Landfast ice, sea ice fastened to coastal land or ice shelves, generally undergoes distinctive thermodynamic growth and less active dynamic deformation due to its prolonged attachment to the land, resulting in a thicker and smoother surface compared to drifting pack ice. In 2019, large landfast ice floes were detached from the Ronne Ice Shelf, and the broken pieces started to drift into the Weddell Sea. This study employs a random forest (RF) machine learning model to detect these ex-fast ice floes using six key variables from the ICESat-2 ATL10 sea ice freeboard product: freeboard, Gaussian width of photon height distribution, standard deviation of freeboard, floe length, modal freeboard, and sea ice concentration. The RF model achieves an overall accuracy of 99 % in detecting ex-fast ice, effectively capturing the drift, freeboard distribution, and size distribution of ex-fast ice floes across the western Weddell Sea in 2019. Among six variables, freeboard, standard deviation of freeboard, and Gaussian width of photon height distribution contribute over 94 % to the model performance. Furthermore, the detection of ex-fast ice improves the quantification of sea ice topographical features derived from ICESat-2, including modal freeboard, ridge fraction, and surface roughness. This study highlights the effectiveness of discriminating heterogeneous ex-fast ice from typical pack ice to enhance sea ice measurements using ICESat-2 satellite altimeter data.

Sea ice plays a critical role in the global climate system and maritime operations, making timely and accurate classification essential. However, traditional manual methods are time-consuming, costly, and have inherent biases. Automating sea-ice type classification addresses these challenges by enabling faster, more consistent, and scalable analysis. While both traditional and deep-learning approaches have been explored, deep-learning models offer a promising direction for improving efficiency and consistency in sea-ice classification. However, the absence of a standardized benchmark and comparative study prevents a clear consensus on the best-performing models. To bridge this gap, we introduce IceBench, a comprehensive benchmarking framework for sea-ice type classification. Our key contributions are three-fold: First, we establish the IceBench benchmarking framework, which leverages the existing AI4Arctic Sea Ice Challenge Dataset as a standardized dataset, incorporates a comprehensive set of evaluation metrics, and includes representative models from the entire spectrum of sea-ice type-classification methods categorized in two distinct groups, namely pixel-based classification methods and patch-based classification methods. IceBench is open-source and allows for convenient integration and evaluation of other sea-ice type-classification methods, hence facilitating comparative evaluation of new methods and improving reproducibility in the field. Second, we conduct an in-depth comparative study on representative models to assess their strengths and limitations, providing insights for both practitioners and researchers. Third, we leverage IceBench for systematic experiments addressing key research questions on model transferability across seasons (time) and locations (space), data downsampling, and preprocessing strategies. By identifying the best-performing models under different conditions, IceBench serves as a valuable reference for future research and a robust benchmarking framework for the field.

Capturing spatial context at multiple scales is crucial for deep learning-based sea ice segmentation. However, the optimal specification of spatial context based on observation resolution and task characteristics remains underexplored. This study investigates the impact of spatial context on the segmentation of sea ice concentration, stage of development, and floe size using a multi-task segmentation model. We implement Atrous Spatial Pyramid Pooling with varying atrous rates to systematically control the receptive field size of convolutional operations, and to capture multi-scale contextual information. We explore the interactions between spatial context and feature resolution for different sea ice properties and examine how spatial context influences segmentation performance across different input feature combinations from Sentinel-1 SAR and Advanced Microwave Radiometer-2 (AMSR2) for multi-task mapping. Using Gradient-weighted Class Activation Mapping, we visualize how atrous rates influence model decisions. Our findings indicate that smaller receptive fields excel for high-resolution Sentinel-1 data, while medium receptive fields yield better performances for stage of development segmentation and larger receptive fields often lead to diminished performances. The fusion of SAR and AMSR2 enhances segmentation across all tasks. We highlight the value of lower-resolution 18.7 and 36.5 GHz AMSR2 channels in sea ice mapping. These findings highlight the importance of selecting appropriate spatial context based on observation resolution and target properties in sea ice mapping. By systematically analyzing receptive field effects in a multi-task setting, our study provides insights for optimizing deep learning models in geospatial applications.

