Holocene deglaciation and glacier readvances on the Fildes Peninsula and King George Island (Isla 25 de Mayo), South Shetland Islands, NW Antarctic Peninsula

To provide insights into glacier-climate dynamics of the South Shetland Islands (SSI), NW Antarctic Peninsula, we present a new deglaciation and readvance model for the Bellingshausen Ice Cap (BIC) on Fildes Peninsula and for King George Island/Isla 25 de Mayo (KGI) ~62°S. Deglaciation on KGI began after c. 15 cal. ka BP and had progressed to within present-day limits on the Fildes Peninsula, its largest ice-free peninsula, by c. 6.6–5.3 cal. ka BP. Probability density phase analysis of chronological data constraining Holocene glacier advances on KGI revealed up to eight 95% probability ‘gaps’ during which readvances could have occurred. These are grouped into four stages – Stage 1: a readvance and marine transgression, well-constrained by field data, between c. 7.4 and 6.6 cal. ka BP; Stage 2: four probability ‘gaps’, less well-constrained by field data, between c. 5.3 and 2.2 cal. ka BP; Stage 3: a well-constrained but restricted ‘readvance’ between c. 1.7 and 1.5 cal. ka BP; Stage 4: two further minor ‘readvances’, one less well-constrained by field data between c. 1.3 and 0.7 cal. ka BP (68% probability), and a ‘final’ well-constrained ‘readvance’ after <0.7 cal. ka BP. The Stage 1 readvance occurred as colder and more negative Southern Annular Mode (SAM)-like conditions developed, and marginally stronger/poleward shifted westerly winds led to more storms and precipitation on the SSI. Readvances after c. 5.3 cal. ka BP were possibly more frequent, driven by reducing spring/summer insolation at 62°S and negative SAM-like conditions, but weaker (equatorward shifted) Westerlies over the SSI led to reduced storminess, restricting readvances within or close to present day limits. Late Holocene readvances were anti-phased with subaquatic freshwater moss layers in lake records unaffected by glaciofluvial inputs. Retreat from ‘Neoglacial’ glacier limits and the recolonisation of lakes by subaquatic freshwater moss after 1950 CE is associated with recent warming/more positive SAM-like conditions.

Ice-marginal and ice-free coastal areas on the NAP and SSI contain an abundance of readily accessible morphostratigraphic features (e.g., raised beaches), lake sediment records, and glacially transported erratics which can be used to constrain past glacier dynamics (Davies et al., 2013). The glacial history of the Fildes Peninsula in the south-west of King George Island (KGI)/Isla 25 de Mayo, the largest of the South Shetland Islands (Figure 1), has been determined by cosmogenic isotope exposure dating of bedrock, radiocarbon and optically stimulated luminescence (OSL) dating of organic remains embedded in raised beaches and moraines of the ~ 1250 km 2 Bellingshausen Ice Cap (BIC; also known as the Collins Ice Cap), as well as radiocarbon ages from lake sediments and terrestrial stratigraphic sequences overlying coarse grained sediments and diamictons ( Figure 1) (Bentley et al., 2005;Hall, 2010;Pallàs et al., 1997;Seong et al., 2009;Simms et al., 2012Simms et al., , 2021Sugden and John, 1973;Watcham et al., 2011).
While some GIA models (e.g., Whitehouse et al., 2012a) infer changes in ice loading that are consistent with RSL reconstructions based on well-dated isolation basin records from the Fildes Peninsula (Roberts et al., 2017;Watcham et al., 2011), the coarse spatial resolution of GIA models means they are not currently able to reproduce smaller scale isostatic responses and RSL reversals implied by field data (Johnson et al., 2022;Simms et al., 2012Simms et al., , 2018Simms et al., , 2021Whitehouse et al., 2012a). It is increasingly clear that spatially variable responses of glacio-isostatic uplift are likely to have occurred across the SSI and AP during the Mid-to Late Holocene (Fretwell et al., 2010;Zurbuchen and Simms, 2019), which requires data collection at high temporal resolution across wide areas, and new methods of constraining glacier readvances if we are to better constrain these variations.
In this study, we examined the deglaciation and readvance history of the BIC outlet on the Fildes Peninsula using a combined geomorphologic, stratigraphic, chronological, palaeolimnological and statistical approach to provide new constraints on glacier readvance(s) during the Mid-to Late Holocene. New and published data were used to test the hypothesis that Holocene retreat of the BIC was discontinuous and interrupted by more than one glacial readvance during the Mid-to Late Holocene, rather than a continuous process of glacial retreat.
To achieve this, first, we obtained new minimum Beryllium-10 ( 10 Be) cosmogenic isotope exposure ages for deglaciation from three granitic erratics on the NW marine platform of the Fildes Peninsula and the BIC foreland ( Figure 1). Second, we obtained new radiocarbon ages from terrestrial and marine macrofossils that were living on the BIC foreland on the Fildes Peninsula before glacier readvances embedded them into moraines; these samples provide maximum ages of glacier advance. Third, we undertook multi-proxy analyses on a new lake sediment core extracted from Kiteschee Lake and compared these results to basal ages and multiproxy data from previously published lake records on the Fildes Peninsula to determine when deglaciation occurred in different sectors (inner, mid, outer; Figure 1) and how Holocene glacier-climate dynamics impacted on lake ecosystems.
With the addition of these new datasets, the amount of chronological data constraining Holocene glacier fluctuations on KGI were large enough to use probability density phase analysis to identify 'probability gaps' when glacier readvances could have occurred. In this paper, we identify eight occasions in the Mid-to Late Holocene when glacier readvances could have occurred and examine the mechanisms and impact of deglaciation and glacier readvance(s) on the Fildes Peninsula and across KGI.

Climate setting
The SSI archipelago is a ~230 km long, ~35 km wide active volcanic arc of 11 major islands, with the Fildes Peninsula located at ~62°S, 58-59°W and ~130 km northwest of the Antarctic Peninsula ( Figure 1b). The SSI are located south of the Polar Front (50-55°S), where the southern limb of the Southern Hemisphere westerly wind (SHW) belt converges with the polar cell ( Figure  1a). Sandwiched between the high pressure subtropic and Antarctic anticyclones, the SSI island chain has a milder and more humid (wetter) maritime climate than the Antarctic continent with up to ~1000 mm of precipitation per year (typically 300-500 mm per year; Michel et al., 2014a), which maintained a stable to slightly negative glacier mass balance on the BIC between 1970 and 1990 CE, with an equilibrium line altitude (ELA) of ~150 m a.s.l. (Bentley et al., 2009;Falk et al., 2018;Turner et al., 2002).
The annual mean air temperature on KGI between 1948 and 2011 CE was −2.5°C (Kejna et al., 2013). Mean annual, summer and winter air temperatures at the Russian Bellingshausen Station, on the Fildes Peninsula, between 1968 and 2021 CE were −2.3 ± 3.1°C, 1.1 ± 0.8°C and −5.9 ± 1.7°C, respectively (BAS READER project -Reference Antarctic Data for Environmental Research; Turner et al., 2004). Mean summer air temperatures above zero and snow-free conditions (at sea level) exist for up to four months between December and March (Michel et al., 2014b).
Intense cyclonic activity through the Drake Passage, associated with the SHW to the north, and the low-pressure circumpolar belt to the south (~66°S) means the SSI experience cloud cover for ~80% of the year (Bañón et al., 2013;King and Turner, 1997). Mean annual, summer and winter wind velocities recorded at Bellingshausen (1968 were 26.5 ± 3.5, 23.8 ± 2.2, 27.3 ± 3.2 km/h, respectively. Storms and ~100 km/h winds are common, and wind directions are predominantly SW/W/NW to NE/E/SE (34% W/NW to E/SE) and rarely along the NE-SW axis (BAS READER project;Turner et al., 2004).
A warming trend across the AP and SSI between 1950 and 1999 CE drove increased rates of glacier retreat, ice-shelf collapse (Meredith and King, 2005;Vaughan et al., 2003) and elevated the ELA on the SSI to greater than 200 m above present sea level (henceforth, m a.s.l.) (Falk et al., 2018). Similar to most other stations on the AP and SSI, temperature data from the Bellingshausen Station exhibit decadal scale variability, which has been linked to the El Niño Southern Oscillation (ENSO) (Oliva et al., 2017). Warming initiated approximately five years prior to El Niño in 1982-1983CE and 1997-1998 CE was followed by a cooling trend to 2015 CE and a return to a warming trend between 2015 and 2021 CE. Although a shift to a cooling trend was observed across the northern AP (NAP) between 1999 and 2015 CE, record summer temperatures have since been measured on the northeastern AP, and attributed to an enhanced Föhn effect (Oliva et al., 2017;Turner et al., 2016).

