Iceberg Detection and Characterisation in Sea Ice using Satellite Radars


The polar regions are unique, but fragile ecosystem that are increasingly threatened by changes to our climate. In this project, we will develop novel techniques for identifying and measuring small icebergs in satellite radar images. As icebergs represent a significant hazard for shipping and offshore activities, any improvement in the ability to detect and track their progress along the Polar regions would be valuable [1,2]. Collisions with icebergs pose a threat not just for lives and goods, but also for the environment, since they may result in oil spills or other ecological disasters. The reduction in sea ice across the Arctic has driven an increase in navigating these waters, especially through the North West and North East Passages which promise large reductions in travel time.

Spaceborne synthetic aperture radar (SAR) systems can image the ocean and their sea ice regions independent of light and cloud conditions and they can allow iceberg monitoring. We will also use artificial intelligence and computer vision. The latter can provid powerful methodologies to identify small objects in satellite images, measure their size and track them in sequential images.

The PhD project will focus on improving detection of icebergs in open and ice-infested waters in surrounding level ice in polarimetric SAR data using algorithms rooted in detection theory, pattern recognition, and machine learning/deep learning. The methods will be tested in Arctic and Antarctic areas under different wind conditions and open water and/or sea-ice clutter situations. In addition to polarimetric characterization of icebergs, their detection and separation from other marine targets, the research will propose to use innovative methodologies as single-pass interferometric SAR observations acquired by TanDEM-X for the retrieval of iceberg freeboard and volume trapped in deformed sea ice or open water in a relatively short time window. In summary, the project aims to

* investigate the characterization of iceberg using polarimetric SAR.
* develop a new iceberg detector in open water and sea ice.
* test the ability of polarimetric SAR to discriminate ships, sea ice, and icebergs using machine learning/deep learning techniques.
* investigate the retrieval of iceberg freeboard and height from single-pass InSAR observations and cross-validate using other source of information.

Click on an image to expand

Image Captions

Figure: (a) Radar scattering mechanisms for iceberg and its surrounding open water and sea ice. (b) Icebergs are seen floating or grounded in open water or trapped in sea ice. (c) Example of RS-2 brightness image of Kongsfjorden, Svalbard. Icebergs are seen floating or grounded on a heterogeneous background. A: area with high iceberg density. B: area with nonuniform wind conditions. C: icebergs surrounded by ice-infested waters. D: one of the seven islands and islets.


The PhD project aims to advance the detection and characterisation of icebergs in heterogeneous background conditions i.e. drifting within regions of open water or embedded within sea ice. Both scenarios present different detection challenges. To this end polarimetric and single-pass interferometric SAR images represent the most advanced data type available from imaging space-borne radar systems. The ambitious, but achievable, goal of this project is (i) the testing and improving of already existing algorithms [3,4] and (ii) the development of novel and robust methods rooted in detection theory, pattern recognition, and machine learning specifically adapted to the targets in question. The application of polarimetric and single-pass interferometric satellite data for iceberg characterization and detection have the greatest potential for success over other data where sea ice conditions are highly complex (e.g., various ice types of different age, topography and thickness), where dense iceberg field exist or where there is a strong contrast between water and ice. Accurate detection and tracking of icebergs are of high impact for the safety of marine traffic and operations, and scientific applications such as studying the interactions between atmosphere, sea ice/icebergs, and ocean. The research aims to address the following points:

1) To characterize icebergs in open and ice-infested waters using polarimetric synthetic aperture radar
The project plans to investigate the application of the polarimetric SAR data for a characteristic analysis of icebergs from the surrounding sea ice or open water under different challenging environmental conditions. The polarimetric characteristics of icebergs and different sea ice types will also be compared with less rich polarimetric datasets, to evaluate the advantages brought by polarimetry. The PhD candidate will evaluate which polarimetric feature or scattering model can do a better job in modelling the appearance of icebergs in images.

2) To detect icebergs in open and ice-infested waters using polarimetric synthetic aperture radar
The use of conventional constant false alarm rate (CFAR) detector has been of great use over the years for detection of marine targets over open water. However, to detect smaller icebergs embedded in sea ice we need more powerful methodologies as the one adopted in [3,4]. The background of SAR images can vary greatly depending on the target they observe, (e.g. different sea ice types and open water). The project aims to use deep convolutional neural networks (CNNs) to improve the detection capabilities. The PhD student will investigate the potential of different polarimetric features with respect to iceberg discrimination from the surrounding sea ice or open water and then use this information for detection of icebergs.

3) To map and estimate icebergs topography and volume using single-pass interferometric SAR
For stationary (grounded) icebergs in fast sea ice, repeat pass monostatic mode is acceptable provided there is no motion of the iceberg between passes. However, for drifting iceberg, bistatic interferometry is required to eliminate iceberg motion effects. The PhD student will address the potential of bistatic single-pass InSAR observations [5] for retrieving topography and volume of icebergs. The results of InSAR-derived iceberg height and volume will be cross-validated using other source of information such as NASA’s ICESat-2 and Operation IceBridge (OIB). The comparison between ICESat-2, TanDEM-X, and OIB for iceberg height is one major objective of this project.

Project Timeline

Year 1

– literature review and skill development, including the use of remote sensing technologies for iceberg characterisation, deep learning/machine learning methods for iceberg detection, and InSAR and altimetry for retrieval of iceberg topography and volume.

– Obtain very high-resolution satellite imagery and manual annotation of images under different background conditions.

Year 2

– SAR remote sensing data processing, feature extraction from polarimetric data for iceberg characterization.

– Machine learning/deep learning for Iceberg detection.

Year 3

– Further analysis and validation of iceberg detection results.

– Iceberg/ice bridge topography and volume retrieval from single-pass InSAR observations.

Year 3.5

The final 6 months will be focused on results dissemination and thesis preparation.

& Skills

– Divisions of Biological and Environmental Sciences and of Computer Science and Mathematics at Stirling University will provide expertise in icebergs, SAR remote sensing and machine learning/deep learning and pattern recognition.

– British Antarctic Survey will work collaboratively with the student to increase their knowledge on sea ice and iceberg processes, provide industry needs/interaction with respect to iceberg detection, and to help with the Calibration/Validation of SAR images including labelling them as input to the deep learning techniques, validation and to provide a supportive environment for the student to interpret their results.

References & further reading

[1] D. I. Benn, C. R. Warren, and R. H. Mottram. Calving processes and the dynamics of calving glaciers. Earth Sci. Rev., 82(3):143-179, June 2007.[2] I. D. Turnbull, N. Fournier, M. Stolwijk, T. Fosnaes, and D. McGonigal. Operational iceberg drift forecasting in Northwest Greenland. Cold Reg. Sci. Tech., 110:1-18, February 2015.[3] A. Marino, W. Dierking, and C. Wesche. A depolarization ratio anomaly detector to identify icebergs in sea ice using dual-polarization SAR images. IEEE Trans. Geosci. Remote Sens., 54(9):5602-5615, September 2016.[4] V. Akbari and C. Brekke. Iceberg detection in open and ice-infested waters using C-band polarimetric synthetic aperture radar. IEEE Trans. Geosci. Remote Sens., 56(1):407- 421, Jan. 2018.[5] D. O. Dammann, L. E. B. Eriksson, S. V. Nghiem, E. C. Pettit, N. T. Kurtz, J. G. Sonntag, T. E. Busche, F. J. Meyer, and A. R. Mahoney: Iceberg topography and volume classification using TanDEM-X interferometry, The Cryosphere, 13, 1861-1875, 2019.

Further Information

Armando Marino, University of Stirling,
Vahid Akbari, University of Stirling,

Apply Now