Conference Agenda

As the core SAR component of the Copernicus Program, Sentinel-1 (S1) has delivered exceptional performance across a broad spectrum of applications for over ten years. Not least, S1 has marked a step change in polar satellite Earth observation since 2014. A key strength of the mission is its systematic acquisition strategy over the polar regions, ensuring consistent and repeated coverage of the Greenland and Antarctic Ice Sheets, as well as other polar ice caps. This systematic acquisition approach has enabled to implement operational monitoring of key ice sheet parameters, including ice velocity and ice discharge—capabilities that were formerly limited to specific glaciers and targeted measurement campaigns.

Current ice velocity products continue to face observational limitations, including data gaps in regions with highly variable surface conditions, rapidly flowing glaciers, and shear zones along glacier margins that lead to signal decorrelation. Moreover, the loss of S1-B in December 2021, which increased the mission’s repeat cycle to 12 days, created further significant challenges for ice velocity monitoring. This stimulated efforts to investigate synergies with other satellite systems to maximize the value of S1 data. This includes complementary observations from other missions, such as the L-band SAR constellation SAOCOM. The activities not only helped to mitigate immediate data gaps but also provided important preparation for new and upcoming L-band missions, including the Copernicus expansion mission ROSE-L and the joint NASA/ISRO NISAR mission.

We will present an overview of key achievements and recent advances from activities carried out within the framework of the European Space Agency Climate Change Initiative (CCI) and Polar+ programs, as well as the Copernicus Climate Change Service (C3S) of the European Union, focusing on ice velocity and ice discharge monitoring using SAR Earth observation data. This includes the development of advanced ice velocity products through the integration of complementary techniques such as InSAR and offset tracking, and the combined use of multi-sensor datasets operating at different frequency bands, including C-band and L-band. The resulting velocity fields, together with ice thickness data, provide the foundation for calculating and analyzing changes in ice discharge, needed to estimate freshwater fluxes and overall ice sheet mass balance.

We show that integrating L-band SAR with S1 C-band data allows the generation of detailed ice velocity maps from crossing-orbit InSAR, exceeding what can be achieved using S1 alone. The deeper penetration of L-band within the snowpack ensures high coherence between image pairs over longer time intervals when surface processes and snowfall, wind drift and melt cause changes on the surface. Moreover, L-band data exhibit lower fringe density within shear margins and over fast-flowing glaciers, which improves phase unwrapping reliability in heavily crevassed areas where C-band data frequently decorrelate. This enhanced performance facilitates more robust retrieval of ice velocities, particularly in rapidly moving regions. However, our results also point out the higher sensitivity of L-band to ionospheric effects that can occasionally reduce performance. In practice, combining L-band with S1 C-band is often the optimal strategy.

Our findings demonstrate the complementarities and opportunities of S1 and other current and upcoming SAR missions ensuring a long-term monitoring of essential climate variables, providing critical data for understanding ongoing changes in the polar regions.

NISAR launched in july 2025 provides a comprehensive coverage of Antarctica in left looking mode. Sentinel-1 d launched in November 2025 has provided an extensive coverage of Antarctica at a one day repeat cycle for six months. iCEYE US and OY operate a constellation of Xband SARs at a one day repeat cycle. In this talk we will show early results of NISAR in grounding line and ice velocity mapping with phase only in areas not well covered in the past, new grounding lines with sentinel-1 and also fine details in key areas with ICEYE. We will show examples of seawater intrusions and close interactions with the subglacial water system.

Frequent and precise monitoring of Antarctic ice shelf calving front locations is essential for constraining ice shelf mass balance and understanding dynamic ice-ocean interactions. Beyond ice dynamic change, recent studies have shown that calving events also have an important role to play in triggering ocean mixing through the genesis of submarine tsunamis. Historically, calving front measurements have been made through manual delineation of satellite images, which is time consuming to produce and reduces the feasibility of making regular repeat measurements over large areas. Developments in the field of deep learning and Artificial intelligence (AI) provides an opportunity to overcome these methodological limitations, enabling automatic delineation of the calving front location. Here, in this study we use a decade of Sentinel-1 Synthetic Aperture Radar (SAR) data combined with AI methods to automatically delineate the location of ice shelf calving fronts in Antarctica.

