To provide additional context for our results and discussion, we first outline sources of uncertainty that need to be accounted for when calculating in situ 14C and 10Be exposure ages, and introduce the concept of paired nuclide diagrams (Granger, 2006). Uncertainty impacts both accuracy and precision at all stages of determining an exposure age of a sample and can be divided broadly into three categories: (i) geologic uncertainty (ii) uncertainty incorporated during sample preparation and isotopic analysis to determine a nuclide concentration and (iii) uncertainty sourced from exposure age calculations.

Cosmogenic nuclide practitioners have least control over geologic uncertainty, which is inherent in a sample from its time of collection in the field and rooted in the limited knowledge we have of a sample's true exposure history and any processes that may have modified production of nuclides following exposure (Dunai, 2010). The two main sources of geologic uncertainty are nuclide inheritance (described above) and post depositional disturbance caused by shielding, erosion, and/or rolling of a sample (Balco, 2011; Gosse and Phillips, 2001). Steps commonly taken to reduce their impact include a robust and detailed geologic interpretation of deposits or depositional features being dated (Balco, 2011) and statistical techniques (Heyman et al., 2016; Johnson et al., 2014). Comprehensive summaries of geologic uncertainty and efforts to quantify it can be found in Balco et al., (2011, 2020b).

The second major source of uncertainty comes from our ability to measure the nuclide concentration accurately and precisely within a sample, which represents the internal uncertainty component of an exposure age calculation (Balco, 2020a). Cosmogenic nuclide dating specialists make efforts to minimise contributions to measurement uncertainty particularly from i) uncertainties introduced during sample preparation, and ii) sample measurement by accelerator mass spectrometry (AMS). Measurement of the cosmogenic nuclide 10Be is now relatively well-established and routine following efforts to reduce sources of laboratory sample preparation uncertainty (Corbett et al., 2016, 2022; Kohl and Nishiizumi, 1992) and improve AMS performance (Merchel et al., 2012; Rood et al., 2010, 2013; Wilcken et al., 2022). These efforts have resulted in 10Be measurement precision on typical quartz interlaboratory comparison materials (e.g., CRONUS-A, CoQtz-N) of between ∼2%-4% (Binnie et al., 2019; Jull et al., 2015; Phillips et al., 2016a). For high in situ 14C concentrations (≥1-2×105atg-1) internal analytical uncertainty is dominated by AMS counting statistics, with a total measurement uncertainty (combined AMS counting error and blank correction) averaging<2% (Goehring et al., 2014; Hippe, 2017). For moderate in situ 14C concentrations in the 104atg-1 range, the uncertainty from AMS counting statistics increases, but typically remains below 10 % and mostly below 5 % (Hippe, 2017). However, for lower in situ 14C concentrations the blank correction increasingly dominates, leading to a rapid increase in total uncertainty (Goehring et al., 2014; Hippe, 2017). These sources of uncertainty have been challenging to quantify despite improvements to in situ 14C extraction from quartz (Fülöp et al., 2010, 2015, 2019; Goehring et al., 2014, 2019a; Hippe et al., 2009, 2013; Lamp et al., 2019; Lifton et al., 2001, 2015b, 2023; Lifton, 1997; Lupker et al., 2019) and the dominance of the blank at lower concentrations illustrates the significant challenges of avoiding contamination from other potential sources of C that impart uncertainty into the final in situ 14C measurement. Laboratory intercomparison reproducibility studies of CRONUS-A (an intercomparison material for cosmogenic nuclides including 14C and 10Be) indicate the coefficient of variation (CoV) of in situ 14C concentration measurements is in the range of 6 %–8 %, and 3 %–4 % for 10Be (Phillips et al., 2016a). There have also been several recent improvements to the in situ 14C extraction process, including identification of potential contaminant sources introduced during quartz purification (Nichols and Goehring, 2019), and automation of 14C extraction lines that reduce risk of atmospheric 14C contamination (Goehring et al., 2019a; Lifton et al., 2015b, 2023; Lupker et al., 2019). Refinements to the stepped heating process to liberate in situ 14C (in the form of CO2) from quartz are also being explored (Lifton et al., 2023) and some extraction facilities now omit the graphitisation stage (that converts CO2 to C) in favour of analysing in situ 14C directly using gas source AMS (e.g., Lamp et al., 2019).

