Asynchronous glacier dynamics during the Antarctic Cold Reversal in central Patagonia

We present 14 new 10 Be cosmogenic nuclide exposure ages quantifying asynchronous readvances during the Antarctic Cold Reversal from glaciers in the Baker Valley region of central Patagonia. We constrain glacier and ice-dammed palaeolake dynamics using a landsystems approach, concentrating on outlet glaciers from the eastern Northern Patagonian Ice ﬁ eld (NPI) and Monte San Lorenzo (MSL). Soler Glacier (NPI) produced lateral moraines above Lago Bertrand from 15.1 ± 0.7 to 14.0 ± 0.6 ka, when it dammed the drainage of Lago General Carrera/Buenos Aires through Río Baker at a bedrock pinning point. At this time, Soler Glacier terminated into the 400m “ Deseado ” level of the ice-dammed palaeolake. Later, Calluqueo Glacier (MSL) deposited subaerial and subaqueous moraines in the Salto Valley near Cochrane at 13.0 ± 0.6ka. These moraines were deposited in an ice-dammed palaeolake uni ﬁ ed through the Baker Valley (Lago Chalenko; 350m asl). The Salto Valley glaciolacustrine landsystem includes subaqueous morainal banks, ice-scoured bedrock, glacial diamicton plastered onto valley sides, perched delta ter- races, kame terraces, ice-contact fans, palaeoshorelines and subaerial push and lateral moraines. Boulders from the subaqueous Salto Moraine became exposed at 12.1 ± 0.6 years, indicating palaeolake drainage. These data show an asynchronous advance of outlet glaciers from the Northern Patagonian Ice ﬁ eld and Monte San Lorenzo during the Antarctic Cold Reversal. These advances occurred during a period of regional climatic cooling, but differential moraine extent and timing of advance was controlled by topography and calving processes. © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Late Quaternary palaeoclimate and the Patagonian Ice Sheet
During the Last Glacial Termination in the Southern Hemisphere, a phase of warming atmospheric and ocean temperatures from~18.0 ka was interrupted by cooling during the Antarctic Cold Reversal (ACR; 14.7 to 13.0 ka) (Jouzel et al., 2001;Morgan, 2009;Putnam et al., 2010;Pedro et al., 2016). The ACR dominates Southern Hemisphere palaeoclimate records up to at least 40 S (Pedro et al., 2016), therefore potentially affecting glaciers of the Northern Patagonian Icefield and Monte San Lorenzo ice cap (NPI and MSL, respectively; Fig. 1) in the Patagonian Andes of South America. Glacier readvances during the ACR have been documented in the tropical Andes (10e20 S) (Jomelli et al., 2014(Jomelli et al., , 2017, for Colonia Glacier of the NPI (Nimick et al., 2016), for Tranquilo Glacier of MSL (Sagredo et al., 2018), in Cordillera Darwin (54 S) (Menounos et al., 2013), and in Torres del Paine (51 S) (Fogwill and Kubik, 2005;García et al., 2012). However, some have argued that a significant readvance also occurred during the Northern Hemisphere Younger Dryas (Ackert et al., 2008;Glasser et al., 2012Glasser et al., , 2016Rasmussen et al., 2014;Nimick et al., 2016). Palaeoclimatic controls on glaciers and the role of the bi-hemisphere seesaw (Pedro et al., 2018) in modulating climate in central Patagonia therefore remain poorly resolved.
Topographic controls and glaciolacustrine dynamics also have the potential to influence the timing of moraine formation (Barr and Lovell, 2014), which may account for the range of published ages from~15.0 to~10.0 ka for moraines at broadly coincident ice margin positions in this region (e.g. Glasser et al., 2012;Nimick et al., 2016). In order to use the geomorphic record of glacier advances as proxy data for reconstructing the structure of these past climatic changes, detailed landform mapping and geochronological data are required from a larger sample of moraines. Furthermore, Fig. 1. A. Regional overview of study area, showing rivers, lakes, glaciers and key geomorphology. Moraines and sandar are modified from Glasser and Jansson (2008). Delta terraces modified from various authors (Bell, 2008;Bell, 2009;Bourgois et al., 2016;Glasser et al., 2016;Bendle et al., 2017b;Thorndycraft et al., 2018). Boulder bars from Thorndycraft et al. detailed mapping of morphostratigraphic relationships between glacial and glaciolacustrine landforms is required to elucidate glacier-glaciolacustrine dynamics.

Aims and objectives
We aim to test the synchronicity of glacier behaviour in the Baker Valley region of the central Patagonian Andes (Fig. 1) during the Last Glacial Termination and elucidate palaeoclimatic and topographic controls on moraine formation through new geomorphic mapping and cosmogenic exposure-age dating from previously undated ice margins of both NPI and MSL outlet glaciers. These moraines are all well inside the Last Glacial Maximum (LGM) limits but beyond moraines dated to ca. 11 ka ( Fig. 1; Glasser et al., 2012), and could be expected to be contemporaneous. Detailed morphostratigraphic analyses are required to infer controls on glacier extent and glacier-glaciolacustrine interactions.
Our objectives are: (1) to constrain glacier extent and icedammed palaeolake dynamics using remote sensing and field mapping; (2) to date post-Last Glacial Maximum (LGM) readvances and the timing of palaeolake drainage events using 10 Be cosmogenic nuclide exposure ages from terminal and lateral moraines both above and below known lake levels; and (3) to evaluate the synchronicity of readvances during the Last Glacial Termination using our new geochronological datasets and previously published moraine ages.

Setting
Our study area is the Baker Valley region of the Andean Cordillera east of the NPI around 47 S (Fig. 1). Río Baker drains Lago General Carrera/Buenos Aires (Lago GCBA) and Lago Cochrane/ Pueyrred on (Lago CP), and flows south of the NPI and into the Pacific Ocean. Ice dams across the Baker valley during the last glaciation resulted in the formation of large ice-dammed lakes east of the NPI. We focus on Soler Glacier, which currently drains east from the NPI and terminates on dry land. The glacier feeds a river, which flows into Lago Plomo, which is separated by a large moraine from Lago Bertrand. Lago Bertrand drains towards the Pacific via Río Baker (Fig. 1). During past glaciations, Soler Glacier blocked the drainage of Lago GCBA into Río Baker. We focus our attention on the Bertrand Moraines, deposited by Soler Glacier above 500 m asl, and therefore above the highest level of Lago GCBA. These moraines mark an advanced postion of Soler Glacier when it discharged into a deep ice-dammed palaeolake (Glasser and Jansson, 2008;Bendle et al., 2017a;Thorndycraft et al., 2018).
We also focus on the Salto valley, which was occupied by Calluqueo Glacier from MSL during past glaciations (Wenzens, 2002). Calluqueo Glacier deposited the Esmeralda and Salto moraines in this valley following recession from the LGM ice margin (Fig. 2). In contrast with the Bertrand Moraines, these moraines are at 350 m asl, close to the elevation of an ice-dammed palaeolake.
MSL (47 35 0 S, 72 18 0 W; 3706 m high) is an isolated granodioritic to granitic massif (Ramos et al., 1982) on the eastern flank of the Andes (Figs. 1 and 2). It lies~70 km east of the southern limit of the NPI (Sagredo et al., 2016). It sustains numerous small glaciers with an ice-covered area of 139 km 2 . These glaciers are currently thinning (Falaschi et al., 2013(Falaschi et al., , 2016. Calluqueo Glacier is the largest glacier (50.9 km 2 ) (Falaschi et al., 2016) (Fig. 2). Calluqueo Glacier and glaciers on the northern flank of the ice mass drain into Río Salto via the Salto and Tranquilo valleys. Río Salto enters Valle Grande at the Salto Waterfall, by the Salto Moraines, where it joins Río Baker and drains towards the Pacific (Figs. 1 and 2).

