Enhanced Moisture Delivery into Victoria Land, East Antarctica During the Early Last Interglacial: Implications for West Antarctic Ice Sheet Stability

The S27 ice core, drilled in the Allan Hills Blue Ice Area of East Antarctica, is located in Southern Victoria Land ~80 km away from the present-day northern edge of the Ross Ice Shelf. Here, we utilize the reconstructed accumulation rate of S27 covering the Last Interglacial (LIG) period between 129 and 116 thousand years before present (ka) to infer moisture transport into the region. The accumulation rate is based on the ice age-gas age differences calculated from the ice 15 chronology, which is constrained by the stable water isotopes of the ice, and an improved gas chronology based on measurements of oxygen isotopes of O2 in the trapped gases. The peak accumulation rate in S27 occurred at 128.2 ka, near the peak LIG warming in Antarctica. Even the most conservative estimate yields a six-fold increase in the accumulation rate in the LIG, whereas other Antarctic ice cores are typically characterized by a glacial-interglacial difference of a factor of two to three. While part of the increase in S27 accumulation rates must originate from changes in the large-scale atmospheric 20 circulation, additional mechanisms are needed to explain the large changes. We hypothesize that the exceptionally high snow accumulation recorded in S27 reflects open-ocean conditions in the Ross Sea, created by reduced sea ice extent and increased polynya size, and perhaps by a southward retreat of the Ross Ice Shelf relative to its present-day position near the onset of LIG. The proposed ice shelf retreat would also be compatible with a sea-level high stand around 129 ka significantly sourced from West Antarctica. The peak insS27 accumulation rates is transient, suggesting that if the Ross Ice Shelf had 25 indeed retreated during the early LIG, it would have re-advanced by 125 ka.

. One way to constrain the sensitivity of ice sheets to climate change is to explore their behavior during past warm periods. The Last Interglacial (LIG) between 129 and 116 thousand years before present (ka) is a geologically recent warm interval with average global temperature 0 to 2 °C above the pre-industrial level (Otto-Bliesner et al, 2013). The LIG could therefore shed light on the response of WAIS to future warming. While the WAIS must have contributed to the LIG sea-level high stand (Dutton et al, 2015a and references therein), quantifying these contributions is challenging and the 35 timing of such WAIS changes (early versus late in LIG) is still debated (e.g. Yau et al, 2016;Rohling et al, 2019;Clark et al, 2020).
As the floating extension of land ice masses, the extent of sea ice and ice shelves can provide important insights into the dynamics of continental ice sheets. For example, as the ocean warms and sea level rises, the loss of ice shelves due to calving and basal melting may lead to further losses of the continental ice they buttress (Pritchard et al, 2012). The Ross Ice 40 Shelf (RIS) is the largest ice shelf in Antarctica, located between the Marie Byrd Land in West Antarctica and the Victoria Land in East Antarctica ( Figure 1). Ice sheet models have suggested that the complete disintegration of the Ross Ice Shelf may have accompanied the collapse of WAIS (DeConto and Pollard, 2016). However, terrestrial evidence is lacking due to subsequent ice sheet growth, and existing marine records do not have enough temporal resolution to resolve the extent of RIS during the LIG. 45 Ice cores provide continuous, well-dated records of local climate information that is sensitive to the extent of nearby ice 50 masses. The position of the ice margin and sea ice extent can impact atmospheric circulation, snow deposition, and isotopic signatures in the precipitation captured in ice cores (Morse et al, 1998;Steig et al, 2015;Holloway et al, 2016). In this study, we use a shallow ice core, Site 27 (S27) from the Allan Hills Blue Ice Area (BIA), to explore RIS changes during the LIG.
The Allan Hills BIA in Victoria Land, East Antarctica, is ideally located near the present-day northwest margin of the RIS ( Figure 1). S27 provides a continuous climate record between 115 and 255 ka (Spaulding et al, 2013). The close proximity of 55 Site 27 to the Ross Sea embayment holds the potential to shed light on the behavior of the Ross Ice Shelf during Termination II (the transition from the Penultimate Glacial Maximum to the LIG) and, by extension, on that of the West Antarctic Ice Sheet.
