Conference Agenda

Shipping in the Arctic is a huge commercial operation. The presence of icebergs therefore poses a hazard to such operations. Of particular interest are icebergs within sea ice, and a need for automated detection methods. In this work, we utilise a convolutional neural network (CNN) for iceberg detection in fast ice environments. Fast ice is a type of sea ice that forms off of coastlines and remains attached to surrounding land or sea floor. This means that fast ice generally remains in place and is not affected by currents or wind. In Arctic seas, fast ice can extend down to 20 m and has a varied topology depending on environment. Fast ice can be distinguished from drift ice since it does not contain large cracks and fractures. We utilise Sentinel-1 SAR data acquired over Franz-Josef region for this work. Although icebergs show up clearly in optical data, the dependence on conditions such as cloud cover means training with optical data would lead to a less robust program. As such, SAR data is used alongside a Sentinel-2 optical dataset. This SAR data is split by horizontal (HH) and horizontal-vertical (HV) polarisation, with icebergs being clearer in HV polarisation. We also make use of a land mask from the Polar Geospatial Center which helped to aid the training process.

A detection filter used to identify icebergs was proposed by Marino et al (2016). This filter is known as the dual intensity polarisation ratio anomaly detector (iDPolRAD) and has been successfully used in a previous study by Soldal et al (2019) to separate, identify and detect icebergs in sea ice environments. For this work, an iDPolRAD filter is applied to the SAR data to produce training images for a YOLO v8 detection model.

We perform training for 50 epochs with a batch size of 16 and a learning rate of 0.001. For YOLO v8, precision, recall, F1 score and mean average precision (mAP) are used for evaluating the detection performance. Precision measures the ratio between true positives and any new detections (false positives), while recall measures the ratio between true positive and objects the model failed to detect (false negatives). The F1 score acts as a ratio between precision and recall and can be used to determine the optimum confidence score for the algorithm. The mAP score is defined as the total accuracy of the model and is found by taking the area under the Precision-Recall (PR) curve. For model evaluation, we obtained a precision of 0.759, recall of 0.706, F1 score of 0.732 and a mAP of 0.789, giving the model an accuracy of 79%. These results are acceptable for feasible operational use.

The main limitations of this work amount to a lack of an available automated iceberg training dataset, which was addressed by creation of a manual dataset and the continued lack of coverage which can be addressed by future SAR missions (ROSE-L and NISAR). It is hoped that our detection system can be further improved in the future for potential commercialisation.

Height information of semi-transparent media is usually derived with interferometric or tomographic SAR techniques, such as forest height or the penetration depth into glaciers. In contrast, BIOMASS data present the opportunity to derive the thickness of icebergs solely from SAR polarimetry. The large penetration of the long-wavelength signals into ice and the quad-pol polarimetry allow to observe signals originating from the bottom of an iceberg, the ice-water interface. This bottom signal appears like a range-delayed replica of the direct signal from the top of the iceberg with distinct polarimetric characteristics.

The targets of interest are tabular icebergs with sizes of often several kilometers. They are calving from ice shelves, as opposed to their non-tabular, irregularly shaped, smaller siblings calving from marine terminating glaciers. Further, tabular icebergs have a more or less rectangular profile, with a flat top, steep sides, and a relatively flat bottom. Even though this rectangular profile degrades over time, it allows to formulate a first-order model of the range delay between the backscatter from the top and the bottom of the iceberg. More specifically, the range distance between the top and bottom signals, as well as the length of the delayed bottom signal, depends on the thickness and the ground-range length of the iceberg. The permittivity of ice is considered for calculating refraction angles and range delays. Using Archimedes principle, the thickness can be separated into freeboard and submerged draft in order to consider the signals penetrating not only through the top of the iceberg, but also through the frontal freeboard wall.

In single-pol backscatter images, it can be difficult to discriminate the iceberg bottom signal from a top signal. Further, the bottom signal can be also hidden in the surrounding sea ice backscatter. The quad-pol data of BIOMASS allows a clear distinction, in most cases, between the top and bottom signals of an iceberg. An interesting polarimetric pattern can be observed: Icebergs that show a surface-scattering mechanism in the top backscatter, with low polarimetric entropy and alpha parameters, have a strong dihedral contribution in the range-delayed bottom signal, with high alpha values and large phase differences between HH and VV. In contrast, icebergs that show a medium entropy and medium alpha scattering mechanism in the top backscatter, have similar scattering properties also in their bottom signal. A first theory is a stronger volume scattering contribution from inside the iceberg causing both the top and the bottom signal to appear volume-like.

So far, the investigation concentrated on one BIOMASS acquisition from the commissioning phase, where channel imbalance phase and Faraday rotation were corrected, so that the phase differences between different channels have been widely calibrated, making the data ready for polarimetric analysis.

A first estimation, with the first-order geometry model and by discriminating top and bottom signals according to their polarimetric characteristics, resulted in a thickness of 130 m. This iceberg is located in front of the Jelbart ice shelf, which has reported ice thicknesses of about 200 m at the calving front. Further investigations will refine the model formulation and the understanding of the polarimetric characteristics, as well as increase the number of estimated icebergs and include validation.