Sea ice extent remained anomalously low in 2024. Annual mean Antarctic and Arctic sea ice extent was 10.38 million km2 and 10.42 million km2, respectively, the 2nd and 7th lowest of the satellite record.

Accurate segmentation and mapping of sea ice types is crucial for safe polar navigation, offshore operations, climate monitoring, and ecosystem analysis. While deep learning has demonstrated strong potential for automating sea ice type segmentation, its success often relies on access to extensive expert-labeled datasets, which is both resource-intensive and time-consuming to create. However, foundation models (FMs), recently developed through self-supervised training on large-scale datasets, have demonstrated impressive performance across a range of remote sensing downstream tasks. Nevertheless, their applicability to sea ice type segmentation based on Synthetic Aperture Radar (SAR) imagery remains uncertain due to the unique challenges posed by sea ice-such as intricate geophysical patterns, pronounced seasonal variability, and SAR-specific artifacts like banding, scalloping, and heterogeneous backscatter-as well as the fact that SAR data in polar regions are often acquired using specialized sensor modes that differ markedly from those used to collect FM training data at lower latitudes, limiting their direct transferability to polar environments. To address this gap, we contribute: (1) Ice-FMBench, a comprehensive benchmark framework for evaluation of the state-of-the-art remote sensing FMs on the sea ice type segmentation task using Sentinel-1 SAR imagery, where Ice-FMBench is composed of a widely used standardized dataset, diverse evaluation metrics, and a representative set of selected remote sensing FM models most suitable for sea ice type segmentation, with the ability to include new models side-by-side the existing models; (2) an extensive comparative evaluation of the representative FMs using Ice-FMBench, with additional case studies to assess performance of the top-performing model in terms of transferability across temporal and spatial domains and sensitivity to training dataset size; and (3) a multi-teacher knowledge distillation approach to address lack of spatiotemporal transferability of the existing FMs by transferring insight from spatially and temporally specialized expert models into a single, efficient student model.

Sea ice motion plays an important role in the seasonal and interannual evolution of the polar sea ice cover. Satellite imagery can be used to track the motion of sea ice via cross-correlation feature tracking algorithms. Such a method has been used for the National Snow and Ice Data Center (NSIDC) sea ice motion product, based largely on passive microwave imagery. This study investigates the use of a new enhanced resolution passive microwave brightness temperature (TB) product to derive ice motion products. The results demonstrate that the new imagery source provides useful daily motion estimates that provide denser spatial coverage and reduced errors. The enhanced TBs yield motions that have a 30% lower Root Mean Square (RMS) difference with motion estimates from buoys. The enhanced resolution TBs will be used in the new version of the NSIDC motion product that is currently in development.

Sea ice type classification is essential for climate change research and maritime safety. Traditionally, this process relies on manual ice charting, which is time-consuming, expensive, and requires expert knowledge, making it difficult to scale up for current demands. Automating sea ice type classification is essential to keep pace with rapidly changing sea ice conditions. However, two main challenges limit the development of effective automated classifiers. First, while ice charts provide valuable labeled data, they only offer large-area (polygon) annotations rather than pixel-level labels, leading to a lack of precise training data. Second, although there are additional datasets with useful sea ice information, effectively combining these different data sources remains difficult. To tackle the first challenge, we employed co-training and label propagation, two semi-supervised learning methods, to learn from a small amount of labeled data and a large pool of unlabeled data, thereby improving the accuracy of sea ice classifiers despite limited labeled data. To address the second challenge, we leveraged co-training’s built-in ability to integrate multiple data sources during the training process for the small labeled data. Additionally, we further enhanced data integration by using an ensemble of these co-trained models after training. Our approach demonstrates significant improvements over traditional supervised methods, showcasing the potential of semi-supervised learning methods in addressing two major challenges in developing automated sea ice classification solutions. Our study shows that semi-supervised learning improved F1 scores by 17% for SAR data and 33% for AMSR2 with limited labels, compared to supervised methods, while ensembling further boosted accuracy by 33%.