Past glacial history
Large outlet lobes from the NPI reached the Argentinian foothills during the LGM (Caldenius, 1932;Mercer, 1976;Hein et al., 2010;Coronato and Rabassa, 2011;Mendelova et al., 2017;Bendle et al., 2017a). The two most significant eastern outlet lobes of the NPI flowed out through Lago GCBA and Lago CP (Caldenius, 1932;Mercer, 1970;Douglass et al., 2006;Hein et al., 2010;Coronato and Rabassa, 2011;Smedley et al., 2016;Mendelova et al., 2017) (Fig. 1B). Outlet glaciers from MSL flowed southeast into Argentina in the Río Belgrano valley (Caldenius, 1932;Wenzens, 2003) and north into the Tranquilo and Salto valleys (Fig. 1), where they joined the larger Lago CP outlet lobe (Wenzens, 2002). Recession occurred rapidly after 17.7 ka (Bendle et al., 2017a). By the mid-Holocene, the ice masses had separated into disparate icefields centred on the main Andean mountain peaks, Lago Jeinemeni Reserve (Douglass et al., 2005) and MSL (Sagredo et al., 2016(Sagredo et al., , 2018. Following initial glacier recession after the LGM, ice-dammed lakes formed behind the moraines in the Lago GCBA and Lago CP valleys . Geomorphic evidence for the lakes includes shorelines and stepped deltas, which suggest that the palaeolakes were stable before abrupt and rapid drops in lake level (Bell, 2008(Bell, , 2009Bourgois et al., 2016), caused by glacier recession opening up new drainage routes through lower elevation cols (Turner et al., 2005;Glasser et al., 2016). There are multiple competing models for ice-dammed palaeolake evolution (Turner et al., 2005;Bell, 2008Bell, , 2009Hein et al., 2010;Bourgois et al., 2016;Glasser et al., 2016;Thorndycraft et al., 2018). It is generally agreed that, at their maximum heights, the ice-dammed palaeolakes drained eastwards towards the Atlantic. Drainage reversals occurred following ice recession, with the lakes ultimately reaching their current extent and draining westwards through Río Baker to the Pacific. These reconstructions used different names for different stages of palaeolake evolution, but have in common the reconstruction of a large, unified palaeolake at 340e350 m asl in the Cochrane and Lago GCBA valleys, which drained through the 350 m asl col at the head of Río Bayo north of Puerto Tranquilo ( Fig. 1) . We use "Lago Chalenko" when referring to the 350 m united ice-dammed lake (cf. Thorndycraft et al., 2018), which is the indigenous name for Lago GCBA. We term the highest level of Lago GCBA the "Deseado Level" (400 m asl), as at this time the lake flowed out through the Deseado col and through Río Deseado to the Atlantic.

Last glacial-interglacial transition in central Patagonia
The West Antarctic Ice Sheet Divide (WAIS Divide) ice cores indicate steady warming in the Southern Hemisphere from 18 ka, which was interrupted by the ACR cooling (Fudge et al., 2013;Wais Divide Project Members, 2015); warming continued in the Southern Hemisphere throughout the period of the Younger Dryas. Palaeoclimate studies from central Patagonia indentify the ACR as a cold period (Hajdas et al., 2003;Massaferro et al., 2014;Moreno and Videla, 2016). Villa-Martínez et al. (2012) recorded a cool climate (2018). Published ages have been recalculated according to our protocols (Turner et al., 2005;Glasser et al., 2012;Villa-Martínez et al., 2012;Boex et al., 2013;Bourgois et al., 2016;Nimick et al., 2016;Sagredo et al., 2016Sagredo et al., , 2018. Glacier extent from Davies and Glasser (2012). Overlain on hillshaded ASTER GDEM. B. Chile, showing current icefields (black), from Davies and Glasser (2012), and extent of Patagonian Ice Sheet at LGM (light blue) (from Coronato and Rabassa, 2011). Red box shows extent of Map A. NPI: North Patagonian Icefield. SPI: South Patagonian Icefield. MSL: Monte San Lorenzo. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) with increased precipitation from 13.4 to 11.8 cal ka BP from Lago Augusta in the Chacabuco Valley (47 S). Palynological records from the nearby Lago Edita (Fig. 1) show increased precipitation between 14.5 and 13.0 cal ka BP (Henríquez et al., 2017).
The Younger Dryas (12.9e11.7 cal ka BP) was characterised by cooling in the North Atlantic (Lowe et al., 2008) and a slow-down in oceanic circulation (Lynch-Stieglitz et al., 2011). However, its expression in the Southern Hemisphere was less pronounced (Singer et al., 1998;Sugden et al., 2005;Douglass et al., 2006;Darvill et al., 2016). After 11.0 cal ka BP, palynological data from the Chacabuco Valley record declines in cold-resistant flora. A rise in Northofagus signal arboreal increase, warming and decreased precipitation (Henríquez et al., 2017).

Methods
We use a landsystems approach combined with cosmogenic nuclide dating. Systematic mapping allows landforms to be grouped into genetically linked sediment-landform assemblages, and the determination of a landsystem (Evans, 2005). Landforms were mapped in ESRI ArcGIS using ESRI Basemap World Imagery (usually Digital Globe and GeoEye images) and high resolution imagery from GoogleEarthPro. Elevation data were derived from ASTER GDEM (30 m resolution) (ASTER GDEM Validation Team et al., 2009). Slope and hillshaded elevation models were used to assist landform interpretation (Smith and Clark, 2005). Remotely sensed mapping was ground-truthed during fieldwork (Novem-bereDecember 2015, December 2016, December 2017). Field mapping utilised a handheld GPS (accurate to ± 8 m) and followed standard procedures (Smith et al., 2011;Otto and Smith, 2013). Moraine transects were conducted using a hand held GPS, an Abney level and tape measure to measure distance and angle to breaks in slope. Large transects were calculated in ArcGIS using ASTER GDEM. Regional landscape mapping built on published geomorphological maps (Turner et al., 2005;Glasser and Jansson, 2008;Bendle et al., 2017b;Thorndycraft et al., 2018). In places, dense forest and steep, unstable slopes precluded mapping.
Glacially transported boulders with abundant quartz situated on moraine ridge crests were targeted for 10 Be exposure-age dating. Rock samples were collected following standard guidelines (Gosse and Phillips, 2001;Cockburn and Summerfield, 2004;Balco, 2011;Darvill, 2013). Where possible, !5 boulder samples were sampled per moraine ridge (cf. Putkonen and Swanson, 2003;Cockburn and Summerfield, 2004). We targeted boulders with a >1 m b axis that were raised well above the local ground surface (cf. Putkonen and Swanson, 2003;Heyman et al., 2016) with signs of glacial erosion (e.g., edge-rounding, faceting), avoiding angular boulders that may have been transported supraglacially, boulders that were very weathered or flaking, or those that may have rolled or been exhumed. Samples (<5 cm thick and weighing >1 kg) were taken using a hammer and chisel (thus minimising environmental impact by making a smaller, more subtle indentation compared with other methods using rock saws) from the centre of the flat top surface of the boulders, away from cracks, edges and joints. Detailed sample descriptions are provided in the Supplementary Information. Samples were processed for cosmogenic 10 Be analysis at the SUERC Cosmogenic Isotope Laboratory. Samples were crushed and sieved to 250e500 mm and quartz was separated in a Frantz ® isodynamic magnetic mineral separator before being repeatedly etched in 2% HF until 30% of the quartz mass was lost (Kohl and Nishiizumi, 1992). Quartz purity was assayed by ICP-OES. Purified quartz was spiked with~0.23 mg of 9 Be and dissolved. Beryllium was extracted and isolated following the methodology described in Child et al. (2000) before being precipitated as Be(OH) 2 and baked to BeO in a quartz crucible. BeO was mixed with Nb and pressed into a copper cathode. Cosmogenic nuclide concentrations include a procedural blank correction of 50,751 ± 11,188 atoms ( 10 Be/ 9 Be ¼ 3.35 Â 10 À15 ± 7.37 Â 10 À16 ). 10 Be/ 9 Be was measured using the 5MW pelletron at SUERC  and normalised to NIST SRM4325 with a 10 Be/ 9 Be ratio of 2.79 Â 10 À11 (Nishiizumi et al., 2007). Cosmogenic nuclide concentrations include a procedural blank correction of 50,751 ± 11,188 atoms. The uncertainties in the cosmogenic nuclide concentrations include the AMS counting statistics and scatter uncertainties from sample, procedural blank, and standards measurements.
The ages were calculated using the CRONUS-Earth online calculator (Balco et al., 2008), version 2.3 (wrapper script 2.3; main calculator 2.1; constants 2.3; muons 1.1), with a local Patagonian production rate of 3.88 ± 0.15 atom/g/yr derived from the calibration data set of Kaplan et al. (2011) from the ICE-D online database (http://calibration.ice-d.org/), the time-varying Lm scaling method (Lal, 1991;Stone, 2000), and a rock density of 2.7 g/cm 3 . We applied a correction for topographic shielding using measurements of the horizon line conducted in the field. As erosion rates are poorly constrained regionally we use an erosion rate of 0 mm/kyr, but also provide ages with 1 mm/kyr erosion rates for comparison. The choice of erosion rate leads to little difference on boulder exposure ages within the timescale of this study, since the difference in calculated age is smaller than the calculated uncertainties. We include no correction for periodic snow cover, which is likely to be negligible in this windy setting and given the size of boulders sampled. No correction for post-glacial uplift has been applied. Published regional exposure ages are recalculated using our protocols. Cosmogenic and optically stimulated luminescence (OSL) ages are presented as "ka".
Normal kernel density estimates ("camel plots") are used to visually evaluate the ages and identify outliers. Code to generate these plots was created in Matplotlib, modified from Balco (2009). Outliers due to isotope inheritance are usually easily identified; however, incomplete exposure biases such as erosion, weathering, rolling, exhumation or burial may not produce obvious outliers, since these processes are differential. Boulder ages may therefore underestimate the timing of moraine formation (Briner et al., 2005;Heyman et al., 2016). However, an 'oldest boulder' approach (cf. Putkonen and Swanson, 2003) is also potentially an over simplification of a complex dataset. We use uncertainty weighted means (UWM) and kernel density plots to evaluate our dataset, since these approaches take into account the range and distribution of ages on a moraine once clear outliers are excluded.

