Drivers of Seasonal Land‐Ice‐Flow Variability in the Antarctic Peninsula

Land‐ice flow in Antarctica has experienced multi‐annual acceleration in response to increased rates of ice thinning, ice‐shelf collapse and grounding‐line retreat. Superimposed upon this trend, recent observations have revealed that land‐ice flow in the Antarctic Peninsula exhibits seasonal velocity variability with distinct summertime speed‐ups. The mechanism, or mechanisms, responsible for driving this seasonality are unconstrained at present, yet detailed, process‐based understanding of such forcing will be important for accurately estimating Antarctica's future contributions to sea level. Here, we perform time‐series analysis on an array of remotely sensed, modeled and reanalysis data sets to examine the influence of potential drivers of ice‐flow seasonality in the Antarctic Peninsula. We show that both meltwater presence and ocean temperature act as statistically significant precursors to summertime ice‐flow acceleration, although each elicits an ice‐velocity response after a distinct lag, with the former prompting a more immediate response. Furthermore, we find that the timing and magnitude of these local drivers are influenced by large‐scale climate phenomena, namely the Amundsen Sea Low and the El Niño Southern Oscillation, with the latter initiating an anomalous wintertime ice‐flow acceleration event in 2016. This hitherto unidentified link between seasonal ice flow and large‐scale climatic forcing may have important implications for ice discharge at and beyond the Antarctic Peninsula in the future, depending upon how the magnitude, frequency and duration of such climate phenomena evolve in a warming world.


Introduction
Extensive remote sensing observations throughout the satellite era have revealed the long-term acceleration of the Antarctic Ice Sheet (Diener et al., 2021;Gardner et al., 2018;Mouginot et al., 2014;Rignot et al., 2008;Shepherd et al., 2018), and in the Antarctic Peninsula, the recent availability of routine, high-temporal-resolution satellite imagery has revealed seasonal ice-flow variability superimposed upon this trend (Boxall et al., 2022a;Seehaus et al., 2016;Wallis et al., 2023).Critically, ice-flow variability over intra-annual timescales has not been accounted for within discharge-based calculations of ice sheet mass balance (Rignot et al., 2011;Shepherd et al., 2012).Thus, depending on the acquisition timing of the observations used, such calculations of Antarctica's annual mass loss may be over-or under-estimated given the hitherto overlooked seasonality in discharge (Boxall et al., 2022a;Wallis et al., 2023).
In situ and remotely sensed data have been used to attribute the multi-decadal ice sheet acceleration observed over the satellite era to a variety of surface and oceanic forcing mechanisms (Cook & Vaughan, 2010;Dutrieux et al., 2014;Joughin, Alley, & Holland, 2012;Paolo et al., 2018;Steig et al., 2012), but the precise processes controlling the Antarctic Peninsula's ice-flow seasonality remain unknown at present.Elucidating the drivers of this seasonality, and their relative contributions, will be important for improving predictions of Antarctica's future dynamic response to surface and ocean changes and for identifying additional locations in Antarctica where iceflow seasonality is likely to occur.Such knowledge will be essential for refining estimates of Antarctica's overall contribution to global sea-level.
Consistent with earlier research examining the drivers of seasonal velocity variability outside Antarctica (Hoffman & Price, 2014;Moon et al., 2014;Vijay et al., 2019), surface and oceanic forcing represent viable potential controls on the ice-flow seasonality recently observed in the Antarctic Peninsula (Boxall et al., 2022a;Wallis et al., 2023).Surface meltwater has been implicated as the key driver of seasonal grounded ice acceleration across several Arctic and Alpine regions (Andrews et al., 2014;Ehrenfeucht et al., 2022;Hoffman & Price, 2014;Hooke et al., 1989;Iken et al., 1983;Kraaijenbrink et al., 2016;Vijay et al., 2019).There, surface-warminginduced melt, and associated supraglacial streams, lakes and firn aquifers, supply water to the subglacial drainage system on a seasonal basis via crevasses and moulins, creating transient increases in basal water pressure and a near-instantaneous acceleration of the ice column (Banwell et al., 2016;Bartholomew et al., 2010;Iken, 1981;Schoof, 2010).In the late summertime, this phenomenon is often followed by a velocity slowdown as the subglacial system evolves to become more efficient (Sundal et al., 2011).Analogous meltwater-induced surface forcing may therefore be applicable to the grounded ice in the Antarctic Peninsula, although observations of meltwater drainage and accompanying velocity speed-up(s) supporting this hypothesis are scarce (Boxall et al., 2022a;Rott et al., 2020;Tuckett et al., 2019).
Oceanic forcing may be an additional, or alternative, control on the Antarctic Peninsula's ice-flow seasonality given that the entirety of the Bellingshausen Sea's continental shelf is flooded in relatively warm (1-2°C), highsalinity (∼34.7 PSU) circumpolar deep water (CDW) (Jenkins & Jacobs, 2008;Schmidtko et al., 2013).The presence of this water mass has in recent decades driven rapid rates of melting beneath the Bellingshausen Sector's adjoining ice shelves (Adusumilli et al., 2020), which in turn, has instigated the long-term land-ice dynamical thinning and acceleration trends detected from satellites (Gardner et al., 2018;Hogg et al., 2017).Previously modeled (Holland et al., 2010) and observed (Webber et al., 2017) seasonal cycles in CDW thickness driven by changing offshore sea-ice conditions, for example, could instigate seasonal variability in land-ice-flow.A dearth of long-term in situ oceanographic observations, however, currently precludes any direct insight into the viability of this mechanism for controlling seasonal flow.In the absence of such observations, detailed examination of all available remotely sensed-, model-and reanalysis-based data sets presents a route by which the relative importance of both surface and oceanic forcing mechanisms may be investigated further.
Previous research has revealed that in addition to local-scale surface and oceanic processes, large-scale climatic phenomena (including the El Niño Southern Oscillation (ENSO), the Amundsen Sea Low (ASL) and the Southern Annular Mode (SAM)) can induce significant atmospheric and oceanic variability across the Pacific-facing margin of Antarctica, including the Antarctic Peninsula (Clem et al., 2016;Hosking et al., 2013;Turner, 2004).Indeed, the influence of, for example, ENSO upon ocean-driven ice-shelf thinning and retreat in West Antarctica is now well documented (Christie et al., 2023;Dutrieux et al., 2014;Paolo et al., 2018;Walker & Gardner, 2017).It is therefore possible that such climatic phenomena might also play a role in driving the seasonal ice-flow variability recently observed in the Antarctic Peninsula.
Here, we examine the possible drivers of the seasonal ice-flow variability identified previously at, and immediately inland of, the grounding line of the outlet glaciers feeding George VI Ice Shelf (GVIIS), Antarctic Peninsula (Boxall et al., 2022a).To do this, we assess the correlation between these seasonal velocity records and a range of local-to large-scale climatic and oceanographic variables, each of which would be expected to presage seasonal speed-up.

