Coastal ice dynamics and event detection from radar imagery

People: Rohith MV and Chandra Kambhamettu

Collaborators: Hajo Eicken, Geophysical Institute, University of Alaska Fairbanks.

Marine radars have been employed to gather data in applications that require near-continuous monitoring and tracking of objects over a wide area from a single viewpoint, independent of weather and light conditions. However, little attention has been paid towards utilizing such systems for the study of long-term phenomena and detecting anomalous environmental events or hazards that may occur infrequently but have potentially significant impacts on coastal populations. Our work builds on infrastructure put in place as part of the Barrow Sea Ice Observatory, a long-term effort to collect sea-ice data in the context of a changing Arctic and provide information on coastal ice conditions of value to local residents and other stakeholders. Barrow, Alaska is the northernmost township in the United States, located on the shores of the Chukchi Sea, with nearby Point Barrow as the country's northernmost point bordering on the Beaufort Sea to the East. Sea ice is a defining feature of the Chukchi Sea, which is ice-covered for much of the year. As part of the Barrow Sea Ice Observatory, we have deployed a land-based Furuno FR7112 10kW, X-band (3.0cm) marine radar for the purpose of observing coastal sea ice dynamics as well as surface vessels in the area. The radar system has an operational range of 11 km, although the effective range varies with atmospheric and ice conditions. We have analyzed ground-based radar imagery of landfast and moving sea ice and developed algorithms to automatically extract tracks of reflectors and derive velocity fields for moving ice, perform landfast edge delineations, detect break-outs and other ice hazard events and identify surface vessels. We have used a combination of dense and sparse optical flow methods to estimate motion in the imagery. Using active contours, the landfast ice delineation is performed with little to no user intervention. The dynamic state of the ice is modeled by Hidden Markov Models and the state transitions are marked out as events associated with ice hazards. Surface vessel characteristics were obtained from principal component analysis of imagery containing known vessel reflectors and were then successfully applied for automated detection of vessels. A comparison of our methods with manual analysis of the imagery shows that the proposed methods perform well on real imagery. In this paper we have proposed some automatic methods for analyzing ground based marine radar imagery. Using a combination of computer vision and machine learning techniques we demonstrated extraction of physical parameters and detection of anomalous events. We are working towards exploiting better temporal sampling in radar imagery and extend our algorithms to improve the accuracy of event prediction and also derive more complex physical attributes of ice motion.

Feature tracking

More results and analysis.

Event detection

More results can be found here.

Edge delineation

Results of manual analysis can be found here.