. Lateral moraines
There are three sets of sloping, sharp-crested lateral moraines (field checked at 380 m, 510 m and 710 m asl) on the north-western flank of Cordon Esmeralda (Figs. 3e6). The 510 m lateral moraine is contiguous with the Esmeralda Moraines (Fig. 4). We did not identify a contiguous terminal moraine with the 710 m lateral moraine (located directly above the Esmeralda Moraine), suggesting that the higher-level glacier was confluent with Colonia ice in the main valley.
Lateral moraines in Baker Valley wrap around Cerro Chacabuco into "Juncal Valley" at 650 m asl, dropping to 400 m asl near the Pared Norte Moraines (Fig. 3). On the side of Juncal Massif, there are lateral moraines above Laguna Larga and Lago Chacabuco at 800 m asl. They are contiguous with the Juncal Moraines (555 m asl) further down the Juncal Valley, above Lago Juncal (Fig. 3).

Subaerial moraines
The Esmeralda moraines (Figs. 3 and 4) comprise two clear sets of moraines on the western and eastern sides of the valley respectively, separated by a deep channel. Those moraines on the east of the channel ("Esmeralda Moraines" on Fig. 4) comprise a clear cross-valley, sharp-crested and arcuate ridge crest at 360e366 m asl, which was sampled for cosmogenic nuclide dating (see Section 6). The ridge is ca. 3 m wide at the crest, 60 m high, with symmetrical proximal and distal slopes both 30 steep ( Fig. 6A and B; profiles E and C in Fig. 5). Inside the main ridge crest, several inset smaller latero-terminal moraines record subsequent glacier recession (Figs. 4 and 6C). The surficial sediments comprise a siltysandy diamicton with numerous erratic cobbles and boulders.
At the distal foot of the moraines are bedded gravelly sands, which grade to well-sorted laminated silty clays (Section 4.2.1) in the basin downstream in the valley. Palaeoshorelines etched onto the hillside in front of and behind the moraines, but below their crestline, indicate that a palaeolake was extant both during moraine formation and following glacier recession.
We interpret these moraines as subaerial push moraines (Bennett, 2001). Similar push moraines in association with glacial lakes have been noted in Iceland (Bennett et al., 2000). A subaerial interpretation precludes any lakes higher than the crest of the moraine (366 m asl) at the time of moraine emplacement, which is in agreement with the lack of deltas or shorelines higher than 350 m asl in this locality. We suggest that the narrower, confined nature of the valley and higher elevation of the bedrock focused sediment delivery, leading to the formation of a higher push moraine where subaerial processes dominated.
The Esmeralda terminal subaerial push moraine links continuously with the 510 m lateral moraine, indicating an ice source from MSL, rather than the NPI, as has previously been suggested . We call this "Calluqueo Glacier". South of the Esmeralda Moraines are the "Moraine Mounds", originally mapped by Glasser et al. (2012) (Fig. 4). They comprise prominent ridges running towards a bedrock high at the lake edge, with hummocky, morainic topography on the ice-proximal side. Granite boulders from their crest yielded exposure ages of 11.7 ± 0.6 and 12.6 ± 0.9 ka . They represent a stillstand during the recession of Calluqueo Glacier.

Laminated silts and clays
North of the Esmeralda moraines are glaciolacustrine sediments (Bendle et al., 2017b) ( Fig. 3) comprising well sorted, laminated silts and clays ( Fig. 7A) with dropstones. They form a well-vegetated area of upstanding relief downstream of the Lago Esmeralda moraines, dissected by palaeochannels and Río Cochrane. Close to the moraines, fans fine upwards from cobbles to silty sediments whilst near Cochrane, laminated silts and clays (some with varve appearance) crop out at elevations of 170e200 m. These sediments were deposited by sediment-laden glacial meltwater entering the lake environment, where they were affected by gravity. The underflows moved away from the ice margin towards topographic lows, with sediment being deposited as momentum was lost (cf. Ashley, 2002). Studies from contemporary glacial lakes in Patagonia have emphasised the importance of underflows in these environments, since water density is controlled by suspended sediment concentrations rather than by water temperature or salinity, as is the case in glaciomarine fjords (Sugiyama et al., 2016). Rhythmic parallel laminated sediments such as these are characteristic of lake environments with sediments deposited by underflows (Ashley, 2002;Carrivick and Tweed, 2013).

Deltas
A series of flat-topped delta terraces are present beyond the Esmeralda moraines, at 350e460 m asl at the foot of Río Estero Elva, which drains into Lago CP (Fig. 3), and at 150 m asl, superimposed on the Salto Moraines (Fig. 8A). Gilbert-type deltas form where streams enter lakes (Ashley, 2002;Longhitano, 2008;Bell, 2009;Slaymaker, 2011;Dietrich et al., 2017). They often form in stepped sequences upstream of modern and actively forming lake deltas (Bendle et al., 2017b). These ice-distal glaciolacustrine deltas were fed by a fluvial network issued from the retreating ice margin (cf. Dietrich et al., 2017).

Palaeoshorelines
Palaeoshorelines are eroded into the diamictic sediments on the hillsides both north of Cochrane at 460 m asl (Fig. 3), and consistently at: 340e350 m asl in Valle Grande; around the Juncal Massif (Fig. 7B); inside the Esmeralda moraines (Fig. 4); and around the southern and western margin of Cerro Ataud, where they are associated with the Salto moraines (Fig. 8B). Shorelines are also cut into the ice-distal face of the Salto Moraine at 280 m asl ( Fig. 4;  Fig. 8A). The shorelines take the form of a gently valley-dipping platform (e.g. Fig. 7B) with a flat long profile, and frequently bear exotic and local boulders. These erosional features, backed by cliffs, are similar to features observed in other parts of Patagonia Glasser et al., 2009;García et al., 2014).
These palaeoshorelines constrain the evolution of ice-dammed lakes in the region. The higher~460 m asl shoreline is located to the north of Cochrane whilst the~350 m asl shorelines extend into the Salto and Juncal valleys. In tandem with the glacial geomorphology (Section 4.1), this demonstrates that Esmeralda moraine formation was coeval with the 350 m asl palaeolake in this valley.