Here, we present a record of the accumulation rate of Site 27 derived from independently constrained ice and gas chronologies. This approach has previously been applied to Taylor Glacier blue ice samples to estimate accumulation rates 60 (Baggenstos et al, 2018;Menking et al, 2019). We take advantage of the fact that the age of the ice is older than the age of the trapped gases at the same depth. This age difference (Δage) results from the process of converting snow into ice (firn densification) and reflects the age of the ice when the firn crosses a threshold density where the gases become isolated in impermeable ice. The evolution of firn density is found to empirically correlate with the ice accumulation rate and surface temperature (Herron and Langway 1980). Subsequent ice thinning and flow do not alter this Δage. 65 The ice chronology of S27 was originally established by matching features in the stable water isotopes (δDice) to those in the EPICA Dome C (EDC) record ( Figure 2; Spaulding et al, 2013). The δ notation here is defined as [(Rsample/Rstandard) − 1] × 1000 ‰, where R is the raw ratio. Similarities between S27 and EDC δDice overall give us confidence in the stratigraphic continuity of the S27 ice core. By contrast, a preliminary gas age timescale is available in Spaulding et al (2013), constructed by matching the δ 18 O of atmospheric O2 (δ 18 Oatm) measured in S27 (sample N = 39) to the δ 18 Oatm record of the Vostok ice 70 core ( Figure S2). This preliminary δ 18 Oatm record, however, did not capture a δ 18 Oatm peak between 170 and 190 ka, hinting that the S27 ice core might not be continuous after all, though δDice in this interval suggests otherwise.
In this study, we extend the existing S27 δ 18 Oatm measurements by adding new δ 18 Oatm values at 45 depths, including one that overlaps with earlier data. This collated δ 18 Oatm record is then correlated with a recently published δ 18 Oatm record of EDC (Extier et al, 2018) to derive a more accurate and complete gas chronology for S27. New measurements of CH4 and CO2 75 from the S27 ice core are also used to further improve the δ 18 Oatm-based age scale. The gas chronology developed here, together with the ice chronology reported in Spaulding et al (2013), yields the Δage, from which the accumulation rate at Site 27 is estimated. We then proceed with interpreting the accumulation rate history in the context of atmospheric circulation and ice shelf/ice sheet stability.  (Figure 1 and Figure S1). MIF ice flows slowly (<0.5 m yr -1 ) from the southwest to the northeast and feeds into the Mawson Glacier before eventually draining into the Ross Sea embayment.
Horizontal ice velocities decrease as the ice approaches the Allan Hills nunatak, from 0.4-0.5 m yr -1 in the upstream portion 90 of the MIF to less than 0.3 m yr -1 near where S27 is located, with the slowest ice flow rate in the area being 0.015 m yr -1 (Spaulding et al, 2012).
The accumulation area of the ice feeding the Allan Hills BIA today lies about 20 km upstream (Kehrl et al, 2018). An accumulation rate of 0.0075 m yr -1 for the past ~660 years is inferred from a shallow firn core drilled near the Allan Hills BIA (Dadic et al, 2015). We regard this value as the present-day accumulation rate for the region where the blue ice at Allan 95 Hills today was originally deposited. Note the accumulation rate of a blue ice record characterizes its original deposition site and is different from the surface mass balance within the blue ice field. Allan Hills BIA in particular is characterized by an ablation rate of 0.02 m yr -1 (Spaulding et al, 2012). This negative mass balance leads to the exhumation of ice older than 100 ka at the surface (Spaulding et al, 2013). S27 δ 18 Oatm samples measured in this work share the analytical procedures described in Dreyfus et al (2007) and Emerson et al (1995) with several modifications. In brief, roughly 20 g of ice was cut from the core and the outer 2-3 mm trimmed away.
The ice was then melted under vacuum to release the trapped air, and the released gases were allowed to equilibrate with the meltwater for four hours (Emerson et al, 1995). After equilibration, the majority of the meltwater was discarded, and the remaining water refrozen at -30 °C. The headspace gases were subsequently collected cryogenically at 4K in a stainless-steel 105 dip tube submerged in liquid helium. During the transfer to the dip tube, H2O and CO2 were removed by two traps in series, the first kept at -100 °C and the second placed inside a liquid nitrogen cold bath.
After gas extraction, the dip tube was warmed up to room temperature and attached to an isotope-ratio mass spectrometer (Thermo Finnigan Delta Plus XP) for elemental and isotopic analysis. δ 15 N of N2, δ 18 O of O2, and δO2/N2 were measured simultaneously. All raw ratios were corrected for pressure imbalance (Sowers et al, 1989). Pressure-corrected δ 15 N and δ 18 O 110 were further corrected for the elemental composition of the O2-N2 mixture (Sowers et al, 1989). Next, δO2/N2 and δ 18 O were normalized to the modern atmosphere and corrected for gravitational fractionation that enriches the heavy molecules in the ice using δ 15 N (Craig et al, 1988). The gravitationally corrected δ 18 O is reported as δ 18 Ograv. δ 18 Ograv is frequently equal to δ 18 Oatm, the δ 18 O of paleo-atmospheric O2. In the case of this study, however, it is necessary to make an additional correction for post-coring gas losses using δO2/N2. Gas losses would lower δO2/N2 and elevate the δ 18 O 115 of O2 trapped in ice and can occur in ice cores stored at or above -50 °C for an extended period of time Suwa and Bender, 2008). δ 18 O of O2 would also be elevated in ice that is extensively fractured (Severinghaus et al, 2009). In S27, δO2/N2 values measured five years apart clearly display the impact of gas losses, both in fractured and non-fractured ice ( Figure S3).
