Glacial lake drainage in Patagonia (13-8 kyr) and response of the adjacent Pacific Ocean

Large freshwater lakes formed in North America and Europe during deglaciation following the Last Glacial Maximum. Rapid drainage of these lakes into the Oceans resulted in abrupt perturbations in climate, including the Younger Dryas and 8.2 kyr cooling events. In the mid-latitudes of the Southern Hemisphere major glacial lakes also formed and drained during deglaciation but little is known about the magnitude, organization and timing of these drainage events and their effect on regional climate. We use 16 new single-grain optically stimulated luminescence (OSL) dates to define three stages of rapid glacial lake drainage in the Lago General Carrera/Lago Buenos Aires and Lago Cohrane/Pueyrredón basins of Patagonia and provide the first assessment of the effects of lake drainage on the Pacific Ocean. Lake drainage occurred between 13 and 8 kyr ago and was initially gradual eastward into the Atlantic, then subsequently reorganized westward into the Pacific as new drainage routes opened up during Patagonian Ice Sheet deglaciation. Coupled ocean-atmosphere model experiments using HadCM3 with an imposed freshwater surface “hosing” to simulate glacial lake drainage suggest that a negative salinity anomaly was advected south around Cape Horn, resulting in brief but significant impacts on coastal ocean vertical mixing and regional climate.

heavy minerals and feldspars were removed on the basis of density using solutions of sodium polytungstate at 2.62 and 2.70 g.cm -3 . Further purification was achieved by immersing the grains in 48% hydrofluoric acid for 40 minutes, etching the outer 10µm from the surface of the quartz grains, and removing any remaining feldspars. All luminescence measurements were made on single grains in order to maximize the opportunity of identifying the grains within a sample that had had their OSL signal reset at the time of deposition (Duller 2008). A Risø TL/OSL-DA-15 equipped with a focused laser system for single grain OSL measurements (Bøtter-Jensen et al., 2003) was used for all samples. The environmental dose rate to samples was calculated using a combination of thick source alpha counting and beta counting. The contribution from cosmic rays was evaluated using the equations provided by Prescott and Hutton (1994).
We collected samples from a range of landforms around the former lake margin ( Fig  1C of the main paper). Seven samples for luminescence dating were obtained from the 300-320 masl lake level, five samples from the 370-440 masl lake level, two samples from the 460-520 masl lake level and one sample from a modern source bordering dune adjacent to a modern delta.
The single aliquot regenerative dose (SAR) protocol was used for equivalent dose (D e ) measurement. Samples were given a preheat of 220°C for 10 seconds after each regeneration dose and a cut heat of 160°C after each test dose (typically ~15 Gy). Acceptance criteria included a recycling ratio within 10% of unity, OSL signal at least 3 times greater than the uncertainty in the background, an uncertainty of less than 20% on measurement of the test dose, and an OSL-IR depletion ratio within 10% of unity (Duller 2003).
Dose recovery experiments were undertaken on three samples to assess the ability of the SAR protocol to accurately measure a known laboratory radiation dose. Grains of quartz were bleached at room temperature for 100 seconds using blue LEDs (470 nm, 35mW.cm -2 ), left for 10ks to allow charge in the 110°C TL peak to escape and then the bleaching procedure repeated a second time. A dose similar to the expected D e was then given to the grains. All three samples were able to recover the given dose within the 10% limits conventionally adopted for such a test, confirming that the SAR protocol can measure a known radiation dose (Tables S1). The overdispersion (σ d ) of the data sets varies from 14% to 26% respectively. This is the scatter in the data that cannot be accounted for by measurement uncertainty and counting statistics. The cause of this scatter is unclear. In the analysis above, a value of 20% has been used for the maximum uncertainty allowed in the determination of the test dose. This is double the value that is normally used and was designed to allow incorporation of data from a larger number of grains in order to help constrain the age models. To test whether this relaxed criteria was inadvertently leading to high overdispersion the dose recovery data were reanalysed using a test dose threshold of 10%. For all samples there was no detectable change in either the value of overdispersion, or the ability to recover the known dose. However in all cases the number of grains accepted fell by 70%. The cause of the high overdispersion in D e values for quartz from this region is not known.
The majority of sediments analysed in this study are from deltas associated with the former lake, and the luminescence signal in the grains is likely to have been incompletely bleached at deposition. Statistical analysis of the dose distributions to isolate the population of grains which was bleached at deposition followed the method described by Duller (2006). This approach used the finite mixture model (FMM) and followed the work of Rodnight et al. (2006). The validity of this approach was demonstrated by Duller (2006) by comparing the OSL ages for a number of glacio-fluvial sediments with independent age control. A small number of samples in this study were collected from beach or from aeolian facies and for these samples the central age model (CAM) was used. Figure S4 shows the D e distributions measured for the samples. A summary of the luminescence data for all the OSL samples is given in Table S2, including the dosimetry data, the number of quartz grains analysed, the D e used for age calculation, and the calculated ages.

