glacier tongues cannot be treated with SIA models.
Although SIA models have limitations, they have added valuable insights on the glacial history of ice sheets in the circumpolar Arctic (e.g. Siegert et al., 1999;Siegert and Dowdeswell, 2004;Tarasov and Peltier, 2004;Stokes and Tarasov, 2010). Comparison of simulated ice sheets using SIA models with landforms, such as MSGL, GZW, and moraine ridges, is difficult because SIA models fail where these landforms typically are produced, i.e. at the marine ice margins where ice shelves and floating glacier tongues interlink the terrestrial and ocean glacial system (Joughin and Alley, 2011).
During the last decade, the ice-sheet modelling community has developed improved ice-sheet models, so-called "higher order" and "full Stokes" models. These account for the slow flow dominating the grounded inland ice motion, the rapid ice flow in ice streams, ice flow across the grounding line, and ice-shelf flow (Pattyn, 2003;Schoof, 2007;Pollard and Deconto, 2012;Seddik et al., 2012;Larour et al., 2012). Full Stokes models are the most accurate ones from a physical point of view, but they are the most costly to run. Furthermore, Simulation results from SIA models, higher order, and full Stokes models have been compared with respect to simplified model problems (Pattyn et al., 2008(Pattyn et al., , 2012Calov et al., 2010).
Simulation of ice dynamics of the Quaternary Arctic Ocean marine glaciers calls for the use of full Stokes models. However, their spatial dimensions and longevity do not yet allow for it, because full Stokes models are still restricted to centennial timescales and smaller spatial domains. Alternatively, full Stokes models and SIA models could be coupled adaptively, i.e. switching to SIA mode where permitted and to full Stokes mode where required. This coupling is technically complicated and requires re-investigation of typically applied scaling relations (Ahlkrona et al., 2013). Of the existing higher order models only two have so far been applied to simulate past glacial conditions in the circum-Arctic (DeConto et al., 2008;Alvarez-Solas and Ramstein, 2011;Koenig et al., 2011). However, even in these applications, coupled ice-sheet/ice-shelf dynamics have either been switched off or could not be resolved by the experimental setup, implying that the higher order ice-sheet model component was run in SIA mode only.
Until ice sheet models of higher-order and full Stokes models are adapted to become more easily applicable, simulations of Arctic Ocean palaeoeice complexes including ice streams and ice shelves have to rely on SIA models, or employ entirely different modelling strategies. Kirchner et al. (2013) proposed a statistical model to investigate likely spatial dimensions of the hypothesized Arctic Ocean MIS 6 ice-shelf complex of Jakobsson et al. (2010b). The statistical model by Kirchner et al. (2013) is designed to specifically assess whether the extensive MIS 6 ice shelf complex could have been a likely source of deep draft icebergs since this has been proposed (e.g. Dowdeswell et al., 2010b;Jakobsson et al., 2010b;O'Regan et al., 2010). The statistical concept is based on establishing relations between contemporary Antarctic ice shelves and their local physical environment, and the assumption that Arctic Ocean MIS 6 ice shelves scaled similarly. Data of contemporary Antarctic ice shelf areas, calving front length, and ice thickness along the calving front are combined with methods from extreme values statistics to derive a probabilistic description of the maximum draft (depth below sea level) along the calving fronts of various hypothesized MIS 6 ice shelf configurations in the Amerasian sector of the Arctic Ocean. The assumption that mapped ice groundings at exceptionally deep water depths should have been generated during extreme events when unusually deep-drafting icebergs were produced is reflected in the specific choice of statistical methodology: concepts from extreme value statistics provide a firm-based statistical framework that is designed to deal with extreme, rather than common events. The results obtained proved robust, and likely sources for icebergs large enough to scour the seafloor at w1000 m depth below present sea level were identified. Furthermore, results from the statistical model also indicate the range of the possible extent of the Amerasian Arctic Ocean ice shelf complex, of which one configuration is suggested in Jakobsson et al. (2010b).

Discussion and conclusions
Advances made since Mercer (1970) presented the hypothesis of a former ice sheet in the Arctic Ocean are numerous, yet we still lack information from key areas required to compile the holistic Arctic Ocean glacial history. This present circumpolar overview of glacial landforms, stratigraphies, and chronologies summarizes the current state of knowledge and identifies a set of outstanding questions arising from this synthesis. It should be noted that several of the questions we identify are easier to answer today than a few decades ago due to the recent development of techniques including new dating methods and much improved high-resolution sea-floor mapping capabilities. This implies that some previously investigated areas of the Arctic Ocean may benefit from being revisited for new surveys. There is also a need for an improved integration of the different datasets, e.g. marine and terrestrial as well as ice-sheet modelling results.
