We measured cosmogenic 3He concentrations in pyroxene separated from samples of Ferrar dolerite. To separate pyroxenes at the University of Maine Cosmogenic Isotope Laboratory, we followed a modified version of the method described by Bromley et al. (2014). We sieved crushed samples to isolate the 125–250 µm grain size fraction, which was boiled for 2 h in 10 % HNO3 to remove Fe oxides and other weathering products. We then removed lighter minerals (mostly plagioclase) using a water-based heavy liquid with a density of 2.94 g cm-3 and leached remaining material in 5 % HF to dissolve adhering plagioclase and remove outer surfaces of pyroxene grains potentially enriched in implanted 4He from U and Th decay (Blard and Farley, 2008; Bromley et al., 2014). Finally, etched pyroxenes were passed through a magnetic separator and handpicked to remove remaining contaminants under a binocular microscope.

We then measured 3He concentrations in clean pyroxene separates at the Berkeley Geochronology Center using the BGC “Ohio” system, which consists of a MAP 215-50 sector field mass spectrometer with updated detectors and counting electronics, coupled to a fully automated gas extraction and purification system. Gas extraction on this system uses a laser “microfurnace” in which ∼15–40 mg aliquots of pyroxene, encapsulated in Ta packets, are heated under vacuum using a 150 W, 810 nm diode laser coupled to a coaxial optical pyrometer in a feedback loop allowing control of the pyrometer temperature. The pyrometer is calibrated by heating a thermocouple in an identical apparatus. However, note that precise temperature measurement is not necessary for this work. In most cases (Table S2 in the Supplement), we extracted helium in an initial 15 min heating step at 1225 ∘C, followed by a second 15 min heating step at 1325 ∘C to ensure complete extraction. The second heating step typically contained 1 %–5 % of total He released. We added additional heating steps for a few representative samples to test for complete extraction and found He signals indistinguishable from blanks. Gases released into the extraction line were purified by reaction with SAES getters and frozen to activated charcoal at 12 K, after which helium was released into the mass spectrometer at 33 K. In all cases, we measured 4He signals on a Faraday cup and 3He on a continuous dynode electron multiplier operated in pulse-counting mode.

We quantified both 3He and 4He sensitivity by peak height comparison between samples and aliquots of custom-mixed helium gas standards, calibrated using direct pressure measurements of both isotopes using Baratron capacitance manometers, containing between 1.57×10-18 and 4.71×10-16 mol of 3He and between 4.39×10-14 and 1.26×10-11 mol of 4He. Ferrar pyroxene has relatively high and highly variable 4He concentrations, and the MAP 215 mass spectrometer displays a significant pressure dependence on He sensitivity (Burnard and Farley, 2000), so accurately quantifying machine sensitivity over a wide pressure range was an important aspect of this work. We addressed this by (i) source tuning at He pressures similar to those expected for sample analyses to improve linearity in the pressure range of interest and (ii) ensuring that observed He pressures in sample analyses were bracketed within the pressure range available from standard analyses. In many cases, this required discarding results of an initial analysis and reanalyzing the sample with a different size aliquot calculated to match sample and standard pressures. Total process blanks measured on empty Ta packets had fewer than 105 atoms 3He and 1010 atoms 4He, which is negligible for all samples discussed here. Reported measurement uncertainties in 3He concentrations include uncertainties from 3He counting statistics (typically 1 %–2 %) as well as the variance in sensitivity inferred from gas standard analyses spanning the pressure range of interest (typically 1 %–3 %).

As additional quality-control measures, we analyzed aliquots of the CRONUS-P pyroxene standard (Blard et al., 2015) together with samples throughout each period of analysis and made replicate analyses of a total of 121 pyroxene samples as well as an additional 21 samples of Ferrar pyroxene from other Antarctic sites (Fig. 2). In each of six distinct measurement periods between 2016 and 2019, we analyzed two to four aliquots of CRONUS-P. Although average measured 3He concentrations in individual measurement periods varied from 4.80±0.30×109 to 5.14±0.1×109 atoms g-1, data from different measurement periods were not distinguishable as separate populations. The mean and standard deviation of 19 measurements during the entire period was 5.03±0.15×109 atoms g-1 (2.9 %), which is indistinguishable from the accepted value of 5.02×109 (Blard et al., 2015). Replicate analyses of other samples had a mean relative standard deviation of 2.2 % (Fig. 2). As expected from counting statistics, replicate scatter varied with 3He concentrations, ranging from 3 % for concentrations <2×109 atoms g-1 to 1.5 % for concentrations >7×109 atoms g-1.