Morainal banks
Asymmetric moraines closely associated with the 350 m asl  palaeolake shorelines are found in the Salto and Juncal valleys. Adjacent to the sharp-crested Esmeralda Moraine there is a wedgeshaped moraine rising to 330 m asl dips back towards Lago Esmeralda at 2 over 2 km (Figs. 4 and 6A,D;profile F in Fig. 5). The distal slope dips at 30 towards Cochrane and is marked by numerous horse-shoe shaped crescentic scars ( Fig. 4; Fig. 6D). Exposures reveal a silty, stone-rich, massive diamicton and interbedded trough cross-bedded sands and gravels (Fig. 6E). The Salto Moraines, which reach heights of up to 350 m asl, are at the edge of a hanging valley, perched above Valle Grande (Fig. 8).
The northern Salto Moraines are sharp-crested at their northwestern flank, near Valle Grande, but as they near Cerro Ataud the ice-proximal face becomes more gradual, and slopes gently inland towards Lago Juncal over a distance of some 1.7 km (Fig. 8B; Profile G in Figs. 4 and 9). The ice distal slope is steep and well defined (Fig. 9, Fig. 8C and D). The gentle ice-proximal slope is interrupted with multiple small recessional moraines. Small exposures of surficial sediments in the ridges reveal a silty, stone-rich diamicton. Where the Northern Salto Moraines meet Cerro Ataud, they are cut by shorelines at 350 m ( Fig. 4; Fig. 8B). Outside and north of the moraines, the shoreline is~50 m wide, littered with locally derived angular gneiss boulders and rounded to subrounded granite clasts (Fig. 8D, E, F), indicative of local talus input.
In the "Juncal Valley", the "Lago Chacabuco Moraines" and "Laguna Larga Moraines" (Fig. 3) have a similar asymmetric profile to the Salto Moraines, with a shallow ice-proximal slope and a steep ice-distal slope. Between the moraine crests are several infilled kettle holes.
We interpret all these landforms as subaqueous morainal banks (cf. Glen, 1954;Rovey and Borucki, 1995;Turner et al., 2005;Batchelor and Dowdeswell, 2015;Bennett et al., 2000) deposited by a valley glacier terminating in water. Following glacier recession from Valle Grande, the glacier that deposited the Salto Moraines was left as a hanging valley with a morainal bank at its mouth. As ice receded from Valle Grande, Lago Chalenko extended into the valley. Lago Chalenko drained through the 350 m col at Rio Bayo, and the lake was therefore confluent in the Lago GCBA and Lago CP valleys . Both the Salto (Fig. 8) and Esmeralda  Lateral moraines indicate that the Lago Chacabuco and Laguna Larga moraines were formed by a glacier that flowed from the Barrancos Mountains, down Barrancos Valley and into Juncal Valley. We call this the "Barrancos Glacier". Shorelines and deltas indicate that the valley was flooded by the 350 m Lago Chalenko following glacier recession. The glacier was likely confluent with ice forming the Salto moraines from the Salto Valley, before receding to the Juncal Valley before the 350 m palaeolake drained, resulting in shorelines being imprinted over the moraines.

Ice-contact fans
On the eastern flank of "Juncal Massif" an accumulation of sediments dips gently towards the valley floor (Fig. 4). The slope of the terrace is steeply incised by numerous gullies and meltwater channels, the largest of which has an abandoned alluvial fan at its base (Fig. 4). The top of the terrace reaches~380 m asl. A rounded break in slope occurs at 365 to 330 m asl. The top surface is~30 m wide, and dips at~10 e17 towards the valley floor. After the break in slope, it steepens to 20 e25 . Exposures in road cuttings show that the sediments of this terrace comprise a stone-rich diamicton with a silty sandy matrix and with abundant fine to coarse gravel. The diamicton is overlain by stratified, poorly sorted, clastsupported sandy gravels with a silty sand matrix.
We interpret the gently sloping top surface of the Juncal Fan as a kame terrace. The more steeply dipping slope (20 ) is interpreted as a subaqueous ice-contact fan, imprinted on top of the kame terrace following glacier recession. Ice-contact fans are sloping sediment accumulations that form subaqueously at the grounding line of glaciers (Powell, 1990;Winsemann et al., 2009;Dowdeswell and V asquez, 2013;Lønne, 1995;Boston and Lukas, 2017). These landforms form at the grounding line when sediment-laden meltwater is released from the glacier as an efflux jet into a glacial lake (Lang and Winsemann, 2013). These high-energy flows build a fanshaped sediment accumulation. The break in slope on the Juncal Fan is contiguous with the 350 m shoreline in the Juncal Valley, providing a clear stratigraphic association with the 350 m Lago Chalenko.

Glacial landforms
A series of moraines on the northern hillside above Lago Bertrand (Fig. 10) were deposited by an advanced Soler Glacier. The Bertrand moraine chronology is also reported by Thorndycraft et al. (2018), where the timing of moraine deposition is used to define a maximum age for Soler Glacier ice blocking Río Baker drainage.
Here we provide the first detailed glacial landsystem analysis of the site, and interpret the glaciological significance of the ages.
Above Lago Bertrand there are four sets of moraines (LBM1 to 4 that we term the Lago Bertrand Moraine Complex (LBMC); Fig. 11). The moraines are located above the limits of the 11 ka Plomo and Lago Negro moraines (Fig. 10) (cf. Glasser et al., 2012), with LBM4, the most northerly and most extensive lobe situated at the col between the Lago Bertrand and Río Canal valleys ( Fig. 10; Fig. 11).
LBM1 moraines extend from 490 to 550 m asl (Fig. 11). The ridge crest runs parallel to the slope towards Lago GCBA. Numerous large rounded granite boulders are visible just south of the ridge, but none are visible north of the ridge, where the local gneiss bedrock is ice-scoured with streamlined roche moutonn ees. LBM1 traces round to LBM2. The more substantial LBM2 moraines reach up to 640 m asl, and are interspersed with small kettle holes, and are situated at the top of a valley, indicating a glacier that flowed up a reverse-bed slope ( Fig. 11; Fig. 12A). The moraine ridges themselves are up to 100 m above the narrow valley floor, with numerous large erratic granite boulders on the crestline (Figs. 11 and 12A, B, D). Icescoured gneiss prevails beyond the moraines (Fig. 11).
LBM3 comprises an inner and an outer crescent-shaped moraine (Fig. 12C, E), both of which wrap around an approximately circular lake (Fig. 12C). The arcuate ridges are~10e20 m high and are steepsided and sharp-crested. Although these moraines are located upice, they are lower, reaching 575 m asl (LBM3 inner moraine) and 555e566 m asl (LBM3 outer moraine). Kettle holes are present between and beyond the moraines.
LBM4 moraines are located further northwest at~550 m asl, with a lateral moraine extending up to at least 700 m asl into the Bertrand valley. They comprise chaotic, discontinuous mounds interspersed with meltwater channels. Large numbers of locally derived angular gneiss boulders are at the terminus of these moraines, suggesting a rockfall event onto the glacier surface (the moraines are not directly overlooked by steep cliffs that could produce these rocks). The arrangement of the moraine crests suggests that LBM4 predate LBM3.
These diamictic, arcuate moraines, interspersed with infilled kettle holes between moraine ridge crests, are typical of activetemperate glaciers (Evans, 2003;Evans et al., 2006). The moraines of the LBMC are formed on reverse-bed slopes and represent small protrusions extending laterally from a large trunk glacier that filled the Lago Bertrand valley (Soler Glacier). LBM1 represents the lateral moraine near the terminus of the glacier at Bertrand col. The moraines record a period of glacier thinning. LBM1, 2 and 4 represent the oldest and highest moraines, when ice was presumably flowing over the col into the Río Canal valley. Thinning ice resulted in recession from LBM2 and the formation of LBM3, with the close-set inner and outer moraines pf the latter likely representing just a small step back of the glacier.
As ice flow is driven by ice-surface slope (Cuffey and Paterson, 2010), we infer that the thickest recorded ice surface (Soler Glacier) in Lago Bertrand at this time was >640 m asl at the location of the LBMC. Given that Lago Bertrand is~90 m deep, this indicates an ice thickness >730 m near the eastern end of Lago Bertrand.