Geology and glaciation
King George Island (KGI), the largest of the South Shetland Islands, has been subdivided into three tectonic regions: the Fildes Block (Fildes Peninsula), Barton Horst (Barton and Weaver Peninsulas) and the Warszawa Block (Potter Peninsula) (Smellie et al., 1984). Covering an area of 38 km 2 , the Fildes Peninsula is the largest ice-free area on the South Shetland Islands. It is composed of predominantly andesitic and basaltic lava bedrock, with some interbedded terrestrial sediments, including shales and conglomerates (Smellie et al., 1984). Most of the Bellingshausen Ice Cap (BIC) foreland was glaciated during the Last Glacial Maximum (LGM) and consists of raised marine platforms up to 180 m a.s.l., formed of basic volcanic rocks of Late Cretaceous to Paleogene-age (Birkenmajer, 1989;John and Sugden, 1971;Smellie et al., 1984). The eastern coastline of the Fildes Peninsula is indented with coves, while the western coast is characterised by sheer sea cliffs and stacks (Hall, 2003).
The elevated island interiors of the SSI are characterised by numerous ice caps and permanent snowfields and the coastal fringes are some of the most extensive ice-free and permafrost areas in Antarctica (Braun et al., 2004;López-Martínez et al., 2012;Michel et al., 2014a;Ó Cofaigh et al., 2014;Figure 1b). Chlorine-36 ( 36 Cl) exposure ages from glacially striated bedrock on nearby Barton Peninsula from above ~50 m a.s.l. show that progressive ice thinning on KGI, SSI started c. 15,000 years ago (Seong et al., 2009).
The Fildes Peninsula contains a variety of glacial geomorphological and periglacial features (López-Martínez et al., 2012;Michel et al., 2014a). López-Martínez et al. (2012) mapped raised marine platforms and an active glacier in the mid-inner central Davies Heights area (Figure 1d), while Michel et al.
Radiocarbon ages obtained from marine-terrestrial transition sediments and raised beaches have been used to reconstruct past changes in local RSL at several sites across the SSI. These indicate that the thickest part of the SSI ice cap was probably centred on the now ice-free area of the Fildes Peninsula during the LGM (Watcham et al., 2011). The SSI ice cap was possibly separated from the AP ice sheet by the Bransfield Marginal Basin during the LGM, but more likely became independent during the Late glacial-Interglacial transition (c. 15-12 ka) (Fretwell et al., 2010;John and Sugden, 1971;Ó Cofaigh et al., 2014;Watcham et al., 2011).
Basal lacustrine sedimentary sequences in the mid-southern part of the Fildes Peninsula have been dated to c. 10 cal. ka BP (e.g., Jurasee Lake (Mäusbacher et al., 1989)) and were formed in over-deepened glacial basins. Several former marine embayments with basal sediment ages of up to c. 12 cal. ka BP (Watcham et al., 2011) were transformed into freshwater (isolation) basins when the rate of isostatic uplift outpaced the declining rate of sea level rise during the Early Holocene (Watcham et al., 2011). A welldefined Holocene marine high-stand of 18-15 m a.s.l. occurred on the Fildes Peninsula between c. 8 and 7 ka (Mäusbacher et al., 1989;Watcham et al., 2011).

Cosmogenic nuclide surface exposure dating of erratics
The cosmogenic isotope concentration accumulated in a superficial rock sample is proportional to the length of time exposed to cosmic rays at the Earth's surface (Balco, 2011). The cosmogenic exposure age produced depends on the production rate (i.e., the concentration of cosmogenic isotopes produced each year per gram of the relevant mineral), which varies spatially and temporally due to variations in atmospheric depth and geomagnetic field effects (Lifton, 2016). Cosmogenic 10 Be has the most wellconstrained production rate and can be measured at low concentrations (Balco, 2011), but glacially striated and quartz-rich, granitic bedrock and erratics required for 10 Be analysis are rare in the basaltic-andesitic volcanic-arc environment of the South Shetland Islands.
Large boulders >50 cm in diameter on the NW BIC glacial foreland were surveyed and classified. These boulders were deposited as glaciers retreated from offshore LGM limits. Differential GPS (dGPS; WGS84 ellipsoid) data for erratic boulders were obtained using a GPS Trimble Pathfinder ProXH. For geodetic reference, we used the landmark DALL 66019M002 (S62°14′16.335″, W58°39′52.364″, ellipsoidal height 39.376 m) at the Argentine Carlini base, ~17 km from the sampled erratics. dGPS precision is better than 10 cm in all axes, but ellipsoid correction errors are larger.
Three large granitic boulders not incorporated into solifluction planes and till sheets of the BIC foreland were considered most suitable for cosmogenic nuclide exposure dating (Table 1). These erratics showed no signs of significant erosion, and their Table 1. Cosmogenic surface exposure dating results. Exposure ages were calculated using version 3 of the online calculator described by Balco et al. (2008). Following standard procedures for the Antarctic Peninsula region, we used the Antarctic ('Ant') pressure flag and the scaling scheme 'LSDn' (Lifton et al., 2014;Stone, 2000). We employed the scaling model 'LSDn' because it derives from an improved understanding of the atmospheric particle flux and neutron spectrum (Lifton et al., 2014), as compared with other scaling schemes using empirical fits to proxy data. A rock density of 2.7 g/cm 3 was assumed for the samples. Exposure ages are derived using the default 10 Be production rate calibration dataset (Borchers et al., 2016) in the v. 3 of the online calculator of Balco et al. (2008). However, for comparison, 10 Be ages were calculated using the mid-latitude southern hemisphere New Zealand (Putnam et al., 2010) and Patagonian calibration datasets (Kaplan et al., 2011) since these are proximal to Antarctica. No snow cover or erosion rate correction was applied. Internal and external uncertainties (Balco et al., 2008) are reported and we use external uncertainties to compare new exposure ages with calibrated AMS radiocarbon ages collected in this study. In the online calculator, we assumed zero erosion, used the AMS standard flag 07KNSTD, an 'ant' elevation pressure flag and a density of 2.7 g/cm 2 . A shielding correction of 0.9999 was applied to all samples as the value is the same for all. Calculations of 10 Be exposure ages using the global (Ww) 'LSDn' scaling (Lifton et al., 2014), the New Zealand-Macaulay (Pt) (Putnam et al., 2010) and the Patagonia calibration dataset (Kp) (Kaplan et al., 2011) are shown. int: internal error; ext: external error; ages have been rounded to the nearest 10 years. Lm production rate scaling from the online calculator (Balco et al., 2008) would give ages that are around 2% older, within the uncertainties on the ages we report using LSDn. For the batch of three samples, the laboratory blank used to correct 10 size and shape meant it is very unlikely they had been overturned. Post-depositional movement was minimised by sampling erratics away from slopes or cliffs. Samples of upper few centimetres of sub-horizontally exposed surfaces were taken with a hammer and chisel ( Figure 2c) and the angle to the skyline was measured to calculate topographic shielding (Balco et al., 2008). Surface shielding due to snow cover was minimised by sampling from wind-exposed localities (cf. Glasser et al., 2014;Johnson et al., 2012Johnson et al., , 2017Johnson et al., , 2020Lindow et al., 2014). Laboratory analysis for 10 Be cosmogenic nuclide surface exposure dating followed the procedures of Kohl and Nishiizumi (1992) and Binnie et al. (2015) ( Table 1 and Supplemental Materials for details, available online).

