Calibrating Antarctic ice sheet mass loss due to millennial-scale ocean thermal forcing

Throughout the Late Pleistocene, millennial-scale cycles in the rate of poleward heat 14 transport resulted in repeated heating and cooling of the Southern Ocean 1 . Ice sheet models 2 15 suggest that this variation in Southern Ocean temperature can force fluctuations in the mass 16 of the Antarctic ice sheet that transiently impact sea level by up to 15 meters. However, 17 current geologic evidence for Antarctic ice response to this ocean thermal forcing is unable 18 to calibrate these models, leaving large uncertainty in how Antarctica contributes to sea level 19 on millennial timescales. Here we present a >100kyr archive of East Antarctic Ice Sheet 20 response to Late Pleistocene millennial-scale climate cycles recorded by transitions from opal 21 to calcite in subglacial precipitates. 234 U- 230 Th dates for two precipitates define a time series 22 for 32 mineralogic transitions that match Antarctic climate fluctuations, with precipitation 23 of opal during cold periods and calcite during warm periods. Geochemical evidence indicates 24 opal precipitation via cryoconcentration of silica in subglacial water and calcite precipitation 25 from the admixture of meltwater flushed from the ice sheet interior. These freeze-flush cycles 26 represent changes in subglacial hydrologic-connectivity driven by ice sheet thickness 27 response Our Our results show the AIM Pleistocene, global sea level has fluctuated by >100 m 8 predominantly due to orbital forcing 5 . Within our current interglacial, sea level fluctuations of this magnitude pose no immediate threat on human timescales, but there remains a possibility that the AIS will respond to more abrupt climate events. Here we show that over the past ~230kyr, millennial-scale ocean temperature fluctuations caused periodic widespread Antarctic ice loss and regrowth. Our results reveal millennial-scale ice sheet thinning in the Ross Sea Embayment, which provides constraints on the magnitude of heat-exchange between the

proxy records indicate that variations on the order of 20m occur on millennial timescales 8 , and in some 44 cases these fluctuations match the Antarctic climate rhythm 9 , implying that the AIS may have exerted 45 considerable control on sea level within these timeframes. Modern observations of ice shelf thinning and 46 ice sheet mass loss from the margins of both the West 10,11 and East 12,13 Antarctic Ice Sheets (WAIS and 47 EAIS) suggest that the processes leading to rapid AIS sea level input may currently be underway.

48
Ice sheet simulations 2 demonstrate the potential for AIS mass fluctuations on short timescales 49 through perturbation of buttressing ice shelves 3,4 brought on by changing Southern Ocean (SO) 50 temperature 14 (hereafter referred to as ocean thermal forcing). Millennial-scale SO temperature oscillations 51 are driven by the rate of Atlantic Meridional Overturn Circulation (AMOC) 15 , which is responsible for 52 cross-equatorial heat transport from the Southern Hemisphere to the Northern Hemisphere (SH and NH).

53
Changes to the intensity of AMOC result in inverse interhemispheric temperature cycles 16   This "bipolar seesaw" results in Antarctic atmospheric temperature shifts of only 1-2˚C 17 , but feedback 61 between AMOC and the strength of the SH westerly winds 18 causes significant fluctuations in the SO 62 temperature structure by modulating the upwelling rate of warm circumpolar deep waters around the AIS 19 .

63
Regional differences in ice bed topography, drainage geometry, and thickness 6 in peripheral sectors 64 of Antarctica lead to geographic variations in grounding line vulnerability to ocean thermal forcing. For 65 example, numerical experiments indicate that major drainages of ice grounded below sea level in the 66 Weddell Sea and Prydz Bay Embayments respond sensitively to even modest ocean temperature increases, 67 while significant ice loss from the Ross Embayment and the EAIS margins require greater thermal forcing 2 68 (Extended Data Fig. 1). In models simulating AIS response to temperature changes in the Southern Ocean, 69 the majority of this sensitivity difference is expressed in the parameter describing heat exchange between 70 the ocean and the ice sheet, which represents one of the largest sources of uncertainty in projections of

71
Antarctic behavior during centennial to millennial-scale climate change 2 . Geologic evidence of millennial-72 scale AIS evolution could calibrate this heat exchange parameter by describing what regions of the ice sheet 73 lost or gained ice mass following a change in ocean temperature. However, existing records are limited to 74 the highly sensitive Weddell Sea sector 20,21 , and therefore cannot confirm a change in mass from the less 75 sensitive Antarctic sectors, which make up the majority of the AIS margin. Given this lack of geologic 76 constraint on millennial-scale AIS response to Southern Ocean temperature fluctuations, best estimates for 77 Antarctic contribution on these timescales is between 0 and 15 meters sea level equivalent 2 . Therefore, a 78 long-term, high-resolution record of AIS evolution from a less temperature-sensitive ice sheet sector, as we 79 document here, will improve our ability to constrain the broader Antarctic contribution to sea level on sub-80 orbital timescales and will help to parameterize projections of future AIS mass loss.