Study Area
Our study area encompasses 16 grounded outlet glaciers draining to George VI Ice Shelf (GVIIS) from Palmer Land and Alexander Island on the western Antarctic Peninsula (Figure 1).There, previous research (Boxall et al., 2022a) has revealed distinct ice-flow accelerations of up to ∼22 ± 1.8 m yr 1 (∼15% of baseline rates) during austral summertime (DJF), superimposed upon multi-annual-to decadal-scale trends of acceleration (Section 1).Analysis of the seasonal dynamics of such glaciers is particularly important due to the ability of ice shelves to buttress vast volumes of inland ice, translating to a sea level equivalent of ∼15 cm in this particular region (Rignot et al., 2019).Similar to Boxall et al. (2022a), only glaciers with a modern-day grounding line (Boxall et al., 2022b;Mohajerani et al., 2021) were included in our analyses (black grounding line; Figure 1) to ensure the examination of land ice given its potential-unlike floating ice shelves-to directly contribute to sealevel rise.Our focus on fully grounded ice also eliminates tidal contamination of the ice-flow signal.To facilitate our examination of ocean conditions immediately offshore from GVIIS (Section 3.2.2), the Marguerite Bay and Ronne Entrance regions of the Bellingshausen Sea (Figure 1) are also incorporated within our study area.

Ice Velocity Records
The land-ice velocities of the outlet glaciers feeding GVIIS were acquired at a high-spatial resolution (200 m) using a combination of coherent and incoherent offset tracking techniques applied to all successive (6-/12-day repeat pass) Sentinel-1 Interferometric Wide (IW) single look complex image pairs acquired between October 2014 and December 2021 (Boxall et al., 2022a;Nagler et al., 2015Nagler et al., , 2021) ) (Figure 2a; Figure S1A in Supporting Information S1).Monthly composites of mean ice flow were produced from these repeat-pass image pairs alongside grids of uncertainty (1σ) and valid pixel count (the number of reliable observations utilized for generating each monthly estimate, excluding any no data values), from which per-pixel observation-weighted standard errors were calculated as in Boxall et al. (2022a) (see also Equation 1 below).Furthermore, following Boxall et al. (2022a), we removed any long-term trends from the monthly resolution time series.To do Translucent black shading shows land-ice regions typically associated with summertime surface layer melt as detected by ASCAT (Section 3.2.1).Ice sheet background is the Reference Elevation Model of Antarctica (Howat et al., 2019).Ocean background is BedMachine Antarctica, with the 500 m depth contour in black and remaining contours spaced by 100 m in gray (Morlighem, 2022).Map projection: ESPG:3031.Inset map shows location.
this, we computed a smoothed time series using a 13-month moving average and differenced it from the original time series to retain the short-term/seasonal signals only.

Local Drivers of Ice-Flow Seasonality
The known controls on multi-annual Antarctic, and seasonal Arctic/Alpine, glacier acceleration (Section 1) suggest that at local scales, Antarctic ice-flow seasonality is likely dominated by surface and oceanic forcing.To investigate the importance of these potential forcing mechanisms, continuous, high-spatial-and high-temporalresolution environmental data are required.Since comprehensive in situ records are historically scarce across GVIIS and its surroundings, we utilized model and reanalysis output and satellite-based observations as our primary data source.For comparability with our velocity records (Section 3.1), monthly data sets were used.In the following sub-sections (Sections 3.2.1 and 3.2.2),we outline the various environmental variables considered.

Surface Drivers
To assess the influence of atmospherically induced surface meltwater forcing on seasonal ice-flow variability, we examined both 2 m air temperature and surface melt extent data.Warm air temperatures, and the associated intensification of surface melt, indicate the enhanced probability of meltwater drainage and thus surface-forced acceleration (Section 1).Monthly 2 m air temperature data were acquired from four (semi-)independent modeled/ reanalysis data sets (Figure S1C in Supporting Information S1): the Regional Atmospheric Climate Model v2.3p2 (RACMO2.3p2;spatial resolution 0.25°× 0.25°; Van Wessem et al., 2018), the European Center for Medium-Range Weather Forecast's fifth-generation atmospheric reanalysis (ERA5; spatial resolution 0.25°× 0.25°; Hersbach et al., 2023), the Modern Era Retrospective-analysis for Research and Applications v2 (MERRA2; spatial resolution 0.625°× 0.5°; Gelaro et al., 2017) and the Japanese 55-year Reanalysis (JRA-55; spatial resolution 1.25°× 1.25°; Kobayashi et al., 2015) (Table S1 in Supporting Information S1).RACMO2.3p2 and ERA5 are considered semi-independent because the former is forced by ERA5.Utilizing these data sets, we then created a monthly 2 m air temperature ensemble mean time series (Figure 2b) (and associated uncertainty; Section 3.2.3) to reduce the weighting of any potential bias(es) associated with the individual data sets (Carter et al., 2022).Using the ensemble mean, average monthly air temperature across the GVIIS region ranged from 24 to 1°C.
To supplement our 2 m air temperature records, we also used a daily, 4.45-km resolution melt data set that utilizes the backscatter time series from ASCAT Enhanced Resolution Scatterometer Image Reconstruction products (Lindsley & Long, 2010) to identify the presence of liquid water in the upper layer of the snowpack.We distinguish melt stages by applying a hierarchical decision tree with dynamic thresholds based on that year's winter reference season and the temporal gradient of the backscatter coefficient.The backscatter signal decreases significantly when the snow surface starts melting (liquid water content of ∼≥1% volume), so a rapid drop in the backscatter coefficient is utilized to identify the occurrence of melt.This data set categorizes surface melt as either "surface layer melt" or "wet snow layer" (Figure 2c; Figure S1B in Supporting Information S1), from which we calculated the monthly average percentage melt extent, and associated uncertainty (Section 3.2.3),for use in our analyses.Across the glaciers, the monthly melt extent ranged between 0% and 100%.Unlike our air temperature and ocean records, no other comparable surface meltwater data sets exist with which to derive an ensemble average product, hence our exclusive use of the ASCAT time series to represent surface meltwater.All other preexisting meltwater products have either a coarser resolution (∼25 km) that prohibits detailed spatial analysis (Johnson et al., 2022) or focus on the detection of supraglacial lakes, thereby excluding any meltwater that does not saturate the snowpack (Dirscherl et al., 2021).

