Extensive MIS 3 glaciation in southernmost Patagonia revealed by cosmogenic nuclide dating of outwash sediments

1 The timing and extent of former glacial advances can demonstrate leads and lags during 2 periods of climatic change and their forcing, but this requires robust glacial chronologies. In 3 parts of southernmost Patagonia, dating pre-global Last Glacial Maximum (gLGM) ice limits 4 has proven difficult due to post-deposition processes affecting the build-up of cosmogenic 5 nuclides in moraine boulders. Here we provide ages for the Río Cullen and San Sebastián 6 glacial limits of the former Bahía Inútil-San Sebastián (BI-SSb) ice lobe on Tierra del Fuego 7 (53-54°S), previously hypothesised to represent advances during Marine Isotope Stages 8 (MIS) 12 and 10, respectively. Our approach uses cosmogenic 10 Be and 26 Al exposure 9 dating, but targets glacial outwash associated with these limits and uses depth-profiles and 10 surface cobble samples, thereby accounting for surface deflation and inheritance. The data 11 reveal that the limits formed more recently than previously thought, giving ages of 45.6 ka 12 ( +139.9 / -14.3 ) for the Río Cullen, and 30.1 ka ( +45.6 / -23.1 ) for the San Sebastián limits. These dates 13 indicate extensive glaciation in southern Patagonia during MIS 3, prior to the well- 14 constrained, but much less extensive MIS 2 (gLGM) limit. This suggests the pattern of ice 15 advances in the region was different to northern Patagonia, with the terrestrial limits relating 16 to the last glacial cycle, rather than progressively less extensive glaciations over hundreds of 17 thousands of years. However, the dates are consistent with MIS 3 glaciation elsewhere in 18 the southern mid-latitudes, and the combination of cooler summers and warmer winters with 19 increased precipitation, may have caused extensive glaciation prior to the gLGM. cosmogenic nuclide depth-profiles through outwash associated with moraine limits can yield 47 robust ages for glacial limits where post-depositional erosion and exhumation may 48 compromise traditional moraine-boulder samples. We present 10 Be and 26 Al dates from two 49 depth profiles through outwash associated with glacial limits of the Bahía Inútil-San 50 Sebastián (BI-SSb) ice lobe on Tierra del Fuego (53-54°S) and use these results to test the 51 established age model for the timing of glacial advance.


21
The terrestrial record of former southern hemisphere ice masses has been used to assess 22 inter-hemispheric synchroneity of glacial advance and retreat (Sugden et al., 2005)  estimates of surface exposure time. We also collected ~ 1 kg samples of mixed lithology 149 pebbles (>0.5 cm and <4 cm) at 25 cm depth intervals (depth error ≤4 cm), including a 150 sample at the base of the section to help calculate inheritance in the profile. Each depth 151 sample was amalgamated and analysed for 10 Be and 26 Al concentrations. One sample 152 8 (FP025cs) consisted half of sand matrix due to insufficient clasts at that depth. In both 153 profiles the lowermost sample consisted of two separate depth samples combined (i.e. an 154 unprocessed weight of ~ 2 kg) due to insufficient quartz; hence the apparent thickness 155 represented by these samples is greater. Detailed sample information is given in Table 1. 156 The nuclide concentration data from the depth profile samples were modelled to yield most 157 probable age, erosion rate and inheritance estimates for the outwash unit. The surface 158 cobble samples were treated independently as exposure age estimates for the outwash 159 surface. 160

Chemical analysis 161
All physical and chemical preparation and 10 Be/ 9 Be and 26  were precipitated and converted to BeO and Al 2 O 3 , before being prepared for AMS analysis. 177 measurements, with nominal ratios of 2.97 × 10 -11 10 Be/ 9 Be and 4.11 × 10 -11 26 Al/ 27 Al,179 respectively. The reported uncertainties of the nuclide concentrations include 2.5% for the 180 AMS and chemical preparation. Blank corrections ranged between 3 and 15% of the sample 181 10 Be/ 9 Be ratios and between 0 and 0.9% of the sample 26 Al/ 27 Al ratios. The uncertainty of the 182 blank measurements is included in the stated uncertainties. All nuclide concentration data 183 are given in Table 1. 184