Ice contact fans
LBM1 moraines trace the ice margin down to the terminus position at the boundary col between Lago Bertrand and Lago GCBA, where there is a thick accumulation of diamicton (Fig. 11). It is incised by a shoreline at 345 m asl. The diamicton grades into a flattopped ice-contact fan at 300 m asl (Fig. 11), formed subaqueously at the terminus of Soler Glacier (Thorndycraft et al., 2018). This feature has previously been described as a moraine (Bourgois et al., 2016) with a single, outlying, cosmogenic nuclide exposure age of 17.5 ± 4.1 ka (at 299 m asl).

Delta terraces and shorelines
Raised deltas and palaeoshorelines around the edge of Lago GCBA include a flight of strandlines from 201 m (the current lake level) to 415 ± 5 m at the eastern end of Lago GCBA (Bourgois et al., 2016). Shorelines are incised into glacial diamicton at 440 m asl near LBM 3, and at 340 m asl through the LBM 1 moraines (Fig. 11).
These data indicate the presence of a palaeolake at 400e440 m asl (Bell, 2008(Bell, , 2009Glasser et al., 2016). The variance in height is due to the differential isostatic rebound at the near and far ends of the lake (Turner et al., 2005;Thorndycraft et al., 2018). This indicates that when Soler Glacier terminated at Bertrand col and formed the LBMC, it calved into the 400 m "Deseado" Lago GCBA. At this time, the glacier dammed drainage south through Río Baker and prevented the formation of a unified ice-dammed lake through the Baker valley.
Evidence of the unified 350 m Lago Chalenko includes a raised delta terrace at 340 m asl at the mouth of the Río Canal valley to the north, and a shoreline superimposed on LBM1 at 345 m asl (Fig. 11). Deltas at 350 m asl are also present above the present-day Lago Bertrand, indicating that Soler Glacier had receded as far as the Plomo Moraine (dated to 10.6 ka; Glasser et al., 2012) at the time that the 350 m Lago Chalenko formed in the Baker, Chacabuco, Lago GCBA and Lago CP valleys (cf. Thorndycraft et al., 2018).

Esmeralda and Salto Moraine chronology
We collected eight samples for 10 Be cosmogenic nuclide exposure-age dating (Table 1 and Supplementary Fig. 1) comprising five samples from the contiguous Esmeralda lateral (510 m asl)/ terminal moraine complex (365 m asl) and three from the Salto Moraines (338e359 m asl). Given the morphostratigraphic relationship of the Esmeralda and Salto moraines to the 350 m shoreline and the geomorphic evidence suggesting the moraines are subaerial/subaqueous, described above, all the boulders from the Esmeralda moraines were located above palaeolake Chalenko, whilst two of the boulders from the Salto Moraines were below the palaeolake, and one just above. The Esmeralda Moraine ages range from 13.3 ± 0.5 to 12.8 ± 0.6 ka. The ages of the two Salto Moraine boulders below the 350 m shoreline are 12.1 ± 0.4 to 12.0 ± 0.4 ka, and one boulder above the shoreline yielded an age of 12.5 ± 0.4 ka; therefore all the boulder exposure ages from the Salto Moraine are within errors of each other (see Table 2). Fig. 13 shows normal kernel density estimate plots, with each age plotted as an individual Gaussian curve using internal uncertainties, for both our ages (Tables 1 and 3) and published ages. The samples from the contiguous and subaerial Esmeralda 510 m lateral and terminal moraines are well clustered, with no outliers (Table 1, Fig. 13), so we calculate the external uncertainty-weighted mean (UWM) age from these moraines together as 13.0 ± 0.6 ka. The UWM of the Moraine Mounds (cf. Glasser et al., 2012) is 12.1 ± 0.6 ka.
The UWM of the two exposure ages from the Salto Moraines that are below the 350 m shoreline is 12.1 ± 0.5 ka (Table 1), likely dating the timing of lake-level fall. Seasonal water-level variations are unlikely as lake depth is controlled by the height of the col (Bayo col at 350 m asl; Glasser et al., 2016). Water shields boulders from cosmic rays in the same way as ice and any other matter (Schaller et al., 2002;Fabel et al., 2010;Balco, 2014;Small and Fabel, 2015); 26 m of water has the same shielding effect as 10 m of granite. Sampling boulders from this morainal bank, which were only exposed when the lake drained, provides a minimum age for lake drainage. In general, samples above the lake water at the Salto and Esmeralda limits together indicate an advance of Calluqueo Glacier from MSL during the later part of the ACR, with palaeolake drainage at ca. 12 ka, when the ice margin was near the Moraine Mounds.

Lago Bertrand moraine complex (LBMC) chronology
The LBMC (Fig. 11) ages constrain an advance of Soler Glacier into Bertrand Valley, where it calved into the 400 m Deseado level of Lago GCBA at Bertrand col. All of these moraines are above the level of any mapped palaeolake in this region (Turner et al., 2005;Fig. 10. Lago Bertrand and surrounding region. Field study site (Fig. 11) is indicated by red box. The Plomo, Negro and Leones moraines are highlighted. Published data includes moraines and sandar (modified from Glasser and Jansson, 2008), cosmogenic nuclide ages (diamonds) Bourgois et al., 2016) and OSL ages (blue circles) . Moraines have been modified and updated. Deltas are based on original mapping and published work (Bell, 2008;Glasser and Jansson, 2008;Bell, 2009;Bourgois et al., 2016;Glasser et al., 2016;Bendle et al., 2017b;Thorndycraft et al., 2018). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)  Bell, 2008;Glasser et al., 2016;Bourgois et al., 2016;Thorndycraft et al., 2018). These moraines must post-date the end of the FCMC-17 varve record (16.9 cal ka BP) to the west of Lago GCBA (Bendle et al., 2017a) and pre-date the lower-elevation Plomo moraine which formed around 11 ka   (Fig. 10). Six samples were collected from the LBMC (Fig. 11) for 10 Be exposure-age dating; these comprise two samples from the LBM2, two samples from LBM3 inner moraine, and two samples from the LBM3 outer moraine (Table 3 and Supplementary Information). The two ages from LBM2 are 14.6 ± 0.5 and 15.9 ± 0.6 ka. The four ages from LBM3 range from 13.7 ± 0.6 to 14.1 ± 0.5 ka. The ages from the inner and outer moraines from LBM 3 are therefore within errors of each other.
The ages from the LBMC are well clustered, with LBM2 being slightly older than LBM3 (Table 3; Fig. 13). LBM2 bears two boulders with a UWM exposure age of 15.1 ± 0.7 ka. Although the oldest age boulder could be an outlier, the older age for this moraine is in agreement with the geomorphology. LBM2 is slightly higher than LBM3 and is down-ice, so this ice margin would reflect a thicker and older ice margin in the Bertrand valley.
We cannot distinguish the LBM3 inner and outer moraines, which have exposure ages within errors of each other, even though the geomorphology indicates that LBM3 is the older ice margin. The combined UWM age of these slightly lower moraines is 14.0 ± 0.6 ka. Together, the ages from LBM2 and LBM3 suggest that the glacier occupied this position in Bertrand valley from 15.1 to 14.0 ka. Since the moraines are overprinted by the 400 m shoreline, the glacier advance occurred during the existence of the 400 m Deseado lake, and then retreated, which enabled the formation of a unified palaeolake through Río Baker.