Ferrar pyroxene is known to contain a nonzero concentration of noncosmogenic (presumably magmatic) 3He. Kaplan et al. (2017), Margerison et al. (2005), and Ackert (2000) obtained maximum limiting concentrations for noncosmogenic 3He of 5–7×106 atoms g-1, which are consistent with an unpublished estimate (Balco, unpublished data) of 3.3±1.0×106 atoms g-1. As this is 1.2 % of the lowest total 3He concentration measured in a Roberts Massif erratic in this study and 0.1 % of the average concentration observed, we disregard it and assume that all observed 3He in pyroxene is cosmogenic.

We purified quartz from sandstone samples using established physical and chemical procedures (e.g., Schaefer et al., 2009) at the University of Maine Cosmogenic Isotope Laboratory. Chemical extraction of beryllium and aluminum and preparation of BeO and Al2O3 targets took place at the University of Maine and Lawrence Livermore National Laboratory (LLNL). Ratios of 10Be/9Be were measured relative to the 07KNSTD standard (Nishiizumi et al., 2007) at LLNL and corrected for background 10Be by procedural blanks with a range of 23 000–44 000 atoms. Al isotope ratios are measured relative to the KNSTD standardization of Nishiizumi (2004) and corrected for a procedural blank of 75 000±75 000 atoms. Note that blank corrections for both 10Be and 26Al are negligible for samples in this study. One measurement of the CRONUS-A quartz standard (Jull et al., 2015) run together with these samples yielded 3.491±0.047×107 atoms g-1 10Be and 1.494±0.030×108 atoms g-1 26Al (Table S5), indistinguishable from accepted values for both nuclides. Reported uncertainties for 10Be and 26Al measurements include uncertainties in AMS isotope ratio measurement, process blanks, and 9Be/27Al concentrations.

We measured 21Ne in the same quartz separates used for 10Be analysis using the BGC Ohio noble gas mass spectrometer system also used for 3He measurements and described above. Aliquots of quartz samples were degassed in two heating steps at 850 and 1100 ∘C, and calculations of excess 21Ne (see below) are based on total Ne released in both heating steps. Ne isotope measurements at BGC use a 39Ar spike to quantify and correct for the 40Ar++ interference on mass 20 and are described in Balco and Shuster (2009). We quantified Ne abundances by peak height comparison between samples and aliquots of an air standard containing between 5×10-16 and 2×10-14 mol Ne and calibrated using a Baratron capacitance manometer. In contrast to helium, neon sensitivity was linear within this range at all times. Corrections for mass discrimination, when necessary, are also based on the air standard and assumed atmospheric 21Ne/20Ne and 22Ne/20Ne ratios of 0.002959 and 0.1020, respectively. A total of 20 analyses of the CRONUS-A quartz standard during the period of this study yielded a mean and standard deviation of 319.8±6.3 Matoms g-1 (2 % RSD) excess 21Ne, indistinguishable from the accepted value of 320 Matoms g-1 (Vermeesch et al., 2015).