Scaling scheme and production rate 186
For consistency, the time-dependent scaling scheme of Lal (1991) and Stone (2000) was 187 this production rate and scaling scheme combination using our surface sample ages 193 calculated using the New Zealand production rate and the Lal (1991) and Stone (2000) time-194 dependent scaling scheme. The global production rate gave ages <17% younger than our 195 ages (irrespective of scaling scheme) but the Patagonian production rate gave ages <6% 196 older or younger than our ages (irrespective of scaling scheme) or <5% older or younger 197 when the same scaling schemes were compared. Using the New Zealand production rate, 198 altering the scaling scheme resulted in <3% older or younger ages. Our choice of production 199 rate and scaling scheme does not alter our conclusions.  . We assumed a 205 density of 2.7 g cm -3 (equivalent to the density of pure quartz) and used a standard, excess 206 thickness of 6 cm for all samples to correct for self-shielding. Topographic shielding was 207 measured in the field using an abney level but this correction was minimal (scaling factor 208 >0.999999). Present day snow and vegetation cover is thin, and is unlikely to have 209 increased significantly during glacial times, so no correction was applied for shielding by 210 snow cover or vegetation. Likewise, no erosion correction was applied given that the quartz 211 cobbles showed no significant signs of surface erosion. As a result of these assumptions, 212 the ages should be considered minimum estimates. 213

Depth profiles 214
The concentration data from the depth samples were modelled using Hidy et al. for uncertainties. It can be constrained using geological parameters to produce a most 218 probable surface exposure age, erosion rate and nuclide inheritance estimate for each 219 outwash unit. For both depth profiles, there were samples that deviated from the theoretical 220 nuclide decay curve: FP150 for the Filaret profile and CP75 and CP150 for the Cullen profile. 221 We used a jack-knifing process to test whether these were outliers by running the model with 222 wide parameters and then excluding all of the samples one at a time. The model would only 223 run with the outliers mentioned above removed from the profiles and they were not included 224 in further modelling. This resulted in normally decreasing nuclide concentrations with depth, 225 though the modelling was constrained by fewer samples. 226 The 26 Al/ 10 Be ratio for CP150 plotted well below the steady state erosion island, normally 227 indicative of a period of burial that results in a lower 26 Al/ 10 Be ratio. However, it is unclear 228 why the FP150 and CP75 samples yielded anomalous results, given that the 26

Surface sample results 248
The four Río Cullen surface sample 10 Be exposure ages range from 23.7 to 43.2 ka (Table  249 2). The oldest sample (CPSS5) yielded a 26

Depth profile modelling 255
There is a paradox involved in modeling cosmogenic nuclide depth profiles. Often, 256 parameters are unknown, but models require some constraint to produce an age. In theory, 257 very wide, even unrealistic, parameters will yield the most reliable estimates of age, erosion 258 rate and inheritance. However, the wider the constraints, the slower the model will run (if at 259 all) and the wider the resulting error ranges. Consequently, a balance must be found 260 between applying constraints to aid modeling and not inadvertently constraining the age, 261 erosion rate and inheritance without good reason. In this section, we outline the conservative 262 constraints that we applied to the Hidy et al. (2010) model. We present χ 2 sensitivity tests to 263 check that the model output was not inadvertently affected and discuss where there is good 264 reason to apply constraint based on a priori knowledge. Model parameters are given in 265 Table 3 and a summary of the 10 Be depth profile results is given in Information for sensitivity results). Importantly, the controlling parameters only reduced the 271 χ 2 maximum age, and did not significantly affect the χ 2 optimum or minimum age estimates. 272 The sensitivity tests demonstrated that there were three model parameters which controlled 273 the χ 2 maximum ages: maximum total erosion, maximum age, and inheritance. Of these, the 274 maximum total erosion is the key determinant given that maximum age can be constrained We ran sensitivity tests with very wide constraints (between 1 and 3 g cm -3 ) and then used 285 the change in maximum age outputs to constrain values slightly, though these were still 286 extremely conservative given the nature of the sediments (between 1 and 2.7 g cm -3 ). 287