Glaciolacustrine landsystem
We present a new landsystem model for the Salto Valley (Fig. 15), with valley glaciers terminating in glaciolacustrine environments forming sub-subaqueous, asymmetric morainal banks and conversely, sharp-crested and arcuate subaerial moraines (Fig. 15A). The glaciolacustrine landform assemblage includes raised deltas, palaeoshorelines at a consistent altitude, and subaqueous fans (Fig. 15B). The moraines formed before large drops in the altitude of the valley floor, indicating that calving and lake depth was one of the main controls on glacier terminus position. These glaciolacustrine landforms are associated with well-sorted laminated silts and clays, such as those deposited north of the Esmeralda moraines, and stratified diamictons and gravels in the morainal banks. The horse-shoe shaped scars on the steep ice-distal slope of the Esmeralda wedge-shaped morainal bank represent sediment gravity flows into the water. This landform assemblage, with sediments deposited subaqueously in front of the ice margin by underflows and turbidity currents, is similar to the glaciomarine landsystem modelled by Dowdeswell et al. (2015) for Chilean fjords in southern Patagonia. However, this more confined environment results in a greater diversity of landforms and sediments, and glaciolacustrine deposition is likely to be more strongly influenced by underflows into a sediment-stratified proglacial lake (cf. Ashley, 2002;Johnsen and Brennand, 2008;Sugiyama et al., 2016;Lang et al., 2017), whereas freshwater bouyant plumes are more dominant in glaciomarine settings (cf. Dowdeswell et al., 2015Dowdeswell et al., , 2016. We suggest that the large asymmetric morainal banks formed during periods of glacier stabilisation, and that ice-contact fans built up along the valley sides during recession. An association of morainal banks and ice-contact/proximal fans is common in temperate glacial environments with abundant meltwater (Dowdeswell et al., 2016). They require an ice margin to be stable for some years to decades . Ice contact fans typically form at the outlet of structural valleys and behind grounding zone wedges (Dietrich et al., 2017), which is the case for the Juncal Fan, situated as it is behind the Esmeralda and Salto morainal banks. The ice contact fan in Lago Bertrand is likewise situated at the outlet of a structural valley.
The Juncal Fan is steeply incised by a deep channel, with an alluvial fan at its base. We interpret this as a subaerial meltwater channel that formed following the drop in lake level. To form a meltwater channel, the glacier must have been very proximal, probably just following recession from the Moraine Mounds directly across the valley mapped by Glasser et al. (2012), possibly to a pinning point at the mouth of the valley. The Juncal Fan then formed between the ice and the hillside. This meltwater channel positions the glacier margin when Lago Chalenko drained.
We surmise that care must be taken in all areas east of the NPI not to sample boulders that may have been underwater, given that ice-marginal lakes reached high elevations, e.g. up to 500 m asl in the more distal parts of the Cochrane valley (Turner et al., 2005;Hein et al., 2010), or to acknowledge that exposure ages from boulders from subaquatic moraines will date the timing of lake drainage.

Regional chronostratigraphy
Beyond our study, there are few other moraine limits dated in the LBMC region. In Figs. 13 and 14, it is clear that the samples below the 350 m Lago Chalenko level from the Chacabuco Moraines Table 1 Sample data used to calculate cosmogenic nuclide exposure ages from the Esmeralda and Salto moraines. See text for analytical details. Uncertainty-weighted means (UWM) are rounded to nearest 100 years and presented as ka. yRecalculated ages from Glasser et al. (2012). Ages referred to in the text are highlighted.

Location
Sample ID  13. Normal kernel density estimates of new and published cosmogenic nuclide ages. Each (grey) Gaussian at the bottom of the plot represents a single age, with the width corresponding to 1 s error in the age uncertainty (internal uncertainties). Each Gaussian has the same area. High, narrow Gaussian distributions therefore have more precise age (also known as the Río Colonia Moraine Mounds; Glasser et al., 2012), and the Leones, Plomo and Negro moraines ( Fig. 10;  Fig. 13; Fig. 14) are substantially younger than the higher samples from the subaerial lateral moraines above, with ages centred on 11e12 ka. We suggest that these ages constrain the timing of lake drainage (cf. Thorndycraft et al., 2018). We note that the higher sample RC3 from Glasser et al. (2012), taken at 350 m asl, yields an older recalculated age of 14.6 ± 0.8 ka, which supports this suggestion that younger, lower ages relate to the draining of Lago Chalenko (see Chacabuco Moraines in Fig. 14). The Lago Negro Moraine is likely synchronous with the ice contact fan that links to the lateral moraines at Bertrand col. The young age of the Lago Negro Moraine therefore is likely to be associated with shielding by lake waters. The Plomo moraine records a later glacier stabilisation once drainage had opened up and the palaeolake had drained to its current level through Río Baker; this advance could be climatically driven, or related to a dynamic change in environment following lake drainage and a change in calving conditions. These younger moraine limits are also lower in altitude, and represent a progressive thinning and recession to key pinning points (Barr and Lovell, 2014).
Further indications of a readvance of the NPI during the ACR come from Lago Colonia (Nimick et al., 2016), who dated a series of moraines (samples ARCO1, 3, 4) from 12.4 to 14.2 ka. The lowerelevation CLARO moraine (samples 1, 2, 3, 5) had boulders with exposure ages ranging from 10.3 to 11.4 ka (these samples are all marginally above 350 m asl, and the context with the 350 m palaeolake Chalenko is poorly constrained). The UWM exposure age of the RT1 moraine from Tranquilo Glacier from MSL (Sagredo et al., 2018) is 13.5 ± 0.2 ka.
The ages yielded from the LBMC are older than those associated with the Colonia, Tranquilo and Calluqueo advances. The weighted means of 15.1 ± 0.7 ka to 14.0 ± 0.6 ka indicate that Soler Glacier reached these moraines during the early part of the ACR, while the Esmeralda Moraines date to the later part of the ACR. We emphasise that the presence of Soler Glacier at Bertrand col prevents the formation of the 350 m asl unified lake in the Lago CP and Baker valley, as this lake drained north through the Bayo col (near Puerto Tranquillo; see Fig. 1), so there is independent geomorphological evidence for these older ages.

Glacial evolution during the Antarctic Cold Reversal and Early Holocene
We use our new geomorphic and chronostratigraphic data alongside published moraines and ages to constrain dynamic ice behaviour and palaeolake-glacier interactions from 16 ka, through the ACR and into the Early Holocene.

Phase A: 16 ka e glacial recession
At 16 ka, the glaciers are well receded from their LGM positions at the far ends of Lago CP and Lago GCBA but are likely still coalesced in the eastern valleys draining the NPI. Between 17.0 and 16.0 ka, ice in the Chacabuco Valley had retreated to the Maria Elena moraines (Fig. 1) (Boex et al., 2013), with the ice-dammed Lago Chacabuco draining over Puesto Tejuela (550 m asl) into Lago CP   (Fig. 16A). Glaciolacustrine influence in Lago Edita in this area had ceased by 19.4 cal ka BP, suggesting that the highest lake in Valle Chacabuco may have drained even earlier (Henríquez et al., 2017); however, we suggest instead that Lago Edita may represent a small ice-dammed lateral lake rather than the regional palaeolake, which drained later.
At this time, ice was more extensive in Valle Grande, with NPI and MSL-sourced glaciers confluent and damming the western end of Lago CP, which therefore drained to the Atlantic via a~500 m outflow (Turner et al., 2005;Hein et al., 2010).
At 16 ka, Barrancos Glacier, Pared Norte Glacier and Colonia Glacier were confluent in Valle Grande, forming a large trunk glacier terminating in 550 m Palaeolake Chacabuco. In the Salto Valley, the 710 m Esmeralda lateral moraine may delimit the ice surface in this valley at this time. Palaeolake GCBA was at the 400 m Deseado level and drained eastwards towards the Atlantic through Río Deseado (cf. Bourgois et al., 2016).