In order to quantitatively correct for gas loss fractionations, we made the following assumptions: (1) δ 18 Oatm samples 120 measured in Spaulding et al (2013) have no gas loss and their δO2/N2 represents the true in situ value; (2) the systematic difference between the δO2/N2 values of the new samples measured in this study and those measured five years earlier is solely due to gas loss; and (3) in both fractured and non-fractured ice, gas loss (registered in the δO2/N2 values) affects δ 18 Oatm in the same proportion, despite the variability of the loss.
Gas loss correction for S27 δ 18 Oatm is given by: 125 where b is the slope of the regression line between the δ 18 Ograv replicate difference versus the δO2/N2 replicate difference observed in new samples measured in this study ( Figure S4). Because all but one of the new S27 samples were measured on depths different from the earlier samples, ΔδO2/N2 cannot be directly computed. We regressed the δO2/N2 values against https://doi.org/10.5194/cp-2021-7 Preprint. Discussion started: 10 February 2021 c Author(s) 2021. CC BY 4.0 License. depth for the new and earlier datasets ( Figure S3). ΔδO2/N2 in Equation (1) was then calculated from the predicted δO2/N2 130 values at the same depth from the two regression lines. The absolute magnitude of this gas loss is on the order of 0.020 ‰.

CO2 and CH4
S27 CH4 was analyzed using a melt refreeze method described by Mitchell et al (2013). In short, ~60-70 g of ice was cut, trimmed, melted under vacuum, and refrozen at about -70 °C. CH4 concentrations in released air were measured with gas chromatography and referenced to air standards calibrated by NOAA GMD on the NOAA04 scale. Precision is generally better than ±4 ppb. CO2 concentrations were measured using a dry extraction (crushing) method described by Ahn et al 140 (2009). 8-15 g samples were crushed under vacuum and the sample air was condensed in a stainless-steel tube at 11 K. S27 CO2 concentrations were measured after equilibration to room temperature using gas chromatography, referenced to air standards calibrated by NOAA GMD on the WMO scale. Samples with cracks and fractures are sensitive to contamination of greenhouse gases in the ambient air. No CO2 and CH4 data from S27 are available below 151 m in S27 due to extensive cracks. CO2 and CH4 samples above this interval were also excluded for age synchronization purposes if fractures were 145 found present. Whenever possible, samples were processed and analyzed in replicate for each depth and results averaged to obtain final CH4 or CO2 concentrations. Only CO2 and CH4 samples with two or more replicates are reported in Supplementary Data Table 2 and used in this study.

Gas age synchronization
We used the EPICA Dome C (EDC) ice core to synchronize the gas records for S27 because EDC has a more recent, higher 150 resolution δ 18 Oatm record available (Extier et al, 2018). We note that the ice chronology of S27 is also based upon EDC (Spaulding et al, 2013). In addition, many Vostok δ 18 Oatm measurements reported by Suwa and Bender (2008) were made on samples stored at -20 °C. They showed appreciable gas losses compared to Vostok samples analyzed during earlier times.
The EDC δ 18 Oatm record, some of which were obtained from samples stored at -50 °C, should be much less affected by gas loss than the Vostok samples are. 155 Figure 3 is the logic diagram outlining the gas age synchronization processes consisting of three steps: (1) match the extrema (either a peak or a trough) in the S27 δ 18 Oatm records to the extrema in EDC δ 18 Oatm records ("peak match"); (2) match the absolute values of the remaining S27 δ 18 Oatm to those of δ 18 Oatm in EDC ("direct match"); and (3) if no match is available from either step (1) or (2), assign age by linearly interpolating the ages of their adjacent δ 18 Oatm points. https://doi.org/10.5194/cp-2021-7 Preprint. Discussion started: 10 February 2021 c Author(s) 2021. CC BY 4.0 License.

Figure 3: Schematic workflow of timescale synchronization by δ 18 Oatm (solid lines) and uncertainty estimates (dashed lines).
Rectangles refer to data, circles include conditional statements, and triangles stand for mathematical operations. Arrows mark the workflow. Boxes in bold lines indicate that we have arrived at the final answer. Definition of ζ(t) is given by Equation (2)  In the first step, an extreme is defined when the δ 18 Oatm sample is higher ("peak") or lower ("trough") than the two adjacent 165 δ 18 Oatm samples. The advantage of this approach is that it relies on the prominent features within the δ 18 Oatm records and is not very sensitive to the systematic offset (if any) between the records. Out of 83 δ 18 Oatm samples from S27, 29 (35 %) were identified as peaks/troughs and matched to corresponding features in the EDC δ 18 Oatm record.