Oceanic Drivers
To evaluate the influence of oceanic forcing mechanisms on ice-flow seasonality, we examined a selection of ocean temperature and sea ice concentration records.Monthly ocean temperature data were acquired from eight (semi-)independent ocean reanalysis products (Figure S1D in Supporting Information S1): four component products of the Copernicus Marine Environment Monitoring Service's Global Ocean Reanalysis Ensemble Product (GREP; spatial resolution 0.25°× 0.25°; vertical resolution 75 layers; Blockley et al., 2014;Garric & Parent, 2017;Storto et al., 2016;Zuo et al., 2019); the Global Ocean Reanalysis and Simulation 12V1 (GLORYS12V1; spatial resolution 0.083°× 0.083°; vertical resolution 50 layers; Lellouche et al., 2021), the Estimating the Circulation and Climate of the Ocean (ECCO; spatial resolution 1°× 1°; vertical resolution 50 layers; Forget et al., 2015), the Biogeochemical Southern Ocean State Estimate (BSOSE; spatial resolution 0.16°× 0.16°; vertical resolution 52 layers; Verdy & Mazloff, 2017) and the Global Ocean Data Assimilation System (GODAS; spatial resolution 0.33°× 0.33°; vertical resolution 40 layers; Behringer et al., 1998) (Table S1 in Supporting Information S1).Similar to our handling of the 2 m air temperature records detailed in Section 3.2.1 and given the variability in the degree of seasonality between ocean reanalysis data sets, we averaged across these eight data sets to create a multi-product monthly mean ensemble of ocean temperature (Figure 2d) and associated uncertainty (Section 3.2.3).In our analyses, we only utilized data residing at depths ≥300 m where CDW is known to reside underneath GVIIS (Holland et al., 2010;Jenkins & Jacobs, 2008) (thereby excluding GODAS due to a lack of data availability (Figure S2 in Supporting Information S1)).This threshold also corresponds to the depth of the grounding line associated with each of the glaciers studied, whose basal elevations range between 250 and 800 m below the mean sea level (Morlighem, 2022).As for our ice velocity records (Section 3.1; Boxall et al., 2022a), long-term trends in ocean temperature (Figure S2 in Supporting Information S1) were removed using a 13-month moving average prior to calculating the ensemble mean in order to isolate seasonal behavior (Figure 2d).The subtle seasonality observed in Figure 2d is attributed to the smoothing applied during the assimilation of limited in situ observations in the production of each ensemble member.
For sea ice concentration, monthly average data were acquired from the Global Ocean Reanalysis and Simulation 12V1 (GLORYS12V1; spatial resolution 0.083°× 0.083°; Lellouche et al., 2021) and from three independently processed remote sensing-based indices: the University of Bremen's AMSR2 ASI sea ice concentration data set (spatial resolution 3.125 km; Spreen et al., 2008), NSIDC's Sea Ice Index (SII; spatial resolution 25 km, Fetterer et al., 2017) and the NOAA/NSIDC Climate Data Record (spatial resolution 25 km; Meier et al., 2021) (Table S1; Figure S1E in Supporting Information S1).Similar to our handling of the air and ocean temperature data described above, we averaged these data sets to produce a monthly mean ensemble time series of sea ice concentration (Figure 2e) and calculated uncertainty as described in Section 3.2.3.Overall, sea ice concentration varied between ∼5% and ∼90% offshore of GVIIS' fronts.

Time Series Production and Uncertainty
Following the generation of two-dimensional monthly time series spanning 2014 to 2021 for ice velocity and each environmental driver described above (Sections 3.2.1 and 3.2.2),we converted these records into onedimensional time series per glacier to facilitate detailed time series analysis (Section 3.4).To produce these time series, a 10 km 2 region located directly upstream of the grounding line of each glacier (where ice-flow seasonality is known to be at its strongest (Boxall et al., 2022a)) was utilized (Figure 1; magenta dashed boxes) within which ice velocity and air temperature were averaged, and mean meltwater percentage extent calculated for each monthly time step.In our ensemble ocean product, temperature is not resolved beneath GVIIS due to a lack of cavity representation in all eight ensemble members, necessitating the spatial averaging of data proximal to the ice front as the best available alternative.To do this, we averaged all ocean temperature records ≥300 m within a 200 km 2 region located in Ronne Entrance adjacent to GVIIS' southern limits (Figure 1; magenta dashed boxes).Ocean characteristics at that location are deemed to be more representative of the conditions at the outlet glaciers' grounding line than those in Marguerite Bay, where waters are believed to be modified by the entrainment of fresh meltwater from beneath GVIIS into the net northerly throughflow within the cavity (Jenkins & Jacobs, 2008).For consistency, our sea ice concentration records were also averaged within the same 200 km 2 region located proximal to Ronne Entrance although for completeness, ocean temperature and sea ice concentration time series were also produced for the 200 km 2 region located in Marguerite Bay (Figure S1D and S1E in Supporting Information S1).
To estimate the uncertainty associated with each of our environmental records, monthly standard error (SE) over each averaging region described above was calculated according to: where σ and n denote the standard deviation and count (i.e., number of values averaged to produce a monthly mean (ensemble) time series), respectively.For ice velocity and meltwater extent, a monthly datapoint was calculated by averaging successive 6/12-day velocity pairs (Section 3.1) or daily ASCAT-derived melt extent (Section 3.2.1),respectively.The standard error was thus calculated from the variability across the sub-monthly data for each month.For air temperature (Section 3.2.1),ocean temperature and sea ice concentration (Section 3.2.2),monthly datapoints were calculated by averaging across the monthly data sets contained within the ensemble.The standard error was thus calculated from the variability across the individual data sets comprising the ensemble.The small variability between air temperature and between sea ice concentration data sets yielded relatively small standard errors.For our ocean ensemble data set, calculated standard error values are excessively high (∼0.8°C)owing to the non-uniform nature of the in situ observations used in the production of each ensemble member through space and time, and so are believed to be non-physical.These records are, however, the best available characterization of the time-varying oceanic conditions near GVIIS and its contributing glaciers; therefore, we retain their use and for clarity do not show their associated error bars in Figure 2.

Ice-Shelf Thickness Change
To supplement our ice velocity and environmental data sets, we examined multi-sensor radar-derived ice-shelf thickness change near the grounding line, which is itself a proxy for the ice-shelf melt needed to initiate seasonal ice-flow variability via external forcing.To do this, we utilized a spatially resolved data set of the rate of iceshelf thickness change available at a 3-monthly resolution between 1992 and 2018 calculated from high-precision ice-shelf height observations corrected for firn air content change (Adusumilli et al., 2020).The temporally coarse nature of this data set prohibited its inclusion in our monthly cross-correlation analyses (Section 3.4).Instead, seasonal anomalies were computed and analyzed directly.A seasonal anomaly was defined as the difference between the mean rate of ice-shelf thickness change for a particular season (e.g., the wintertime average between 1992 and 2018), and the mean rate across the entire 26-year time series.Using this methodology, the resulting anomalies emphasize the degree to which any thickness change observed during a given season is distinct from the long-term (i.e., complete time series) average.

Large-Scale Drivers of Ice-Flow Seasonality
To understand the role of large-scale climate as a driver of the observed seasonal ice-flow behavior, we also examined the relationship between SAR-derived ice velocity (Figure 2a) and the occurrence of ENSO, ASL and SAM utilizing several established climate indices: the Oceanic Niño Index (ONI) (Huang et al., 2015; Figure 2f), the Amundsen Sea Low (ASL) Climate Index (longitudinal location variable; Hosking et al., 2016; Figure 2g) and the Marshall Southern Annular Mode (SAM) Index (Marshall, 2003; Figure 2h), respectively.The ONI is the rolling 3-month average temperature anomaly averaged over the central equatorial Pacific (5°N-5°S, 170°W-120°W) and is preferred over other commonly reported metrics (such as the 5-month averaged Niño 3.4 index) for our purposes.Index values exceeding 0.5°C, sustained over at least five consecutive months, indicate the occurrence of an El Niño event in the ONI index.The longitudinal location variable of the ASL Climate Index was selected given its importance in controlling the surface climate over West Antarctica versus, for example, its latitudinal location and depth (Hosking et al., 2013).The ASL longitude is identified from the location of the ERA-Interim mean sea level pressure minima within the ASL sector (80°-60°S, 170°-298°E).Finally, the SAM index indicates the zonal pressure difference between the latitudes of 40°S and 65°S, whereby positive values correspond with stronger-than-average westerlies over the mid-high latitudes.