Phase B: 15 ka e Soler Glacier stabilisation
Soler Glacier stabilised at Bertrand Col, therefore still blocking Lago GCBA drainage through Río Baker, resulting in separate lakes at different elevations. An arm of Soler Glacier was also present at the Negro Moraines, which formed underwater as an ice-contact fan (Thorndycraft et al., 2018), and an arcuate moraine denotes ice extent in Chinito valley (Fig. 16B). It is possible that ice stabilisation initially occurred due to a narrow bedrock pinning point (cf. Barr and Lovell, 2014) at Bertrand Col. The early ACR ages from the LBM3 moraines (14.0 ± 0.3 ka) suggest that climate cooling helped maintain Soler Glacier at this position for over a thousand years. Our new ages from the LBMC are essential for timing the damming of the Bertrand Col and the separation and unification of the Lago CP and Lago GCBA ice-dammed lakes.
There is no equivalent ice margin mapped from Barrancos Glacier or Calluqueo Glacier. The mapping of the Barrancos spillway at the head of the Juncal valley (Thorndycraft et al., 2018), with a lower 420 m col and a higher spillway at~460 m asl, required Barrancos Glacier to recede sufficiently in the Juncal valley to allow drainage over the Barrancos col and in to the lower Baker valley. The opening of this spillway is dated by the abandonment and cessation of glaciolacustrine influences at Lago Augusta at 15.7e14.8 cal ka BP (Villa-Martínez et al., 2012;Thorndycraft et al., 2018). This is supported by independent evidence suggesting that the lake level in Lago CP dropped below~500 m asl at 15.5 ka (Hein et al., 2010).
The northward-flowing Barrancos Glacier had an accumulation area in the Barrancos Mountains immediately south of Monte San Lorenzo. This glacier was therefore independent of the NPI. The opening of the Barrancos spillway and recession of Barrancos Glacier indicates that NPI and ice from the eastern massifs (MSL and Barrancos Mountains; Fig. 1) were separating. The early recession of Barrancos Glacier, before Calluqueo Glacier, may have been caused by its smaller accumulation area and lower altitude, rendering it sensitive and vulnerable to both climatic changes and calvinginduced recession.
It is likely that Calluqueo Glacier was advanced north of the Esmeralda Moraine, calving into the 460 m palaeolake in the Lago CP valley. A lack of shorelines above 350 m asl on Cerro Ataud and geomorphic evidence from the Esmeralda Moraines suggests that the glacier was beyond the Esmeralda limits.
There are no dated ice limits of Nef Glacier at this time, however, the~460 m asl shorelines can be traced in the Baker valley to Soler Glacier so we hypothesise that by the time the 420e460 m palaeolake formed, Nef Glacier had receded towards its valley confluence with Río Baker (Fig. 16B). Calluqueo Glacier stabilised at Cerro Ataud forming the Esmeralda and Salto moraines at 13.0 ± 0.6 ka. The lake level had fallen to the~350 m asl Lago Chalenko elevation. Soler Glacier had retreated up the Bertrand Valley allowing Lago GCBA and Lago CP to unify through the Baker valley, with drainage to the Pacific through the by the now opened Bayo col following recession of Exploradores Glacier and other northerly glaciers of the NPI from the Río Norte/ Bayo valley (see Fig. 1) . We argue that our multiple cosmogenic ages and the morphostratigraphic relationships we have identified between the moraines and Lago Chalenko are more reliable than the single OSL date used to infer a later opening of the Bayo valley . The presence of Lago Chalenko in Valle Grande and the Cochrane basin indicates that NPI ice was still blocking Río Baker, likely in the sector from the Colonia to LosÑadis confluences.
Tranquilo Glacier on the northern side of Monte San Lorenzo also underwent a readvance at this time (UWM age of 13.5 ± 0.2 ka and 13.0 ± 0.4 ka respectively) (Sagredo et al., 2018) (Fig. 1; Fig. 13;  Fig. 14). Table 3 Details for Lago Bertrand cosmogenic nuclide samples. See text for analytical details. Uncertainty weighted means (UWM) are rounded to nearest 100 years and presented as ka. Ages referred to in the text are highlighted (0 cm/yr erosion rate; internal uncertainties).
Fig. 14. Age-altitude scatter plot of our new ages and recalculated published ages Nimick et al., 2016;Bourgois et al., 2016;Sagredo et al., 2016Sagredo et al., , 2018. Blue bar indicates duration of ACR, black dashed vertical line indicates elevation of Lago Chalenko. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) Fig. 15. A: 3D plan view and along-profile 2D views illustrating glaciolacustrine sedimentary processes. Following glacier advance and stabilisation, a large subaqueous morainal bank is built. Whilst the lake is dammed at the higher level, glaciolacustrine terraces and shorelines develop beyond the glacier limits. Modified from Powell (1990), Longhitano (2008), Dowdeswell and V asquez (2013), Dowdeswell et al. (2015Dowdeswell et al. ( , 2016 and Carrivick and Tweed (2013). B: 3D plan view and 2D along-profile view illustrating the formation of an ice-proximal fan during glacier recession. These features typically form under moderate meltwater and a high sediment discharge scenario. Modified from Powell (1990), Lønne (1995) and Batchelor and Dowdeswell (2015). Glacier recession resulted in the formation of the recessional Moraine Mounds (365 m asl)  at 12.1 ± 0.6 ka. Lago Chalenko was still extant at this time, imprinting shorelines at 350 m asl around Cerro Ataud, Lago Esmeralda, and in the Juncal Valley (Figs. 3 and 4). On the opposite side of the Salto Valley, the Juncal Fan was forming. The Juncal Fan is dissected by a subaerial meltwater channel, suggesting the lake drained soon after the fan formed, while the ice remained in a proximal position, most likely at or near the Moraine Mounds. This fits, within the dating uncertainties, the exposure age from the subaqueous boulders from the Salto Moraines (12.1 ± 0.5 ka), and a Baysian modelled age of 12.4e11.8 ka for the timing of lake drainage (Thorndycraft et al., 2018). Pulsed drainage is suggested by the lower stepped shorelines imprinted upon the Salto Moraines (Fig. 4). The drainage of Lago Chalenko indicates that the Pared Norte Glacier had receded southwards, opening up the Río Baker drainage route to the Pacific Ocean.

Phase E: 10e11 ka e Early Holocene ice limits
A stabilisation of some outlet glaciers during the Younger Dryas may have occurred, with clusters of ages around 10e11 ka (e.g. see Fig. 16). Although some of these ages represent boulders sampled from under water, moraines with boulders above the surface of Lago Chalenko and well inside the ACR limits are dated to this time for Colonia Glacier (Nimick et al., 2016) and Tranquilo Glacier (Sagredo et al., 2018), as well as at the Moraine Mounds in the Salto Valley . A limit congruent with the Younger Dryas but well inside the ACR limit has also been observed at the Herminita moraines in Lago Argentino Kaplan et al., 2011) and Tierra del Fuego (Menounos et al., 2013). These data suggest that although the ACR advance was much more significant, there may have been a smaller Younger Dryas readvance in Patagonia as well.
By 10 ka, the lakes and drainage routes around the NPI had reached their present-day configuration (Fig. 16E). The glaciers had receded up their valleys towards modern-day positions. Ages of 10.2 ± 0.5 to 11.1 ± 0.6 ka at elevations of 1120 m asl from a lateral moraine of Nef Glacier  suggest that Nef Glacier had retreated substantially up-valley. Lateral moraines only 14 km from the ice margin around Lago Colonia (349e375 m asl) yield exposure ages of 10.1 ± 1.0 to 11.2 ± 1.1 ka (Nimick et al., 2016). Soler Glacier was at the Plomo Moraine (ages of 10.4 ± 0.5 and 10.6 ± 0.6 ka; Glasser et al., 2012).