Radiocarbon dating of moraines and stratigraphic sections
Moraines adjacent to the BIC were mapped and interpreted from field observations and satellite images (DigitalGlobe, Catalogue ID: 1030010020C900; Google Earth, 2006 and 2011). Contour lines are derived from the Antarctic Digital Database with elevation data (±5 m) obtained from the Instituto Antártico Uruguayo (1997). Radiocarbon ages from moraines and stratigraphic sections were obtained by Accelerator Mass Spectrometry (AMS) dating of marine mollusc shells, terrestrial mosses and seaweed layers embedded in sediments and represent maximum ages for BIC glacier readvance (Table 2). To reduce the risk of contamination with modern material, all samples for radiocarbon dating were taken from fresh and previously cleaned outcrops or sediment cores, packed in zip-lock plastic bags, and stored at 4°C. Sample preparation prior to radiocarbon dating analysis was undertaken at the Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research and the British Antarctic Survey. Sample measurements and corrections for 13 C/ 12 C ratio and calculation of Conventional Radiocarbon Ages were undertaken by ETH Zürich and SUERC, Scotland (Table 2 and Supplemental Materials for details, available online). Calibration of radiocarbon ages from marine samples was undertaken in Oxcal v. 4.4 using the Marine20 calibration curve (Bronk Ramsey, 2009;Heaton et al., 2020Heaton et al., , 2022, and a new local marine reservoir age offset (ΔR) of 666 ± 76 14 C years for the South Shetland Islands, which represents the weighted mean ΔR of four radiocarbon-dated marine samples collected prior to 1950 CE from the northern Antarctic Peninsula and Signy Island in the CALIB online Marine20 database (http://calib.org/marine/) ( Table 2; Heredia Barión et al. (2023)). Terrestrial samples, Kiteschsee Lake samples and previously published data from Yanou Lake (Roberts et al., 2017), were (re)calibrated using the Southern Hemisphere SHCal20 calibration curve (Hogg et al., 2020) in Oxcal v. 4.4 (Bronk Ramsey, 2009). Post-bomb (>1950 CE) ages were corrected according to 13 C/ 12 C isotopic ratios from measured pMC with the 'present day' pMC value defined as 107.5% (2010 CE) and calibrated using the SHCal13 SH Zone 1-2 Bomb curve in CALIBomb Reimer and Reimer, 2004). There were no significant statistical differences (i.e., beyond twosigma error ranges) between Holocene radiocarbon ages calibrated using the SHCal13/Marine13 (Hogg et al., 2013) and the ShCal20/Marine20 curves (ShCal13 vs 20: n = 19, mean difference ±1σ = −4 ± 8 years; Marine: n = 8, mean difference ±1σ = −45 ± 8 years); hence, comparisons with published data and age models calibrated using the SHCal13 and Marine13 curves remain valid.
New and published chronological and chronostratigraphic data constraining Holocene glacier readvances from the Fildes Peninsula (n = 43), Potter Peninsula (n = 29) and KGI (n = 76) were compiled and a non-parametric phase model (i.e., a probabilistic version of the Oxcal SUM command) was applied to each dataset using the Bchron v. 4.7.6 in R (Haslett and Parnell, 2008;Parnell, 2021). This analysis provides a Gaussian mixture prior distribution density phase age-range estimate, at 0.68 and 0.95 probability level, constraining periods when glacier readvances could have occurred on the Fildes, Potter and Barton Peninsulas on KGI (FP, PP and BP in Figure 3; Supplemental Table S2, available online). Density phases produced by similar analysis of published SSI-wide data were less well-defined, likely reflecting local glacier influences and larger errors for some data, and not considered further. The Rcarbon package v. 1.4.2 (Bevan, 2021) was also used for data analysis and plotting. Mapping was undertaken in Arc-GIS and data was analysed and plotted using the Tidyverse suite of packages in R v. 4.1.0/Rstudio v. 1.4.1717, Sigmaplot v. 14.0, C2 (Juggins, 2007) and MATLAB v. R2021a.

Kiteschsee Lake record
Kiteschsee Lake (62°11′36.55″S, 58°57′59.93″W) is located at ~15 m a.s.l. in the eastern-central region of the Fildes Peninsula, ~550 m from the east coast and ~2.6 km from the margin of the BIC (Figures 1d and 2b). It is adjacent to the Davies Heights (>100 m a.s.l.), a previously glaciated part of the former raised marine platform in the mid-central area of the Fildes Peninsula, which supplies meltwater directly into the eastern end of the lake ( Figure 1d). Marine sediments at the base of a previously studied sedimentary record from Kiteschsee Lake (Mäusbacher et al., 1989) show that much of the low-altitude central area of the Fildes Peninsula remained below sea-level until c. 7-6 ka (Mäusbacher et al., 1989).
Lithology, geochemistry and chronology. We undertook multiproxy analyses (diatom, grain size, geochemical and sedimentological analysis) on a 77 cm-long sediment record extracted from the flat-bottomed eastern basin depocentre of Kiteschsee Lake ( Figure 4 and Supplemental Materials S1-S3, available online) and compared data obtained with published lake records from the Fildes Peninsula. High resolution X-ray fluorescence (ITRAX™) core scanning (XRF-CS) of bulk sediment was performed at Aberystwyth University contiguously at 0.02 cm (200 µm) intervals following procedures in Davies et al. (2015) and Roberts et al. (2022). Gamma (bulk) wet density, magnetic susceptibility (Bartington Instruments MS2E point sensor, 10 s count time; true χvol data), P-wave amplitude and velocity, and resistivity were measured at 5 mm intervals on a Geotek ® multi-sensor core logger (MSCL) at Durham University following standard procedures (Gunn and Best, 1998) (Supplemental Materials for details, available online).
A chronology for the Kiteschsee Lake sediment record was based on two subaquatic freshwater moss ages, six bulk sediment AMS radiocarbon ages, and four tephra correlation ages and established using Bayesian age-depth modelling in BACON v. 2.5 in R (Blaauw and Christen, 2011) ( Figure 4 and Table 2). Volcanic glass-shard counts, and electron probe microanalysis (EPMA) data and tephrostratigraphic correlation with the nearby Yanou Lake record were used to improve the final age-depth model (Blockley et al., 2005;Roberts et al., 2017; Figure 4 and Supplemental Materials for details, available online). Reporting of radiocarbon data and age-depth modelling parameters follows recommendations in Lacourse and Gajewski (2020). Calibrated ages in the text have been rounded to the nearest 100 years to reflect realistic total (internal and external) uncertainties in radiocarbon dating.
Diatom analysis. Since the early Holocene isolation history of Kiteschsee Lake is well-established (Mäusbacher et al., 1989), and diatom smear-slide screening tests revealed that samples from 77 to 47 cm depth had very low diatom concentrations,  Table 1. (e and f ) Radiocarbon dating samples from locations ART-01 and ART-03 embedded in moraine sediments (Table 2). (g) View from the crest of the Shetland I moraine in the Artigas Beach ART01/03 area looking approximately south towards the Uruguayan Station. (h) The small moraine on top of a raised beach at 7-6 m a.s.l. on Artigas Beach at ART02 (open black box) and inset close ups of the moraine and ART02 reworked moss deposits sampled (red circles), looking approximately SW towards the Uruguayan Artigas Station, with the Davies Heights beyond. ± 100 14 C year lake reservoir effect was applied to bulk sediment radiocarbon ages from units 7-9 in Kiteschsee Lake, based on the average age offset between ages for the T1 tephra layer in Kiteschsee Lake and Yanou Lake (Supplementary Materials for details).  Figure 1; units for the probability density phase analysis are cal. ka/BP; data sources are shown in subtitles and described in code input files.
quantitative diatom analysis focussed on reconstructing high-resolution species compositional changes in the uppermost 47 cm (Late Holocene) where diatom preservation was sufficient for meaningful environmental interpretation ( Figure 4). Quantitative diatom counting and analysis was undertaken at 1-2 cm intervals on sediments between 47 and 5 cm depth (3680 to −63 cal. a BP) and related to XRF-CS data using. Species identifications and ecological preferences followed Jones and Juggins (1995) Figure 4, we used the diatom-chlorophyll-a training set developed for the Antarctic Peninsula using data from 61 lakes (Jones and Juggins, 1995). Diatom-based transfer functions were applied in C2 (Juggins, 2007) using simple weighted averaging (WA) and weighted partial least squares (WA-PLS) algorithms. The WA-Inverse transfer function was chosen to reconstruct chlorophyll-a as the RMSE, and average bias were low and displayed a strong relationship (R 2 ) between measured and diatom-inferred chlorophylla (Supplemental Figure S10 and Supplemental Materials for details, available online). Constrained cluster analysis (CONISS) with broken stick analysis was applied to square root transformed datasets using R packages Vegan and Rioja to define significant lithofacies units and diatom zones and their boundary positions (Juggins, 2022;Oksanen, 2014).

Cosmogenic nuclide surface exposure dating of erratics
The three erratic granitic boulders on the NW marine platform shown in Figure 3 gave an error-weighted mean ± internal (external) error cosmogenic surface nuclide exposure age of 6780 ± 220 (590) years (chi-squared, p-value 0.9718), based on the nearby Patagonian production rate (Kp in Table 1) calibration dataset, and 6580 ± 209 (442) years (chi-squared p-value 0.9715) based on the global calibration dataset (Ww in Table 1) (Borchers et al., 2016, both using the 'LSDn' scaling scheme (Table 1). The relatively small standard deviation of the three ages of ±60 years is a favourable indicator of low analytical uncertainties but omits any systematic bias in the age related to, for example, production rate. Henceforth, we base our discussions on the LSDn scaled age estimates (±external errors) and their weighted mean age of 6580 ± 440 years ( Figure 2 and Table 1), as this provides a minimum age constraint for ice retreat coupled with a realistic estimate of external uncertainty, taking production rate uncertainty into account. However, the close agreement of the three exposure ages and the overlap between ages produced using the three different production rate calibration datasets shown in Table 1 implies the samples had negligible pre-exposure and that they were exposed simultaneously on the NW foreland during a period of (potentially) rapid Mid-Holocene glacier retreat.