Antarctic sea ice extent was the lowest on record in 2023, with an annual mean of 9.81 million km2, beating the previous minimum of 2022. Arctic sea ice extent was also low, with an annual mean of 10.49 million km2, but did not break any records.

Sea ice, crucial to the Arctic and Earth's climate, requires consistent monitoring and high-resolution mapping. Manual sea ice mapping, however, is time-consuming and subjective, prompting the need for automated deep learning-based classification approaches. However, training these algorithms is challenging because expert-generated ice charts, commonly used as training data, do not map single ice types but instead map polygons with multiple ice types. Moreover, the distribution of various ice types in these charts is frequently imbalanced, resulting in a performance bias toward the dominant class. In this article, we present a novel GeoAI approach to training sea ice classification by formalizing it as a partial label learning task with explicit confidence scores to address multiple labels and class imbalance. We treat the polygon-level labels as candidate partial labels, assign the corresponding ice concentrations as confidence scores to each candidate label, and integrate them with focal loss to train a convolutional neural network. Our proposed approach leads to enhanced performance for sea ice classification in Sentinel-1 dual-polarized SAR images, improving classification accuracy (from 87% to 92%) and weighted average F-1 score (from 90% to 93%) compared to the conventional training approach of using one-hot encoded labels and categorical cross-entropy loss. It also improves the F-1 score in four out of the six sea ice classes.

This study quantifies the state of the art in the rapidly growing field of seasonal Arctic sea ice prediction. A novel multimodel dataset of retrospective seasonal predictions of September Arctic sea ice is created and analyzed, consisting of community contributions from 17 statistical models and 17 dynamical models. Prediction skill is compared over the period 2001-20 for predictions of pan-Arctic sea ice extent (SIE), regional SIE, and local sea ice concentration (SIC) initialized on 1 June, 1 July, 1 August, and 1 September. This diverse set of statistical and dynamical models can individually predict linearly detrended pan-Arctic SIE anomalies with skill, and a multimodel median prediction has correlation coefficients of 0.79, 0.86, 0.92, and 0.99 at these respective initialization times. Regional SIE predictions have similar skill to pan-Arctic predictions in the Alaskan and Siberian regions, whereas regional skill is lower in the Canadian, Atlantic, and central Arctic sectors. The skill of dynamical and statistical models is generally comparable for pan-Arctic SIE, whereas dynamical models outperform their statistical counterparts for regional and local predictions. The prediction systems are found to provide the most value added relative to basic reference forecasts in the extreme SIE years of 1996, 2007, and 2012. SIE prediction errors do not show clear trends over time, suggesting that there has been minimal change in inherent sea ice predictability over the satellite era. Overall, this study demonstrates that there are bright prospects for skillful operational predictions of September sea ice at least 3 months in advance.

Due to the growing volume of remote sensing data and the low latency required for safe marine navigation, machine learning (ML) algorithms are being developed to accelerate sea ice chart generation, currently a manual interpretation task. However, the low signal-to-noise ratio of the freely available Sentinel-1 Synthetic Aperture Radar (SAR) imagery, the ambiguity of backscatter signals for ice types, and the scarcity of open-source high-resolution labelled data makes automating sea ice mapping challenging. We use Extreme Earth version 2, a high-resolution benchmark dataset generated for ML training and evaluation, to investigate the effectiveness of ML for automated sea ice mapping. Our customized pipeline combines ResNets and Atrous Spatial Pyramid Pooling for SAR image segmentation. We investigate the performance of our model for: i) binary classification of sea ice and open water in a segmentation framework; and ii) a multiclass segmentation of five sea ice types. For binary ice-water classification, models trained with our largest training set have weighted F1 scores all greater than 0.95 for January and July test scenes. Specifically, the median weighted F1 score was 0.98, indicating high performance for both months. By comparison, a competitive baseline U-Net has a weighted average F1 score of ranging from 0.92 to 0.94 (median 0.93) for July, and 0.97 to 0.98 (median 0.97) for January. Multiclass ice type classification is more challenging, and even though our models achieve 2% improvement in weighted F1 average compared to the baseline U-Net, test weighted F1 is generally between 0.6 and 0.80. Our approach can efficiently segment full SAR scenes in one run, is faster than the baseline U-Net, retains spatial resolution and dimension, and is more robust against noise compared to approaches that rely on patch classification.