The final major source of uncertainty comes from transforming a measured nuclide concentration into an exposure age. This requires estimating the production rate due to secondary spallation reactions, which accounts for the majority of surface production (Dunai, 2010), and muons (Balco, 2017). Production rate uncertainties have been incrementally reduced via improvements in scaling models, especially more recent models based on particle-physics simulations (Argento et al., 2015a, b; Lifton et al., 2014). Estimates of the 10Be production rate uncertainty from spallation are currently in the range of 6 % (Borchers et al., 2016; Marrero et al., 2016). However, in the case of in situ 14C, a spallogenic production rate uncertainty could not be fitted to calibration data because of scatter in excess of an assumed measurement uncertainty of 7.3 % for in situ 14C concentrations at selected calibration sites (Borchers et al., 2016). Muons account for a much smaller proportion of total cosmogenic nuclide production at the surface than spallation, but this quantity differs between 10Be and in situ 14C. The proportion of 10Be production by muons at the surface is between 1.5 %–2 %, which translates to a maximum scaling uncertainty of only 0.5 % for estimating total 10Be surface production by muons. However, for in situ 14C, production by muons accounts for ∼20% of total in situ 14C nuclide production at the surface (Lupker et al., 2015). Therefore, for in situ 14C the same 10 %–25 % uncertainty on computing a production rate by muons equates to between a 2 % and maximum 5 % uncertainty on the total surface production rate estimate (Balco, 2017).

Due to the inherently different systematics of production and radioactive decay of in situ 14C and 10Be, paired nuclide diagrams (Fig. 2) represent a useful method of visualizing and interpreting exposure/burial histories, and can help to identify or explain uncertainty and scatter in a dataset (see Granger, 2006 for a detailed description of paired nuclide diagrams). Paired nuclide plots generated from exposure age pairs (including 14C–10Be pairs) can be classed into three distinct types: Type 1 for samples with simple exposure history (only one period of exposure), Type 2 for samples with a complex exposure history (multiple periods of exposure and burial), and Type 3 for samples with an impermissible concentration ratio (where an ellipse plots above the line of constant exposure in the “impermissible” zone). The Type 3 scenario can indicate analytical inconsistencies, for example, 14C contamination increasing in situ 14C concentrations (Nichols and Goehring, 2019) or could reflect application of an incorrect production rate to one or both nuclides. In certain cases, a Type 3 nuclide ratio may be explained geologically because the constant exposure line assumes a surface production rate rather than subsurface production. However, because the cosmogenic nuclide production rate by muons as a proportion of total surface production is an order of magnitude higher for in situ 14C than for 10Be, the 14C–10Be production ratio increases with depth below the surface (Hippe, 2017). For example, a sample that is buried under a thin layer of rock, ice, till or other material and then rapidly exhumed by plucking can, therefore, exhibit seemingly “impermissible” paired 14C–10Be concentrations due to differences in the 14C–10Be total production ratio at the surface versus at depth (Hippe, 2017; Rand and Goehring, 2019).

In summary, sources of geologic, sample preparation, and exposure age calculation uncertainty impact the accuracy and precision of Holocene deglaciation chronologies. An increase in paired 14C–10Be measurements in the recent ∼5 years, driven by greater 14C extraction throughput (Goehring et al., 2019a; Lifton et al., 2015b, 2023) provide many new data to make an assessment of the application of both nuclides and investigate sources of uncertainty, particularly of in situ 14C. In the following sections, we investigate the cause of concordant and discordant paired in situ 14C–10Be exposure ages at Mt Murphy and potential causes for the large amounts of scatter in reported in situ 14C measurements using new in situ 14C data from Mt Murphy and existing paired in situ 14C–10Be data extracted from ICE-D.

Mount Murphy is a large volcanic edifice adjacent to Thwaites Glacier in the Amundsen Sea Embayment (Fig. 1a). Along its western flank, adjacent to Pope Glacier, there are several smaller nunataks, many of which host erratic cobbles and boulders that are well-rounded and of exotic lithology, indicating transport to the site by ice. We selected nine samples from erratics (Table 1) for in situ 14C analysis. These had previously been measured for 10Be (Adams et al., 2022; Johnson et al., 2020), with the resultant thinning history implying exposure during the Holocene. We selected four of the nine samples for repeat in situ 14C measurements to determine if measurement uncertainty may have contributed to conflicting exposure histories suggested by initial in situ 14C concentrations from our samples.