Cross-Correlation
Using our one-dimensional time series for each glacier (Figure 2; Figure S1 in Supporting Information S1), we cross-correlated ice velocity with each of the local and large-scale environmental variables detailed above, excluding ice-shelf thickness change.We computed the Pearson standard correlation coefficient (McKinney, 2010) to assess the strength of the linear relationship between variables.While we acknowledge that perfect linearity may not exist between ice velocity and each of the environmental variables examined, previous work in Svalbard and Greenland has found linear relationships between atmospheric or oceanic drivers and glacier behavior (Bartholomew et al., 2010(Bartholomew et al., , 2011;;Cowton et al., 2018;Joughin et al., 2020;Luckman et al., 2015;Ultee et al., 2022), suggesting that linearity is a valid assumption for the purpose of this study.Moreover, should a nonlinear relationship exist, the linear correlation coefficient would underestimate the strength of the relationship, negating the possibility of overstating the connection between variables.A time lag (k) was introduced to establish how the magnitude of correlation changed as each driver was shifted in time with respect to the ice velocity.For this purpose, we incrementally offset the environmental series backwards in time (0 ≤ k ≤ 12); as such, a strong correlation at a positive lag (k > 0) signified that any change within the environmental variable time series preceded a change in the velocity signal by k months.Assuming no autocorrelation, the coefficient was significant with 95% confidence if it exceeded: where N and k denote series length and time lag, respectively (after Ultee et al., 2022).
Journal of Geophysical Research: Earth Surface 10.1029/2023JF007378 BOXALL ET AL.
Since our time series were autocorrelated at 1 month (implying that they correlate to versions of themselves lagged by 1 month; Figure S3 in Supporting Information S1), we corrected the significance limits accordingly.Following Dean and Dunsmuir (2016), the assigned significance limit was altered by a Correction Factor (CF) where a and b denote the autocorrelation at a 1-month lag in the velocity and environmental series, respectively.
For each glacier, we next identified the statistically significant correlation coefficient with the largest magnitude, and its corresponding lag, (k), for each environmental variable.It was not possible to supplement this analysis with, for example, season-specific correlations as seasonal subsets of the time series did not contain enough datapoints from which robust conclusions could be drawn.Finally, using the lagged time series that yielded the largest coefficients, we calculated partial correlations between ice velocity and each environmental variable.Partial correlations were calculated as the correlation between the residuals when two variables (ice velocity and an environmental variable) were each regressed with a third variable (another environmental variable).In other words, partial correlations determine whether a relationship exists between ice velocity and a given environmental variable once any influence from another environmental variable is removed.The lack of a significant partial correlation where a significant standard correlation coefficient exists therefore suggests that the relationship between velocity and a given environmental variable is indirect due to the influence of an additional variable.

Ice Flow Response to Local Surface and Oceanic Forcing
The outlet glaciers draining to GVIIS exhibit clear seasonal ice-flow variability with distinct austral summertime (December-February) speed-ups constituting a ∼15% acceleration relative to baseline rates of flow (Figure 2a; Figure S1A in Supporting Information S1; Boxall et al., 2022a).Similar summertime maxima are found in all local environmental driver time series (Figures 2b-2e and Figure S1B-S1E in Supporting Information S1) apart from sea ice concentration, which exhibits minima centered around February/March commensurate with increased radiative forcing.The highest statistically significant, lagged correlation coefficients between ice velocity and each environmental variable are summarized in Table 1, and all statistically significant coefficients are provided in Table S2 in Supporting Information S1.

Surface Forcing
Surface meltwater extent typically possesses the strongest correlation with ice velocity (mean (x) = 0.59; Table 1).All 16 glaciers have their strongest correlations with meltwater extent at either a 0-or 1-month lag.The highest correlation occurs most consistently at a 1-month lag, implying that trends in surface meltwater extent precede those in ice velocity by 1 month (blue shaded panels; Figure 2c).For some glaciers, particularly those on Alexander Island (Figure 1), surface meltwater extent and ice velocity behave synchronously, as demonstrated by the largest correlations occurring with no lag (Table 1).Similarly, strong positive correlations are found between ice velocity and air temperature (x̄= 0.5; Table 1).Indeed, 10 of the 16 outlet glaciers possess their strongest correlation between ice velocity and air temperature at a 0-, 1-, or 12-month lag (where a 12-month lag is interpreted as no lag given the 12-month seasonal cycle), indicating that air temperature trends either occur coincidently with those in ice velocity or precede them by 1 month (blue shaded panels; Figure 2b).Similar to meltwater extent, the synchronous behavior tends to occur on Alexander Island's glaciers.The relationship between velocity and air temperature is indirect given few significant partial correlations exist once the effect of meltwater is removed (Table S3 in Supporting Information S1).Conversely, the relationship between velocity and surface meltwater extent is direct, given that consistent significant partial correlations exist when the effect of air temperature is removed (Table S3 in Supporting Information S1).

Oceanic Forcing
Turning to the influence of oceanic forcing mechanisms on ice-flow, the largest correlations between ice velocity and sea ice concentration are consistently negative, with moderate strength for all outlet glaciers (x̄= 0.54; Table 1).Such correlation often exists with no lag, demonstrating that low sea ice concentrations occur concurrently with ice-flow acceleration in the austral summertime (Table 1).For glaciers fringing Alexander Island, the associated lag is 10-11 months, implying that low sea ice concentrations occur 10-11 months prior to ice acceleration or, alternatively, 1-2 months after peak velocity (i.e., February-March).Perhaps more importantly, we further note that significant positive correlations exist between ice velocity and sea ice concentration when the latter is offset backwards by 6 months (x̄= 0.34; Table S2 in Supporting Information S1), suggesting that sea ice maxima is also a significant precursor to summertime speed-up (red shaded panels; Figure 2e).
Similar to the relationship between ice velocity and sea ice concentration maxima, ice velocity and 300+ m ocean temperature are mostly positively correlated with moderate strength at a 2-5-month lag (x̄= 0.52; Table 1).In the minor number of instances where these variables are negatively correlated, the strength is comparable (x̄= 0.46) and the associated lag is 7-10 months.Given that the ocean temperature time series is approximately sinusoidal with a 12-month period (i.e., the maxima follows the minima by approximately 6 months), we note that glaciers exhibiting 7-10-month lags are, however, by default also associated with positive correlations in ocean temperature at 2-5-month lags (Table S2 in Supporting Information S1).Collectively, these results demonstrate that high ocean temperatures lead ice acceleration by 2-5 months and low ocean temperatures precede ice acceleration by 7-10 months.This ultimately suggests that oceanic forcing, governed by the timing of ocean warming at depth, is at its most dominant across all glaciers during late winter/early spring, that is, 2-5 months prior to the observed summertime speed-up in ice velocity (red shaded panels; Figure 2d).For the majority of glaciers, however, a significant partial correlation does not exist between ice velocity and either ocean temperature (once the effect of sea ice concentration is removed) or sea ice concentration (once the effect of ocean temperature is removed) (Table S4 in Supporting Information S1), signifying that these relationships are indirect, with some of the variance being explained by additional variable(s).
The lack of any significant partial correlation between ice velocity and ocean temperature or sea ice concentration noted above suggests the influence of an additional variable controlling ice-flow seasonality.Figure 3 indicates that this additional variable is likely to be ocean-driven ice-shelf thickness change (Section 3.2.4).Indeed, satellite radar altimetry data reveal that the greatest rate of ice-shelf thinning over GVIIS occurs during austral springtime, that is, several months prior to summertime ice-flow acceleration, and concurrent with the late winter/ early springtime oceanic warming observed (Figure 3).At the ice shelf-scale, the average springtime thickness anomaly is 0.35 m yr 1 , compared to 0.19 m yr 1 , 0.12 m yr 1 and -0.17 m yr 1 for summer, autumn, and winter, respectively.The concurrence of the greatest rates of ice-shelf thinning in springtime alongside ocean warming (Table 1; Figure 2d) strongly implicates the role of enhanced ocean-induced basal melting as the dominant control upon this phenomenon.Such ice-shelf thinning might also imply intensified surface melting and runoff (Banwell et al., 2021); however, negligible surface melt extent during springtime over each glacier (Figure 2c; Figure S1B in Supporting Information S1) suggests that these processes are likely insignificant.We also disregard ice divergence (dynamical thinning) as a significant contributor to the observed ice-shelf thickness change, given that the observed anomaly (Figure 3) is similar in magnitude to the basal melt term as documented by Adusumilli et al. (2020).

