Optical imagery is used to identify supraglacial lakes and drainage events. Sentinel-2 Level-1C (S2) and Landsat 8 Collection 2 Tier 2 TOA reflectance (L8) images are assessed via GEE and filtered by a maximum cloud coverage of 30 % and minimum sun elevation of 20°, following a similar approach as . Median image mosaics are produced over time periods of 8 days for L8 and 10 days for S2 by taking the pixel-wise median value of all images within the respective period – stitching and combining individual overpasses into one domain-covering image – with resolutions of 30 m and 10 m, respectively. For the 2018–2019, 2019–2020, and 2020–2021 melt seasons, each November to March, the mosaics are sorted by their assigned date, which is defined as halfway between first and latest date of all containing images.

Synthetic aperture radar (SAR) data is utilized for damage detection in the same year as the lake detection. Sentinel-1 GRD (S1) images (instrument mode: “EW” and polarization: “HH”) are used, with images collected between 1 and 10 November to minimize the influence of meltwater on damage detection. S1 images are processed individually to detect damage and combined in postprocessing to generate one damage map per year (see Sect. ).

Ice flow velocity observations for 2019 are sourced from the ITS_LIVE campaign and used for the calculation of the activeness metric (Sect. ). The ice flow velocity data is downsampled from its original resolution of 240 m to a resolution of 300 m using an average resampling method.

The grounding line and ice shelf front are obtained from .

The location and depth of supraglacial lakes are determined using a threshold-based method on the collected mosaics. developed the method and thresholds, which are further refined and automated by . Their method involves combining different bands of the satellite images, applying thresholds to obtain masks for meltwater, rocks and clouds, and finally estimating the depth by taking into account albedo and reflectance of water. These steps are executed within GEE, and produce meltwater lake masks and meltwater depth output. The lake volume is then obtained by multiplying the depths with their respective pixel areas . Outliers are removed in a subsequent step if they are not located on ice mass or have a misinterpreted depth of less than 0 m . To minimize noise, lakes with a surface area of less than 1800 m2 (2 or 18 pixels of L8 and S2 imagery, respectively) are removed as suggested by . For each melt season, the maximum lake extent is derived by aggregating all lake masks generated during that period.

To identify lake drainages, the optical time series are analyzed. A drainage event is identified if a supraglacial lake lost at least 80 % of its area between two consecutive images, provided that no more than 10 d elapsed between those images (following ). While lower thresholds (e.g. 50 %–70 %) can also indicate partial drainage, our observations suggest that these cases are often ambiguous and more difficult to classify confidently as true drainage events, as surface changes may equally reflect refreezing, snow cover variations, or meltwater redistribution. Therefore, we retain the 80 % criterion as a conservative and literature-consistent threshold.

Further, we only retain those lakes that meet the following conditions: surface area greater than 54 000 m2 , median lake depth greater than 0.65 m , standard deviation of lake depth greater than 0.3 , and distance to the nearest masked cloud greater than 500 m. These filtering steps focus on large, deep lakes with a defined depth profile, while avoiding misinterpretations due to faulty cloud masks or blue ice regions.

As a final step, we perform a manual visual inspection of the detected drainage events utilizing all available non-mosaic imagery from L8, and S2. 31 out of 56 events are removed, as they are judged to be refreezing lakes rather than draining lakes.

The Normalised Radon transform Damage detection (NeRD) method is used to detect surface damage features on the ice shelf from strong linear contrasts in S1 images (line detection). In short, the NeRD method consists of the following steps: (i) create cut-out windows from the image, (ii) apply the Normalised Radon transform to these windows, (iii) extract dominant feature signal strength and orientation for every window, (iv) quantify the damage signal by removing noise from the signal and (v) postprocessing . In the post-processing step we clipped the product to the ice shelf bounds. The S1 images are resampled to the same L8 resolution of 30 m and processed with a window size of 10 pixels within the NeRD algorithm, yielding 300 m-resolution damage maps with a damage signal between 0 and 0.5 (low to high damage indication) and their orientation (-90 to 90°). By using the NeRD method, we detect features at the ice shelf surface only – all visible crevasses, fractures and rifts – and group these under the umbrella term “damage”, distinguishing from the damage parameter commonly used in continuum mechanics literature . Basal crevasses are not explicitly included, but might be detected if their concurrent surface depression is distinct. The damage maps represent the presence of large features (roughly >100 m) within the 300 m range, rather than the delineation of the feature itself.