Radiocarbon dating of moraines and stratigraphic sections
Twenty-five new radiocarbon ages were obtained from 10 samples in the Artigas Beach-Valle Norte sections of the Shetland I moraine and its associated glaciofluvial and marine sediments (Figures 2, 3a-c and Table 2). Shells embedded in marine sandy to silty sediments reworked into the central Valle Norte sector of the Shetland I moraine (Figure 2b) returned near-infinite radiocarbon ages >40 cal. ka BP, implying significant reworking of older glaciomarine deposits by the BIC (Table 2). In contrast, shells reworked into till in the Artigas Beach southern sector ( Figure 2b) have a mean calibrated age range of 6.6-5.4 cal. ka BP (Table 2), with two distinct 0.95 probability density age phases of 6.8-6.5 and 6.3-5.3 cal. ka BP (Figure 3c and Supplemental Table S2, available online). The oldest of these ages are broadly coeval with cosmogenic exposure ages of the granite erratics (Figure 3c), which represent a minimum retreat age from the northern and central sectors of the inner BIC foreland, while the youngest are consistent with basal (deglaciation) ages from the inner Fildes Peninsula lake sediment records in the Valle Norte-Artigas Beach sectors (Figure 3c and e) (Roberts et al., 2017;Watcham et al., 2011).
Terrestrial moss fragments in glaciofluvial and glacial sediments from the Shetland I moraine dated in this study have mean age ranges of 2.0-1.2 cal. ka BP and two well-defined 95% probability density phases of 2.1-1.7 and 1.4-1.1 cal. ka BP (n = 13) (Figures 2b, 3a and d and Supplemental Table S2, available online), suggesting emplacement by limited Late Holocene glacier readvances between 1.7 and 1.4 and/or after 1.1 cal. ka BP (Figure 3a).
Non-parametric 95% probability density phase modelling of all maximum age constraints from terrestrial mosses and shells emplaced in Shetland I moraines from this study (n = 25) and that of Hall (2007) (n = 43 in total) revealed up to six 'probability gaps' when the readvance of the BIC on the Fildes Peninsula could have occurred (i.e., when no shells or mosses were living): 7.3-6.8, 5.3-3.5, 3.1-3.0, 2.6-2.2, 1.7-1.4 and <0.2 cal. ka BP (Figure 3a-c and Supplemental Table S3, available online).
Similar analysis of all new and existing chronological constraints for glacier advance on the Fildes Peninsula (Hall, 2007; this study), the Potter Peninsula (Heredia Barión et al., 2023) and the Barton Peninsula (Oliva et al., 2019) (n = 80) identified up to eight 95% 'probability gaps' when readvances could have occurred (7.4-6.6, 5.3-4.8, 4.5-3.9, 3.3-3.0, 2.6-2.5, 1.7-1.5 and <0.7 cal. ka BP) and one 68% 'probability gap' (1.3-0.7 cal. ka BP) (Figure 3d and e and Supplemental Table S3, available online). Subaquatic freshwater moss with post-bomb 14 C ages is a common feature at sediment-water interfaces in lakes on Fildes Peninsula lakes (Figure 1) and likely reflects warmer conditions on KGI since the mid-C20th. Late Holocene 'readvance probability gaps' are broadly anti-phased with maximum to minimum calibrated 14 C age ranges constraining subaquatic freshwater moss deposition in lake records from Fildes Peninsula (red bars in Figure 3e). Bayesian best fit age-depth model for the Kiteschsee Lake record. Sample thicknesses were 0.5 cm or 1 cm, and no zero-depth surface age was set in the age-depth model. Run settings and priors are shown in Supplemental Figure S4, available online. Modelled ages and errors were derived from the MCMC 'best-fit' weighted mean ages from the BACON age-depth model. Initial age model runs were fine-tuned by tephrastratigraphic correlation to the nearby Yanou Lake record and the inclusion of radiocarbon ages of subaquatic freshwater moss immediately below and above the 'T5', between the 'T3a' and 'T3b' tephra layers, and immediately below the 'T1a' layer in the Yanou Lake record (Supplemental Materials for details, available online). (c) Age depth model and tephrastratigraphic correlation for the Yanou Lake record. (d) An updated Bayesian age-depth model (2020 calibration) for the Yanou Lake record (cf. Roberts et al., 2017). All measured radiocarbon ages were included in age-depth models.