We ensured that paired 10Be and 14C exposure ages cover a wide elevation range by selecting samples from three different locations around the Mt Murphy massif (Notebook Cliffs, samples collected from 893–834 m a.s.l. (metres above sea level), Turtle Rock, 696–438 m a.s.l., and a scoria cone adjacent to Kay Peak, 239–178 m a.s.l.). Notebook Cliffs comprises basaltic lava flows overlying thick sequences of hyaloclastite (Adams et al., 2025; Smellie, 2001). A few granite erratics and SSE–NNW trending bedrock striations are present, indicating past ice-cover (Johnson et al., 2020). Turtle Rock, situated adjacent to Pope Glacier, is primarily composed of hyaloclastite and consists of a broad flat lower terrace (438–452 m a.s.l.), which hosts the highest number of erratics observed at Mt Murphy (Johnson et al., 2020). Turtle Rock rises at its northern end to ∼710m a.s.l. and consists of several superimposed sequences of basalt and hyaloclastite, with erratics collected from three smaller terraces up to 696 m a.s.l. (Johnson et al., 2008, 2020). The scoria cone is located>15km downstream of Notebook Cliffs and Turtle Rock and is less than 1 km from the grounding line of Pope Glacier. The site consists of two small outcrops comprised of rubbly oxidised scoria bounded on one side by a moraine (Adams et al., 2022; Nichols et al., 2024). Cobbles deposited on the outcrops are generally well-rounded suggesting long distance transport (Adams et al., 2022). Detailed geological and geomorphological descriptions of these sites are provided by Johnson et al. (2020) and Smellie (2001). Geomorphic descriptions and supporting information of the nine samples with paired 14C–10Be measurements are provided in Sect. S1 and Table S2 in the Supplement.

We obtained purified quartz necessary for in situ 14C extraction of initial and replicate samples by performing mineral separation on our whole rock samples in the CosmIC Laboratory at Imperial College London (UK), largely following methods specified in Corbett et al. (2016). We omitted the froth flotation step (used to separate feldspars and quartz) following recommendations made by Nichols and Goehring (2019) and instead performed 3×1% HF/HNO3 etches to isolate the quartz (Kohl and Nishiizumi, 1992). Quartz purity was determined using Inductively Coupled Plasma Optical Emission Spectrometry (ICP-OES), after which ∼10g of purified quartz from each sample was sent to Tulane University Cosmogenic Nuclides Laboratory (Tulane CNL, New Orleans, USA) for in situ 14C extraction. Extraction of in situ 14C was performed using the fully automated Tulane University Carbon Extraction and Graphitisation System (TU-CEGS) following methods presented in Balco et al. (2023), modified from Goehring et al. (2019a). Sample aliquots of purified quartz ranging from 3–5 g were loaded into platinum crucibles and fused with lithium metaborate (LIBO2) flux to ensure sample dissolution and complete liberation of in situ 14C (Lifton et al., 2001). The sample was heated in a stable high purity O2 atmosphere for 30 min at 500 °C to remove atmospheric 14C and organic contaminants. Following evacuation of the furnace and addition of new high purity O2, the sample was further heated to 1100 °C for 3 h to completely dissolve the quartz and liberate in situ 14C (in the form of CO2). Liberated CO2 was cryogenically purified before being collected in a measurement chamber, quantified monometrically and diluted with 14C-free CO2 to ensure a measurable sample size (Goehring et al., 2019a). CO2 was graphitized using standard H2 reduction methods over an Fe catalyst (Santos et al., 2004, 2007; Southon, 2007). Several changes were made to the configuration of the TU-CEGS prior to the replicate measurements (n=4). Alterations included the introduction of a new compact borosilicate coil trap held at liquid nitrogen temperature (-196°C) for trapping evolved CO2 following quartz dissolution (Lifton et al., 2015b, 2023, 2001; Pigati et al., 2010), which replaced the previously installed variable temperature trap (Goehring et al., 2019a). A new mullite tube was also used for 14C extraction due to failure of the previous tube; mullite tubes at Tulane CNL have previously been observed to undergo a “break in” period, during which initial 14C blanks are higher but often fall with continued use (Goehring et al., 2019a). 14C/12C isotope ratios were measured by AMS at the National Ocean Sciences Accelerator Mass Spectrometry Facility (NOSAMS) (Woods Hole, USA) using the methods described in Longworth et al. (2015). A small aliquot of 2–3 µg of C was removed for δ13C analysis at the University of California Davis Stable Isotope Facility (see Table S1 in the Supplement). Data reduction to convert 14C/12C ratios to 14C/Ctotal followed methods outlined in Hippe and Lifton, (2014). We applied the blank correction reported from the Tulane CNL of 4.53±0.24×104atoms for initial 14C measurements (n=9) and 7.14±0.30×104atoms for replicate measurements (n=4, Table S1), respectively, to total measured in situ 14C concentrations. Prior to calculating exposure ages, we assigned a 6 % (1σ) uncertainty to each in situ 14C measurement concentration reported by AMS. This 6 % uncertainty exceeds the reported analytical uncertainty for all in situ 14C measurements made for this study and reflects the reproducibility of replicate measurements of CRONUS-A extracted at Tulane CNL, which is reported as 6.12±0.32×105 at g⁻¹ (n=10) (Goehring et al., 2019a). This 6 % uncertainty has been routinely applied by studies where in situ 14C extraction was carried out at Tulane CNL, e.g., (Nichols et al., 2019).