While Sentinel-1 SAR data provides the continuous, weather-independent imagery necessary for this task in the polar regions, automated delineation is frequently hindered by the ambiguous scattering surface at the ice-ocean boundary. Dense ice mélange (sea ice and broken iceberg mix) often mimics the backscatter properties of the crevassed ice shelf surface, making the boundary difficult to delineate independent of the method used. In contrast, most delineation methods perform well when identifying the calving front boundary where an ice shelf meets open ocean, as the boundary between these two features is clearly distinct. To overcome these complex physical variations, deep learning approaches require extensive, manually labelled training datasets, which creates a significant bottleneck when considering continent-scale monitoring.

To overcome the training data bottleneck, we present a novel, weakly supervised workflow for calving front delineation. Rather than relying on manual delineation, our approach automatically generates pseudo-labels for training by using existing coastline datasets combined with an unsupervised clustering approach using Gaussian Mixture Models (GMMs) to differentiate distinct surface types (e.g., solid ice, open water, and ice mélange). We engineer a feature space to emphasise the textural differences between these surface types and improve the accuracy of the annotation. We train a U-Net segmentation architecture on these automatically generated labels, where the calving front boundary is initially masked out. Our results show that the model learns robust feature representations of the distinct ice and water classes, while ignoring the high-uncertainty boundary zones. This enables the model to accurately infer the boundary when exposed to the full, unmasked SAR scenes.

Our results show that this weakly supervised framework successfully circumvents the need for manual annotation while effectively mapping complex ice-water margins. We investigated the performance of this architecture quantitatively by evaluating its boundary delineation accuracy against a manually annotated validation subset of Sentinel-1 imagery and a high-resolution dataset from ICEYE. We extend the intercomparison to also compare our satellite derived results to in-situ imagery of the calving front acquired during a field campaign, which enables us to more accurately time stamp the calving events to understand more about the changes in surface characteristics before and after the calving occurs.

By automating both the training data generation and the inference pipeline, this scalable approach paves the way for continent-wide, high-temporal-resolution calving front tracking in Antarctica, providing vital boundary conditions for predictive ice sheet and ice shelf modelling work. Future studies should apply this method more widely to other regions in Antarctica, in order to deliver a continent-wide monitoring system for ice shelf calving front locations.

Ice cover provides an effective and economical base for seasonal roads. These roads facilitate valuable land transport links to e.g. communities and industries in isolated cold regions of Canada and other northern countries. Current methods to ascertain the trafficability of ice roads involve in situ measurements and are therefore time-consuming, hazardous and expensive. Climate change adds to a need for alternative methods because it diminishes the value of historical expertise. This paper focuses on a section of the Tibbitt-to-Contwoyto Winter Road (TCWR) in the Northwest Territories of Canada. The TCWR is dedicated to the transport of goods that are critical for the year-round operation of diamond and gold mining industries that contribute to Canada’s GDP in a major way.

The suitability of an ice road to traffic depends on several variables. This includes ice cover characteristics as well as other variables such as weather conditions. Logically, the thickness of the ice is of great importance. However, the composition and integrity of the ice also play a role. For example, an ice cover comprised of columnar ice (aka black ice) has a larger loading capacity than an ice cover that includes snow ice (aka white ice)—assuming both ice covers are equally thick. Similarly, the presence of e.g. cracks can negatively affect an ice cover’s trafficability. Earlier studies have shown that conventional, high frequency, dual-polarization (or better) radar systems can provide valuable information regarding the distribution of ice types and the presence of hazardous features such as cracks and ridges. The application of SAR technology to estimate the total ice thickness remains under development. This paper will describe the initial results of a study that aims to assess and develop the utility of Pol-InSAR and SAR Tomography for the mapping of ice cover composition, i.e. its vertical structure due to the presence of different ice types and/or ice impurities such as gas bubbles.

The present study reports preliminary results achieved by means of data that were acquired by the F-SAR system of the German Aerospace Center (DLR) in the context of the PermASAR19 Campaign. Two tomographic flight lines composed of 9 and 6 passes, with relative horizontal baselines ranging from 0 to 100 m and 55m respectively, and two other InSAR flight lines (e.g. two passes separated by a 15 m horizontal baseline). For each SAR flight pass, fully polarimetric acquisitions were taken at X, C and L bands, where incident angle ranging from 27^o to 56^o along SAR range.