Kiteschsee Lake record
The new Kiteschsee Lake record shown in Figure 4a and b provides a continuous and well-defined Mid-to Late Holocene record of palaeoenvironmental change and volcanic activity. It is largely devoid of soliflucted ash and catchment reworking that disrupts sedimentation in other lakes on the Fildes Peninsula and KGI, for example, Yanou Lake ( Figure 4c) (Roberts et al., 2017;Watcham et al., 2011).
Lithology, geochemistry and chronology. Lithofacies units were defined by cluster analysis of XRF-CS data and are summarised as follows: Units 1-3, 77-60 cm (7.5-5.6 cal. ka BP, mean modelled ages) are composed of poorly sorted olive-brown silt with three prominent fine black ash layers in Unit 2. After deposition of the T5 tephra (Unit 4: 60-58 cm; 5.6-5.5 cal. ka BP), there is a distinct change to a light orange-brown silty-clay at the base of Unit 5. A shift to light-grey clay deposition began as early as 1.7 cal. ka BP, the basal age of Unit 6, at 34 cm. After a return to olive-brown silt in Unit 7 (c. 1.2 cal. ka BP), there is a shift to a light grey clayeymud with silt and sand-rich layers (Unit 8: 20-12 cm depth; 1.2-0.5 cal. ka BP). Unit 9 (12-5 cm; 0.78 to −0.005 cal. ka BP) is composed of laminated olive-brown silt, similar to Units 5 and 7. Unit 10 is a thick subaquatic freshwater moss composed entirely of Drepanocladus l. sp. of post-bomb age (1950-2011 CE) ( Table  2). Lithofacies units and chronological descriptions for Kiteschsee Lake are summarised in Figure 4 and Table 2. Radiocarbon ages from the 77 cm long sediment record from Kiteschsee Lake are in chronological order, except for a calibrated age of 9900 ± 240 cal. a BP near the base of the fine grey (glaciofluvial/glaciolacustrine) silt and clay deposits of Unit 8 at 21-22 cm depth (Table 2). Bulk sediments at 7-8 cm, 54.5-55 cm, and 64.5-65 cm depth could not be dated due to insufficient carbon. In summary, the Kiteschsee Lake record has a basal age of 7470 cal. a BP at 77 cm depth (weighted mean modelled age) (7820-7110 cal. a BP min.-max. 95% confidence age range). The mean age-depth model 95% confidence age range is 470 years, with a minimum 95% confidence age range of 4 years at 4.5 cm, in the post-bomb era and a maximum 95% confidence age range of 1230 years at 51.5 cm; 83% of the measured ages overlap with the age-depth model (95% ranges).
Elevated dry mass accumulation rates and increased sand deposition within Unit 6 and the glaciofluvial/glaciolacustrine Unit 8 are indicative of increased erosion and meltwater input from c. 1.7 cal. ka BP onwards (Supplemental Figures S3 and S8, available online). The radiocarbon age of 9380 ± 30 14 C years from 21 to 22 cm depth at the base of Unit 8 had a heavily depleted δ 13 C value of −32.6‰, likely due to reworking of old (possibly marine) carbon. The uppermost sediments, immediately below the living subaquatic freshwater moss, have a measured radiocarbon age of 1350 ± 30 14 C years (1.2 ± 0.8 cal. ka BP) implying that bulk sediment radiocarbon ages obtained from Unit 9 are either significantly less influenced by old carbon, sedimentation rates declined, or a hiatus exists between Units 9 and 10. K1-8 are radiocarbon ages shown in Table 2; T1a, T3a, T5 and T7 are correlation ages of tephras in the Yanou Lake record (Supplemental Table S1, available online).
Downcore changes in XRF-CS Ti, K, Ca and sand content reflect catchment erosion of (volcanic) bedrock and tephra deposition. These, and tephra glass-shard counting, revealed the position of several prominent visible volcanic ash (tephra) layers (T1-T7 in Figure 4 and Supplemental Materials S1-S4, available online). Incoherence/coherence scatter ratios (inc./coh.) are a proxy for organic content and were also used to define the precise positions of the visible tephra layers in Figure 4 and Supplemental Materials S2 and S3, available online. Glass shard-specific electron probe microanalysis of the two most prominent tephra horizons showed that a tephra peak at 33 cm depth (modelled mean age ±95% confidence interval: 1690 + − 410 230 / cal a BP) had a broadly bi-modal basaltic-rhyolite glass shard geochemical composition, and the tephra peak at 58 cm depth (modelled mean age ±95% CI: 5550 + − 150 200 / cal. a BP) consisted solely of rhyolitic glass shards. The latter was found in a ~6 cm thick tephra deposit between 56 and 62 cm depth and was part of a large, explosive multi-stage eruption between 6-5 cal. ka BP (Figures  4b-d, 5c and Supplemetal S1 and S5, available online; Roberts et al., 2017). The shard geochemistry of glass shards from both layers aligns with the magma evolution trend of the Deception Island volcano and is characteristic of post-caldera eruptions from Deception Island (cf. Geyer et al., 2019) (Supplemental Figures  S1-S5, S14-S16, Table S5 and Supplemental Materials for details, available online). 5. Regional palaeoclimatic context for Holocene glacier fluctuations on the South Shetland Islands, Antarctic Peninsula. (a) Probability density phase model for glacier readvances Stages 1-4 on King George Island between 12 and 0 cal. ka BP and outer Fildes Peninsula freshwater subaquatic moss layers (median age and two-sigma age range) (Figure 3d and e for details); BP: Barton Peninsula; FP: Fildes Peninsula; PP: Potter Peninsula; units for the probability density phase analysis are cal. ka/BP. (b) Diatom-inferred reconstructions and key indicator diatom species for Kiteschsee Lake. Lower DCCA scores reflect more turbid conditions, and lower Gomphonema sp. percentages represent less aerophilic conditions, and longer seasonal ice-cover. (c) Glycerol dialkyl glycerol tetraether (GDGT) mean summer air temperature (MSAT) anomaly reconstruction and tephra layers from the Yanou Lake record (Foster et al., 2016;Roberts et al., 2017; Supplemental Materials for details, available online). The dark red plot and its 6 ka mean (0.06 ± 1.50°C: dark red dotted line) are a new MSAT anomaly reconstruction (relative to the 250-1000 cal. a BP mean (±SD) value of 1.74 ± 1.44°C) for Yanou Lake, with RMSE of 1.65°C shaded in grey (updated 2020 age calibration). This reconstruction is more realistic than the light red plot and its 6 ka mean (1.17 ± 3.46°C: light red dotted line) in Roberts et al. (2017). Airfall tephra layer ages are plotted above as modelled mean ages (white circles) and 95% confidence intervals (grey bars), and classified according to layer thickness in the Yanou Lake record. The timing of the caldera collapse (CC) eruption event from Antoniades et al. (2018) is also shown (black circle -mean age with two-sigma error within). The age-depth models in Roberts et al. (2017) and the caldera collapse eruption event age in Antoniades et al. (2018) have been updated to 2020 calibrations; Supplemental Table S1B and Supplemental Materials for details, available online). (d) Maxwell Bay total organic carbon (%TOC) record (see Figure 6 for location) with lower values interpreted by Milliken et al. (2009) as reduced sea-ice concentration and more open water; the light blue plot is the published data; the dark blue line is a 100-yr interval LOESS regression with standard errors shaded in grey (LOESS: locally weighted sum of squares with tricube weighting and polynomial regression applied to the published %TOC data). (e) Holocene warm periods on the Antarctic Peninsula (red bars; EHWP: Early Holocene Warm Period; HH: Holocene Hypsithermal; MCA: Mediaeval Climate Anomaly; RRR: Recent Rapid Regional warming; Bentley et al., 2009) and peaks greater than the 10 ka mean in the Total Solar Irradiance deviation (ΔTSI) (orange stars; see Figure 7a for explanation). (f ) Probability density phase model for glacier advance (blue) and retreat (red) data on the northern Antarctic Peninsula in Kaplan et al. (2020); units for the probability density phase analysis are cal. ka/BP. (g) Temperature anomaly relative to pre-industrial reference period (1850-1900 CE) from the James Ross Island (JRI) ice core record, NE Antarctic Peninsula (Mulvaney et al., 2012); grey shading represents the published error range; the dotted line is the 12 ka mean. (h) Holocene palaeotemperature compilation showing the median temperature anomaly for the 60-90°S data stack (dark red line) and the temperature anomaly for the global data stack (black line) relative to the 'pre-industrial' reference period (1850-1900 CE) (Kaufman et al., 2020a(Kaufman et al., , 2020b. The grey dotted line is the average median 12 ka temperature anomaly value of 0.02 ± 0.14°C for the 60-90°S dataset; 95% CI errors in both datasets are, on average, +1.51°C and −2.20°C, mostly out of range and not plotted for clarity. (i) Climate syntheses for the Antarctic Peninsula (Bentley et al., 2009;Ingólfsson et al., 2003).
Diatom analysis. The Kiteschsee Lake diatom record is characterised by short-term changes in diatom communities superimposed upon the steady ecological and environmental evolution of the lake during the Mid-to Late Holocene. Although certain species found respond directly to climatic variables (temperature, precipitation and wind strength), the overall community structure response in Kiteschsee Lake was primarily associated with changes in lake-ice cover and turbidity, reflected in β-diversity DCCA scores (Figure 4a).
Variations in key indicator species and the reconstructed chlorophyll-a were used to examine the impact of climate-driven glacier readvances in the mid-southern BIC foreland (Figures 4 and  5). A notable shift in XRF-CS geochemistry exists at ~60 cm depth ( Figure S1), but we found no clear bio-geochemical evidence of (glacio) marine sedimentation. Although diatom concentration in the lowermost zone was too low for quantitative analysis, all sedimentary units younger than c. 6 cal. ka BP (~63 cm) contained freshwater diatoms. The δ 13 C value from 73.5-74.5 cm depth is also consistent with a freshwater sedimentary environment, implying that it is younger than c. 6.5 cal. ka BP. The increase in diatom accumulation rate from c. 1.9 cal. ka BP is linked to the increased input of minerogenic sediments and dry mass accumulation rate between c. 1.9 and 1.6 cal. ka BP (Supplemental Figure S8, available online). Redundancy analysis (RDA) using key geochemical variables associated with tephra from XRF-CS analysis showed that tephra deposition did not have a significant impact on the diatom community composition in the Kiteschsee Lake record (p = 0.596) (Supplemental Figures S9 and S10, available online).
Using constrained cluster analysis (Figure 4a and Supplemental Figure S9, available online), we divided the diatom record from 47 cm upwards into five zones (Supplemental Materials for details, available online). Diatom Zones 4 and 5 are of most interest to this study because they correspond to a decline in aerophilic and littoral diatom species and concurrent increases in species associated with more turbid conditions (e.g., Stauriserella pinnata), and lower DCCA scores (Figures 4a, 5b and Supplemental Figure S9, available online). The transition to increased minerogenic input in lithofacies Unit 8 coincides broadly with the start of Diatom Zone 4 (22-13 cm: 1.3-0.8 cal. ka BP). Trends initiated in Diatom Zone 4 continued in Zone 5 until ~5 cm depth (mid-late C20th onwards), when there was a significant shift to a subaquatic freshwater moss (Drepanocladus longifolius (Mitt. Paris) sp.) dominated environment (Lithofacies Unit 10).

Discussion
In this section, we assess: (1) The Holocene deglaciation and glacial readvance history of the BIC, of KGI, and relationships to regional-global change; (2) Mechanisms driving Holocene climate-glacier dynamics on the AP identified in Bentley et al. (2009), principally, solar insolation, westerly winds, the Southern Annular Mode (SAM); (3) Impacts of Holocene deglaciation and glacier readvances.