We calculated exposure ages for the new in situ 14C measurements, as well as for the published 10Be measurements, using version 3 of the online calculators (https://hess.ess.washington.edu/math/v3/v3_age_in.html, last access: 10 September 2025) with the “LSDn” production rate scaling method for neutrons, protons, and muons (following Lifton et al., 2014 and summarised in Balco, 2017). For 10Be, exposure ages were calculated based on the CRONUS-Earth primary production rate calibration data set of Borchers et al. (2016). For in situ 14C, we used the long-term CRONUS-A measurements from extractions performed at Tulane CNL presented in Goehring et al. (2019a) for production rate calibration. When comparing in situ 14C and 10Be exposure ages at Mt Murphy, we used the external uncertainty, which includes the 6 % (1σ) measurement uncertainty propagated in quadrature with the production rate and scaling scheme uncertainties. We report all exposure ages assuming no erosion or snow cover (making them “apparent” exposure ages) and a sample density of 2.7 g cm-3 to maintain consistency with Johnson et al. (2020) and Adams et al. (2022). In situ 14C AMS data and corresponding calculated exposure ages are available from the NERC UK Polar Data Centre, 10.5285/dbb30962-bbf3-434a-9f27-6de2f61a86e2 (Adams et al., 2024).

Thirteen in situ 14C measurements (including four replicate measurements) were performed on nine erratic samples recovered from Notebook Cliffs, Turtle Rock, and the unnamed scoria cone (all from surfaces situated between 179 and 893 m a.s.l.). Exposure ages calculated from nuclide concentrations (Table 1) are reported with 1σ internal uncertainties unless otherwise stated. Initial in situ 14C exposure ages from the nine samples range from 9.0±1.0ka to 3.1±0.2ka (Table 1 and Fig. S9 in the Supplement), with an average exposure age of 6.0±2.1ka (mean and standard deviation). At the unnamed scoria cone, in situ 14C exposure ages exhibit considerable scatter over a small elevation range (180–240 m a.s.l.) with ages ranging from 9.0±1.0ka to 3.4±0.3ka. The spread of in situ 14C exposure ages calculated from initial measured in situ 14C concentrations is ∼6kyr, and some samples at higher elevations yield younger exposure ages than samples from lower elevations, which is the inverse of the expected age-elevation pattern associated with ice-sheet thinning through time. Apparent exposure ages from Notebook Cliffs (850–900 m a.s.l.) are 2 kyr younger than those from the scoria cone (180–240 m a.s.l.).

Repeat in situ 14C measurements (n=4) were made on samples TUR-117 and TUR-132 from Turtle Rock (450–650 m a.s.l.), and samples CIN-108, and CIN-112 from scoria cone (180–240 m a.s.l.) to determine if measurement reproducibility contributed to the two conflicting exposure histories suggested by initial in situ 14C concentrations. Exposure ages calculated from the four repeat measurements range from 8.2±0.8ka to 7.2±0.7ka (Table 1). Only one exposure age derived from replicate measurements reproduce within internal measurement uncertainties (at 1σ), whilst three do not. Ages from TUR-117 and CIN-112 do not reproduce within internal uncertainty at 1σ (Table 1) or indeed 2σ, with ages calculated from initial 14C concentrations resulting in ages 5–3 kyr younger than those calculated from replicate in situ 14C measurements. In other words, TUR-117-R and CIN-112-R are 165 % and 112 % older, respectively, than initial ages from the same samples, and exhibit significant scatter in excess of their internal uncertainties. CIN-108-R (7.8±0.8ka) is 24 % older than CIN-108 (6.3±0.6ka) and does not reproduce within 1σ internal uncertainty. TUR-132-R does, however, reproduce within 1σ internal uncertainty. Neither Turtle Rock nor scoria cone sites show a systematic bias in terms of reproducibility, with each site yielding one unreproducible in situ 14C exposure age. There is no correlation between sample lithology and in situ 14C reproducibility, as ages derived from both granite and gneiss samples do not reproduce within internal uncertainties. Notably, initial analyses of samples from Notebook Cliff (n=3) and TUR-123 from >600ma.s.l. all yield younger ages than those obtained from repeat measurements on samples below 500 m a.s.l.; this is inconsistent with the expected age-elevation pattern associated with ice thinning.