To support the airborne campaign, in situ field measurements were collected during the same week of the airborne acquisitions. 11 GPR (Ground Penetration Radar) transects, 54 ice cores (extraction and characterization i.e. thickness of each ice type (clear ice, snow ice, or slush ice) and snow cover thickness) were collected. Snow dielectric measurements were taken routinely and the local air temperature, at the time of the field visit, was recorded occasionally.

PolInSAR coherence diagrams derived over the ice‑core sites revealed clear polarization‑dependent variations in both coherence amplitude and phase across all three radar wavelengths. Depending on the local incidence angle and perpendicular baseline, several local polarimetric coherence diagrams follow the Random Volume over Ground (RVoG) scattering model, where polarimetric interferometric phase diversity increases at shorter wavelengths and with greater ice‑layer thickness. However, X‑band coherence results become noticeably noisier when the perpendicular InSAR baseline exceeds ~40 m.

A series of tomograms was produced using both the Capon and MUSIC algorithms for six polarimetric configurations: HH, VV, HV, RR, RL, and HH–VV. The MUSIC method consistently yields sharper vertical spectral peaks than Capon, enabling clearer identification of scattering‑center displacement across polarizations. While tomographic peak widths show partial correlation with ice thickness, no consistent relationship with ice type is observed. Nonetheless, spatially unstable tomographic profiles suggest the presence of vertically layered ice columns (e.g., snow ice, slush ice, and clear ice). As observed in the PolInSAR results, X‑band tomograms are more sensitive to large spatial baselines, which introduce noisier coherence terms into the tomographic covariance matrix. Reducing the range of spatial baselines helps refine the resulting vertical structure profiles.

Overall, the analysis of PolInSAR and SAR tomographic data over lake ice remains challenging due to spatial and vertical heterogeneity, combined with the inherently low dielectric constant of pure ice and the relatively thin ice layers (50–158 cm). These factors reduce both top‑interface interactions and volumetric scattering contributions. More advanced quantitative analyses and scattering‑model evaluations will be presented at the conference.

Landfast sea ice is a critical component of Arctic coastal ecosystems, indigenous subsistence activities, maritime operations, tourism, and over‑ice logistics. Despite its importance, consistent seasonal and long‑term monitoring of fast‑ice extent and mechanical stability remains limited. We are developing a new operational InSAR‑based workflow funded by the ESA Arctic Phi‑lab, with additional support from the Svalbard Integrated Arctic Observing System (SIOS) and the Svalbard Environmental Protection Fund. The system provides fully automated, large‑scale mapping of landfast sea ice using Sentinel‑1 coherence estimates, enabling reliable detection of fast‑ice presence during the winter season. Multiple orbits are utilized to generate a Svalbard‑wide dataset with an effective 6‑ or 12‑day repeat cycle.

Beyond ice‑edge detection, we derive fast‑ice deformation and mechanical response to environmental forcing to infer stability. This is achieved by translating phase gradients into localized strain estimates on the order of centimetres per kilometre. The result is a near‑operational system capable of routinely characterizing both the extent and dynamic behaviour of landfast sea ice. The aim is for the system to be used routinely by ice analysts at the Norwegian Sea Ice Operational Service by the end of the project to improve landfast ice mapping for maritime safety and to evaluate decadal‑scale changes in fast‑ice extent and stability.

Validation efforts is planned to include comparison with airborne measurements of ice and snow thickness as well as ice roughness collected by the Alfred Wegener Institute (AWI), as well as evaluation against optical and SAR‑based mapping carried out by the Norwegian Sea Ice Operational Service. While InSAR techniques for fast‑ice mapping have been demonstrated previously, our focus is on developing a fully automated end‑to‑end processing chain designed for continuous, regional‑scale, operational monitoring, with potential for future expansion to other relevant Arctic coastal regions.

With the processing chain now completed, early outputs reveal substantial spatio‑temporal variability in winter ice mobility, including localized regions experiencing stronger dynamic responses. Systematic, Svalbard‑wide archives are currently being generated, enabling long‑term analyses of fast‑ice regimes and their response to climate‑driven environmental forcing.

This project demonstrates the feasibility and value of scaling Sentinel‑1 InSAR for continuous monitoring at an operational level and for generating new data products with widespread relevance to Arctic research and operations, spanning marine‑mammal ecology, biological and cryospheric systems, climate‑driven environmental change, maritime activities, coastal community resilience, and interdisciplinary scientific research.