East Antarctic Ice Sheet Subglacial Hydrologic Response to Southern Ocean Temperature 82
In this study, we present observations from a novel archive of subglacial thermal and hydrologic 83 evolution recorded by chemical precipitates that formed beneath the EAIS. These samples are deposited in 84 aqueous systems beneath the ice sheet on the EAIS side of the Transantarctic mountains (TAM), and are 85 transported to the surface in basal ice, which erodes them from subglacial environments and deposits them 86 in supraglacial moraines. Results are presented here from two samples: PRR50489 formed in the David from 55 to 42ka (Fig. 1f). Depth profiles of Si and Ca concentration, collected using Energy Dispersive X-94 ray Spectroscopy, provide a continuous representation of sample mineralogy: with high Ca areas 95 representative of calcite and low Ca areas representative of opal. We construct depositional age models for 96 each sample by using linear interpolation between dated layers, and pair these age-models with Ca 97 concentration to create timeseries describing the oscillations of precipitate mineralogy. Mineralogic 98 timeseries reveal a temporal cyclicity in opal deposition, with PRR50489 containing opal layers precipitated 99 every 8-10kyr between marine isotope stages (MIS) 7 and 6, and MA113 containing opal layers precipitated 100 every 2-4kyr during MIS 3. To investigate a possible link between cycles of precipitate mineralogy and 101 climate, we compare mineralogic timeseries for each precipitate (Fig. 1f, i) with climate proxies in both 102 Antarctic 17,22 (Fig. 1a,b) and Greenland 23 (Fig. 1c) ice cores. Comparison between the Ca-spectra and

103
Antarctic temperature proxies shows a strong similarity, with calcite formation (high Ca wt%) during warm 104 AIM peaks, and opal formation (low Ca wt%) during Antarctic cold periods (Fig. 1). These data indicate 105 that thermal and hydrologic conditions beneath the EAIS respond to bipolar-seesaw-driven Antarctic 106 climate cycles on sub-orbital timescales.

107
To understand the link between ocean-atmosphere-cryosphere teleconnections and the composition 108 of subglacial waters, we first utilize geochemical and isotopic measurements to characterize the waters from 109 which opal and calcite originate.

Millennial-Scale Ice Sheet Thickness Changes Driven by Ocean Thermal Forcing 187
The key finding from our subglacial precipitate archive is that millennial-scale ocean-atmosphere-188 cryosphere teleconnections cause a hydrologic response beneath the EAIS where AIM warm phases drive 189 increased melting and enhanced hydrologic connectivity between the ice sheet margin and interior. the subglacial aqueous system at the ice sheet edge to become hydrologically isolated from the ice sheet 192 interior. We investigate how climate cycles may drive subglacial hydrologic change using a reduced supplementary text for full description). This model assumes that the ice thickness at the foothills of the 195 TAM is a linear function of the isotopic records of climate from either EDC (PRR50489) or WDC (MA113) 196 ice cores (Fig. 1 a, g). This temporal forcing feeds into changes of ice surface slope, basal shear stress, ice   (Fig.1d), corroborating the idea that bipolar seesaw driven SO temperature changes cause ice 223 sheet thickness variations during AIM cycles 2 . Therefore, we conclude that the freeze-flush cycles that lead 224 to opal-calcite cycles in subglacial precipitates reflect ocean-forced AIS thickness changes (Fig. 2).

225
Simulations of AIS sea level contribution during millennial-scale climate cycles suggest that SO 226 warming can lead to Antarctic input ranging from 0 to 15 m depending on SO temperature, the duration of                                               Fig. 2) were 529 produced using Energy Dispersive X-ray Spectroscopy (EDS) measured on the Thermoscientific Apreo

530
Scanning Electron Microscope (SEM) housed at UCSC. EDS data were generated using an Oxford

531
Instruments UltimMax detector and were reduced using AZtecLive software. To quantify the opal-calcite 532 transitions in the samples, Si and Ca concentration data were produced from line scans across precipitate 533 layers (Extended Data Fig. 2). For sample MA113, detritus within two calcite layers results in Si peaks that 534 do not correspond to opal. These areas are identified using Al and are corrected to reflect a calcite 535 composition. An age versus depth model was produced for each precipitate by linear interpolation between 536 each dated layer. In sample MA113, no dates were produced for layers corresponding to the timeframe 537 between 48ka and 52ka, so the primary age versus depth model is derived by applying an accumulation rate 538 for opal and calcite based on the two accumulation rates from the rest of the sample (Fig. 1f, orange curve).

