Polar remote sensing of the sea ice covered Oceans was transformed when researchers at the Centre for Polar Observation and Modelling (CPOM) at UCL first realised that the shape of the electromagnetic echo returns emitted by the satellite and travelling back and forth to the Earth surface via the atmosphere could be used to distinguish sea ice from open ocean or leads (cracks of exposed water in the ice). From an altitude of several 100s km it was then possible to measure the elevation difference – or freeboard – between the sea ice and the level water and using Archimede’s principle deduce the total ice thickness.
Using this technique it has been possible to produce for the first time maps of the sea ice thickness and sea surface topography from several radar altimeters (i.e. ERS (1990s), ENVISAT (2000s), and CryoSat-2 (2010s)). One of the major sources of uncertainty in these estimates remains the sub-optimal classification of the returned echoes into sea ice, open ocean, and lead categories. Currently this classification relies on simple shape characteristics (skewness, peakiness) completed by visual image recognition of SAR imagery.
In this project you will explore the use of state of the art Deep Learning algorithms developed and implemented in the context of exoplanet atmosphere characterization in the Physics and Astronomy Department. Such algorithms will allow an efficient classification of each individual radar echoes at different levels of the radar processing chain developed at CPOM. Training will be achieved with independent imagery from various instruments including from the KuKa radar deployed as part of MOSAiC expedition.