Causal modeling offers a principled foundation for uncovering stable, invariant relationships in time-series data, thereby improving robustness and generalization under distribution shifts. Yet its potential is underutilized in spatiotemporal Earth observation, where models often depend on purely correlational features that fail to transfer across heterogeneous domains. We propose RIC-TSC, a regionally-informed causal time-series classification framework that embeds lag-aware causal discovery directly into sequence modeling, enabling both predictive accuracy and scientific interpretability. Using multi-modal satellite and reanalysis data-including Sentinel-1 microwave backscatter, Sentinel-2 and Landsat-8 optical reflectance, and CARRA meteorological variables-we leverage Joint PCMCI+ (J-PCMCI+) to identify region-specific and invariant predictors of supraglacial lake evolution in Greenland. Causal graphs are estimated globally and per basin, with validated predictors and their time lags supplied to lightweight classifiers. On a balanced benchmark of 1000 manually labeled lakes from two contrasting melt seasons (2018-2019), causal models achieve up to 12.59

Supraglacial lakes form on the surface of the Greenland Ice Sheet during the summer months and can directly impact ice sheet mass balance by removing mass via drainage and runoff or indirectly impact mass balance by influencing ice sheet dynamics. Here, we utilize the growing inventory of optical and microwave satellite imagery to automatically determine the fate of Greenland-wide supraglacial lakes during 2018 and 2019, a cool and warm melt season respectively. We use a machine learning time series classification approach to categorize lakes into four different categories: lakes that 1) refreeze, 2) rapidly drain, 3) slowly drain, and 4) become buried lakes at the end of the melt season. We find that during the warmer 2019, not only was the number of lake drainage events higher than in 2018, but also the proportion of lakes that drained was greater. By investigating mean lake depths for these four categories, we show that drained lakes were, on average, 22% deeper than lakes that refroze or became buried lakes. Interestingly, drained lakes had approximately the same maximum depth in 2018 and 2019; however, lakes that did not drain were 29% deeper in 2018, a cooler year. Our unique two-year dataset describing the fate of every Greenland supraglacial lake provides novel insight into lake drainage and refreeze in a relatively warm and cool year, which may be increasingly relevant in a warming climate.

Supraglacial lakes on the Greenland Ice Sheet (GrIS) can impact both the ice sheet surface mass balance and ice dynamics. Thus, understanding the evolution and dynamics of supraglacial lakes is important to provide improved parameterizations for ice sheet models to enable better projections of future GrIS changes. In this study, we utilize the growing inventory of optical and microwave satellite imagery to automatically determine the fate of Greenland-wide supraglacial lakes during 2018 and 2019; cool and warm melt seasons respectively. We develop a novel time series classification method to categorize lakes into four classes: 1) refreezing, 2) rapidly draining, 3) slowly draining, and 4) buried. Our findings reveal significant interannual variability between the two melt seasons, with a notable increase in the proportion of draining lakes in 2019. We also find that as mean lake depth increases, so does the percentage of lakes that drain, indicating that lake depth may influence hydrofracture potential. However, we also observe that non-draining lakes are deeper during the cooler 2018 melt season, suggesting that additional factors may predispose lakes to drain earlier in a warmer year. Our automatic classification approach and the resulting two-year ice-sheet-wide dataset provide unprecedented insights into GrIS supraglacial lake dynamics and evolution, offering a valuable resource for future research.

The Greenland Ice Sheet (GrIS) has emerged as a significant contributor to global sea level rise, primarily due to increased meltwater runoff. Supraglacial lakes, which form on the ice sheet surface during the summer months, can impact ice sheet dynamics and mass loss; thus, better understanding these lakes' seasonal evolution and dynamics is an important task. This study presents a computationally efficient time series classification approach that uses Gaussian Mixture Models (GMMs) of the Reconstructed Phase Spaces (RPSs) to identify supraglacial lakes based on their seasonal evolution: 1) those that refreeze at the end of the melt season, 2) those that drain during the melt season, and 3) those that become buried, remaining liquid insulated a few meters beneath the surface. Our approach uses time series data from the Sentinel-1 and Sentinel-2 satellites, which utilize microwave and visible radiation, respectively. Evaluated on a GrIS-wide dataset, the RPS-GMM model, trained on a single representative sample per class, achieves 85.46% accuracy with Sentinel-1 data alone and 89.70% with combined Sentinel-1 and Sentinel-2 data. This performance significantly surpasses existing machine learning and deep learning models which require a large training data. The results demonstrate the robustness of the RPS-GMM model in capturing the complex temporal dynamics of supraglacial lakes with minimal training data.