Ice Flow Response to Large-Scale Forcing
In addition to the local-scale surface and oceanic forcing relationships detailed in Section 4.1, our analyses further reveal that ice velocity inland of GVIIS has a consistent and significant relationship with the longitudinal location of the Amundsen Sea Low (ASL; Table 1).At 10 out of 16 of the outlet glaciers, the longitude of the ASL (Figure 2g) has a weak positive correlation with ice velocity (Figure 2a) (x̄= 0.34; Table 1), with the ice velocity signal lagging by 0-3 months.Conversely, ice velocity across all 16 glaciers is only sporadically correlated with ONI (Figure 2f) and the SAM Index (Figure 2h) (Table 1).

Discussion
Our results demonstrate that local-scale surface and oceanic drivers are both significantly correlated with GVIIS outlet glacier velocity, but the response time between external forcing and ice-velocity acceleration varies.Iceflow acceleration occurs (near-)simultaneously with heightened surface meltwater extent and air temperature (lag = 0-1 month), but lags warm ocean temperatures by 2-5 months and sea ice maxima by 6 months (Table 1; Section 4.1).Our results also reveal the potential importance of large-scale forcing mechanisms in controlling iceflow seasonality in the Antarctic Peninsula, particularly the longitude of the ASL (Section 4.2).
On the basis of our findings, we next elaborate upon the physical processes responsible for driving the intraannual ice-flow variability observed at GVIIS' outlet glaciers, and discuss the probable interactions at work between local-and large-scale driving mechanisms.To augment our discussion, Figure 4 summarizes the physical processes discussed in the following sections.

Surface Forcing
The increase in surface melt extent and (near-)simultaneous summertime speed-up we observe inland of GVIIS mirrors closely the (near-)instantaneous ice-flow acceleration that occurs in response to supraglacial meltwater inputs on Arctic and Alpine glaciers (Andrews et al., 2014;Das et al., 2008;Ehrenfeucht et al., 2022;Hoffman & Price, 2014;Hooke et al., 1989;Iken et al., 1983;Joughin et al., 2008;Kraaijenbrink et al., 2016;Vijay et al., 2019).As alluded to in Section 1, there, summertime supraglacial meltwater input to the subglacial drainage system drives a reduction in effective pressure and an associated increase in basal sliding, which in turn leads to the seasonal acceleration of the ice column.In Greenland, such meltwater input is often attributed to the growth of supraglacial lakes, which instigates a process of hydrofracture and subsequent lake drainage (Das et al., 2008;Doyle et al., 2013;Stevens et al., 2015;Tedesco et al., 2013).
In comparison to the Greenland Ice Sheet, the presence of ponded surface water inland of the Antarctic Peninsula's grounding zone is minimal, as determined from both optical-and radar-based satellite imaging (Dirscherl et al., 2021;Figure S4 in Supporting Information S1).Inland of GVIIS, significant ponding is not initiated as insufficient meltwater is generated to exceed the pore space available within the snowpack (van Wessem et al., 2023).The presence and drainage of perennial firn aquifers -similar to that detected in Greenland (Koenig et al., 2014;Poinar et al., 2017), near the northern ice front of GVIIS (Miller et al., 2022), and on neighboring Wilkins Ice Shelf (Miller et al., 2023;Montgomery et al., 2020) -could provide an alternative means of hydrofracture and meltwater delivery to the bed.We note, however, that the topography of GVIIS' contributing glaciers may be too steep to allow for the formation and persistence of such aquifers (van Wessem et al., 2021).
Despite the lack of significant meltwater ponding at the surface of GVIIS' outlet glaciers noted above and in Figure S4 in Supporting Information S1, our ASCAT-derived measurements reveal clear signals of summertime meltwater presence beneath the snow surface (Figure 2c).The lack of surface ponding ultimately implies that this meltwater presence is limited, and thus we expect that it is unlikely to initiate hydrofracture without a pre-existing store of additional meltwater (Krawczynski et al., 2009;van der Veen, 2007).Perennial englacial drainage features beneath the snowpack, such as water-filled crevasses that persist year-round, could provide an important store for such meltwater (Catania & Neumann, 2010) and would remain undetected in satellite imagery of the icesheet surface (Figure 4a).Inspection of our daily ASCAT-derived observations reveal that both melt and refrozen layers exist within the snowpack throughout the summer (Figure S5 in Supporting Information S1), suggesting an intermittent cycle of melt and refreezing on a sub-monthly timescale.Such melt, when not obstructed by the presence of a refrozen layer, could drain to -and subsequently fill -any such underlying englacial drainage features throughout the summer (or possibly over several consecutive summers).Once an increasingly water-filled englacial drainage feature nears its threshold for hydrofracture, the summertime contribution of only limited surface layer melt from the snowpack could act as a catalyst for the initiation of hydrofracture (Krawczynski et al., 2009;Van der Veen, 1998, 2007;Weertman, 1973).Ultimately, we expect that this phenomenon would facilitate the rapid delivery of surface layer melt -alongside stored meltwater -to the bed, causing (near-) instantaneous glacier speed-up (Figure 4b).A high density of such englacial drainage features would allow such acceleration to occur even if only a small proportion of the persistent englacial features surpass the threshold for drainage each year.