Climate modelling experiments
Potential effects on regional ocean circulation were investigated using the UK Hadley Centre global climate model, HadCM3. HadCM3 consists of coupled atmospheric model, ocean model and sea ice model components, (Pope et al., 2000;Gordon et al., 2000). The resolution of the atmospheric model is 2.5° in latitude by 3.75° in longitude by 19 unequally spaced vertical levels. The spatial resolution over the ocean in HadCM3 is 1.25° by 1.25° by 20 unequally spaced layers in the ocean extending to a depth of 5200 m. The model contains a range of parameterizations, including a detailed radiation scheme that can represent the effects of minor trace gases (Edwards and Slingo, 1996). The ocean model uses the Gent-McWilliams mixing scheme (Gent and McWilliams, 1990). The sea ice model is a simple thermodynamic scheme and contains parameterizations of ice drift based on surface ocean currents (Cattle and Crossley, 1995). The model version used here incorporates the MOSES1 land surface scheme (Cox et al., 1999) and fixed vegetation. HadCM3 was forced with prescribed changes in orbit (altering the seasonal and latitudinal distribution of solar insolation), greenhouse gases, sea level, and ice-sheet evolution appropriate to 10 kyr BP. For further details on these boundary conditions see Singarayer and Valdes (2010). We performed three simulations: (1) a control simulation (CNTL) had 10 kyr BP boundary conditions and no freshwater hosing, (2) 500,000 km 3 freshwater added to the west coast of Patagonia ( Fig S5) over a 1-year "hosing" simulation (YEAR1S), (3) the same hosing rate as (b) but over a 5-year period (YEAR5S); this was to enable further statistical evaluation of the climatic impact.

Further details of the sedimentology, geomorphology and OSL sample context
Perched deltas here are typically flat-topped, and a well-developed shoreline indicates that the lake level was stable for a considerable length of time at this elevation before falling abruptly (Fig. S1). Numerous, closely spaced shorelines and associated sloping deltas indicate that this lake level lowered more or less continually throughout its existence and was not stable at this level for any length of time. In all cases, sediments are typically coarse gravels and sands commonly containing 1-2 m thick medium to fine sand lenses (Fig. S2). Rounded and wellrounded clasts dominate the sediments. Sections are typically capped by massive fine sand, interpreted as a wind-blown deposit.
A summary of the glacial lake stages is presented in Fig. 2 of the main paper and Table S3.