The question of huge ice shelves in the central Arctic Ocean and how these were fed from circumpolar margins is far from resolved. Although the ice-grounding events reaching >1000 m below present sea level are currently dated to MIS 6 (e.g. Jakobsson et al., 2010b), other expansive but thinner ice shelves likely formed throughout the Quaternary (Polyak et al., 2007). In fact, given evidence for dynamic growth and collapse of Arctic ice shelves through the Holocene (Antoniades et al., 2011), and considering the Antarctic as an analogue to the glacial Arctic, it seems plausible that ice shelves would have developed in several parts of the Arctic Ocean. However, these enigmatic features of the glacial Arctic, including their extent and links to terrestrial ice sheets, are difficult to establish. Identification of subice-shelf sedimentary facies in the Arctic could provide a critical way to constrain their past distribution, but few 10e20 m long sediment cores collected from the marginal sites of the Arctic extend back to the LGM, where these subice-shelf sediments would be most obviously preserved. Whether the lack of LGM sedimentation in numerous Arctic Ocean cores from the Western Arctic is attributable to the existence of large ice shelves remains an open question. Similarly, assertions that these ice shelves may have extended over the central Lomonosov Ridge are not conclusive (Jakobsson, 1999;Polyak et al., 2001). This area is one that should be revisited and fully mapped with modern high-resolution swath bathymetry. Furthermore, the southern Lomonosov Ridge off Greenland is another hot spot for the central Arctic Ocean glacial history as evident from seismic lines collected in this region showing extensive ice erosion (Kristoffersen and Mikkelsen, 2006) and terrestrial landforms that suggest the passage of large ice shelves during the LGM and possibly earlier glaciations.
The Barents Sea is likely the best known area in the high Arctic regarding the glacial history, although large gaps in data coverage still remain. Outstanding questions are therefore on a more detailed level. One such question is how the SvalbardeBarents Sea ice sheet behaved dynamically, from the LGM to the time of its retreat (e.g. Ingólfsson and Landvik, 2013). Did ice domes undertake major reorganizations during the rapid retreat phase, which likely was strongly coupled to oceanographic and sea-level changes? This could explain why marine and terrestrial data from the eastern Svalbard region are still from many points of view contradictory. The new discoveries of landforms suggesting a highly dynamic post-LGM retreat of the SvalbardeBarents Sea ice sheet, with the ice leapfrogging back, is of particular interest considering the present day concerns regarding instabilities of marine ice sheets, in particular the Pine Island Bay area of West Antarctica (e.g. Katz and Worster, 2010).
Ice-stream dynamics in the Beaufort Sea are a key to unravelling the glacial history in the western Arctic Ocean. The advances of the north-western LIS appear to be linked to glaciogenic features on the sea floor along the Alaskan shelf and Chukchi Borderland (Northwind Ridge and Chukchi Plateau). Recent research indicates that only two ice-advances reached the Mackenzie Trough in the late Quaternary, and one of these took place during LGM (Murton et al., 2010;Batchelor et al., 2013). This raises the question whether there have been additional large ice advances onto the Canadian Beaufort Sea shelf that are not yet mapped, and if so, did they contribute to the growth of ice shelves in the western Arctic Ocean? The stratigraphic record from the Chukchi supports this possibility, as an ice erosional event there is associated with Laurentide-sourced ice in MIS 4 (Polyak et al., 2007). Alternatively, was the growth of huge ice shelves limited to MIS 6, as the central Arctic Ocean records seems to imply, with erosional events in the Chukchi Sea area linked to independent glacial activity in Beringia? One key area that may be worth revisiting to further investigate links between the LIS and Chukchi is the Alaskan Beaufort where Engels et al. (2008) suggested the presence of glacial lineations from side-scan imagery. New high-resolution swath-bathymetric mapping and coring in this area may provide important pieces of information. Finally the widespread evidence for IRD events in central Arctic sediments, and as far away as the Fram Strait, suggest multiple and large icesurging events of the CAA ice streams, from which at least seven ice streams discharged from the IIS and northwest LIS during MIS 2 (Blasco et al., 1990;Stokes et al., 2005;England et al., 2006;Stokes et al., 2009;MacLean et al., 2010;Fritz et al., 2012). Earlier glaciations from the CAA will require higher-resolution and deeper penetrating offshore records proximal to the ice sources.
The recent mapping of the Chukchi Borderland and easternmost East Siberian Sea margin raises several outstanding questions regarding the glacial history in this part of the Arctic Ocean (Dove et al., 2014). Previous assumptions of an ice free cold tundrasteppe stretching from the easternmost Taymyr Peninsula in the west to east of Wrangel Island during LGM will have to be revisited. New data seems to suggest that ice-sheet centres indeed existed on the Chukchi and/or outer East Siberian shelf (Dove et al., 2014); even on the northern part of the East Siberian Islands glacial impacts seems to be preserved (Basilyan et al., 2010). Much more field work will be required in these areas before a holistic view of the Quaternary glacial components in this area can be established.