However, not all points are at peaks or troughs. To maximally utilize the rest of the data, we proceed with step (2) and constructed ζ, a function of time t, defined below: 170 (2) Note δ 18 Oatm(t), EDC here is linearly interpolated between individual δ 18 Oatm points reported in Extier et al (2018). We seek the age t that satisfies ζ(t) = 0, in which case a "direct match" is deemed successful and a corresponding EDC age assigned to the S27 sample. 51 samples (61 %) have their ages assigned this way.
Finally, if a δ 18 Oatm(t), S27 point is neither at a peak or trough nor successfully matched to EDC δ 18 Oatm, the age of this data 175 point is constrained by the ages of its adjacent δ 18 Oatm points, as in step (3). Only three points (4 %) fall into the final category. The age assignment method and result of each S27 δ 18 O datum are listed in Supplementary Data Table 3, along with their uncertainties. The final reported uncertainties associated with the gas chronology consists of three parts: the analytical uncertainties of δ 18 Oatm, the relative uncertainties of S27 chronology to EDC chronology, and the inherent uncertainties of the EDC chronology itself. Readers are referred to the Supplement for a more detailed discussion. 180

Firn densification inverse modeling
Firn densification models typically use accumulation rate and mean annual surface temperature to calculate Δage (see Lundin et al, 2017 for a more in-depth review). In our case, however, we seek to do the opposite, using the temperatures inferred from the isotopic composition of the ice (δDice), and the Δage to estimate accumulation rates for S27. Δage is calculated by subtracting the gas age (obtained according to Section 2.4) from the ice age [from Spaulding et al (2013)] of 185 the same depth.
Here, we employ an empirical firn densification model from Herron and Langway (1980), abbreviated as H-L hereafter. H-L has the merits of transparency and simplicity, as only three properties are involved. In any case, densification models are trained against similar data and simulate similar relations between temperature, accumulation rate, close-off depth, and close-off age (Δage). A limitation of empirical firn densification models is that their range of calibration may not include 190 low-accumulation sites. To evaluate the performance of the H-L model, we compare the model output with present-day accumulation rate in the vicinity of S27 from Dadic et al (2015). The H-L model divides firn densification into two stages. In the first stage (firn density < 550 kg m -3 ), the densification process is independent of accumulation rate and is a function of surface temperature. At the threshold density of 550 kg m -3 , the elapsed time since snow deposition on the surface, t0.55, is given by: 195 k0 is a temperature-dependent rate constant [k0 = 11*e (-10160/8.314/T) ], in which T equals temperature, in Kelvin, to be inferred from δDice), A is the accumulation rate (m yr -1 ), ρi is the density of ice (917 kg m -3 ), and ρ0 is the density of surface snow, which we assume to be 360 kg m -3 (Herron and Langway, 1980). In the second stage, t, the total elapsed time since snow deposition, is calculated from firn density (ρ) using the following relationship: 200 (4) where k1 is another temperature-dependent rate constant [k1 = 575*e (-21400/8.314/T) ], and ρ is the firn density at this depth (Herron and Langway 1980). A key step in the firn densification process is bubble close-off, at which point gases can no longer diffuse within the firn and become "locked" between ice grains. For S27, bubble close-off is assumed to occur when the firn density reaches 830 kg cm -3 . At this density, t in Equation (4) equals Δage. 205 An additional parameter needed to solve accumulation rate A from Equation (4) is the site temperature, T. In order to derive historic Site 27 temperatures, we use δDice reported in Spaulding et al (2013) and a regional isotope-temperature sensitivity of 4.0 ‰ °C -1 established at the nearby Taylor Dome (Steig et al, 2000). We acknowledge that this isotope-temperature relationship could change over time, but it is not well-constrained in Southern Victoria Land (Steig et al, 2000).
Nevertheless, increasing the isotope-temperature sensitivity by 50 % to 6.0 ‰ °C -1 reduces the accumulation rate estimates 210 by no more than 20 %. Main conclusions of this paper would remain unchanged. Modern-day Allan Hills surface δDice of -257 ‰ (Dadic et al, 2015) and mean annual temperature of -30 °C (Delisle and Sievers, 1991) are used to calculate past temperatures.
Finally, we ran a Monte-Carlo simulation for each single Δage datum with 100,000 iterations to derive the distribution of accumulation rate estimates given the Δage uncertainties (See Supplement for its derivations). The reported accumulation 215 rate comes from the value with the highest number of occurrences (the mode) and its 95 % confidence interval is bracketed by the values at 2.5 th -and 97.5 th -percentile, respectively ( Figure S5).