Oceanic Forcing
Following Section 4.1.2,the velocities of the outlet glaciers draining to GVIIS correlate significantly with ocean temperature and sea ice concentration, implicating oceanic forcing as another key potential driver of their seasonality.Our findings show that summertime ice acceleration occurs 2-5 months after increased ocean temperatures at depth (and associated maximum ice-shelf thinning rates; Figure 3), and around 6 months after rapid seasonal sea-ice growth and concentration maxima.Minimum sea ice concentration is also significantly correlated with summertime acceleration (Table S2 in Supporting Information S1) although, as discussed later, we do not believe that this correlation represents causality.Collectively, our findings signal the importance of sea-icemodulated oceanic forcing in driving the observed seasonal accelerations.Following Boxall et al. (2022a), we discard atmosphere-ocean variability as a controlling factor on ocean temperature in Ronne Entrance (and, by implication GVIIS' cavity) due to the relative insensitivity of the Bellingshausen Sea to atmospherically induced influxes of CDW (Christie et al., 2023;Holland et al., 2010).
Our ensemble data set reveals that deep (≥300 m) ocean temperatures within the Ronne Entrance (Figure 1) are consistently greater than 1°C (Figure S2 in Supporting Information S1).Such ocean conditions align closely with in situ oceanographic observations acquired offshore from GVIIS (Jenkins & Jacobs, 2008;Schmidtko et al., 2013), and are indicative of the widespread presence of warm, highly saline CDW around and underneath the ice shelf.As indicated in Section 4.1, the presence of such water can facilitate the vigorous ocean-driven melting of ice-shelves from below, as witnessed throughout the satellite era in West Antarctica (Adusumilli et al., 2020;Shepherd et al., 2018).
With the above in mind, previous ocean modeling experiments have revealed the significance of sea ice in modulating CDW thickness on the Bellingshausen Sea continental shelf near GVIIS (Holland et al., 2010), a process which presents a plausible mechanism with which to explain the seasonal acceleration patterns observed.Holland et al. (2010) show that wintertime sea-ice growth and associated brine rejection result in enhanced convection throughout the ocean mixed layer, leading to a thickening of the CDW layer beneath.This mechanism is further supported by in situ observations in the neighboring Amundsen Sea that demonstrate the uplift of warm waters in response to sea ice presence (Dotto et al., 2022).Thus, we interpret the seasonality in the ensemble ocean temperature presented in Figure 2d and Figure S1D in Supporting Information S1 to be attributable to the changing depth of the CDW layer throughout the year (Figure S2 in Supporting Information S1), similar to that observed in the Amundsen Sea (Webber et al., 2017).In effect, CDW thickness follows the trend in sea ice concentration, whereby the warmest CDW temperatures at depth occur during late wintertime/early springtime, which is shortly after sea ice concentration, and the volume of brine rejected, are at their maximum (Figure 4a).
The 2-5-month lag between ocean warming and ice-velocity response (Section 4.1) can be explained by preexisting knowledge of CDW circulation underneath GVIIS.Specifically, Boxall et al. (2022a) inferred that a sub-shelf, south-to-north CDW throughflow of ∼8-16 cm s 1 would be required over the course of 1-2 months to induce the summertime acceleration events observed along the length of GVIIS' coastline (assuming linear and laminar flow and near-synchronous ocean warming and land-ice-acceleration).For the 2-5-month lag times between sea-ice-induced ocean warming and ice-flow acceleration presented in this study, this equates to flow rates ranging between ∼3 and 8 cm s 1 : values towards the lower end of the estimate made by Boxall et al. (2022a), and in alignment with earlier throughflow estimates (∼2.5 cm s 1 ) derived from temporally limited in situ records at GVIIS's southern ice front (Jenkins & Jacobs, 2008).Together, these findings provide important evidence for the influence of sea-ice-enabled oceanic forcing upon GVIIS' seasonal ice-flow behavior.
At shorter lags, the (near-)simultaneous occurrence of ice acceleration and sea ice minima is, by comparison, not likely to represent a causal relationship.Unlike other ice shelves in the Antarctic Peninsula, GVIIS' stress regime is predominantly compressive, meaning that it is unlikely to respond instantaneously to reduced summertime seaice buttressing (Christie et al., 2022;Massom et al., 2018).Moreover, the instantaneous summertime sea iceocean-land ice interactions required theoretically to induce such rapid accelerations (Section 4.1) would be expected to coincide with the greatest rates of ice-shelf basal melting, which is inconsistent with our observations (Figure 3).As alluded to above, throughflow rates of ∼16 cm s 1 would also be required to initiate an ice acceleration response to sea ice minima over the course of one month.While such CDW throughflow rates are not unfeasible on the basis of observations elsewhere in Antarctica (Boxall et al., 2022a;Jacobs et al., 2013;Jenkins et al., 2018), we expect that the time required to induce ice-shelf melting, and associated debuttressing-induced acceleration, would take much longer (Joughin, Smith, et al., 2012;Rignot et al., 2004).

Surface Versus Ocean Forcing
Sections 5.1 and 5.2 present two mechanisms by which the surface and the ocean, respectively, may drive the observed ice-flow seasonality of GVIIS' outlet glaciers (Figure 4).Each mechanism is supported by statistically significant correlations (Table 1), but given their similarities in magnitude, the relative importance of either mechanism is difficult to determine from the correlation coefficients alone.In reality, we expect that the observed ice-flow seasonality is likely to be driven by both mechanisms acting in concert, as discussed next.
The importance of surface forcing is signified by the similarity in waveform between meltwater presence and near-contemporaneous ice acceleration (i.e., rapid onset and relatively short-lived peaks) (Figure 2; Figure S1 in Supporting Information S1).Assuming that these summertime meltwater increases lead to enhanced basal lubrication via hydrofracture, then surface forcing can also explain the aforementioned earlier onset of ice acceleration on Alexander Island relative to Palmer Land (Section 4.1; Figure S1 in Supporting Information S1; Boxall et al., 2022a), given that the relatively thinner glacier ice on Alexander Island (Morlighem, 2022) would require less meltwater, and thus less time, to hydrofracture.As noted in Section 5.1, however, the scarcity of ponded meltwater above the grounding line (Dirscherl et al., 2021;Figure S4 in Supporting Information S1) renders the creation of hydrofracture-enabled surface-to-bed connections improbable without some existing store of meltwater, and cannot easily explain the distinct seasonal velocity signals observed over successive seasons without fail.
Reconciling the above, we expect that ocean-driven basal melting at the grounding line augments the effects of surface forcing on glacier flow by: (a) reducing ice-shelf buttressing, (b) enhancing ice-shelf flexure and crevassing/damage and, thus, (c) increasing the propensity for hydrofracture (and, by implication, the ability of perennial, water-filled englacial drainage features to propagate downwards to the grounding line with only minimal additional summertime surface meltwater input; Section 5.1).Oceanic forcing also likely contributed to the earlier speed-up of Alexander Island's glaciers compared to those fringing Palmer Land (Section 4.1; Figure S1 in Supporting Information S1; Boxall et al., 2022a).As noted in Section 3.2.3, the strongest CDW inflows beneath GVIIS occur from Ronne Entrance at the northern margin of the southern ice front (Jenkins & Jacobs, 2008), where Coriolis-driven deflection of CDW toward Alexander Island is believed to maximize ice-shelf melting along the island's grounding line (Boxall et al., 2022a).
While the evidence presented above implicates the ocean as an important potential forcing mechanism in its own right, mass conservation calculations (Figure S6; Table S6 in Supporting Information S1) reveal that contemporary ice-shelf thinning rates (Figure 3; Adusumilli et al., 2020) are an order of magnitude smaller than those required to induce the observed acceleration.This finding suggests that oceanic forcing likely only contributes to ∼10% of the seasonal acceleration signals observed at GVIIS, further emphasizing the likely importance of combined surface and oceanic forcing in driving the seasonality of GVIIS's outlet glaciers.