Published 10Be exposure ages and initial in situ 14C exposure ages from samples TUR-132, CIN-102 and CIN-108 overlap within their respective 1σ external uncertainties making them concordant (Fig. 3a). However, most of the paired 14C–10Be ages (n=6, including all Notebook Cliff samples, TUR-117, TUR-123, and CIN-112) are discordant and have apparent exposure ages that are mid-late Holocene (5–3 ka). Where in situ 14C and 10Be ages are concordant, the in situ 14C age is systematically older and early- to mid- Holocene (9–6 ka). Ages calculated from three of four replicate measurements overlap at 1σ external uncertainty with the corresponding 10Be exposure ages (Fig. 3b). However, the exposure age calculated from the in situ 14C concentration of sample CIN-108-R is discordant with the 10Be exposure age from the same sample. We note the initial in situ 14C measurement of CIN-108 (6.3±0.7ka), and average of the two in situ 14C measurements of CIN-108 (7.1±0.9ka, n=2) resulted in a concordant 14C–10Be exposure age pair.

Based on their initial in situ 14C concentrations, the samples from Notebook Cliffs, Turtle Rock, and scoria cone (Fig. 4a) can be classified as Type 1 and Type 2 nuclide ratios on a paired nuclide diagram (see Sect. 1.1). TUR-132, CIN-102 and CIN-108 plot within the steady-state erosion island (Type 1) and display concordant in situ 14C and 10Be exposure ages (Fig. 4a). The remaining samples (NOT-103, NOT-104, TUR-123, TUR-117, and CIN-112) yield paired 14C and 10Be nuclide concentrations that plot below the steady state erosion line, suggesting complex exposure histories (Type 2). Samples plotting below the steady erosion line (n=6) include all the young in situ 14C ages, which are discordant with respect to the 10Be exposure age from the same sample. Conversely, all in situ 14C concentrations measured in replicates result in older 14C exposure ages and yield 14C–10Be ratios which plot within the steady state erosion island (Fig. 4b), suggesting samples CIN-112 and TUR-117 instead experienced a simple (Type 1) rather than complex (Type 2) exposure history. The 14C–10Be ratio of CIN-108-R, and to some extent TUR-117-R, CIN-102 and TUR-132, suggest Type 1 exposure but border on an impermissible Type 3 exposure history (see Fig. 2).

Three of the four replicate in situ 14C measurements yielded exposure ages that do not overlap within internal uncertainty (1σ). TUR-117-R and CIN-112-R also do not overlap at 2σ internal uncertainty with the corresponding initial 14C exposure ages, and are also discordant with 10Be exposure ages from the same samples. The six systematically young initial in situ 14C exposure ages measured in samples ranging from 150–900 m a.s.l. appear to contradict previous interpretations of ice surface lowering to ∼150ma.s.l. at Mt Murphy by 6 ka (Adams et al., 2022; Johnson et al., 2020). In addition, the young discordant in situ 14C exposure ages from higher elevations (Notebook Cliffs and Turtle Rock upper terrace; TUR-123) and older reproducible in situ 14C ages from lower elevations (scoria cone; CIN-102, CIN-108) contradict the age-elevation pattern expected with ice thinning. The two samples measured for in situ 14C that did reproduce within their uncertainties at 2σ (TUR-132 and CIN-108) were also concordant with respect to 10Be exposure ages from the same sample (Fig. 3b). From the Mt Murphy replicate measurements (n=4), the two young in situ 14C ages are not reproducible at 2σ internal uncertainty, but both older exposure ages are reproducible at 2σ internal uncertainty. In summary, Mt Murphy paired 14C–10Be exposure ages display both concordance and discordance across multiple sites. Concordant exposure ages are consistent with Type 1 14C–10Be ratios and discordant exposure ages consistent with Type 2 14C–10Be ratios. Concordant 14C–10Be exposure ages exhibit in situ 14C ages which are reproducible at 2σ internal uncertainty whereas discordant 14C–10Be exposure ages do not. This lack of reproducibility suggests the stated 6 % (1σ) measurement uncertainty (based on replicability measurements of CRONUS-A at the in situ 14C extraction laboratory) that is assigned to in situ 14C concentrations prior to calculating an exposure age may be underestimated for our study.