The H-L model also produces estimates of the depth at which firn density crosses the bubble close-off threshold. The interval from the close-off depth to the surface contains three components: the lock-in zone where ice layers are impenetrable and vertical transport is inhibited (hLIZ); the height of the diffusive column where the gravitational separation of heavy isotopes 220 occurs (hdiff); and the height of the convective zone where vigorous mixing by convective air motions prevents the establishment of gravitational profiles (hconv; Severinghaus et al, 2010).
We compare the δ 15 N profile predicted by the H-L model to the measured values, assuming the hLIZ and hconv to be 3 m and 0 m, respectively. The following barometric equation is used to link the diffusive column height to the δ 15 N enrichment (Sowers et al, 1989): 225 where ∆m is the difference between the molecular weight of 15 N 14 N and 14 N 14 N (0.001 kg mol -1 ), g is the gravitational acceleration constant (9.8 m s -2 ), R is the ideal gas constant (8.314 J mol -1 K -1 ), and T is the temperature (in Kelvin). Figure 4 shows the result of synchronization between the S27 and EDC via δ 18 Oatm. Each of the δ 18 Oatm minima and maxima associated with orbital-scale insolation variations between 105 and 245 ka is successfully identified in S27, including the δ 18 Oatm peak around 180 ka that was previously missing in Spaulding et al (2013). Overall, the strong similarities between the two δ 18 Oatm series give confidence to the stratigraphic integrity of the S27 gas record. The match between the S27 and EDC δ 18 Oatm is particularly tight between 105 and 145 ka, which corresponds to the depth interval of 7 m to 145 m in S27. 235 δ 18 Oatm samples older than 145 ka show more offsets between S27 and EDC. This is noticeably evidenced by the scattering of S27 δ 18 Oatm data around the EDC curve between 202 and 210 ka. One possible explanation for the increased scatter is a decline in core quality in S27, where ice below ~145 m is heavily fractured and visibly characterized by uneven surface cracks. This explanation is supported by more variable S27 δO2/N2 values below 150 m ( Figure S3), suggesting a critical point between 145 and 150 m, below which depth data reproducibility deteriorates substantially. This variability would be 240 accompanied by more noise in the δ 18 Oatm record despite the corrections for gas loss in the deeper part of the ice core. CH4 (sample N = 12) and CO2 (N = 17) measured at S27 are plotted on the δ 18 Oatm-derived timescale in Figure 5. Also shown for comparison are an EDC CH4 (Loulergue et al, 2008) record and a composite high-resolution CO2 record built upon multiple Antarctic ice cores (Bereiter et al, 2015 and references therein). Below we describe the process of fine-tuning the δ 18 Oatm-derived gas chronology to better match greenhouse gas measurements in the reference records. We also tabulate 245 chosen tie-points between the S27 and EDC CH4, as well as S27 and the composite CO2, in Supplementary Data Table 2.

A new gas chronology for S27 230
The timescale adjustment below only applies to the interval between 115.7 and 147.2 ka: at 115.7 ka, both S27 CO2 and CH4 agree well with the co-eval values observed in the reference records, and S27 and EDC δ 18 Oatm values were both extrema at 147.2 ka. They are selected as "anchor points" that do not involve any adjustment.

Figure 4: δ 18 Oatm measured in S27 (red) was matched to a high-resolution δ 18 Oatm record from EDC (black) between 100 and 250 ka by Extier et al (2018): (a) the whole record between 114 and 255 ka and (b) a close-up view between 123 and 133 ka. S27 δ 18
Oatm data include those reported in Spaulding et al (2013) and additional δ 18 Oatm samples measured in this work. Error bars represent 95% confidence interval of the combined EDC and S27 δ 18 Oatm measurements, following the approaches described in the Supplement. The most prominent feature in Figure 5 is the greenhouse gas peak at ~128 ka. There is a 2-ppm offset in this CO2 peak observed in the S27 record compared to the composite record at this peak differ from by 2 ppm (within the analytical 260 uncertainty). There is an offset of only 163 years between the ages of the CO2 peaks recorded at S27 and in the composite record. We therefore tied the CO2 peak at 128.6 ka in S27 to the peak at 128.5 ka in reference time-series. In the ice below, both CH4 and CO2 in S27 show a clear increasing trend with time going upward towards the maximum, corresponding to the deglacial rise of greenhouse gases. We tied the S27 CH4 data point at 144.2 ka with the EDC CH4 point at 144.8 ka. We acknowledge that the low sampling resolution of greenhouse gases below 140 m precludes more rigorous evaluation of the 265 δ 18 Oatm-derived gas chronology.