Antarctic sea ice prediction has garnered increasing attention in recent years, particularly in the context of the recent record lows of February 2022 and 2023. As Antarctica becomes a climate change hotspot, as polar tourism booms, and as scientific expeditions continue to explore this remote continent, the capacity to anticipate sea ice conditions weeks to months in advance is in increasing demand. Spurred by recent studies that uncovered physical mechanisms of Antarctic sea ice predictability and by the intriguing large variations of the observed sea ice extent in recent years, the Sea Ice Prediction Network South (SIPN South) project was initiated in 2017, building upon the Arctic Sea Ice Prediction Network. The SIPN South project annually coordinates spring-to-summer predictions of Antarctic sea ice conditions, to allow robust evaluation and intercomparison, and to guide future development in polar prediction systems. In this paper, we present and discuss the initial SIPN South results collected over six summer seasons (December-February 2017-2018 to 2022-2023). We use data from 22 unique contributors spanning five continents that have together delivered more than 3000 individual forecasts of sea ice area and concentration. The SIPN South median forecast of the circumpolar sea ice area captures the sign of the recent negative anomalies, and the verifying observations are systematically included in the 10-90% range of the forecast distribution. These statements also hold at the regional level except in the Ross Sea where the systematic biases and the ensemble spread are the largest. A notable finding is that the group forecast, constructed by aggregating the data provided by each contributor, outperforms most of the individual forecasts, both at the circumpolar and regional levels. This indicates the value of combining predictions to average out model-specific errors. Finally, we find that dynamical model predictions (i.e., based on process-based general circulation models) generally perform worse than statistical model predictions (i.e., data-driven empirical models including machine learning) in representing the regional variability of sea ice concentration in summer. SIPN South is a collaborative community project that is hosted on a shared public repository. The forecast and verification data used in SIPN South are publicly available in near-real time for further use by the polar research community, and eventually, policymakers.

Information on snow depth on sea ice and bulk sea ice density is required to convert CryoSat-2 radar freeboard ( $F_{r}$ ) into sea ice thickness (SIT). It is difficult to obtain their information on an Arctic basin scale; therefore, most CryoSat-2 SIT products largely rely on the distributions of snow depth and bulk sea ice density derived from parameterizations, which are based on sea ice type and climatological values. Several observational studies have found that the distributions of parameterized variables are inaccurate compared to the actual distributions. This study aims to develop a new type of retrieval algorithm for snow depth, SIT and bulk density, and ice freeboard in the Arctic winter by synergizing active CryoSat-2 with passive microwave and infrared measurements. Two parameterizations for the snow–ice thickness ratio and bulk sea ice density were combined with the hydrostatic balance and radar wave speed correction equations. Consequently, solutions for the four target variables were obtained and applied to different CryoSat- $2~F_{r}$ , derived from empirical and waveform-fitting (WF) retracker algorithms. The retrieved thickness-related parameters based on $F_{r}$ from the Lognormal WF retracker algorithm showed good agreement with the airborne snow depth, total freeboard, and mooring ice draft measurements. The retrieved multiyear sea ice bulk density was significantly higher than the value of 882 kg $\cdot ~\text{m}^{-3}$ , which was used in the previous density parameterization, showing a higher agreement with values from in situ measurements. The spatial and interannual variabilities of SIT increased when the results from this study were compared with those based on previous parameterizations.