Large-Scale Forcing
Beyond local-scale influences, the significant positive correlations we detect between the ASL and the velocity of GVIIS' outlet glacier (Section 4.2) present strong evidence for the additional role that large-scale climatic forcing can play in controlling Antarctic ice-flow seasonality.We note, however, that the average strength of these correlations (x̄= 0.34; Table 1) is less than those observed between ice velocity and the local environmental drivers examined (∼0.4-0.6;Table 1).Thus, our findings are suggestive of an indirect relationship between ice velocity and large-scale climatic forcing, whereby the latter influences local environmental factors (Dutrieux et al., 2014;Marshall et al., 2011;Raphael et al., 2016;Raphael & Hobbs, 2014;Steig et al., 2012;Turner, 2004;Turner et al., 2017;Yadav et al., 2022;Yuan, 2004) which, in turn, provide the principal control on ice-flow seasonality.
The positive relationship between ASL longitude (Figure 2g) and ice velocity (Figure 2a) is explained by the seasonal migration of the ASL.In general, the ASL's zonal position is centered to the west (∼200°E) over the Amundsen Sea during wintertime, before shifting eastwards toward the Bellingshausen Sea and the Antarctic Peninsula (to ∼260°E) during the summertime (Hosking et al., 2016; Figure 2g).The longitude of the ASL provides an important control on atmospheric meridional circulation in West Antarctica and the Peninsula, and therefore the climate of these regions more broadly (Hosking et al., 2013).During austral wintertime, when the ASL is located westwards, continental intrusions of cold and dry air occur over the Antarctic Peninsula (Hosking et al., 2013).Conversely, conditions are warmer and wetter over the Peninsula during the austral summertime when the ASL resides farther east.In this way, the intra-annual zonal migration of the ASL plays a significant role in driving the seasonal behavior of our proposed local environmental drivers.Specifically, the easterly location of the ASL during summertime results in warmer air temperatures (Figure 2b) and thus increased meltwater extent (Figure 2c) over the western Peninsula (Hosking et al., 2013), while its westerly wintertime location results in increased sea ice concentration (Figure 2e).The relationship between the ASL and our proposed local-scale surface drivers explains the observed lag of 0-3 months between its longitude (Figure 2g) and GVIIS' outlet glacier velocity response, given that this lag includes the time required to initiate local-scale surface forcing, which, in turn, drives the ice velocity speedups observed after a 0-1 month lag (Section 4.1).
In terms of larger-scale climate circulation, the lack of any consistent, significant correlation between ice-flow seasonality and ENSO (Section 4.2) is explained by the intra-decadal timescale over which ENSO operates (Philander, 1989).That is, unlike the seasonal migration of the ASL, ENSO operates over a period of 3-7 years, and only one El Niño event occurred within the time span of this study (2014)(2015)(2016)(2017)(2018)(2019)(2020)(2021).Moreover, ENSO is believed to play only a limited role in controlling Antarctic climate seasonality through its interaction with the ASL, given that the phase of ENSO prompts no statistically significant difference in the zonal location of the ASL (Turner et al., 2013).The ENSO phase (and amplitude) does, however, exert some control over ASL depth (whereby anomalous pressure deepening occurs over West Antarctica during La Niña events; Raphael et al., 2016), although the resulting Antarctic climate fluctuations induced by ASL depth variability are not statistically significant (Hosking et al., 2013).The timescale of ENSO-induced forcing is thus distinct from the seasonal forcing provided by local drivers and the intra-annual migration of the ASL.Similar to ENSO variability, SAM operates over timescales of decades to centuries (Fogt & Marshall, 2020) and exerts control over ASL depth but not its zonal location (Clem et al., 2016), potentially explaining its limited influence on seasonal climate and, by extension, ice flow at GVIIS (Section 4.2).
Despite the lack of any consistent seasonal ice-flow response to ENSO and SAM, a connection between these climatic indices and Antarctic ice flow cannot be disregarded entirely.That is, a distinct, isolated wintertime velocity acceleration event (green shading, Figure 2a) exists on most outlet glaciers fringing GVIIS in 2016 (Figure S1 in Supporting Information S1) that can be traced to the unprecedented magnitude El Niño event of the same year (green shading, Figure 2f).This velocity response is likely explained by the increased moisture, heightened air temperature and strengthened westerlies resulting from ENSOinduced teleconnections that developed between the tropics and the Antarctic Peninsula (Figure 5) during austral wintertime when ENSO teleconnections are known to be strongest (Karoly, 1989;Yiu & Maycock, 2019).As was the case during the initiation of the 2016 El Niño event (Figure 2h), the strength of ENSO teleconnections is typically also modulated by the phase of SAM (Hosking et al., 2013), with the strongest connection to El Niño events occurring when SAM is weak and negative (Fogt et al., 2011;Paolo et al., 2018).
Mechanistically, sustained positive ONI anomalies, denoting an El Niño event, represent warm sea surface temperatures across the equatorial Pacific (5N-5S, 170W-120W; Trenberth & Stepaniak, 2001).Such anomalies act to drive increased convective activity in the central Pacific, generating Rossby waves which travel poleward and establish a teleconnection between the tropics and high-latitude regions (Turner., 2004).This characteristic Rossby "wave train" response -comprising alternating bands of high and low geopotential height anomalies between the central Pacific toward Antarctica (Dutrieux et al., 2014;Steig et al., 2012) -is presented in Figure 5a for austral wintertime 2016.The fingerprint of this El Niño signal at high latitudes is manifest predominantly as a positive 500 hPa geopotential height anomaly over the Peninsula (Figure 5a), together with a positive zonal wind anomaly (Figure 5b) and a negative vertically integrated moisture divergence anomaly (Figure 5c).The El Niño also caused the 2016 wintertime 9°C isotherm (also referred to in the literature as the "thermal limit of ice-shelf viability"; Morris & Vaughan, 2003;Cook & Vaughan, 2010) to reside at a location typically associated with its longer-term, annually averaged location (Figure 5d).These four anomalies signify higher than normal air pressure, enhanced westerlies, precipitation intensification and warmer air temperatures over the Antarctic Peninsula, respectively.
The anomalous conditions noted above (Figure 5) triggered unusual behavior in all examined local-scale environmental drivers either during or in the months preceding winter 2016 (green shading in Figure 2; see also Nicolas et al., 2017;Stuecker et al., 2017).Specifically, anomalously warm wintertime air temperatures (Figure 2b) and associated, enhanced surface meltwater extent (Figure 2c) occurred over GVIIS and its feeder glaciers alongside enhanced surface meltwater extent, which we expect facilitated wintertime hydrofracture and subsequent glacier acceleration.Similarly, the temperature of the ocean at 300+ m was slightly warmer than usual during winter 2016 compared with all other winters on record (Figure 2d), which is attributable to the unusually high sea ice survival rate offshore of GVIIS during the previous summer when concentration reduced to only 20% (mean = 12%; Figure 2e).This phenomenon is, in turn, explained by the enhanced westerlies over the Antarctic Peninsula region (Figure 5b), which acted to advect (and, by implication, compact) sea ice concentration immediately offshore of GVIIS.Following Section 5.2, such compaction (and associated, more persistently high sea ice concentrations overall) likely resulted in a more brine-rich water column facilitating the creation of a thicker volume of CDW (Figure 4a; Holland et al., 2010).