A bootstrap linear regression analysis (see Sect. S3 in the Supplement) of in situ 14C and 10Be exposure age datasets from the scoria cone adjacent to Kay Peak indicate that the chronologies derived from each dataset are broadly similar with respect to the timing of deglaciation, implying they are equally accurate (see Fig. S14 in the Supplement). There is, however, excess scatter of ∼1849 years in the in situ 14C ages (Sect. S3, Table S6 in the Supplement) that cannot be accounted for by the nominal 6 % 1σ internal measurement uncertainty for in situ 14C that has been adopted in many studies (Balco et al., 2019; Nichols et al., 2019). Together, our new in situ 14C exposure ages, the existing 10Be exposure data and the corresponding 14C–10Be ratios from Mt Murphy (Figs. 3 and 4) raise two questions: (1) Is there a way to explain why 3 of the 4 replicate in situ 14C analyses from Mt Murphy do not reproduce at 1σ internal uncertainties? and (2) Is there a geological explanation for the coexistence, often at the same elevation, of concordant and discordant paired in situ 14C and 10Be exposure ages? We discuss answers to these questions in the following sections.

First, we explore if discordant paired in situ 14C–10Be ages and in situ 14C concentrations that do not reproduce within reported uncertainties can be explained by a comprehensive examination of blank and CRONUS-A data. For this, we describe results of a series of sensitivity analyses using quality control data from Tulane CNL (Balco et al., 2023).

For initial in situ 14C concentrations, we apply a new blank correction of 7.44±4.16×104atoms (n=6, cf. Table S5 Balco et al., 2023) calculated from the mean and standard deviation of process blanks reported from 2 March–10 April 2021, which brackets the extraction dates of initial in situ 14C measurements. For replicate in situ 14C concentrations (extracted week beginning 19 April 2022), we apply a revised blank correction of 7.14±3.50×104atoms, which is the same as reported from Tulane CNL. We also propagate a larger blank uncertainty of 35 043 atoms based on the standard deviation of all process blanks measured at Tulane CNL from 2019–2021 (Table S5, Balco et al., 2023). The blank uncertainty we use for the in situ 14C replicate analyses is an order of magnitude larger than the blank uncertainty of ∼3000atoms originally reported by Tulane CNL. See Sect. S2 in the Supplement, for a further sensitivity analysis using an alternative blank correction of 3.99±1.25×104atoms (n=5).

Prior to calculating in situ 14C exposure ages, we also assign a 10 % (1σ) uncertainty to in situ 14C concentrations calculated from the standard deviation of in situ 14C measured in CRONUS-A extracted at Tulane CNL from 22 December 2015–12 March 2021 (5.88±0.59×105atg-1, n=18, cf. Table S5 Balco et al., 2023). For three in situ 14C measurements, the 10 % value is exceeded by our combined AMS and recalculated blank uncertainty, which we use instead for the 14C exposure age calculations (see Table 2 and Table S4 in the Supplement). The 10 % uncertainty we assign to in situ 14C concentrations from the Mt. Murphy samples includes more recent measurements of CRONUS-A from Tulane CNL and is larger than the 6 % typically assigned to in situ 14C measurements from that laboratory in other studies (e.g., Balco et al., 2019, Nichols at al., 2019, Rand et al., 2025).

Using different blank corrections (Table 2) to calculate the in situ 14C exposure ages results in some improvements in reproducibility. Older initial in situ 14C ages and replicate in situ 14C ages in the ∼8-6ka range now all reproduce at 1σ and are concordant with previously published 10Be ages. However, young initial in situ 14C ages spanning from ∼5-3ka (n=6) do not reproduce at 1σ (or 2σ) and remain discordant with published 10Be ages (Fig. 5). The mean value of all initial in situ 14C exposure ages lowers from 5.2±2.1ka to 4.8±2.0ka when the larger blank correction is applied (7.44±4.16×104 atoms). This shift to younger in situ 14C ages increases the mismatch between TUR-117 and TUR-117R by 6.8 %, and CIN-108 and CIN-108-R by 10 %.