Palaeoenvironmental history and change
Early Holocene deglaciation c. 12-8 cal. ka BP. Previously published lake sediment records show that the terrestrial deglaciation of the southern-mid part of the Fildes Peninsula from the LGM ice limit dates to c. 12-9 cal. ka BP (Mäusbacher et al., 1989;Watcham et al., 2011;Figure 1), and was broadly coeval with other areas of the SSI (Palacios et al., 2020). New radiocarbon data from the BIC foreland and Potter Cove show the ice cap had thinned and retreated substantially during the early Holocene, with large parts of the Fildes Peninsula and Potter Peninsula becoming ice free by c. 8 cal. ka BP. The BIC had also retreated from the central area of Fildes Peninsula by c. 8 ka (Figures 1, 6a and 7e). New chronological evidence from the BIC foreland and the Shetland I moraine in this study is consistent with the deglaciation of Mondsee Lake (Figure 2), 45 m a.s.l., and ~1500 m away from the present glacier margin (Schmidt et al., 1990). Recession progressed towards the higher elevation areas in the eastern sector, located above the C20th equilibrium line altitude (ELA) of ~150 m a.s.l., into areas located in depressions next to the former ice limit, and triggered an increased rate of isostatic uplift and falling relative sea level (Figure 7e and f) (Watcham et al., 2011).
Stage 1 readvance/standstill. Using evidence from marine-freshwater transitions in isolation basins between 14 and 16 m a.s.l., Watcham et al. (2011) established a Holocene marine limit of ~16 m a.s.l. at c. 8.0 cal. ka BP and that the short-lived fall in RSL and return to that level at c. 7.0 cal. ka BP on the Fildes Peninsula was indicative of a restricted readvance or standstill in glacier retreat. New cosmogenic exposure ages from erratics in this study show the NW marine platform close to the present BIC ice front on the Fildes Peninsula became ice-free by c. 6.5 ka ( Figure  7e and Table 1). We found no evidence for cosmogenic nuclide inheritance; therefore, the dated erratics had not been reworked or eroded from pre-Pleistocene formations on the SSI and were most likely glacially transported from the AP prior to (or during) the LGM, then fully reset by glacial reworking prior to deglaciation. Our new data are consistent with isolation basin evidence for a restricted readvance/standstill at c. 7 cal. ka BP on the Fildes Peninsula and evidence of a marine transgression at c. 7.5-7.0 cal. ka BP in Potter Cove followed by a glacier readvance at <7.0 cal. ka BP on Potter Peninsula (Strelin et al., 2014) (Figures  3, 5a, 6a and 7e-g).
Based on previous geomorphological mapping ( López-Martínez et al., 2012;Michel et al., 2014a) and the retention of snow patches even during late summer on some satellite imagery, parts of the Davies Heights were probably glaciated during the Stage 1 readvance and colder phases of the Holocene (Figure 6a and c).
New cosmogenic exposure ages and radiocarbon ages from bivalve shells in marine sediments from the Shetland I moraine dated to between 6.8 and 5.3 cal. ka BP (95% probability phase range; Figure 5a) constrain (renewed) deglaciation across the BIC foreland to c. <7 cal. ka BP. Persistently warmer atmospheric temperatures across the Peninsula between 7 and 6 cal. ka BP is reflected in reduced sea-ice and warmer-than-Holocene-average sea-surface temperatures in marine sediment records from Maxwell Bay and the Palmer Deep between 8.2 and 5.9 cal. ka BP (Figure 5d) (Etourneau et al., 2013;Milliken et al., 2009;Mulvaney et al., 2012). Overall, the marine-terminating BIC was less extensive during this shell colonisation phase between 6.8 and 5.3 cal. ka BP (Fretwell et al., 2010;Hall, 2007). Tiefersee Lake and Hochlandsee Lake, located ~800 and ~1500 m from the present-day ice cap, became ice free (Figures 2 and 3e) as the BIC retreated to within its present-day limits by c. 6.0 cal. ka BP on land (Mäusbacher, 1991;Mäusbacher et al., 1989) and its marineterminating glacier front (re)occupied the inner bays of Collins Harbour in Maxwell Bay (Chu et al., 2017;Milliken et al., 2009;Simms et al., 2011;Tatur et al., 1999;Watcham et al., 2011;Yoon et al., 2006), and at the same time as ice sheets and caps thinned, and ice shelves retreated elsewhere on the AP and around Antarctica ( Johnson et al., 2022 -J22) used to determine the elevation of the coastlines in panels A-C; BIC: Bellingshausen Ice Cap; DH: Davies Heights; KL: Kiteschsee Lake, which is a marine embayment in A; the blue star is the location of Maxwell Bay marine records referred to in the text (Milliken et al., 2009;Simms et al., 2011); white areas (90% transparency) bordered by solid black lines in A-C are schematic representations of BIC extent on the Fildes Peninsula for each stage; present-day moraines mapped in this study are shown in D; black arrows represent schematic glacier flow directions, except in B where they represent retreat close to (or within) present day limits between 6.8 and 5.3 ka BP, followed by minor readvances close to present limits between 5.3 and 2.2 ka BP; blue arrows are main meltwater pathways. Question marks in A and B indicate tentative locations of potential glacier fronts above 100 m a.s.l.; lakes sediment basal and/or freshwater transition ages shown as white text are in cal. ka BP, labelled as in Figure 1. Black shaded areas around the BIC in C and D are the last readvance moraines, as shown in Figure 2  (a) Total Solar Irradiance, which is the power per unit area received from the Sun as measured, or reconstructed, on Earth, where ΔTSI represents the deviation from its present-day value (Steinhilber et al., 2009) compared with solar insolation received at 62°S during the Holocene (11.75 ka) (Laskar et al., 2004); annual insolation = dotted black line, austral spring/summer (SONDJF) insolation = red line; winter ( JJA) insolation = blue line; the open dark red circle is the mean ± 1σ ΔTSI value for the last 10 ka; orange stars mark ΔTSI peaks greater than the 10 ka mean value. (b) Ultra-high resolution (69 μm) hyperspectral (SPECIM) R850/R900 data, a proxy for mineral input into the Emerald Lake, Macquarie Island at 54°S (Saunders et al., 2018; age model updated to Sh20.cal), which reflects changes Southern Hemisphere westerly wind (SHW) strength during the Mid-to Late Holocene; the dark grey horizontal dotted line is the R850/R900 dataset mean; dark green line is 100-year interval LOESS smoothing of the R850/R900 (see code for details). (c) Hypothetical representation of changes in the mean annual latitudinal position of the core SHW intensity belt (dark grey line) and approximate 1σ latitudinal range of enhanced precipitation (light blue stipple) (Ariztegui et al., 2010;Quade and Kaplan, 2017). In reality, the SHW are more intense and focused during positive (warmer) phases of the SAM, but weaker, latitudinally Stage 2 readvances c. 5.3-2.2 cal. ka BP. There are four probability 'gaps' when minor readvances could have occurred between c. 5.3 and 2.2 cal. ka BP (5.3-4.8, 4.5-3.9, 3.3-3.0 and/or 2.6-2.2 cal. ka BP). Evidence for glacier readvances on KGI and SSI is limited across the Mid-to Late Holocene transition, but it seems likely that the general warming transition into the Mid-Holocene Hypsithermal (MHH) as defined by Bentley et al. (2009) (now referred to as the Holocene Hypsithermal, HH) was interspersed by at least one short-lived or restricted readvance (Figures 3d and 5).
Marine sediments and bivalve shells dated to between 6.8 and 5.3 cal. ka BP were reworked into till at the extreme east of the Shetland I moraines most likely between 5.3 and 3.9 cal. ka BP as RSL fell below 15 m a.s.l. (Figures 2, 6b, 7e, Table 2 and Supplemental Material S3, available online), but we found no evidence of BIC readvance on the Fildes Peninsula beyond its present-day limits at this time. Moraines were located close to the former sealevel prior to reworking, and Holocene raised beaches up to ~16-18 m a.s.l. extended under the BIC in the Artigas Beach area may have survived relatively unmodified beneath the ice cap (Hall, 2003;John and Sugden, 1971).
A Stage 2 readvance probability 'gap' between 5.3 and 4.8 cal. ka BP (Figure 5a) overlaps with persistently colder temperatures in the Yanou Lake GDGT-temperature record between 6.0 and 3.7 cal. ka BP as well as a hypothesised local readvance in the Stranger Point area of Potter Peninsula at c. 5 cal. ka BP (Emslie et al., 2019) and glacier readvances elsewhere on the NAP (Figure 5c, f and i). Surface temperatures decreased. Sea-ice cover also increased in Maxwell Bay between c. 5.9 and 2.6 cal. ka BP and summer sea-ice occupied the inner Collins Harbour of Maxwell Bay until at least c. 1.7 cal. ka BP (Simms et al., 2011).
Stage 3 readvance (Late Holocene Neoglacial) c. 1.7-1.3 ka. The BIC readvanced only by tens of metres relative to its present position in the Artigas Beach area in the second half of the Late Holocene and most of the present-day Shetland I moraine was established during Stage 3 and/or Stage 4 readvances.
New mean radiocarbon ages from moss fragments incorporated into a small moraine on top of a raised beach at 7-6 m a.s.l. on Artigas Beach, indicate a two stage readvance of the BIC in this area during the Late Holocene: the first, after the formation of the raised beach and sometime between c. 1.7 and 1.4 cal. ka BP (Stage 3 readvance) and a second at <0.7 cal. ka BP (Stage 4 readvances). New radiocarbon ages also constrain the initial formation of the present-day Shetland I moraine from Valle Norte towards its eastern extension (see Figure 2 for location), where moraines are still ice-cored and in contact with the present ice. There, we found moraine crests several metres above the icelevel, similar to those found in Valle Norte (Hall, 2007), which post-date c. 1.2 cal. ka BP, the age of the youngest new sample dated. This age means that the eastern part of the BIC was landward of its present limit, with mosses colonising its foreland until at least 1.2 cal. ka BP.
The Stage 3 readvance occurred after the start of the 'Neoglacial' on the AP, defined as 2070 ± 50 cal a BP by Čejka et al. (2020). The most significant downturn in the Yanou Lake reconstructed GDGT-temperature record occurred at c. 1.3 cal. ka BP and coincided with a shift to more turbid conditions in Kiteschsee Lake (Figure 5b and c and Supplemental Figure S11, available online). The Davies Heights could have been ice-covered at this time, even with a relatively minor reduction in temperature as they are close to the present-day ELA (Figure 6c). The Stage 3 readvance was broadly synchronous with two phases of glacier expansion and the reformation of ice shelves on the north-eastern AP (Figure 5f), and in Southern South America between c. 1.5 and 1.1 cal. ka BP (Balco and Schaefer, 2013;Kaplan et al., 2020;Moreno et al., 2018;Strelin et al., 2014) and occurred when a colder conditions existed on the north-eastern AP (Figure 5g) (Mulvaney et al., 2012), but also when composite ice core records from elsewhere in Antarctica show a predominantly warmer phase (Masson et al., 2000).
Stage 4 readvance(s) c. 1.0-0.7 cal. ka BP and after <0.7 cal. ka BP. Probability 'gaps' in the Shetland I moraine radiocarbon ages ( Figure 3) suggest that readvances could have occurred at c. 0.7, 0.4 and/or <0.2 cal. ka BP. Hall (2007) determined that glacier readvances of the Shetland I moraines occurred at Valle Norte and Valle Klotz (Figure 1 for locations) between c. 1.0 and 0.7 cal. ka BP (Figures 2, 3c and 5a). Ice in these areas advanced in lobes, which spread out into ice-marginal depressions, forming several small thrust moraines, ~300 m from the prominent moraine ridges (dated in this study). These ages broadly align with persistently low DCCA Axis 1 (β-diversity) values in Diatom Zone 5 of Kiteschsee Lake, reflecting resuspension of sediment in a more persistently-turbid lake environment, indicative of reduced lake ice, increased meltwater and/or stronger winds (Figures 4 and 5b and Supplemental Figures S5F and S11, available online). We found no new chronostratigraphic data constraining a readvance <1 cal. ka BP, but Hall (2007) described the 'last readvance' at <0.2 cal. ka BP as the most extensive phase of glacier readvance of the last c. 3500 years.
broader and less focused during negative (colder) phases of the SAM, shown in (d). The light grey horizontal dashes are 54°S (Macquarie Island) and 62°S (South Shetland Islands); vertical dashed lines and arrows highlight the key hypothesised latitudinal shifts of the Holocene and their relationship to the global irradiance profile shown in A; ENSO: El Niño Southern Oscillation. (d) Reconstructed Holocene SAMindex variability between positive (red) to negative (blue) SAM-like conditions based on Northern Arboreal Pollen data from Southern Chile (Moreno et al., 2018) compared with ΔTSI > 10 ka mean peaks (orange stars), and warm periods on the Antarctic Peninsula (EHWP: Early Holocene Warm Period; HH: Holocene Hypsithermal; MCA: Mediaeval Climate Anomaly; RRR: Recent Rapid Regional warming; Bentley et al., 2009). (e) Summarised relative sea level (RSL) envelope for the SSI (W11; light blue shading; Watcham et al., 2011) compared with the RSL curve in Johnson et al. (2022) (J22; dark solid and dotted blue lines) and the W12a GIA model (Whitehouse et al., 2012a(Whitehouse et al., , 2012b; see original references for data and further details. (f) Cosmogenic nuclide exposure ages and key lake basal age constraints on deglaciation for terrestrial ice-free peninsulas on KGI at different altitudes from the Fildes Peninsula (this study), the Potter Peninsula (Heredia Barión et al., 2023), and the Barton Peninsula (Oliva et al., 2019;Seong et al., 2009)   The Stage 4 readvances, and increased lake turbidity, occurred during a phase of sustained colder reconstructed summer air temperatures in the nearby Yanou Lake record (Figure 5c and Supplemental Figure S11, available online) (Roberts et al., 2017) after the Mediaeval Climate Anomaly (MCA) (c. 1.2 and 0.8 cal. ka BP - Bentley et al. (2009);1250-950 CE -Mann et al. (2009. No evidence of substantial glacial readvance exists on KGI or the northern AP during the MCA (Charman et al., 2018;Emslie et al., 2019;Kaplan et al., 2020;Oliva et al., 2019), and marine sediment records from Maxwell Bay and Potter Cove imply no substantial increases in sea-ice at this time (Figure 5d) (Milliken et al., 2009;Monien et al., 2011). Increased fine-grained sedimentation in Potter Cove during the MCA has been linked to warmer conditions, including warmer sea surface temperatures from the western AP (Etourneau et al., 2013;Hass et al., 2010;Milliken et al., 2009;Watcham et al., 2011). Some Southern Hemisphere proxy records between 60 and 90°S (Figure 5h) show warmer conditions during the MCA (Kaufman et al., 2020a(Kaufman et al., , 2020bMarcott et al., 2013;Neukom et al., 2014), but no sustained upturn in temperature exists in the James Ross Island ice core record or the Yanou Lake GDGT palaeotemperature record (Figure 5c and Supplemental Figures S11 and S12, available online) (Björck et al., 1993;Kaplan et al., 2020;Mulvaney et al., 2012;Roberts et al., 2017).