In addition to detecting the presence of surface damage features, we infer if the features are likely actively developing and opening (“activeness”) or if they are passive (apart from advecting with the ice flow). The obtained damage orientation is used to identify the potential activeness by comparing the damage orientation to local ice flow angle, to infer if the feature is in a position that favors crevasse opening. First, the ice flow velocity was resampled to match the resolution of the detected damage orientation map. Then, an area was considered likely to be “active” if the damage orientation and flow angle have a difference of 45° ± 15°, which occurs mainly in the shear zones of the ice shelf or for mixed mode opening fractures. This way we disfavor large open rifts near the ice front that are perpendicular to the ice flow, which cut completely through the ice and hence do not support lake formation and subsequent lake drainages, as well as other orientations likely representing inactive fractures misaligned with current stress fields (e.g. due to advection) . This procedure produces a binary activeness mask (values 0 or 1) at 300 m resolution. new small fracture and basal crevasses. From these observational products we cannot prove causality between individual damage features and specific drainage events. Furthermore, it's possible for the drainage to occurs through a fracture that is not visible in the 300 m maps. We therefore use the damage and activeness maps as an indication of a general structurally weakened ice zone, which we hypothesize to favor lake drainages through undetected or new small fracture and basal crevasses.

For this reason, we downsampled the damage and activeness maps to inspect the overall integrity of the ice for a larger area surrounding drainage events. Both 300 m maps are downsampled with a factor of 10 using an average resampling method, and normalized with their respective maxima, resulting in 3000 m resolution rasters with values between 0 and 1.

To investigate how ocean tides influence drainage events, we utilize the Circum-Antarctic Tidal Simulation (CATS2008) model to compute daily tidal amplitudes. These amplitudes are determined by calculating the difference between the maximum and minimum tidal heights for each day. Due to low tidal amplitude variations across Shackleton ice shelf (Fig. S5, and ), the tidal amplitudes for this study are determined at the central location of lake I (Fig. ). The CATS2008 model is a high-resolution barotropic tide model specifically designed for the Antarctic continental shelf, incorporating bathymetry data from various sources and assimilating tide gauge and satellite altimetry observations to enhance its accuracy. The model solves the depth-integrated shallow water equations on a finite difference grid, accounting for the effects of sea ice cover and ice shelf cavities, which are crucial for accurately representing tidal dynamics in polar regions .

Activeness and damage distribution on the Shackleton Ice Shelf, including the distribution of lake drainages listed in Table S3 in the Supplement. The limits are indicated as introduced in Table S2. Refer to Sect.  for details on damage and activeness classifications and for yearly evolution of spatial distributions of damage, activeness and strain refer to Figs. S2, S3, and S4.

The detected damage maps (Fig. a) show the spatial distribution of damage features on the ice, with high values indicating areas characterized by rifts or crevasses. A visual comparison with the detected lakes (Fig. ) shows that the distribution of the lakes are often coinciding with detected damage patterns. We categorized damage levels into three distinct groups with values specific to Shackleton ice shelf: not/low damaged (values < 0.02, 70 % of the area, no lake draining events), medium-damaged (0.02 < values < 0.16, 20 % of the area, eight lake draining events), and highly damaged (values > 0.16, 10 % of the area, 17 lake draining events). Due to the strongly skewed data distribution (Fig. h, and Table  S2) we discretized the damage signal values in bins of unequal width, containing progressively less data samples (damaged pixels) to favor the representation of the minority, high values, class.

The results show that lake drainages predominantly occur in areas classified with medium to high relative levels of damage (Fig. a and e). The three lake drainages in relatively medium-damaged regions are particularly concentrated around Scott Glacier and Bunger Hills (drainages H, L, and M). Although relatively high-damaged areas cover only a small portion of the Shackleton Ice Shelf, they account for the majority of lake drainages, primarily located further west along the grounding line (drainages A to I, excluding G), with one event occurring on the glacier tongue (drainage G). In contrast, no lake drainages have been recorded in relatively low-damaged areas, despite these areas covering the majority of the ice shelf.