Glacier dynamics in Patagonia during the ACR
Antarctic ice cores indicate that the ACR was a period of strong cooling (Fudge et al., 2013;Wais Divide Project Members, 2015). Numerical modelling studies have suggested that the Patagonian Ice Sheet stabilised during the ACR (14.7e13.0 ka) (Hubbard et al., 2005), which is in line with palaeoclimatic studies that suggest that the influence of the ACR extended to at least 40 S, with cooling strongest in the South Atlantic (Pedro et al., 2016). Marine sediment cores also support cooling in the South Pacific near Chile at 41 S during the ACR (Kaiser et al., 2007). This cooling resulted in advances of glaciers from the tropical Andes (Jomelli et al., 2014(Jomelli et al., , 2017, NPI (this study; Turner et al., 2005), MSL (this study; Sagredo et al., 2018), and Torres del Paine (50 S) (Fogwill and Kubik, 2005;Moreno et al., 2009;García et al., 2012). Palynological data from Lado Edita, near Lago Augusta, suggest that this period of time was also associated with an increase in precipitation (Henríquez et al., 2017), which, in combination with lower temperatures associated with the ACR, would have encouraged glacier advance. However, although there is emerging evidence of wide-spread glacier advances during the ACR in Patagonia, the exact timing of the moraine formation is spatially variable, even in the same region. Our data show that topographic constraints, pinning points and calving dynamics in ice-dammed lakes exerted a significant control on the glaciers.
The exposure ages from the LBMC latero-terminal moraines and associated ice-contact fan, which lie beyond the limits of the Plomo and Negro moraines, suggest that Soler Glacier stabilised at the Lago Bertrand col during the early part of the ACR (Jouzel et al., 2001;Morgan, 2009;Putnam et al., 2010;Pedro et al., 2016). Soler Glacier occupied the LBMC from 15.1 ± 0.7 to 14.0 ± 0.6 ka, when it was calving into the 400 m Deseado level palaeolake. Topographical constraints encouraged ice-margin stabilisation and moraine formation (Barr and Lovell, 2014) against a background of a cooling climate.
The Esmeralda Moraine was formed subaerially at 13.0 ± 0.6 ka, at which time there was a unified 350 m Lago Chalenko in both the Lago CP and Lago GCBA valleys. Our geomorphological mapping independently demonstrates that the moraines cannot have formed synchronously, because Soler Glacier at the Bertrand Moraines blocks the formation of the unified 350 m Lago Chalenko. Lago Chalenko likely had a significant influence on the ice-margin stabilisation, with ice margins forming at topographic highs at the Esmeralda and Salto margins. Deep water beyond these moraines encouraged calving and limited the size and extent of the glacier.
Glaciers on the northern flank of MSL underwent smaller advances at 13.5 ± 0.2 ka (Sagredo et al., 2018), highlighting the important role of glacier catchment and aspect on driving glacier advances. Sagredo et al. (2018) reconstruct a glacier occupying the RT1 moraines in Río Tranquilo Valley with a length of~14 km and an ELA 260 ± 20 m below modern values (1815 ± 10 m) (Falaschi et al., 2013;Sagredo et al., 2018). We do not calculate ELAs at this time for Calluqueo Glacier, as it and Soler Glacier are calving into deep water, which affects mass balance (cf. Rea, 2009). However, Calluqueo Glacier was clearly much more extensive than Tranquilo Glacier during the ACR. This differential could reflect changes in the ice divide location, and the additional accumulation area afforded to Calluqueo Glacier by the Barrancos Mountains (Fig. 16C). Today, glaciers in Río Tranquilo valley are less extensive than Calluqueo Glacier (50.9 km 2 ; 2013; Falaschi et al., 2016), which has a much larger accumulation area (Fig. 2). Glaciers on the western flank of Monte San Lorenzo also have presentday ELAs that are 1700e1750 m asl, whilst the eastern, drier sectors have higher ELAs >1800 m asl (Falaschi et al., 2013). Depressed ELAs on the western portion of MSL could contribute to a larger readvance in the Salto Valley. However, the interpretation by Sagredo et al. (2018) is consistent with our argument that a large readvance occurred during the ACR in MSL. We also note that Barrancos Glacier receded before Soler Glacier and Calluqueo Glacier, likely due to its smaller catchment in the Barrancos Mountains (cf. Fig. 2) combined with a lake-terminating margin. By the time of the Younger Dryas, the glaciers had all substantially retreated from their ACR positions, but there was a stabilisation of the glaciers leading to moraine formation at this time; exposure ages from Colonia Glacier moraines show that glaciers were not far beyond their Little Ice Age Limits by~12.1 ka (Nimick et al., 2016). These data highlight the importance of not assuming that similar-sized moraines in adjacent valleys necessarily relate to synchronous advances. A detailed landsystems approach is recommended as a methodology to understand the context of moraine formation, and to infer past glacial dynamics from moraine age.

Conclusions
In this study, we used geomorphological mapping and cosmogenic nuclide dating from outlet glaciers of the eastern Northern Patagonian Icefield and the Monte San Lorenzo ice cap in central Patagonia, South America, to provide new, detailed, process-based interpretations for landform formation in a glaciolacustrine landsystem. We highlight for the first time the detailed mechanisms by which subaqueous morainal banks formed during the deglaciation of central Patagonia, and we constrain the timing of ice-lake interactions. We argue that recognition of subaqueous landforms and environments is critical to developing a strategy for dating past glacial fluctuations.
Our cosmogenic exposure-age data demonstrate that the largest readvance since the LGM in central Patagonia occurred during the ACR, with further glacier recession and lake drainage (12.0 ± 0.6 ka) occurring rapidly during warming after the ACR. Our data show that whilst this readvance was broadly climatically driven, the timing of individual outlet glacier advances was asynchronous and spatially variable, subject to external controls such as proglacial lakes and calving, aspect and accumulation area size, and bedrock pinning points.
We mapped and dated moraines from above known palaeolake levels of Lago GCBA, identifying an advance of Soler Glacier to the col between Lago Bertrand and Lago GCBA, where it formed a subaqueous ice-contact fan as it calved into a 400 m high palaeolake. Cosmogenic nuclide ages from these moraines yielded uncertainty-weighted mean ages of 15.1 ± 0.7 ka for the oldest, highest moraines, and 14.0 ± 0.6 ka for the inner moraines, indicating that the glacier was likely in this position for some 1000 years. The highest and oldest moraine ages pre-date the ACR indicating that the local topography with a narrow valley exiting into the western embayment of Lago GCBA encouraged topographic pinning here (cf. Barr and Lovell, 2014).
By contrast, we date the most distal ice margin of Calluqueo Glacier, terminating at the Esmeralda push moraine, to towards the end of the ACR at 13.0 ± 0.6 ka. The ice margin here was contemporaneous with Lago Chalenko (350 m asl). Subaqueous morainal banks delimit the ice margin, with a glaciolacustrine landsystem that includes ice-scoured bedrock, glacial diamicton plastered onto valley sides, perched delta terraces, kame terraces, ice-contact fans, palaeoshorelines etched into glacigenic sediments and subaerial push and lateral moraines.
Our data demonstrates that cosmogenic nuclide exposureage dating can be applied to constrain lake level falls where ice-lake interactions are well understood. Sampled boulders from Salto moraine, from below the Lago Chalenko lake level, were dated to 12.0 ± 0.6 ka, closely matching the age of recessional moraines at the mouth of the Salto valley  and providing independent support for a Bayesian modelled age of 12.4e11.8 ka for lake drainage (cf. Thorndycraft et al., 2018). Our study adds to the growing body of evidence that there was a strong ACR signal controlling glacier fluctuations in central Patagonia, but with the caveat that this was modulated and contolled by spatially variable topographic and glaciolacustrine controls on glacier extent.

Acknowledgements
Davies and Thorndycraft undertook geomorphological fieldwork and collected cosmogenic nuclide samples in November/ December 2015. Geomorphological mapping was also conducted by Davies, Thorndycraft andMartin in November/December 2016 and. Fabel undertook analysis of cosmogenic nuclide samples. Fieldwork was funded by the Royal Holloway University of London Research Strategy Fund (RHUL-RSF) and NERC DTP PhD award NE/L002485/1 awarded to Martin. We thank the RHUL-RSF, Quaternary Research Association, British Society for Geomorphology, and the Geologists' Association for each part-funding analytical costs of cosmogenic nuclide analysis. We thank Maria Miguens-Rodriguez (SUERC Cosmogenic Isotope Laboratory) for the preparation of the cosmogenic nuclide samples, and staff at the SUERC AMS Laboratory, East Kilbride, are thanked for 10 Be measurements. We thank all landowners for allowing access to their property during fieldwork. ASTER GDEM is a product of NASA and METI. Landsat-8 images used in mapping are provided courtesy of the U.S. Geological Survey. BD and VT were supported in the field at the Bertrand moraines by Esteban, the local shepherd. All supporting data for this publication are included either in the paper or in the Supplementary Information. We thank two anonymous reviewers whose comments helped improve and clarify the manuscript.