<p>Over the past decade, Antarctic sea ice extent exhibited a sequence of record maxima, followed by a rapid decline in 2015/16, and record minima since. In this presentation, we show that this sudden and remarkable ice loss marks an abrupt transition from a high to a low ice state that cannot be explained by year-to-year variability. Instead, it is most likely associated with a longer term variability arising from ice–ocean feedbacks. The abrupt transition was preceded by a multi-decadal increase in persistence and variance of the sea ice anomalies, an increasing upper Southern Ocean density stratification, and an accumulation of heat at the subsurface; suggesting a decoupling of the surface from the subsurface ocean. During this period, the sea ice anomalies shifted from being structured predominantly regionally and seasonally to a largely circumpolar and interannual regime. In 2015/16, the upper ocean density stratification in the ice-covered region suddenly weakened, leading to a release of heat from the subsurface, contributing to the sea ice decline during winter. Our analysis suggests that the sudden sea ice loss in 2015/16, and the persisting low ice conditions since, arose from a systematic change in the physical state of the coupled circumpolar ice–ocean system. This change will have wide implications for global climate, ecosystems, and the Antarctic Ice Sheet.</p>

We have developed a technique to track individual Arctic sea ice parcels as they advect through the Arctic Ocean. This method utilizes our sea ice motion product to create a lagrangian dataset containing hundreds of sea ice parcel trajectories with weekly latitude/longitudes on a 25km EASE-grid. As the sea ice parcels drift, they are periodically overflown by NASA’s ICESat-2 satellite. A gridded sea ice freeboard product derived from the ICESat-2 altimeter allows for the tracking of freeboard for tracks that intersect a significant number of freeboard grid cells. We show a collection of tracked sea ice parcels over a recent year, and track the ICESat-2 sea ice freeboard product to assess the rate of sea ice growth. Simultaneous tracking of ice surface temperature and ice concentration will show the influence of these properties on sea ice growth rates.

We assess the skill of forecasts of Arctic September sea ice in the Sea Ice Outlook over 2008–2022. The multi‐model median June initialized forecast of September sea ice extent (SIE) is slightly more skilled (RMSE = 0.48 million km 2 ) than a damped anomaly forecast, but July and August initialized forecasts (RMSE = 0.52 and 0.36 million km 2 respectively) do not beat this benchmark. The skill of individual dynamical and statistical SIE forecasts is lower than the multi‐model median forecast skill. Overall skill is lower than expected from retrospective forecasts. Several forecasts initialized in early September 2021 and 2022 imply physically improbable values. Spatial forecasts of sea ice concentration show multi‐model forecast skill and an improvement in individual forecast skill in recent years. Initial conditions show large spread in sea ice volume and a positive correlation between initialized sea ice volume and September SIE forecast. Summer weather has an impact on forecast error.

Sea ice is an essential component of the Arctic climate system. The Arctic sea ice cover has undergone substantial changes in the past 40+ years, including decline in areal extent in all months (strongest during summer), thinning, loss of multiyear ice cover, earlier melt onset and ice retreat, and later freeze-up and ice advance. In the past 10 years, these trends have been further reinforced, though the trends (not statistically significant at p <0.05) in some parameters (e.g., extent) over the past decade are more moderate. Since 2011, observing capabilities have improved significantly, including collection of the first basin-wide routine observations of sea ice freeboard and thickness by radar and laser altimeters (except during summer). In addition, data from a year-long field campaign during 2019–2020 promises to yield a bounty of in situ data that will vastly improve understanding of small-scale processes and the interactions between sea ice, the ocean, and the atmosphere, as well as provide valuable validation data for satellite missions. Sea ice impacts within the Arctic are clear and are already affecting humans as well as flora and fauna. Impacts outside of the Arctic, while garnering much attention, remain unclear. The future of Arctic sea ice is dependent on future CO2 emissions, but a seasonally ice-free Arctic Ocean is likely in the coming decades. However, year-to-year variability causes considerable uncertainty on exactly when this will happen. The variability is also a challenge for seasonal prediction.