Neon isotope ratios, as observed in previous studies for TAM sandstones, were indistinguishable from the atmospheric–cosmogenic mixing line (see Supplement, Table S3). However, Balco et al. (2019) and Middleton et al. (2012) have also shown that significant concentrations of nucleogenic 21Ne produced by decay of trace U and Th are present in quartz from this lithology. To calculate cosmogenic 21Ne concentrations in quartz samples, therefore, we first calculated excess 21Ne with respect to atmospheric composition, following Balco et al. (2019) in assuming that excess 21Ne consists of both cosmogenic and nucleogenic 21Ne, and estimated nucleogenic 21Ne concentrations using the following procedure. First, we measured excess 21Ne concentrations in a set of six sandstone samples from ice-proximal sites at Upper Roberts Massif that have apparent 10Be exposure ages of less than 10 kyr and one additional sample with an apparent 10Be exposure age of 75 kyr. Assuming that these samples have experienced a single period of exposure, we calculated the 21Ne concentration attributable to this exposure and subtracted it from total excess 21Ne concentrations to obtain estimates of nucleogenic 21Ne; the resulting mean and standard deviation for nucleogenic 21Ne estimates in these samples are 10.5±2.8 Matoms g-1, similar to but slightly higher than estimates for Beacon Supergroup sandstones in the Dry Valleys region (Balco et al., 2019; Middleton et al., 2012). We then measured U and Th concentrations in quartz and computed apparent (U-Th)/21Ne closure ages as described in Balco et al. (2019); excluding one outlier attributed to a spurious Th measurement, the mean and standard deviation of apparent closure ages is 603±110 Myr. If we assume that all other sandstone erratics from Roberts Massif that we analyzed in this study have a similar source and therefore a similar apparent closure age, we can estimate nucleogenic 21Ne concentrations using U and Th concentrations and this closure age estimate. Note that this apparent closure age is older than the depositional age of the Beacon Supergroup. If these sandstone samples are derived from the Beacon Supergroup, therefore, it is most likely inaccurate as a cooling age. However, the provenance of the sandstone erratics is unknown, and in any case this inaccuracy would not affect the assumption that Roberts Massif sandstone erratics have a single characteristic apparent closure age. Table S4 shows the results of this procedure. For samples with less than 200 Matoms g-1 total excess 21Ne, we measured U and Th concentrations in individual samples and applied the mean closure age inferred from the ice-proximal samples, which resulted in subtraction of up to 20 % of total excess 21Ne as nucleogenic and had a significant effect on results. For samples with higher 21Ne concentrations, the uncertainty in the nucleogenic 21Ne estimate is negligible and we used an average value rather than measuring U and Th in individual samples. For example, for samples from the Southwest Col on Misery Platform, discussed below, estimated nucleogenic 21Ne is less than 0.5 % of total excess 21Ne. Reported uncertainties for 21Ne measurements, as for 3He, are derived from counting statistics as well as reproducibility of the gas standards.

For the majority of samples, we made replicate 3He and 21Ne measurements and performed chi-squared tests on replicate sets with the null hypothesis that all measurements on the same sample belong to a single population and disagree only because of measurement uncertainty. If we could not reject the null hypothesis at 95 % confidence, we took the error-weighted mean of replicate analyses as the true nuclide concentration and the standard error as the uncertainty. If the null hypothesis was rejected, we used the arithmetic mean and standard deviation. A caveat to this procedure, however, is that we found that our 3He results from CRONUS-P during the period of this study did not pass a chi-squared test (p=0.02), indicating that our internal uncertainty estimates for individual 3He measurements are underestimating the true scatter in multiple measurements of the same sample. Thus, we adjusted calculated uncertainties upward when necessary such that no 3He concentration has a relative uncertainty of less than 2.9 %, the relative standard deviation of CRONUS-P measurements. 21Ne results from CRONUS-A, on the other hand, passed the chi-squared test (p=0.35), so we did not make a similar adjustment to 21Ne data. However, cosmogenic 21Ne concentrations do include an additional uncertainty derived from nucleogenic 21Ne subtraction after averaging of replicates.

In the southern portion of the Lower Roberts area, a complex of faults forms a deep, back-tilted basin named “The Bowl” by Hambrey et al. (2003). With the exception of a 100 m relief bedrock hill, referred to here as the Central Rise, and the Bowl, the Lower Roberts area exhibits relatively gentle topography (Fig. 4). Dolerite bedrock surfaces outcrop at several locations throughout Roberts Massif and commonly exhibit glacial polish, striations, and molding consistent with erosion beneath a wet-based glacier. Most of these bedrock outcrops are directly overlain by semilithified, poorly sorted pockets of sediment (several meters thick in places), containing deeply striated gravel- to cobble-sized clasts of heterogenous, non-native lithologies embedded in an olive-gray, clay-rich matrix (Figs. 4 and 5). We interpret these sediments as lodgement tills associated with the Sirius Group. First described by Mercer (1972), the Sirius Group occurs throughout the upper (>∼1500 m elevation) TAM as erosional remnants of clay-rich diamicton that are correlated with at least one period of past temperate glaciation. An in-depth sedimentological study of glacially eroded bedrock surfaces and Sirius Group tills at Roberts Massif and other locations along the upper Shackleton Glacier is provided by Hambrey et al. (2003).