of main paper)
This glacial lake stage is constrained by sample LTH01 (11.8 ± 1.2 ka), collected at 452 masl from a section in palaeo-beach sands and gravels at Los Tres Hermanos. Sample MCS02 was collected at 444 masl from aeolian sand capping a c 5m-wide paleo-beach at Las Mercedes. It would be expected that this sample would be well bleached. The central age model (CAM) for the sample gives an age of 6.5 ± 0.7 ka, and indicates that the lake had drained by this time and subaerial aeolian deposition was taking place. However, the distribution of D e values for this sample is broader than would be expected for an aeolian sediment. No exposure was present in this palaeo-beach and the sample was collected from a pit dug to a depth of 0.5m. Because the sampling depth is very shallow, it is possible that the broad distribution of D e values results from the introduction of younger grains by bioturbation. A distinct population of older grains can be observed in the sample and these give an age of 12.0 ± 1.3 ka, which is far closer to the age obtained from LTH01. This glacial lake stage is also constrained by BC02 (12.9 ± 2.1 kyr), collected at 396 masl from a section in the shoreface platform of a delta fragment in the col at Bertrand; RII01 (10.3 ± 0.9 kyr), collected from a shallow pit excavated on the upper surface of a flattopped feature at Río Ibáñez; and PII02 (11.1 ± 4.2 kyr), collected at 396 masl from a section in the delta front foresets in one of a staircase of steeply dipping deltas near Puerto Ibáñez. Sample WCC01 (16.8 ± 1.5 kyr) was collected at 403 masl from a section in the shoreface platform of a delta fragment west of Chile Chico but is poorly bleached so is not included in this calculation of the mean age. A further sample (BC01) was collected at 396 masl from a section in the shoreface platform of a delta fragment in the col at Bertrand but yielded insufficient data to calculate a reliable age.
Glacial lake stage C (Fig. 2 of main paper) The elevation of lake levels above the Lago Tranquilo col at 350 masl was controlled by the moraines at the eastern end of LGC/LBA and LC/LP, and the level of this lake fell slowly as the lake outlet channel progressively incised to form an outflow through the moraines. The lake initially drained to a level controlled by a rock-cut channel cut into a col at 350 masl at Lago Tranquilo in the Rio Bayo Valley and not via the Rio Baker as previously suggested (Turner et al., 2005). This lake level was stable because the col at Lago Tranquilo remained ice-free during deglaciation.
The final drainage route and the lowering of the lake level below 260 masl occurred through the Rio Baker valley. A younger age from aeolian sand capping a large flattopped delta at Puente Santa Marta at 333 masl (Sample PSM01; 8.0 ± 0.5 kyr) indicates that the lake had drained by this time and subaerial aeolian deposition was taking place.
Glacial lake stage D and E (Fig 2 of main paper) As the ice receded into the Nef valley, a third drainage route for the 330-260 masl paleo-lake developed southeastward through the Rio Bravo and Pascua systems. Dates that constraine this glacial lake stage are: Samples PL01 (10.2 ± 0.7 kyr) and RII02 (9.5 ± 0.8 kyr) collected at 301 masl and 315 masl from sections in the shoreface platform of large flat-topped deltas on Peninsula Levican and the Rio Ibanez respectively. Sample MCS01 (8.5 ± 0.9 kyr), collected at 330 masl from a section in palaeo-beach sands and gravels at Las Mercedes, provides a similar, but slightly younger, age. Sample FD02 (11.0 ± 0.7 kyr) was collected at 303 masl from a section in the shoreface platform of the delta at Fachinal. This sample is important because here there is independent age control from cosmogenic isotope dates of boulders on moraines built on top of the delta (Douglass et al., 2005). The cosmogenic isotope dates indicate that local mountain glaciers advanced onto the delta surface to build moraines at ca. 8.5 and 6.2 kyr so the delta itself must be older than these ages, which is indeed the case. Sample BMG02 (10.6 ± 0.9 kyr) was collected at 315 masl from a section in the delta front foresets in a large flat-topped delta at Bahia Murta. Sample JV02 (9.8 ± 0.7 kyr) was collected at 303 masl from a section in the shoreface platform of a delta fragment in the Jenemeni Valley. Fig S1. Photographs of key landforms relating to the ice-dammed lakes. A. Multiple lake delta surfaces at Puerto Ibáñez. B. Valley-mouth deltas at Rio Claro. C. Clearly defined depositional shorelines (arrowed) at un-named site near Lago Negro. D. Lake shorelines forming erosional notches (arrowed) in bedrock above Lago Negro.   Tables   Table S1. Dose recovery data using the OSL signal from single grains of quartz. Table S2. Analytical data associated with OSL dating from samples of paleobeaches and deltas above Lago General Carrera/Lago Buenos Aires. Table S3. Summary of the evidence for former lake levels of Lago General Carrera/Lago Buenos Aires. Former lake levels are based on dGPS elevations collected in the field for geomorphological features including perched deltas, paleo shorelines, terraces, beaches and drainage cols. Lake area, lake volume and drainage volume were calculated by overlaying former lake elevations onto SRTM data in Arc-Map GIS. Stages A to E correspond to stages A to E in Figure 2 of the main paper.

Supplementary Figures
Ivins, E. and T. James, 1999. Simple models for late Holocene and present-day Patagonian glacier fluctuations and predictions of a geodetically detectable isostatic response.