Ages of the data points in between the anchor and tie points were re-sampled by linear interpolation. Uncertainties of the gas chronology are assumed to be unaffected by this fine-tuning. The new, complete gas chronology for S27 is presented in Supplementary Data Table 3. We emphasize the effect of this fine-tuning on the gas chronology is at most 600 years (at 144.2 ka), and in many cases much smaller. Even with the timescale solely derived from δ 18 Oatm, the conclusions of this 270 study remain the same.

Ice age-gas age difference (Δage)
Below we evaluate Δage calculated by subtracting gas age from ice age. Gas age comes from the δ 18 Oatm-derived, CH4-and CO2-adjusted gas chronology from this work. Ice age comes from the δDice-based ice chronology established in Spaulding et al (2013). All chronologies discussed here have been converted to AICC2012, the most up-to-date Antarctic ice core 275 timescale Bazin et al, 2013).
The relationship between the depth and the ice and gas age in S27 is shown in Figure 6. The ice age is younger than the gas age between 192 and 204 m. This result is glaciologically impossible given the presence of a diffusive column and the measured positive δ 15 N values ( Figure S6). Such discrepancies could arise from the ambiguous matching of δDice, severe impact of gas losses on δ 18 Oatm, or both. 280 The interval between 115 and 140 ka in S27 is where the gas and ice age scales are both well-constrained (Figure 2), corresponding to a depth range of 10.05 and 134.55 m. Ice in this section is not affected by visible fractures and cracks, and we therefore limit our subsequent discussion of Δage to the interval between 115 and 140 ka. Here, δDice values represent deglacial warming, and the cooling after the LIG, allowing unambiguous feature matching (e.g. the distinct MIS5e peak around 128.2 ka). Δage of S27 between 115 and 140 ka is plotted along with the Δage estimates in Talos Dome, EDC, and 285 Vostok (Figure 7). Apart from the similarities in the shape of the Δage curve across Termination II between the four records, a prominent feature here is the very low Δage of S27 during the LIG, reaching its minimum value of 145 years (95 % CI: 27-300 years) at 128.2 ka.    Beginning at 132.2 ka, S27 accumulation rate increased by an order of magnitude within 4,000 years and reached its 305 maximum value at 0.092 m yr -1 (95 % CI: 0.056~1.18 m yr -1 ) at 128.2 ka. The peak in S27 accumulation rates coincides with ~128 ka peak warming in Antarctica, as well as with the maximum accumulation rate recorded in three other East Antarctic ice cores. We acknowledge the large uncertainty here, as high accumulation rate estimates are associated with a very small Δage values and hence large relative errors. That said, this particular small Δage is a robust estimate, because the precise match between δDice peaks around ~128 ka puts a firm constraint on ice age (Figure 2), and the monotonic deglacial δ 18 Oatm 310 change means small gas age uncertainty (Figure 4b). In addition, this estimate agrees with the peak LIG accumulation rate at the nearby Taylor Dome deduced from 10 Be activities in the ice (0.074 m yr -1 ; Steig et al, 2000). Importantly, even the most conservative accumulation rate estimate of 0.056 m yr -1 (the lower bound of the 95 % CI) means a six-fold increase in the LIG S27 accumulation rate. 315 Figure 8: Accumulation rate between 115 and 140 ka in S27 (red), Talos Dome (TALDICE; blue), Vostok (brown), and EDC (black). Note the y-axis is plotted on logarithm scales. Accumulation rates of EDC, TALDICE, and Vostok are from Bazin et al (2013) and the references therein. Error bars represent the 95% confidence interval for S27 accumulation rate estimates. The dashed line in red represents the present-day accumulation rate (0.0075 m yr -1 ) in the vicinity of Allan Hills (Dadic et al, 2015).

Accumulation rates
The elevated snow accumulation during the LIG at S27 was a transient phenomenon as by 125.5 ka, accumulation rates had 320 already dropped below 0.02 m yr -1 , and further declined to a baseline value of less than 0.01 m yr -1 after 120 ka. Except for the 7,000-year interval between 132 and 125 ka, accumulation rates at S27 are comparable to accumulation rates estimated for ice cores drilled further inland such as Vostok and EDC located in the East Antarctica Plateau during glacial periods (<0.02 m yr -1 ). In modern settings, sites with low accumulation rates are often characterized by a deep convective layer in the https://doi.org/10.5194/cp-2021-7 Preprint. Discussion started: 10 February 2021 c Author(s) 2021. CC BY 4.0 License. firn column and by δ 15 N values lower than values predicted by firn densification models under the assumption of 325 gravitational fractionation (Severinghaus et al, 2010). This observation can explain H-L model estimates for δ 15 N values at S27 which are systematically higher than observations, consistent with the presence of a convective column ( Figure S6). We note, however, that the occurrence of deep convection in the firn column does not impact accumulation rate estimates from Δage because we are not relying on δ 15 N values to reconstruct lock-in depths.