Instances of concordant and discordant 10Be and 14C exposure ages (Balco et al., 2019) or seemingly impermissible 14C–10Be concentration ratios have been explained in previous studies by invoking geological processes (Balco et al., 2019; Rand and Goehring, 2019). First, we examine the Notebook Cliffs, Turtle Rock, and scoria cone sites (see Figs. S1–S5 in the Supplement) to determine if localised glacial-geological changes at Mt Murphy permit the existence of paired in situ 14C–10Be discordant exposure ages at the same elevation as paired in situ 14C–10Be concordant exposure ages. A USGS trimetrogon aerial (TMA) photograph shows that in 1966, in contrast to today, the lower scoria cone outcrop was almost completely buried by ice (see Fig. S7 of Nichols et al., 2024). This finding indicates that samples CIN-112 and CIN-108 were shielded by ice for a non-zero time between 6.4 ka and present (Adams et al., 2022; Balco et al., 2023). A discordant initial in situ 14C age for CIN-112 (3.4±0.3ka, 179 m a.s.l.) that is younger than the 10Be exposure age (6.6±0.4) from the same sample and other in situ 14C ages from higher elevation scoria cone samples supports the interpretation that such burial occurred during the late Holocene. The in situ 14C replicate measurement, CIN-112-R, however, yielded an exposure age of 7.2±0.9ka, which is in agreement with the corresponding 10Be age. Both in situ 14C exposure ages determined from measurements of sample CIN-108 (collected from the same outcrop and elevation as CIN-112) are early-mid Holocene (CIN-108 – 6.3±0.7ka; CIN-108-R – 7.8±1.0ka). With the exception of the initial 14C exposure age from sample CIN-112, all exposure ages from the lower scoria cone outcrop (Adams et al., 2022) suggest that ice cover during the late Holocene was short-lived.

At both Turtle Rock and Notebook Cliffs, there is little evidence to suggest prolonged cover or burial of samples. At Turtle Rock, discordant in situ 14C–10Be exposure ages of TUR-117 and TUR-123 could be due to individual samples being partially shielded by till or ice debris cover during the Holocene, but the preferential sampling of topographic highs makes this less likely (Johnson et al., 2020). Furthermore, the in situ 14C exposure age of TUR-117-R (8.2±1.1ka) agrees with the sample's existing 10Be exposure age. There is no geological explanation for measurements of the same nuclide (in situ 14C) on the same two samples (TUR-117, CIN-112) yielding different exposure ages. At Notebook Cliffs, all in situ 14C exposure ages (n=3) are late Holocene and discordant with existing 10Be ages, implying inheritance in 10Be and prolonged burial of all three samples. The three Notebook Cliffs in situ 14C exposure ages contradict evidence from lower elevations of Mt Murphy that indicate early to mid-Holocene deglaciation from 9–6 ka (Adams et al., 2022; Balco et al., 2023; Johnson et al., 2020). In situ 14C ages from Notebook Cliffs could be reconciled with the currently accepted Mt Murphy deglaciation history if a localised ice dome had persisted atop Notebook Cliffs, shielding samples until the late Holocene. The flat top of the Notebook Cliffs site would favour persistence of a post-glacial ice dome; however, there is no physical evidence for this having occurred (Johnson et al., 2020)

Except for late Holocene ice cover of samples CIN-108 and CIN-112 at the lower scoria cone outcrop, evidence for localised geological and topographical drivers of repeated burial and exposure of samples at Mt Murphy are lacking. Instead, differing exposure and transport histories of erratics prior to deposition might explain the concordant and discordant paired 14C–10Be ages observed at the same elevation at Mt Murphy. A similar mechanism, whereby some erratics are initially exposed at higher elevation (and thus subjected to a higher nuclide production rate) has been used to explain the presence of concordant older 14C–10Be ages and discordant young 14C ages at comparable elevations at another site, Shark Fin Nunatak, adjacent to Tucker Glacier in the Ross Sea Embayment (Balco et al., 2019). This hypothesis is supported by both extensive weathering of older samples at Shark Fin Nunatak, and the presence of cliffs of the same lithology upstream, from which erratics could have originated (Balco et al., 2019). However, exposed outcrops with lithologies matching the aplite, granite, and gneiss lithologies of erratics observed at Mt Murphy and surrounding nunataks (Adams et al., 2022; Johnson et al., 2020) are absent in the near vicinity. The nearest outcrop upstream, Mt Takahe (Figs. S6 and S7a in the Supplement) ∼100km to the south, is composed of mafic extrusive rock (Ohio State Polar Rock Repository; https://prr.osu.edu/collection/, last access: 15 February 2024). Such prior exposure and transport of erratics from nunataks with the same lithologies as those found at Mt Murphy would necessitate dramatic past ice flow re-organization, for which there is no evidence (see Supplement S1 and Figs. S6–S8 in the Supplement).