Antarctic firn is critical for ice-shelf stability because it stores meltwater that would otherwise pond on the surface. Ponded meltwater increases the risk of hydrofracture and subsequent potential ice-shelf collapse. Here, we use output from a firn model to build a computationally simpler emulator that uses a random forest to predict ice-shelf effective firn air content, which considers impermeable ice layers that make deeper parts of the firn inaccessible to meltwater, based on climate conditions. We find that summer air temperature and precipitation are the most important climatic features for predicting firn air content. Based on the climatology from an ensemble of Earth System Models, we find that the Larsen C Ice Shelf is most at risk of firn air depletion during the 21st century, while the larger Ross and Ronne-Filchner ice shelves are unlikely to experience substantial firn air content change. This work demonstrates the utility of emulation for computationally efficient estimations of complicated ice sheet processes.

Antarctic firn is critical for ice-shelf stability because it stores meltwater that would otherwise pond on the surface. Ponded meltwater increases the risk of hydrofracture, and subsequent potential ice-shelf collapse. Here, we use output from a firn model to build a computationally simpler emulator that uses a random forest to predict ice-shelf firn air content (FAC) based on climate conditions. We find that summer air temperature and precipitation are the most important climatic features for predicting FAC. By 2100, Larsen C Ice Shelf is most at risk of firn air depletion, while the larger Ross and Ronne-Filchner are likely to experience little FAC change. Variability in Earth System Model projections is the largest uncertainty source regarding future FAC predictions. In fact, the model uncertainty, described by the firn air depletion spread from CMIP6 models in the SSP3-7.0 and SSP5-8.5 emission scenarios, is larger than the firn air depletion range across the emission scenarios.

Earth system models are essential tools for understanding the impacts of a warming world, particularly on the contribution of polar ice sheets to sea level change. However, current models lack full coupling of the ice sheets to the ocean and are typically run at a coarse resolution (1∘ grid spacing or coarser). Coarse spatial resolution is particularly a problem over Antarctica, where sub-grid-scale orography is well-known to influence precipitation fields, and glacier models require high-resolution atmospheric inputs. This resolution limitation has been partially addressed by regional climate models (RCMs), which must be forced at their lateral and ocean surface boundaries by (usually coarser) global atmospheric datasets, However, RCMs fail to capture the two-way coupling between the regional domain and the global climate system. Conversely, running high-spatial-resolution models globally is computationally expensive and can produce vast amounts of data. Alternatively, variable-resolution grids can retain the benefits of high resolution over a specified domain without the computational costs of running at a high resolution globally. Here we evaluate a historical simulation of the Community Earth System Model version 2 (CESM2) implementing the spectral element (SE) numerical dynamical core (VR-CESM2) with an enhanced-horizontal-resolution (0.25∘) grid over the Antarctic Ice Sheet and the surrounding Southern Ocean; the rest of the global domain is on the standard 1∘ grid. We compare it to 1∘ model runs of CESM2 using both the SE dynamical core and the standard finite-volume (FV) dynamical core, both with identical physics and forcing, including prescribed sea surface temperatures (SSTs) and sea ice concentrations from observations. Our evaluation reveals both improvements and degradations in VR-CESM2 performance relative to the 1∘ CESM2. Surface mass balance estimates are slightly higher but within 1 standard deviation of the ensemble mean, except for over the Antarctic Peninsula, which is impacted by better-resolved surface topography. Temperature and wind estimates are improved over both the near surface and aloft, although the overall correction of a cold bias (within the 1∘ CESM2 runs) has resulted in temperatures which are too high over the interior of the ice sheet. The major degradations include the enhancement of surface melt as well as excessive cloud liquid water over the ocean, with resultant impacts on the surface radiation budget. Despite these changes, VR-CESM2 is a valuable tool for the analysis of precipitation and surface mass balance and thus constraining estimates of sea level rise associated with the Antarctic Ice Sheet.

Antarctic ice‐shelf stability is threatened by surface melt, which has been implicated in several ice‐shelf collapse events over recent decades. Here, we first analyze cumulative days of wet snow/ice status (“melt days”) for melt seasons from 1980 to 2021 over Antarctica's ice shelves using passive and active microwave satellite observations. As these observations do not directly reveal meltwater volumes, we calculate these using the physics‐based multi‐layer snow model SNOWPACK, driven by the global climate‐reanalysis model Modern‐Era Retrospective analysis for Research and Applications Version 2. We find a strong non‐linear relationship between melt days and meltwater production volume. SNOWPACK's calculation of melt days shows agreement with observations of both cumulative days, and spatial and interannual variability. Highest melt rates are found on the Peninsula ice shelves, particularly in the 1992/1993 and 1994/1995 austral summers. Over all ice shelves, SNOWPACK calculates a small, but significant, decreasing trend in both annual melt days and meltwater production volume over the 41 years.