Bedrock and Sirius Group tills are blanketed by patchy glacial drift, comprising primarily angular, cobble-to-boulder-sized clasts with little to no fine-grained material (Fig. 5). Ferrar dolerite is the most abundant lithology, although this drift includes the occasional sandstone boulder, as well as rounded cobbles reworked from the underlying tills described above. A key feature of this drift deposit is the abundance of open-work boulder moraines, which we targeted for surface-exposure dating (Fig. 6). These low-relief (1–2 m high) ridges are composed primarily of large, angular dolerite boulders and are oriented subparallel to the modern ice edge, marking former marginal positions of the EAIS to the south and the unnamed spur of the Zaneveld glacier to the north. The sediments of these drifts and associated boulder-belt moraines exhibit characteristics typical of cold-based glaciation, being thin, patchy, and clast-supported with little to no fine-grained material (Figs. 5 and 6; Atkins, 2013). Furthermore, clasts are generally angular and lack the striations, polish, and molding associated with erosive wet-based ice.

We identified and sampled for surface-exposure dating 15 moraines throughout the Lower Roberts area. We focused on the most prominent, laterally continuous moraines, which comprise accumulations of stacked boulders, and avoided the numerous discontinuous moraine mounds and isolated erratic boulders, from which former ice marginal positions are difficult to reconstruct. The stratigraphically oldest moraine in the Lower Roberts sequence, the Ringleader moraine (informal name) encircles the summit of the Central Rise, indicating that north- and south-flowing ice masses once converged to form a continuous ice surface across the Lower Roberts area at least ∼170 m higher than the modern ice margin to the north. From the Ringleader moraine, at the highest position in the Lower Roberts site, we sampled northern (extending from Ringleader to A in Fig. 4) and southern (extending from Ringleader to A′ in Fig. 4) moraine transects. Listed in stratigraphic order, the northern transect included the BAS, HDY, SSU, WBK, POS, AND, and NLO–NLI moraines (moraine initials correspond to informal names and sample ID suffixes listed in the ICE-D:ANTARCTICA online database and Table S1); from the southern transect we sampled the BBY, BGE, WAL, WIN, MON, and MNM moraines. Notably, the POS moraines constitute a complex of three main ridges, while the NLO–NLI moraines comprise two distinct ridges spaced only by ∼5 m.

The youngest deposit at Roberts Massif comprises a thin layer of sandstone and dolerite debris that extends several tens of meters beyond the current ice margins. Clasts are relatively unweathered (i.e., exhibit minimal staining and/or exfoliation) and exhibit fresh scuff marks (abrasions formed as cold-based ice drags entrained boulders across underlying surfaces; Atkins et al., 2002; Fig. 5f). With the exception of a few discontinuous segments, this unit generally is not associated with distinct moraines. Based on strong similarities in position, morphology, and relative weathering with deposits reported from other TAM sites (e.g., Todd et al., 2010), we correlate the youngest drift unit at Roberts Massif with the most recent late Quaternary expansion of Shackleton Glacier and EAIS and do not discuss it further.

Outboard of this relatively unweathered limit, drift and moraine boulders become progressively more weathered with distance from and elevation above the modern ice. For instance, dolerite boulders belonging to the outermost deposits of the HDY, BAS, and Ringleader moraines (up to 3 km from and 170 m above the modern ice margin) exhibit dark red staining, pitting of up to ∼0.5 cm depth, exfoliation up to ∼4 mm, and weathering rinds 1–2 mm thick, while the presence of sandstone clasts is increasingly rare (Fig. 6d). In contrast, dolerite boulders that we sampled on the innermost moraines were generally blue-gray in color and lacked significant weathering characteristics, such as staining or pitting (Fig. 6c). Although the boulders on the outermost moraines at Roberts Massif display more pronounced weathering than those on the inner moraines, the characteristics described here represent relatively minimal surface weathering compared to slightly warmer and wetter Antarctic locations, such as the McMurdo Dry Valleys. There, ∼3 Ma clasts, which are similar in age to those on the HDY, BAS, and Ringleader moraines (Sect. 3.4), display pitting of greater than 4 cm depth (Swanger et al., 2011). Additionally, we did not observe any cross-cutting relationships between moraine crests throughout Lower Roberts, either on the ground or in satellite imagery. Therefore, we conclude that moraines at this site increase in age with distance away from and elevation above the modern ice sheet surface. Altogether, these surfacemost deposits indicate that the Lower Roberts area records >15 prior expansions of cold-based ice.