Snow water equivalent (SWE) is a critical parameter of seasonal snow cover for meteorology and hydrology in northern China and other high-latitude or high-altitude regions with abundant snow resources. However, our ability to accurately measure and monitor SWE change from satellite remote sensing remains a challenge. Traditional passive microwave remote sensing provides daily and large-scale SWE observations, but is limited by its coarse spatial resolution, which is typically tens of kilometers in scale. Repeat-pass Interferometric Synthetic Aperture Radar (InSAR) offers a promising approach to obtaining SWE change at high spatial resolution and accuracy. For this technique, low-frequency (e.g., L-band) radar signals and shorter revisit times are essential for minimizing temporal decorrelation in frequent snowfall regions. This technique has been available until recently due to its limited observations with the optimal radar frequencies and temporal repeat intervals.

This study presents the first demonstration of spaceborne repeat-pass L-band InSAR observations from the Chinese Lutan-1 mission for retrieving SWE changes at Altay, Xinjiang Province, during the winter of 2023–2024. Consecutive 4-day and 8-day repeat-pass interferometric pairs were processed to phase changes, and then related to SWE variations. An InSAR processing chain was developed, including atmospheric phase delay correction (both ionospheric and tropospheric effects), orbital error removal, filtering parameter optimization, and phase calibration. These procedures establish a comprehensive workflow for time-series InSAR SWE retrieval using L-band Lutan-1 data.

The retrieved SWE change shows a good agreement with in-situ SWE observations during the dry snow period (January 12 to February 9, 2024), yielding a root mean square error (RMSE) of 9 mm and a correlation coefficient (R) of 0.48 for the 4-day temporal baselines (p-value << 0.05). However, the accuracy decreases significantly for the 8-day baselines (February 17 to March 28, 2024), mainly due to temporal decorrelation associated with snowfall and snowmelt events. A heavy snowfall observed from February 9 to 17, 2024, induced severe decorrelation, leading to phase unwrapping errors and preventing the retrieval of SWE. This finding emphasizes the necessity of using shorter temporal baselines, such as 4 days, in regions characterized by rapid snow accumulation and ablation processes.

Overall, this study demonstrates the capability of spaceborne repeat-pass L-band InSAR with short revisit intervals to effectively retrieve SWE change under appropriate snow cover conditions. The results also highlight the potential and challenges of operational SWE monitoring from existing and upcoming L-band SAR missions, such as JAXA’s ALOS-4, NASA’s NISAR, and ESA’s ROSE-L, which feature short repeat cycles, wide swath coverage, and high spatial resolution. Future work will focus on improving SWE retrieval accuracy by investigating the impacts of meteorological and environmental factors on InSAR phase.

Plastic pollution represents a major threat to marine ecosystems, leading to biodiversity loss and posing risks to human health and safety. The detection of floating plastic objects through in-situ surveys and direct observations remains a challenging task, as these materials are in constant motion across vast and often inaccessible marine regions. In this context, satellite data can play an important role for monitoring and detecting plastic accumulations thanks to their frequent temporal sampling and broad spatial coverage. While most existing methods for detecting floating plastic islands rely on satellite optical data, the use of Synthetic Aperture Radar (SAR) images for plastic marine litter monitoring has so far been limited, although they offer the advantage of being acquired regardless of sunlight or weather conditions. This study aims at investigating the capability of multi-frequency and dual-polarization SAR data to identify a small floating plastic island, approximately 30 m x 3 m in size, deployed in a controlled marine environment in the Gulf of La Spezia (Liguria, Italy) in April 2025. During the experimental campaign, X-band dual-polarization COSMO-SkyMed Second Generation (CSG) and C-band dual-polarization Sentinel-1 (S1) images were acquired with different spatial resolutions and imaging geometries, providing the opportunity to assess the detectability of plastic marine litter using different SAR configurations. The dual-polarization covariance matrix was extracted from both datasets, and different polarimetric techniques were applied, including the H-alpha decomposition and the m-chi decomposition. The objective of the study is to identify an optimal polarimetric parameter capable of detecting the floating plastic island by distinguishing it from the surrounding water. In fact, while calm water surfaces are expected to reflect most of the radar signal away from the sensor, the presence of plastic objects increases the surface roughness, resulting in the signal being scattered in multiple directions. The results of the analysis, which is still ongoing, will be presented during the workshop. However, the preliminary findings already provide promising indications regarding the potential to extract different parameters suitable for this challenging task, also considering the very limited number of literature studies that have explored the use of SAR data for similar applications.

The primary source of ocean surface signatures in SAR images is the surface wave field generated by local wind stress. The normalized radar cross section (NRCS) wave spectrum offers a statistical representation of the surface roughness induced by these wind waves and is closely linked to wind speed, wind direction, and overall sea state. Wind-generated waves affect the backscattered radar signal through three main mechanisms: specular reflection, Bragg scattering, and wave breaking. The relative contribution of each mechanism to the received signal depends on the radar observing frequency. For instance, at P-band wavelengths, Bragg scattering is expected to be the dominant mechanism, primarily interacting with wave features on the order of one meter. Comparing wave spectra retrieved across different frequencies enhance understanding of upper ocean dynamics under varying sea state conditions. Since different radar wavelengths are sensitive to different ocean wave scales, multi-frequency analyses also reveal how wave energy is distributed across scales, improving interpretation of sea surface processes.

In this project, we will analyze the NRCS wave spectrum to investigate ocean surface roughness and the structure of the wave field. The Biomass mission offers a unique opportunity to investigate radar observables at longer wavelengths, enabling assessment of NRCS signatures in P-band SAR observations over the ocean. This analysis allows testing the assumption that Bragg scattering dominates the received signal at P-band. In addition to Biomass, we perform multi-frequency comparisons of wave spectra retrieved from Sentinel-1 (C-band) and SAOCOM-1 (L-band). This analysis will further improve understanding of frequency-dependent scattering behavior over the ocean.