Mechanisms of Holocene climate-glacier dynamics. Deglaciation
of the Fildes Peninsula and other peninsulas on KGI to present day limits occurred in two broad phases during the Early to Mid-Holocene. The first phase between c. 15 and 8 cal. ka BP occurred during a period of rising spring/summer insolation at 62°S and relatively elevated global solar irradiance in the transition into the present (Holocene) interglacial (Figure 7a, d and e) (Baggenstos et al., 2019;Reynhout et al., 2019;Steinhilber et al., 2009). This coincided with a period of sustained warmth and persistently positive Southern Annular Mode (SAM)-like conditions (Figure 7d). Meanwhile, data from the BIC foreland moraines and other readvance constraints across KGI suggest that ice caps on the SSI likely responded to millennial-to centennial-scale changes in solar forcing (insolation and irradiance), westerly wind strength and the SAM in the Mid-to Late Holocene.
The SAM is the primary mode of annual to centennial scale variability in atmospheric circulation in the Southern Ocean (Marshall, 2007) and reflects the zonal mean sea level pressure difference between Antarctica (>65°S) and the mid latitudes (40°S); hence, the SAM index is, in effect, a measure of the longitudinal mean SHW strength (Marshall, 2007). Positive SAM phases are associated with lower pressure at higher latitudes, more poleward focussed SHW, with stronger Westerlies in summer and stronger Easterlies in winter over the NAP and SSI (Martin et al., 2021). During the Holocene, stronger and more poleward SHW are thought to be the centennial-to millennial-scale expression of decadal-centennial warming and positive SAM-like conditions (Charman et al., 2018;Moreno et al., 2018Moreno et al., , 2021Perren et al., 2020). Conversely, phases of enhanced El Niño drive warmer interannual global mean temperature variability and negative phases of the SAM, which are characterised by colder and more humid conditions on the NAP, SSI and sub-Antarctic islands, and have increased over the last 4 kyr (Figure 7c) (Kaplan et al., 2020;Reynhout et al., 2019;Turner et al., 2016;Verfaillie et al., 2021;Wang and Cai, 2013).
Since 1957 CE, the longer-term pattern of interannual variability in the SAM has been altered by the ozone hole over Antarctica, which, combined with increased greenhouse gases and global temperature, has led to more positive SAM-like conditions, similar to the early Holocene, and enhanced the ENSO (Marshall, 2007) (Figure 7d). Increased cyclonic activity in the Drake Passage between 1996 and 2005 CE has been linked to a more poleward focussed SHW and more positive SAM, driving sea-ice poleward and increasing the advection of warm air across the NAP and SSI during winter (Marshall et al., 2017;Oliva et al., 2017). Between 2006 and 2015 CE colder winters pushed sea ice north, increasing snowfall across the NAP and doubling snow cover on some SSI islands between 2009 and 2014 CE (de Pablo et al., 2013;Oliva et al., 2017). In addition, the SAM and a deepening Amundsen Sea low have been interacting in recent decades, bringing warmer air from the Atlantic and north-easterly winds with cooler air from the Weddell Sea Kaplan et al., 2020;Liu et al., 2005). As a result, the NAP and SSI became colder and wetter, and its glaciers have gained mass (Falk et al., 2018;Goodwin et al., 2016;Oliva et al., 2017;Thomas et al., 2008).
Accordingly, longer-term changes in temperature and precipitation that drove glacier readvances on the NAP and SSI during the Holocene were most likely the result of annual to millennial scale variations in the position and intensity of the SHW, modulated by changes in the SAM (Bentley et al., 2009;Lamy et al., 2010;Moy et al., 2008;Varma et al., 2012).
The Stage 1 readvance occurred during a downward trend in global solar irradiance between c. 7.5 and 7.0 cal. ka BP which coincided with the transition to colder and more negative SAMlike conditions as the SHW migrated marginally equatorward (Figure 7a-c and g). Glacier readvance(s) on KGI are regionally coherent with readvances on the eastern AP (Kaplan et al., 2020;Figure 5f) and in Patagonia at c. 7 ka (Reynhout et al., 2019;Strelin et al., 2014), implying a common forcing mechanism.
Following the Stage 1 readvance, increasing spring/summer insolation over the SSI and global solar irradiance, weaker or marginally equatorward westerlies, and reduced storminess drove glacier retreat on the Fildes Peninsula and the Potter Peninsula between c. 6.6 and 5.3 cal. ka BP (Figure 7a-d, g and h) (Heredia Barión et al., 2023;Bentley et al., 2009;Quade and Kaplan, 2017;Reynhout et al., 2019;Saunders et al., 2018). Retreat on KGI coincided with sustained temperatures around the 0-12 ka mean in the Southern Hemisphere high latitudes and with the 'Holocene global thermal maximum', centred on 6.5-6.0 cal. ka BP (Figure 5h). Although the latter is more clearly defined in Northern Hemisphere proxy records, average global temperatures were 0.7°C warmer than the 'pre-industrial' reference period, defined by the IPCC as 1850-1900 CE (Kaufman et al., 2020a(Kaufman et al., , 2020b. Windier conditions further north at this time in the Southern Ocean have been linked to rising temperatures (Kaufman et al., 2020a(Kaufman et al., , 2020bMarcott et al., 2013;Saunders et al., 2018;Figures 5h, 7b and c).
While transitions into and/or out of short-lived phases of marginally more positive SAM-like conditions in the Mid-to Late Holocene could have increased snow accumulation on the SSI, the combination of generally colder and more negative SAM-like conditions, stronger westerly winds, and an enhanced ENSO from c. 4 ka onwards (Garreaud, 2007;Yang et al., 2019) led to deteriorating regional climatic conditions that lowered the ELA across the SSI and drove Late Holocene Stage 2, 3 and 4 readvances on KGI (Figures 5c, g, 7a-d, g and h and Supplemental Figure S12, available online).
Stage 3 and 4 readvances occurred as spring/summer insolation at 62°S declined and around global irradiance minima in the last 2000 years and during persistently colder conditions in lake records from the mid-outer areas of the Fildes Peninsula ( Figure  5c) (Roberts et al., 2017). Late Holocene readvances are antiphased with aquatic moss layers in lake records from outer Fildes Peninsula lakes that were not directly affected by glaciofluvial inputs. Generally stronger westerly winds and colder/more negative SAM-like conditions likely drove marginal increases in snow accumulation and greater turbidity in Kiteschsee Lake (Figures 4,  5b and 7b), but an equatorward shift in the core westerly wind belt storm belt to <55°S likely reduced storminess (and precipitation), restricting readvances to around or within present day limits (Figures 4, 5b, 7b and c).
From the mid-late C20th onwards, the mode of sedimentation in Kiteschsee Lake, other lakes on the mid-outer Fildes Peninsula and lakes elsewhere on King George Island that are now detached from direct glaciofluvial influence has shifted dramatically to one dominated by sub-aquatic moss (Figure 4a and Supplemental Material S1, available online). We link this dramatic ecological shift to the Recent Rapid Regional warming (RRR) on the northern AP and SSI (Bentley et al., 2009) and increasingly positive SAM-like conditions that have moved stronger SHW poleward in the late C20th and early C21st (Figure 7d), associated with anthropogenic activity (e.g., the ozone hole, increasing greenhouse gas emissions).