Earth system models (ESMs) allow us to explore minimally observed components of the Antarctic Ice Sheet (AIS) climate system, both historically and under future climate change scenarios. Here, we present and analyze surface climate output from the most recent version of the National Center for Atmospheric Research's ESM: the Community Earth System Model version 2 (CESM2). We compare AIS surface climate and surface mass balance (SMB) trends as simulated by CESM2 with reanalysis and regional climate models and observations. We find that CESM2 substantially better represents the mean-state AIS near-surface temperature, wind speed, and surface melt compared with its predecessor, CESM1. This improvement likely results from the inclusion of new cloud microphysical parameterizations and changes made to the snow model component. However, we also find that grounded CESM2 SMB (2269 ± 100 Gt yr−1) is significantly higher than all other products used in this study and that both temperature and precipitation are increasing across the AIS during the historical period, a trend that cannot be reconciled with observations. This study provides a comprehensive analysis of the strengths and weaknesses of the representation of AIS surface climate in CESM2, work that will be especially useful in preparation for CESM3 which plans to incorporate a coupled ice sheet model that interacts with the ocean and atmosphere.

Abstract. Antarctic Ice Sheet (AIS) mass loss can be mitigated by an increase in surface mass balance (SMB) which is impacted by the ice sheet's surface climate. Because of Antarctica’s remoteness, in-situ observations of its surface climate are spatially and temporally sparse, limiting our understanding of how the surface climate, and therefore SMB, are changing. To that end, Earth System Models (ESMs) fill an important gap, allowing us to explore components of the AIS climate system, both historically and under future climate change scenarios. Here, we present and analyze output of the most recent version of the National Center for Atmospheric Research’s ESM: the Community Earth System Model version 2 (CESM2). We compare AIS surface climate and SMB as simulated by CESM2 with regional climate models and observations. We find that CESM2 substantially better represents AIS near-surface temperature, wind speed, and surface melt compared with its predecessor, CESM1, which is likely a result of the inclusion of new cloud microphysical parameterizations and changes made to the snow model. However, CESM2 SMB is shown to be biased high, particularly because of excessive precipitation. ESMs such as CESM2 are used as forcing for ice sheet models, which provide estimates and projections of ice sheet contribution to sea level rise. Thus, it is important to fully understand the limitations and biases of this climate model forcing. This study provides a comprehensive analysis of the strengths and weaknesses of CESM2 representation of AIS surface climate, which will be especially useful in preparation for CESM3, which plans to incorporate a coupled Antarctic Ice Sheet that interacts with the ocean and atmosphere.

Abstract. The Thwaites Eastern Ice Shelf (TEIS) buttresses the eastern grounded portion of Thwaites Glacier through contact with a pinning point at its seaward limit. Loss of this ice shelf will promote further acceleration of Thwaites Glacier. Understanding the dynamic controls and structural integrity of the TEIS is therefore important to estimating Thwaites' future sea-level contribution. We present a ~20-year record of change on the TEIS that reveals the dynamic controls governing the ice shelf's past behavior and ongoing evolution. We derived ice velocities from MODIS and Sentinel-1 image data using feature tracking and speckle tracking, respectively, and combined these records with ITS_LIVE and GOLIVE velocity products from Landsat 7 and 8. In addition, we estimated surface lowering and basal melt rates using the REMA DEM in comparison to ICESat and ICESat-2 altimetry. Early in the record, TEIS flow dynamics were strongly controlled by the neighboring Thwaites Western Ice Tongue (TWIT). Flow patterns on the TEIS changed following the disintegration of the TWIT in ~2008, with a new divergence in ice flow developing around the pinning point at its seaward limit. Simultaneously, the TEIS developed new rifting that extends from the shear zone upstream of the ice rise and increased strain concentration within this shear zone. As these horizontal changes occurred, sustained thinning driven by basal melt reduced ice thickness, particularly near the grounding line and in the shear zone area upstream of the pinning point. This evidence of weakening at a rapid pace suggests that the TEIS is likely to fully destabilize in the next few decades, leading to further acceleration of Thwaites Glacier.

The Greenland Ice Sheet (GrIS) rapid mass loss is primarily driven by an increase in meltwater runoff, which highlights the importance of understanding the formation, evolution, and impact of meltwater features on the ice sheet. Buried lakes are meltwater features that contain liquid water and exist under layers of snow, firn, and/or ice. These lakes are invisible in optical imagery, challenging the analysis of their evolution and implication for larger GrIS dynamics and mass change. Here, we present a method that uses a convolutional neural network, a deep learning method, to automatically detect buried lakes across the GrIS. For the years 2018 and 2019 (which represent low- and high-melt years, respectively), we compare total areal extent of both buried and surface lakes across six regions, and we use a regional climate model to explain the spatial and temporal differences. We find that the total buried lake extent after the 2019 melt season is 56 % larger than after the 2018 melt season across the entire ice sheet. Northern Greenland has the largest increase in buried lake extent after the 2019 melt season, which we attribute to late-summer surface melt and high autumn temperatures. We also provide evidence that different processes are responsible for buried lake formation in different regions of the ice sheet. For example, in southwest Greenland, buried lakes often appear on the surface during the previous melt season, indicating that these meltwater features form when surface lakes partially freeze and become insulated as snowfall buries them. Conversely, in southeast Greenland, most buried lakes never appear on the surface, indicating that these features may form due to downward percolation of meltwater and/or subsurface penetration of shortwave radiation. We provide support for these processes via the use of a physics-based snow model. This study provides additional perspective on the potential role of meltwater on GrIS dynamics and mass loss.