Inheritance 288
Inheritance was essentially unknown. We ran sensitivity tests to assess the effect of 289 inheritance on maximum age outputs and then selected wide constraints. Given that we had 290 deep samples in both profiles, we could also back-check the modelled inheritance in all 291 model runs with the deep-sample nuclide concentrations. In all cases, our maximum 292 inheritance parameters were well in excess of the measured deep nuclide concentrations. these extreme upper limits for sensitivity tests and then took 1100 ka as a more reasonable, 298 but still highly conservative, maximum age limit for all other modelling. We applied no lower 299 age limit during sensitivity tests, but then used an age of 14.3 ka for all other modelling. This 300 is from a well dated Reclus tephra layer, known to have been deposited after the deposition 301 The maximum total erosion is the total amount of surface erosion that the model will allow, 312 and may limit the erosion rate over time if the threshold is low but the erosion rate is high. 313 Sensitivity tests showed that the maximum total erosion strongly affected age outputs, but is 314 an unknown. It was, therefore, the key determinant in constraining maximum modelled age. 315

Approach to modelling 316
To provide the most reliable estimates of age, erosion rate and inheritance from the depth 317 profile modelling, we ran three models for each profile. Firstly, we ran the model 318 'unconstrained' using very wide parameter values from the χ 2 sensitivity tests. All of these 319 parameters were essentially unrealistically wide (e.g. up to 100 m of erosion and 2.7 g cm -3 320 density) but this was useful to gauge if constraining the maximum total erosion altered the 321 age results. Next, we constrained the maximum total erosion to 4 m to test whether there 322 had been significant surface deflation similar to the moraine exhumation of Kaplan et al. and San Sebastián glacial limits suggest that these surfaces are substantially younger than 353 previously thought. For the depth profiles, the optimum ages are 45.6 ka ( +139.9 / -14.3 ) for the 354 Río Cullen limit and 30.1 ka ( +45.6 / -23.1 ) for the San Sebastián limit (Figure 4). The surface 355 samples yield apparent mean ages of 27.2 ± 3.7 ka for the Río Cullen limit and 25.9 ± 1.3 for 356 the San Sebastián limit, which suggests that there has not been substantial deflation of the 357 outwash surfaces that would otherwise result in a scatter of ages. Moreover, the depth 358 erosion (Kaplan et al., 2007). Rather, we show that the Río Cullen and San Sebastián limits 361 were deposited during the last glacial cycle (MIS 4-2), with optimum ages during MIS 3. 362 These new constraints radically alter the glacial chronology of the BI-SSb lobe and 363 demonstrate that it was more extensive during the last glacial cycle, but prior to the gLGM. 364 As noted, high moraine exhumation and boulder erosion rates have been invoked to suggest 365 that exposure ages from moraine boulders on these glacial limits underestimated their age 366 (Kaplan et al., 2007). Our data suggests surface deflation rates of 48.7 mm ka -1 and 0.59 367 mm ka -1 for the Río Cullen and San Sebastián outwash, respectively. The former is relatively 368 high because the age and erosion rates are not well constrained, which is due to fewer 369 samples and our conservative modelling constraints. In contrast, the San Sebastián outwash 370 age and deflation rate estimates are well-constrained. Crucially, all modelled erosion rates 371 are substantially lower than those required for the limits to be hundreds of thousands of 372 years old (Meglioli, 1992), and the close agreement of the depth and surface ages suggests 373 that deflation has not substantially lowered our ages. 374

Geomorphic considerations 375
Our modelling does not support erosion rates consistent with the loss of metres of surface 376 sediment that might be expected if significant deflation of the outwash surface has occurred. 377 However, our erosion rates are assumed to be steady over time, and do not consider rapid,

Comparison to other glacial chronologies 398
Our BI-SSb chronology is unusual because none of the preserved glacial limits of the BI-SSb 399 lobe pre-date the last interglacial (MIS 5) and two major limits were deposited during MIS 3, 400 ~ 100 km beyond the gLGM limit (Figure 1