Discussion 330
Today, moisture transport into Allan Hills vicinity is primarily in the form of synoptic-scale low-pressure weather systems (Cohen et al, 2013) modulated by the position and intensity of the Amundsen Sea Low and the austral westerlies (Bertler et al, 2004;Patterson et al, 2005). In this context, one way to interpret the pronounced increase in S27 accumulation rates during Termination II is a transient reorganization of large-scale atmospheric circulation due to the poleward shift in the westerly wind belt associated with the deglacial warming. This mechanism is commonly invoked to explain the CO2 rise 335 during ice age terminations (Toggweiler et al, 2006). Atmospheric CO2 during Termination II began to increase around 140 ka and peaked around 128.5 ka, coinciding with the accumulation rate peak in S27 (Figure 9). The contraction of the westerlies would push the storm tracks further into the Antarctic continent. The results would be increased precipitation at otherwise low-accumulation sites, and the peak in accumulation concomitant with the maximum atmospheric CO2.
In addition to large-scale circulation shifts, local changes to climate boundary conditions must also be at work for two 340 reasons. First, peak accumulation rate at S27 during the LIG (0.092 m yr -1 ; 95 % CI: 0.056~1.18 m yr -1 ) is an order of magnitude larger than the average S27 accumulation rate between 140 and 133 ka. This difference is at least three times larger than the doubling or tripling in accumulation rates recorded in other Antarctic ice cores ( Figure 8). Second, the S27 accumulation rate started to increase at 132 ka and apparently lagged the Antarctic warming and CO2 increase (Figure 9).
What other factors could affect the moisture delivery into Allan Hills vicinity? We call attention to the similar magnitude 345 and timing of the LIG accumulation rate maximum in S27 and Talos Dome (Figure 8). Talos Dome is situated in the Ross Sea sector in Northern Victoria Land ( Figure 1) and regarded as a "coastal site" due to its proximity to the open ocean, characterized with high accumulation rate (~0.09 m yr -1 ; Bazin et al, 2013) and small Δage today (~500 yr; Buizert et al, 2015). We thus hypothesize that the peak S27 accumulation rate at 128 ka may arise from S27's transition into a coastal site analogous to the present-day Talos  interpreted as a minimum extent of sea ice and therefore the proximity of Mt Moulton ice field to an open ocean at 128 ka 355 (Korotkikh et al, 2011). A spike in sea salt concentration similar to that in the Mt Moulton ice record should also be visible in the S27 record if this hypothesis is correct. We note, however, that no aerosol record from S27 is available at present.
Moreover, Holloway et al (2016) demonstrate that the retreat of winter sea ice in the Southern Ocean is fully capable of explaining the distinctive 128 ka δ 18 O isotope peak observed in Antarctic ice cores, although it should be noted that the inferred sea ice retreat in the Ross Sea is minimal in Holloway et al (2016). 360 Figure 9: Paleoclimate records during Termination II and the Last Interglacial. (a) S27 accumulation rate (this study) deduced from ice without any visible fractures. (b) δ 15 N of N2 in S27 (this study). (c) S27 stable water isotope (δDice) original (light pink) and smoothed (pink) records (Spaulding et al, 2013). (d) EPICA Dome C δDice records . (e) Atmospheric CO2 (Bereiter et al, 2015 and references therein). (f) Greenland surface temperature change inferred from ice core δ 18 Oice relative to the past millennial average, with the 365 dashed lines marking the standard error range (NEEM community members, 2013). (g) Relative sea level inferred from δ 18 O of Red Sea planktonic foraminifera with 95% confidence interval marked by the dashed lines (Rohling et al, 2019). Shaded zone marks the time interval when S27 accumulation rate is greater than 0.02 m yr -1 . The black bar marks Heinrich event 11 (H11). The second hypothesis concerns the Ross Ice Shelf (RIS). Since the Last Glacial Maximum (LGM) the RIS has shrunk in size (Ship et al, 1999;Yokoyama et al, 2016), and recent work on glacial deposits in the southern Transantarctic Mountains 370 reveals rapid grounding line retreat in the central and western Ross Sea in the early Holocene (Spector et al, 2017). Morse et al (1998) first proposed that the elevated topography and the expansion of RIS during the LGM drove the storms heading towards Victoria Land northward, supported by later studies such as Aarons et al (2016). If the RIS has been capable of exerting influence on the synoptic weather systems over the last 13,000 years, the same underlying mechanism could also be operating in the LIG. That is, a further retreat of RIS led to the southward displacements of storm tracks and the 375 enhancement of moisture transport into Site 27's accumulation region. Indeed, surface airflow into Victoria Land via the Ross Sea is enhanced in some numerical simulations where WAIS and its adjacent ice shelves are removed (Steig et al, 2015). McKay et al (2012) in addition suggest the absence of ice shelf cover in the western Ross Sea sometime in the past 250 thousand years, a scenario compatible with the LIG retreat of RIS inferred from the S27 accumulation rate record. We acknowledge that the hypothesized response of atmospheric circulation to the absence of the RIS and reduced sea ice extent 380 requires more examination by climate models. suggest that the Greenland Ice Sheet (GIS) was not responsible for the sea-level high stand at that time (Yau et al, 2016). In this case, the elevated sea level beginning ~129 ka would have to significantly result from mass losses in West Antarctica, according to a recent LIG sea-level reconstruction with high temporal resolution and precise chronological controls ( Figure  390 9; Rohling et al, 2019). This reconstruction is further reinforced by basin-wide ice losses in the Weddell Sea as early as 129 ka deduced from a blue ice record from Patriot Hills, West Antarctica (80° 18′S, 81° 21′W; Figure 1; Turney et al, 2020).