To investigate potential sources of sample preparation uncertainty, we focus on evaluating in situ 14C measurement reproducibility using both our new data from Mt Murphy and existing datasets from the ICE-D database for samples where two or more measurements had been made, excluding measurements of laboratory intercomparison materials such as CRONUS-A. In addition to the Mt Murphy results, a further 25 samples with repeat in situ 14C measurements are available in ICE-D, and we present a further two samples with replicate in situ 14C measurements from the Leymon High Core (Lupker et al., 2015) bringing the total number of samples with replicate in situ 14C measurements to 31. The majority of replicate measurements are reported from samples sourced from the Antarctic Peninsula (Balco and Schaefer, 2013), Weddell Sea Embayment (Nichols et al., 2019), Promontory Point (Pleistocene Lake Bonneville), Utah (Lifton et al., 2015a), and the Northwest Highlands, Scotland (Borchers et al., 2016). To each in situ 14C concentration, we assign a measurement uncertainty of 6 % of the total concentration reported from AMS measurements for each replicate. Using a 6 % uncertainty is appropriate here because the majority of replicate measurements presently in the ICE-D database were measured at Tulane and 6 % is the published, and conventionally used, measurement uncertainty from that laboratory based on reproducibility of in situ 14C measured in CRONUS-A from 2015–2018 (Goehring et al., 2019a) cited in numerous studies (e.g., Balco et al., 2019, Nichols et al., 2019, Rand et al., 2025). In cases where in situ 14C replicates were reported from a different extraction laboratory, and that laboratory reports measurement uncertainty exceeding 6 %, we assign the larger value to each in situ 14C concentration.

From our reproducibility assessment of 31 replicate samples, 18 display one or more in situ 14C measurements that do not replicate within 6 % (1σ) measurement uncertainty (Fig. 8). There is a slight increase in reproducibility at 2σ, but 15 samples still exhibit one or more in situ 14C measurements that are not reproducible (Fig. S13 in the Supplement). These results include the 3 out of 4 in situ 14C concentrations reported from Mt Murphy which do not replicate within the 6 % (1σ) measurement uncertainty, and 2 of 4 in situ 14C concentrations do not replicate at 2σ (see Sect. 3.1). Notably, replicate measurements included in our Holocene filter analysis from the Antarctic Peninsula (Sjögren and Drygalski Glaciers, n=5; see Sect. 4.3.1 and Fig. 8) also yielded many impermissible paired 14C–10Be ratios, suggesting a possible link between in situ 14C reproducibility and Type 3 14C–10Be concentration ratios. However, the lack of an apparent geologic explanation for irreproducibility of in situ 14C measurements from field samples (see Sect. 4.3) suggests that the assumed measurement uncertainty may be too low. Such an issue has been noted previously by Borchers et al. (2016) where scatter of in situ 14C concentrations from calibration sites (including the Northwest Highlands) exceeded stated measurement uncertainties.

The findings presented in this paper suggest that routine laboratory uncertainties reported with our samples from Mt Murphy likely underestimated the true measurement uncertainty of in situ 14C for our dataset. This result is consistent with previous findings from the CRONUS-Earth Project (Borchers et al., 2016; Phillips et al., 2016a) and other studies (Hippe, 2017; Jull et al., 2015) where issues regarding the interlaboratory variation in reported CRONUS-A in situ 14C concentrations (∼15%) and in situ 14C blank variability were shown to impact the accuracy and precision of in situ 14C measurements. A seemingly isolated issue associated with the initial in situ 14C extractions likely resulted in systematically young ages inconsistent with both the replicate measurements and previously published 10Be exposure ages from Mt Murphy, as well as with records of the deglacial history of the Amundsen Sea Embayment more widely. The nature of the apparent loss of in situ 14C from most of our samples during the initial extractions is not understood and is atypical of the considerable number of in situ 14C measurements reported from Tulane CNL. Nevertheless, complexities in our dataset highlight the value of routinely conducting replicate analyses not just for in situ 14C, but for all cosmogenic nuclides, especially if a dataset displays systematic offsets that cannot be accounted for by reported uncertainties. Ongoing developments, including automation of in situ 14C extraction (Goehring et al., 2019a; Lifton et al., 2023; Lupker et al., 2019), will help facilitate analysis of the additional replicates and process blanks needed to improve the precision of in situ 14C measurements. With a focus on improving in situ 14C analytical reproducibility and precision, we therefore make the following suggestions for future work, which will ultimately contribute to the provision of robust combined 14C–10Be chronologies:

Routinely undertake and report more in situ 14C replicate measurements. This will provide a check on quality control.