The activeness parameter (Fig. b and f) provides insights into the dynamic behavior of the shelf, with relatively high values indicating areas where the ice and local fractures have a high likelihood to be under active development due to the flow of the ice. Similar as the damage values, we categorized the activeness in the following groups to favor the tails of the distribution: not active (values < 0.24, 20 % of the area), moderately active (0.24 < values < 0.62, 60 % of the areas), and highly active (0.60 < values, 20 % of the area), as detailed in Table S2.

High activeness is found most clearly where fast flowing ice experiences shear stress from adjacent slower-moving or stationary ice masses, leading to mix-mode crevasse opening . Areas close to the Northcliffe, Denman, Scott and Apfel glaciers serve as the clearest examples of areas exhibiting high activeness on Shackleton ice shelf.

All, but one, of the detected lake drainage events occur in areas of the ice shelf classified as medium to highly active relatively to Shackleton ice shelf (20 and four lake draining events respectively; Fig. b and Table S2), indicating that areas without damage development are not likely to accommodating lake drainage.

Considering both damage and activeness, several notable patterns emerge. First, drainages A to F, on the grounding line on West Shackleton are cases where damage is relatively high and activeness is moderate at the lake drainage sites. This indicates that areas along the grounding line are prone to such events due to the combined effect of significant damage and moderate activeness. Second, east of the glacier tongue (drainages J and M), lake drainage events occurred in areas of high activeness, despite relatively medium levels of detected damage. This suggests that the interaction between the small glaciers and the Bunger Hills creates a highly active section of the ice shelf, where accumulated meltwater rapidly drains even in moderately fractured regions. Third, we also detect a few drainages in areas with high activeness (drainages G and M) but with relatively high levels of detected damage. Taken together, the results suggest that while moderate damage appears to be a necessary precondition, activeness acts as an additional driver that can amplify the likelihood of drainage events. Drainages are more likely to occur if either detected damage or activeness is high, or both.

Compared to the vulnerability metric of (Fig. c and j), which indicates the (in)stability of detected fractures to inundation with meltwater, we see that drainages that occur on “vulnerable” areas also have moderate to high activeness (7 out 25, Fig. j). Intriguingly, we also detect 15 drainages that occur in areas where indicated “no-hydrofracturing” or outside of classification. These drainages are in areas of both moderate and high activeness (14 out 15), mostly in combination with relatively high damage areas (9 out of the 14 in areas of both moderate and high activeness). This highlights the role of ice dynamics in the behavior of hydrofracturing, and suggest that vulnerability estimates based only on fracture mechanics is not sufficient.

Overall, these results suggest that both damage and activeness are important indicators of the likelihood of supraglacial lakes draining rapidly. Although the exact relationship between these metrics requires further investigation across different ice shelves and with more drainage events, it seems that the activeness of detected damage adds insights into the where lake drainage occur in otherwise unsuspected areas.

Given the tidal amplitudes and the geometry of the Shackleton Ice Shelf, large areas experience cyclical stress states driven by tidal oscillations . As tidal amplitudes surge towards their spring maxima, the ice shelf undergoes enhanced flexure, potentially generating complex tensile stress fields at both its surface and base, which can lead to crevasse opening and so facilitate rapid lake drainages.

We compare the drainage time-windows to tidal data (Fig. ), and indeed find a clear pattern: the majority of drainage episodes aligns with the ascending phase of tide cycles (17 out of 25). Although we cannot determine the exact drainage date, only a snapshot before and after the event, few drainages seem to have occurred on the descending phase.

This pattern suggests that amplified tidal flexure may serve as a catalyst, either initiating new cracks or reactivating dormant weaknesses within the ice shelf structure. Once this process begins, the rapid influx of draining water could further propagate these fractures, facilitating complete lake drainage in a cascade of events reminiscent of those described by .

Our observations of several drainage events indicate that tidal forcing plays an important role in modulating supraglacial lake drainage across the ice shelf. These results are consistent with previous findings .