An open Ross Sea at 128 ka inferred from the S27 ice record presented in this work supports the proposed early collapse of the WAIS in the LIG, and underscores the vulnerability of Antarctic ice shelves and ice sheets to rising ocean temperatures during Termination II, possibly linked to Heinrich event 11 between 135 and 130 ka (Figure 9; Marino et al, 2015). For 395 example, a chain of events could be that meltwater discharge in the North Atlantic weakened the Atlantic Meridional Overturning Circulation and consequently reduced the northward cross-equatorial heat transport, resulting in Southern Hemisphere warming, southward shifts of the intensified westerlies, and enhanced CO2 ventilation in the Southern Ocean (Menviel et al, 2018). Even if the S27 accumulation rate increase since 132 ka was not accompanied by the collapse of the WAIS in the early LIG, our reconstruction appears at odds with some model predictions that the collapse of WAIS only 400 began at 127 ka while the RIS remained largely intact (e.g. Clark et al, 2020).
Finally, regardless the cause(s) of the LIG spike in S27 accumulation rate, it has significant glaciological implications in terms of ice flow modeling for the Allan Hills BIA, where ice older than 2 million years (Ma) has been discovered in sites disconnected from the main ice flow line (Yan et al, 2019). Recent ice-penetrating radar surveys and ice flow modeling by Kehrl et al (2018) have suggested the potential preservation of a stratigraphically continuous ice record, with 1 Ma ice 405 located 25 to 35 m above the bedrock. The modeling efforts by Kehrl et al (2018) assume no-higher-than-present accumulation rates in the past and constant sublimation rates that control the exhumation of ice along the flow line. In light of the discovery of this work, models with the time-dependent accumulation rate constrained by observations could better predict the age-depth profile. Nonetheless, S27 itself provides a continuous, readily available ice record for Termination II and the LIG with a nominal resolution of 285.7 yr m -1 in the upper 145 m with good ice core quality, making the Allan Hills 410 BIA an appealing archive for paleoclimate investigations targeting the LIG.

Conclusion
We present an improved gas chronology for a shallow blue ice record (S27) drilled in the Allan Hills Blue Ice Area, East Antarctica, located in close proximity (~80 km) to the current northwest margin of the Ross Ice Shelf (RIS). The new S27 gas chronology is derived from the δ 18 O of O2 trapped in the ice. Complementary CH4 and CO2 measurements validate and 415 refine the gas chronology, paving the way for future utilization of S27 samples. Calculation of accumulation rate on the basis of the ice age-gas age differences between 115 and 140 ka in S27 reveals a dramatic increase in accumulation rate since 132 ka and peaking at 128.2 ka, coinciding with the peak LIG Antarctic warming and atmospheric CO2.
We hypothesize that in addition to changes in the large-scale atmospheric circulation affecting precipitation on the Antarctic continent, sea ice and ice shelf extent could alter local meteorological boundary conditions and lead to the observed spike in 420 accumulation rate. A greater reduction in size of the RIS would cause the storm tracks that bring substantial precipitation to Victoria Land today to migrate farther south. The ice shelf retreat would also be compatible with a high sea stand around 129 ka sourced from the collapse of the West Antarctica Ice Sheet near the onset of Last Interglacial period (Yau et al, 2016;Rohling et al, 2019). If this was the case, an early collapse of WAIS (along with the RIS) in the LIG would underscore its vulnerability to rising temperatures. 425 Our data suggests that, soon after the conclusion of peak warming, the open ocean conducive to high accumulation rate near S27's accumulation region became once again covered by ice by 125 ka. The depositional site of S27 returned to its previous conditions characterized by low accumulation rates similar to those today. We conclude that if the Ross Ice Shelf indeed collapsed early in the LIG, it would have quickly re-advanced by 125 ka.