539
A secondary age versus depth model was also produced between 48ka and 52ka using linear interpolation 540 ( Fig. 1f, grey, dashed curve), but sparce data coverage during this timeframe likely makes this curve less 541 accurate for this period. Timeseries in figure 1f and 1i were then generated by plotting the age versus depth 542 model, against Si and Ca concentration spectra. 2σ uncertainties on each individual U-series age are 543 between 0.5% and 2% for PRR50489 and between 1% and 5% for MA113; we interpolate across the mean 544 age for each layer. Within these error bounds, we are not able to estimate leads or lags between the 545 subglacial hydrologic system and climate cycles. Nonetheless, stratigraphic consistency between dated 546 layers, the regular frequency of mean ages, and the striking similarity between our mineralogic timeseries 547 and climate proxy records supports our conclusion of a link between climate teleconnections and subglacial 548 hydrology. Furthermore, as was previously mentioned, calcite layers form rapidly upon introduction of 549 carbon-rich, alkaline waters from the EAIS interior to the marginal system, and the system then slowly 550 transitions back to opal precipitation after hydrologic connectivity is shut off and the waters freeze.

551
Therefore, it is possible that there is missing time between calcite layers that is not accounted for by our 552 linear interpolation. However, based on the regularity of opal depositional cycles, and the similarity 553 between precipitate opal-calcite cycles and climate proxies, these unconformities do not represent enough 554 time to disrupt the millennial-scale cyclicity of the precipitate mineralogy.

556
Geochemical analyses    Fig. 5b). Therefore, the addition of carbon-rich, alkaline 679 meltwaters to opal precipitating, Ca-Cl-rich brines can trigger calcite supersaturation driving rapid calcite 680 growth, consistent with our geochronologic outputs and calcite morphology. Collectively this modeling 681 effort, along with the timescale data presented in figure 1, and the fibrous crystal textures suggests that 682 calcite forms rapidly after meltwaters to the subglacial aqueous system. However, the relative volume of 683 meltwater added is unclear from these results alone and requires further isotopic constraints.

698
Calcite data, in δ 13 C vs δ 18 O space, define a trend that suggests they form through admixture with 699 an isotopically lighter water with respect to both carbon and oxygen. The carbon concentration dependent 700 mixing curve that best fits that calcite data alone requires that the isotopically light endmember, what we'll 701 refer to here as the "meltwater" endmember carries a higher DIC than the isotopically heavier water, which 702 we'll now refer to as the brine endmember. In figure 3, we assume that isotopically heaviest opals record 703 the δ 18 O composition of the brine endmember and that the δ 13 C composition matches marine carbon derived 704 from the substrate (-0 ‰), the latter of which is recorded by sodic carbonates suspected of forming from 705 brines in the Lewis cliff area 41 . Under such assumptions the carbon ratio between meltwater and brine is 706 97:3 for PRR50489 (Fig. 3a) and 80:20 in MA113 (Fig. 3b), a result that is consistent with the calcite 707 precipitation model presented above, whereby the addition of a carbon rich, oxidized meltwater, to a 708 reduced or intermediate CaCl2 brine, triggers calcite precipitation. As shown in extended data figure 6, the 709 calcite data imply formation when there is >30% of meltwater in the mixture. The array of calcite data can 710 also be fit by a mixing model that assumes an isotopically lighter carbon composition (δ 13 C = -15 ‰). While 711 feasible, this is a less appealing solution as a δ 13 C of -15 ‰ does not match the composition of any specific 712 carbon source and would require a mixture of waters. The data presented here suggests that over the ~100ka 713 of sample precipitation there are two consistent endmember waters: a reduced brine that is locally derived 714 (star in second quadrant, Fig. 3), and an oxidized meltwater that is from the polar plateau (star in third 715 quadrant, Fig. 3).

717
Ocean-ice heat exchange and sea level input calculations

718
Our reduced-complexity model of ice sheet thermodynamics requires that ice thickness changes reach an 719 amplitude of a few hundred meters on AIM timescales in the two parts of Ross Embayment proximal to 720 our sample locations. While thickness change adjacent to the David Glacier catchment occurs at moderate 721 levels of ocean-ice heat exchange (> 5 m a -1 K 1 ; Extended Data Fig. 1c), thickness change in the vicinity of