Impacts of Holocene deglaciation and glacier readvances.
Relative sea level change is the most obvious impact of ice (un)loading on the SSI during the Holocene. Data constraining RSL change on the SSI following deglaciation are more complex than other locations across Antarctica, and several different RSL curves have been suggested (Bentley et al., 2005;Hall, 2010;Johnson et al., 2022;Pallàs et al., 1997;Roberts et al., 2011;Simms et al., 2012Simms et al., , 2021Watcham et al., 2011). Bentley et al. (2005) initially proposed that RSL declined from an undated Early to Mid-Holocene marine limit of ~16-18 m a.s.l. in a discontinuous manner as the SSI deglaciated, and was interrupted at 14.5-16 m a.s.l. by a readvance between 5.8 and 3.0 cal. ka BP. In contrast, and with more data, Hall (2010) and Watcham et al. (2011) both suggested RSL fell continuously after 6 cal. ka BP, implying glacier readvances were restricted and had limited isostatic impact.
A readvance/standstill at c. 7 ka is consistent with evidence of marine-freshwater transitions in isolation basins between 14 and 16 m a.s.l. on the Fildes Peninsula (Watcham et al., 2011). There is limited evidence for RSL change associated with Stage 2 readvance(s), but some Stage 3 and 4 readvance(s) have been linked to increased isostatic rebound followed by an accelerated rate of RSL decline between 1.5 and 0.5 ka BP (Figure 7f) (Johnson et al., 2022;Simms et al., 2012). Raised beaches at <10 m a.s.l., are thought to be contemporaneous with Late Holocene readvances on KGI and Livingston Island between c. 0.5 and 0.25 cal. ka BP (Hall, 2010;Hall and Perry, 2004;Sugden and John, 1973). Restricted Late Holocene readvances on the Fildes Peninsula are consistent with the limited Late Holocene ice-loading scenario of the W12a GIA model (Whitehouse et al., 2012a(Whitehouse et al., , 2012b. Greater 'Neoglacial' ice-loading is broadly supported by field data, implying more substantial readvances and/or differential ice-mass (un)loading elsewhere on the SSI during the Late Holocene (cf. Fretwell et al., 2010).
Thick airfall tephra layers and >0.5 m thick gravity flow deposits in lake sediments from across the SSI dated to between c. 5.6 and 3.8 cal. ka BP have been linked to the Deception Island caldera-forming event (Figure 5c) (Antoniades et al., 2018;Roberts et al., 2017; Supplemental Material for details, available online). Interestingly, rhyolitic tephra associated with the three most explosive Deception Island eruptions of the Holocene (T7, T5, CC in Figure 5c) has been found in lake sediments from Fildes Peninsula shortly after the BIC had deglaciated close to/ within its present-day limits at c. 8.0-7.5 cal. ka BP (T7: 6.5-7.2 cal. ka BP), after c. 6.0 cal. ka BP (T5: 5.2-5.6 cal. ka BP), and after the c. 4.2 cal. ka BP readvance (CC: 3.98 ± 0.13 cal. ka BP) (Supplemental Table S1B and Supplemental Material for details, available online). More than one Deception Island caldera-forming event in the Holocene is considered unlikely due to the amount of time needed to fully recharge the magma chamber (Geyer et al., 2019). Instead, explosive (rhyolitic) eruptions from Deception Island generated by magma injection processes (Geyer et al., 2019) could have been triggered by increased crustal and magma chamber stress, associated with isostatic rebound, and increased magma interaction with sea water, associated with localised changes in sea level, during the Mid-to Late Holocene (cf. Forte and Castro, 2019;Maclennan et al., 2002;Praetorius et al., 2016;Satow et al., 2021). Further work is needed on the South Shetland Islands to investigate the relationship between Holocene deglaciation and volcanic activity.

Conclusions
We reconstructed the Mid-Late Holocene deglaciation and readvance history of the Fildes Peninsula, King George Island/Isla 25 de Mayo (KGI), South Shetland Islands, NW Antarctic Peninsula, using new cosmogenic exposure ages from glacial erratics on the foreland of the Bellingshausen Ice Cap (BIC), radiocarbon dating of terrestrial mosses and marine Laternula sp. bivalve shells found in moraines on the Artigas Beach and Valle Norte sectors of BIC foreland.
Deglaciation on KGI began after c. 15 ka and had progressed to within present-day limits on the Fildes Peninsula, its largest ice-free peninsula, between c. 6.6 and 5.3 cal. ka BP as spring/ summer insolation at 62°S gradually increased, leading to generally warmer and more positive SAM-like conditions during the early Holocene.
Using a novel statistical approach, we developed a new deglaciation and four-stage Holocene readvance model for the BIC on the Fildes Peninsula.
Probability density phase analysis of new (n = 25) and existing chronological data constraining Holocene glacier advance on the Fildes Peninsula and other ice-free peninsulas of KGI (n = 80 in total) revealed up to eight 95% probability 'gaps' when readvances could have occurred, which we grouped into four stages: • • Stage 1: a readvance and marine transgression, well-constrained by field data, between c. 7.4 and 6.6 cal. ka BP. • • Stage 2: four less well-constrained readvance probability 'gaps' between c. 5.3 and 4.8, 4.5-3.9, 3.3-3.0 and 2.6-2.2 cal. ka BP. • • Stage 3: a well-constrained readvance between c. 1.7 and 1.5 cal. ka BP when terrestrial mosses were embedded in the Shetland I moraine and the BIC extended landward of its present position on its eastern flank. • • Stage 4: two further readvances, one less well-constrained by field data between c. 1.3 and 0.7 cal. ka BP (68% probability) when moss fragments and Mid-Holocene marine sediments were translocated into the moraines surrounding the present-day BIC and a prolonged phase of elevated turbidity existed in Kiteschsee Lake, and then a well-constrained 'final' readvance after <0.7 cal. ka BP, consistent with the 'last readvance' on James Ross Island, NE Antarctic Peninsula.
Similarities in the timing of Mid-to Late Holocene glacier readvances on the SSI, the northern Antarctic Peninsula, some sub-Antarctic Islands and in southern South America imply underlying regional-hemispheric drivers were responsible. The Stage 1 readvance occurred at the Holocene global solar irradiance minimum as colder/more humid (negative) Southern Annular Mode (SAM)-like and stormier conditions developed, with marginally stronger (poleward shifted) westerly winds over the SSI.
Readvances were more frequent after c. 5.3 cal. ka BP, driven by reducing spring/summer insolation at 62°S and more persistent negative SAM-like conditions. However, weaker (equatorward shifted) westerlies and reduced storminess over the SSI restricted readvances close to present day limits.
Late Holocene readvances occurred around global solar irradiance minima and are anti-phased with aquatic moss layers in lake records from outer Fildes Peninsula lakes unaffected by glaciofluvial inputs.
Recovery from Late Holocene 'Neoglacial' conditions in Kiteschsee Lake and other mid-outer lakes on the Fildes Peninsula in the post-bomb era (>1950 CE) is characterised by the recolonisation of well-developed aquatic moss at the lake sediment-water interface. This was driven by late 20th Century/early 21st Century warming of the northern Antarctic Peninsula and more positive SAM-like conditions on KGI.