Graph-based Method for Multitemporal Segmentation of Sea Ice Floes from Satellite Data

Automated segmentation of the sea ice evolution would allow scientists studying climate change to build accurate models of the sea ice meltdown process, which is a sensitive climate indicator. In this paper, we propose a novel approach which uses shape analysis and graph-based optimization for segmentation of a multiyear ice floe from time series of satellite images. Differently of the state-of-theart sea ice segmentation techniques, the new method does not rely on the coherence of the intensity values between successive time moments, but only on the coherence of the shape. We successfully validated the performance of the proposed approach on a set of AMSR-E and MODIS images and estimated the area of a sea ice floe as a function of time.