The Upper Roberts site is situated on a steep, west-facing slope of Arena Peak, directly adjacent to the northward-flowing lobe of the EAIS that ultimately flows over the Bowl headwall (Fig. 7). Here, we mapped glacial drift and moraines identical in character to those at Lower Roberts, indicating deposition by a cold-based EAIS. Similar to observations at Lower Roberts, a fresh-looking drift of sandstone and dolerite boulders extends several tens of meters beyond the modern ice edge. At the Upper Roberts site, that fresh deposit is associated with a low-relief (∼1.5 m) ridge. We attribute this deposit to the most recent expansion of the EAIS during the Late Quaternary and do not discuss it further in this paper. We focused on five moraine ridges located along a vertical transect between ∼60 and 150 m above and oriented subparallel to the modern ice surface (2150 m). In order of descending elevation, we identified and sampled the Arena (2300 m), Eine (2260 m), Kleine (2240 m), Nacht (2220 m), and Musik (2220 m) moraines (informal names). Additionally, we mapped moraine segments preserved both within and above (up to ∼2500 m elevation) this transect, but, owing to lateral discontinuity and poor preservation on high-gradient slopes, we did not sample these limits for surface-exposure dating. As at Lower Roberts, the general increase in boulder-surface weathering (Fig. 6) and the absence of cross-cutting moraine stratigraphy (determined from field observations and satellite imagery; Fig. 7) suggest that glacial deposits at Upper Roberts become older with increasing elevation above the modern EAIS.

Misery Platform is a broad, gently sloping platform in the southwest part of Roberts Massif (Figs. 3 and 8). Comprising the top surface of the hanging-wall block of a large normal fault, Misery Platform is bounded to the south by a ∼300–340 m high fault scarp. At the base of the scarp, we mapped a series of arcuate moraine ridges (here termed the Misery moraines), four of which we sampled for exposure-age dating (Figs. 8 and 9). The southern edge of the footwall block, which includes Misery Peak (2723 m elevation), drops steeply to the EAIS surface at ∼2200 m elevation and exhibits south-facing, amphitheater-shaped valleys that are occupied partially by north-flowing lobes of the EAIS (Fig. 8). The largest of these valleys is located directly south of the Misery moraines, and its extension above the current surface of the EAIS suggests that this lobe of ice was significantly thicker in the past. Further, a thin drift of glacial erratics atop the footwall block at ∼2550 m elevation marks where a north-flowing lobe of the EAIS overtopped the broad slopes east of Misery Peak and cascaded down the escarpment, where it deposited the Misery moraines on the platform below. This interpretation requires that the Misery moraines (a) postdate the formation of the fault scarp and (b) were deposited by an EAIS that was sufficiently thick (>300 m above the current surface) to overtop the footwall block. Although the Misery moraines are similar in elevation to those sampled at Upper Roberts, they represent the highest former ice surface elevation of the EAIS examined in this study.

Compared to moraines at Lower and Upper Roberts, the Misery moraines are relatively broad and high-relief (∼2–5 m high) and comprise finer matrix material (silt-to-gravel). Moraine crests are mantled with angular dolerite boulders exhibiting pronounced weathering features, including deep red-to-purple staining, 2–3 mm thick weathering rinds, and ventifaction pits of up to 2 cm depth. On the basis of these physical characteristics, they appear older than the outermost moraines at both Lower and Upper Roberts. Therefore, we interpret the Misery moraines as cold-based ice-marginal features marking the ostensibly oldest and most extensive EAIS terminus positions that we documented at Roberts Massif. We used cross-cutting relationships of the Misery moraines to determine their stratigraphic order. From outermost (oldest) to innermost (youngest), we sampled boulders on the following moraine crests: Misery D, Misery A, Misery B, Misery C (note that the designations A–D are field designations reflecting the sequence of sample collection, not the stratigraphic order; Fig. 8). Importantly, we avoided sampling adjacent to overlapping moraine segments.