Summary and Implications
Using time series analysis, we show that both surface oceanic forcing mechanisms can explain the seasonal land-ice-flow variability recently observed in the Antarctic Peninsula, with the latter likely augmenting the effects of the former.More generally, our results reveal that an array of local-scale environmental drivers act as statistically significant precursors to summertime ice-flow acceleration.Each driver, however, elicits an ice-velocity response after a distinct lag.Specifically, the observed ∼15% mean summertime speed-up of GVIIS' outlet glaciers occurs (near-)synchronously with an increase in air temperature and surface layer melt extent and several months after maximum sea ice concentration and associated warming of the ocean at depth.Downward propagation of water-filled perennial englacial drainage features (and associated increases in basal lubrication) likely provides the primary forcing mechanism driving these seasonal acceleration signals, augmented by enhanced basal melt-driven glacier debuttressing enabled by sea-ice-induced thickening of the CDW layer.Dedicated oceanographic data collection beneath GVIIS and geophysical measurements atop its feeder glaciers are, however, desperately required to further elucidate the precise local-scale mechanism(s) responsible for controlling this seasonal ice-flow behavior.
In addition to local-scale forcing, our results also provide evidence for the effects of large-scale climatic forcing on Antarctic ice flow.The longitude of the Amundsen Sea Low is found to indirectly stimulate Antarctic ice-flow seasonality through its influence on surface climate seasonality over the Antarctic Peninsula, while ENSO and Southern Annular Mode (SAM) provide no significant forcing over seasonal timescales.Despite this latter finding, our observations do, however, present evidence for a hitherto undocumented relationship between ENSO and ice-sheet velocity, whereby the unprecedented magnitude El Niño event of 2015/16 elicited an isolated and highly anomalous wintertime ice-flow response.Similar ENSO behavior may ultimately have important implications for ice discharge at and beyond the Antarctic Peninsula in the future, depending upon how the magnitude, frequency and/or duration of El Niño events change in a warming world.

Figure 1 .
Figure 1.Mean land-ice velocity of Palmer Land and Alexander Island derived from Sentinel-1A/B synthetic-aperture-radarderived observations acquired between 2014 and 2021.Numbered flowlines demarcate the centerline of the outlet glaciers draining to George VI Ice Shelf, where observations of the modern-day (2018-2020) grounding line exist (black; Boxall et al., 2022b).The 10 km 2 dashed boxes immediately inland of the glaciers' grounding line delimit the averaging regions used for ice velocity, surface meltwater extent and air temperature, while the 200 km 2 dashed boxes located in Ronne Entrance and Marguerite Bay indicate the averaging region utilized for ocean temperature and sea ice concentration (Section 3.2.2).Translucent black shading shows land-ice regions typically associated with summertime surface layer melt as detected by ASCAT (Section 3.2.1).Ice sheet background is the Reference Elevation Model of Antarctica(Howat et al., 2019).Ocean background is BedMachine Antarctica, with the 500 m depth contour in black and remaining contours spaced by 100 m in gray(Morlighem, 2022).Map projection: ESPG:3031.Inset map shows location.

Figure 2 .
Figure 2. Ice velocity and environmental forcing time series for Glacier 1 (see Figure 1 for location).(a) Detrended SARderived observations of ice velocity; (b) Ensemble mean 2 m air temperature; (c) ASCAT-derived surface meltwater percentage extent; (d) Detrended, ensemble mean 300+ m ocean temperature at Ronne Entrance (see Figure S1D in Supporting Information S1 for Marguerite Bay); (e) Ensemble mean sea ice concentration at Ronne Entrance (see Figure S1E in Supporting Information S1 for Marguerite Bay); (f) Oceanic Niño Index, dashed line indicates 0.5°C (above which indicates the occurrence of an El Niño event); (g) Amundsen Sea Low (ASL) Climate Index, longitudinal location variable, horizontal lines indicate 200 and 260°E; (h) Marshall Southern Annular Mode (SAM) Index.(a-c) are calculated over 10 km 2 region immediately inland of Glacier 1, while (d-e) are calculated over 200 km 2 region in Ronne Entrance (Figure 1).Colored shading bounding the time series (a-c, e) denote the standard error (Section 3.2.3).Shaded panels highlight (a) summertime (DJF) ice-flow acceleration or (b-e) the timing of seasonal forcing of each environmental driver.In panel (f), green shading denotes an anomalous El Niño event in 2015/16 and in panels (a-e), corresponding El Niño-driven anomalies in the velocity and environmental driver time series.Width of all shaded panels for (b-e) is determined by the range of lags that yield a significant positive correlation with ice velocity across all glaciers (Section 4.1).The time series associated with all other glaciers are available in Figure S1 in Supporting Information S1.

Figure 3 .
Figure 3. Seasonal ice-shelf thickness change anomalies across George VI Ice Shelf (GVIIS) derived from 3-monthly radarderived ice-shelf height observations spanning 1992 to 2018(Adusumilli et al., 2020).Positive anomalies (blue) indicate the magnitude of ice-shelf thickening and negative anomalies (red) indicate the magnitude of ice-shelf thinning in a given season relative to the long-term (all seasons) average.Springtime is defined as September-November; summertime, December-January; autumntime, March-May; and wintertime, June-August.Background is the Reference Elevation Model of Antarctica(Howat et al., 2019).Map projection: ESPG:3031.Inset map shows location.

Figure 4 .
Figure 4. Schematic diagrams illustrating the role of ocean and surface forcing in driving Antarctic Peninsula seasonal landice acceleration.

Figure 5 .
Figure 5. Wintertime 2016 climate anomalies derived from ERA5.Wintertime anomalies are defined as the difference between the 2016 wintertime mean and the mean of all wintertime records over the Sentinel-1 observational period (2014-2021).(a) 500 hPa geopotential height anomaly; (b) Zonal wind anomaly; (c) Vertically integrated moisture divergence anomaly (whereby moisture convergence is related to precipitation intensification); (d) Mean surface air temperature in June 2016.The red contour represents the 9°C isotherm in wintertime (June) 2016.Other dashed contours represent the average annual (orange), summertime (green) and wintertime (blue) locations of the ERA5-derived 9°C isotherm between 2014 and 2021.Map projection: ESPG:4326.
This research was undertaken while KB was in receipt of a United Kingdom Natural Environment Research Council PhD studentship awarded through the University of Cambridge C-CLEAR Doctoral Training Partnership (Grant NE/ S007164/1).This work was also produced with financial assistance (to FDWC) of the Prince Albert II of Monaco Foundation, and (to ICW) from the United Kingdom Natural Environment Research Council awarded to the University of Cambridge (Grant NE/T006234/1).TN, JW and SS acknowledge support from the European Space Agency through the Antarctic Ice Sheet Climate Change Initiative program (ESA/Contract No. 4000126813/19/I-NB) and the 4DAntarctica project (ESA/ Contract No. ESA/AO/1-9570/18/I-DT).We further acknowledge JM van Wessem and MR van den Broeke for sharing the RACMO2.3p2atmospheric data set.