Between 1992 and 2017, the Antarctic Ice Sheet (AIS) lost ice equivalent to 7.6 ± 3.9 mm of sea level rise. AIS mass loss is mitigated by ice shelves that provide a buttress by regulating ice flow from tributary glaciers. However, ice-shelf stability is threatened by meltwater ponding, which may initiate, or reactivate preexisting, fractures, currently poorly understood processes. Here, through ground penetrating radar (GPR) analysis over a buried lake in the grounding zone of an East Antarctic ice shelf, we present the first field observations of a lake drainage event in Antarctica via vertical fractures. Concurrent with the lake drainage event, we observe a decrease in surface elevation and an increase in Sentinel-1 backscatter. Finally, we suggest that fractures that are initiated or reactivated by lake drainage events in a grounding zone will propagate with ice flow onto the ice shelf itself, where they may have implications for its stability.

The evolution of surface and shallow subsurface meltwater bodies across Antarctic ice shelves has important implicationsfor their (in)stability, as demonstrated by the 2002 rapid collapse of the Larsen B Ice Shelf. Ice-shelf break-up may be triggered by stress variations associated with meltwater movement, ponding and drainage, causing ice-shelf flexure and fracture. We have recently begun a four year, jointly-funded US-NSF / UK-NERC project that will provide important geophysical insights into the stability of the George VI Ice Shelf on the Antarctic Peninsula, where hundreds of surface lakes form each summer. In November 2019, we deployed global positioning systems, pressure transducers, automatic weather stations, and in-ice thermistor strings to record ice-shelf flexure, surface water depths, and surface and subsurface melting, respectively, in and around several surface lakes. Next austral summer (November 2020), we also plan to record fracture seismicity with a passive seismometer deployment, and to conduct ground penetrating radar surveys to detect subsurface water. Instruments, which are all within ~30 km of BAS's Fossil Bluff Station, will remain on the ice shelf until January 2022, resulting in a 27-month observational record in total. Here, we report results of satellite image analysis of surface and shallow subsurface meltwater bodies, together with preliminary field and modelling results associated with our project. Using NDWIice thresholds applied to Landsat 8, Sentinel-2 and WorldView optical imagery, we show how patterns of surface meltwater evolve within and between summer melt seasons. Using Sentinel-1 SAR imagery and a convolutional neural network technique, we detect and track bodies of shallow subsurface water and show how they relate to patterns of surface water. We also report on field reconnaissance surveys made to two dolines (drained lake basins) on the ice shelf, and present a simple model to describe the process of doline formation. Throughout the project, we will combine field and remotely sensed data to extend and validate our existing approach to modelling ice-shelf flexure and stress, and possible ‘Larsen-B style’ ice-shelf instability and break-up at less geographically confined ice shelves.

The presence of meltwater influences Antarctic ice shelf dynamics in a way that is poorly understood. In addition to surface meltwater, subsurface meltwater lakes have been discovered close to the ice shelf grounding line. Drainage and collapse of these subsurface lakes may induce hydrofracturing and poses a potential threat to ice shelf stability. Here, we present direct observations of the near-surface firn and ice shelf structure before and after the collapse of a subsurface meltwater lake near the grounding line of the Roi Baudouin Ice Shelf (RBIS). In February 2016, ground penetrating radar (GPR) data were collected of the subsurface lake, highlighting its depth and extent. Surprisingly, when the field team returned to the site in December 2017 to repeat the GPR surveying, they found that the lake had collapsed. These unique GPR and GPS observations highlight the heterogeneity of the lake structure after collapse and allow us to see structural differences before and after collapse. Continued geophysical monitoring and analysis could provide important information to estimate the volume and hydrodynamics of the interglacial lake (e.g. horizontal vs. vertical drainage). In addition to field data, we use a regional climate model and remote sensing observations to provide an analysis of the climate forcing that may have contributed to the lake collapse. We show that anomalously high surface melting in the summer season of 2016-2017 likely contributed to the collapse. Our results shed light on the impact of subsurface lake collapse on the ice shelf structure, dynamics, and surface height changes, which is essential to understand the impact of meltwater drainage on ice shelf stability.