Immediately outside of, and stratigraphically underlying, the Misery moraines, the weathered bedrock surface is mantled with a thin patchy ablation till, dominated by dolerite boulders and a small number of sandstone clasts, and associated with a coarse-grained sand and gravel deflation surface. We observed this unit throughout Misery Platform and collected samples for surface-exposure dating from boulders on Southwest Col, located approximately 1.5 km northwest of the Misery moraine complex and 400 m above the modern surface of Shackleton Glacier (Fig. 8). Here, the ablation till (“Southwest Col drift”) mantles a bedrock surface of heavily stained and deeply exfoliated coarse-grained dolerite. In places, granular sediments fill joints and depressions in the bedrock. These sediments are characterized by red-stained silt-to-gravel-sized grains, which may derive from the disintegration of the dolerite bedrock, and gravel-to-cobble-sized clasts of various lithologies. In contrast to the Sirius Group deposits observed elsewhere at Roberts Massif, boulders comprising Southwest Col drift are predominantly dolerite (as opposed to a broad mix) and generally more angular.

We sampled three dolerite clasts (one boulder and two cobbles) and four sandstone clasts (three boulders and one cobble), all of which are perched on bedrock and/or interstitial sediments, for surface-exposure dating. The surface of the dolerite boulder (15-ROB-28-COL) exhibits deep red staining and evidence of significant wind abrasion, except on the lee side where there is a thick red-brown weathering rind (Fig. 9). The sandstone boulders (15-ROB-32-COL, 15-ROB-33-COL, and 15-ROB-34-COL) exhibit orange-to-red staining, surface varnish, and ventifaction of up to 4 cm depth. Based on the thin nature of this deposit, we interpret the Southwest Col drift as a cold-based ablation till deposited by the EAIS. Owing to its weathering state, we suggest that this deposit is the oldest glacial unit in our record. Surface-exposure ages from this site therefore provide a minimum-limiting age for temperate glaciation at Roberts Massif.

We mapped three primary surfaces at Roberts Massif (listed in stratigraphic order): glacially molded and striated dolerite bedrock, temperate-style tills belonging to the Sirius Group, and cold-based drifts associated with openwork boulder moraines. All samples collected for surface-exposure dating are derived from the cold-based deposits marking former positions of the EAIS. At both the Lower and Upper Roberts sites, weathering patterns and the lack of cross-cutting moraines suggest that relative moraine ages increase with distance from, and elevation above, the modern ice sheet margin. Deposits on Misery Platform (the Misery moraines and the Southwest Col drift) exhibit more advanced subaerial weathering than our other sites, indicating that these deposits are significantly older. In Sect. 4.2, we describe results from cosmogenic-nuclide measurements made on samples from 23 separate moraine ridges and one drift sheet.

As a majority of landforms at Roberts Massif are several million years old, quantifying the magnitude of surface erosion is key to accurate exposure dating. Here, we summarize geochemical data and field observations that allow us to place limits on long-term erosion rates. Four sandstone erratics at Southwest Col have 10Be concentrations close to predicted production-erosion saturation values, and apparent 21Ne exposure ages of 9–12 Myr. As these samples have nearly the highest concentrations of these nuclides yet measured on Earth, concentration measurements are correspondingly (and unusually) precise, making it possible to use the paired 10Be–21Ne data to simultaneously infer exposure ages and surface erosion rates from these samples (Fig. 10; Gillespie and Bierman, 1995; Lal, 1991). Given the assumption that these samples have experienced continuous exposure at a steady erosion rate, the 10Be–21Ne data imply true exposure ages in the range of 12–15 Myr but varying surface erosion rates in the range of 0.5–3 cm Myr-1. These low erosion rates are consistent with our field observations pertaining to surface erosion of these sandstones as described in Sect. 3.1.3.

Apparent 3He exposure ages from three dolerite clasts also located on Southwest Col, and which therefore should have the same true exposure age as the sandstone clasts, are 8.6, 10, and 11 Myr. Assuming that the true exposure age of the deposit is no greater than 14.5 Myr, as implied by the two-nuclide data for the highest-nuclide-concentration sandstone (15-ROB-032-COL) shown in Fig. 10, this implies maximum erosion rates for the dolerite clasts of 3.8, 2.7, and 1.9 cm Myr-1, respectively. Further assuming that the dolerite clast with the highest 3He concentration (15-ROB-028-COL) has been exposed at the drift surface for the longest period, and has therefore experienced mainly surface weathering rather than exhumation from till, we propose that ∼2 cm Myr-1 is likely a maximum limit on rock surface erosion rates for dolerite surfaces in our study area. The assumption that this clast has been exposed at the surface is supported by the fact that 15-ROB-028-COL is a boulder, while the rest of the dolerite surfaces we sampled on Southwest Col are cobbles. If the deposit is younger than 14.5 Myr, an even lower erosion rate would be implied. Although this is an extremely low surface weathering rate by global standards, it is nonetheless consistent with the polar desert climate and the field observations described in Sect. 3.1.3 (i.e., angular clasts with surface varnish and minimal pitting).

As on Southwest Col, we also measured multiple nuclides (10Be and 21Ne and, in one case, 26Al) in several sandstone boulders on the Ringleader, WIN, MON, AND, and NLO moraines at Lower Roberts (Fig. 11). Although sandstone clasts are rare on these moraines, these data provide some insight into the exposure history of these boulders that we can use to assess the importance of inheritance and postdepositional disturbance for moraine exposure ages.

In general, a boulder that has experienced a single period of exposure that is equal to the emplacement age of the moraine should display concordant 10Be, 21Ne, and 26Al ages that are the same as those of other boulders on the moraine. For the Ringleader moraine (Fig. 11), 10Be–21Ne–26Al measurements are concordant at 2.8–3 Ma, therefore consistent with simple exposure at negligible erosion, and lie in the center of the range of 3He ages from dolerite clasts on the same moraine (Fig. 12). These observations suggest that (i) the sandstone boulders have experienced a single period of exposure with minimal postdepositional exhumation or weathering, which is consistent with our field observations as described in Sect. 3.1.1; (ii) their exposure age most likely represents the true emplacement age of the moraine; and (iii) two outliers in the 3He age distribution can likely be attributed to both inheritance (one ∼4 Myr age) and postdepositional disturbance (one ∼2 Myr age).

In contrast, paired 10Be–21Ne measurements on four boulders on the MON moraine and one on the WIN moraine (Fig. 11), both adjacent to the Bowl and emplaced by ice from Upper Roberts overflowing the Bowl headwall (Fig. 4), display discordant apparent ages. Additionally, apparent exposure ages from both sandstone and dolerite boulders at these moraines are relatively scattered (coefficient of variance >20 %). The 10Be–21Ne data (Fig. 11) could be explained either (i) by an extended period of steady erosion at an ice-free site prior to entrainment and deposition of the clasts or (ii) by repeated exposure and ice cover of the samples prior to emplacement. Both of these conditions are likely if these boulders were sourced from the adjacent outcrop area of sandstone on the Bowl headwall (Fig. 4a). Thus, we consider it most plausible that the apparent exposure ages of these sandstones reflect prior exposure and, thus, overestimate the true age of the moraine. In general, these results imply that high scatter in exposure ages for moraines in the Bowl are most likely explained by inherited nuclide concentrations in clasts sourced from the adjacent headwall, and the true ages of the moraines are therefore likely close to the young end of their age distributions.

Finally, paired 10Be–21Ne measurements from the AND and NLO moraines (Fig. 11), both at the ice-proximal end of the northern Lower Roberts transect, fall within the “erosion island” on the two-nuclide diagram, indicating that their true exposure ages are older than the apparent ages for either nuclide. In addition, these clasts have apparent ages higher than most 3He ages from these moraines (Fig. 12). Again, this is best explained if the scatter exhibited by these moraines is largely the result of inheritance.

Overall, although we have a relatively small number of multiple-nuclide data from sandstone boulders, our results demonstrate that (i) inheritance is unequivocally present in some moraine boulders and (ii) inheritance is likely most significant at moraines where boulders are likely sourced from a combination of far-traveled EAIS subglacial debris and cliff fall within the massif itself. These scenarios are also consistent with the observation that boulders on moraines at Upper Roberts, which can only be derived from beneath the EAIS, exhibit substantially less scatter than moraines at Lower Roberts (Table 1 and Fig. 12), where additional input from rockfall is likely. Overall, while none of our observations exclude postdepositional disturbance as a potential source of scatter, they do show that inheritance is likely a more important contributor. However, the small number of young outliers in our dataset (Fig. 12) may reflect the ablation of an ice core from the moraine.