Bayesian Image Classification at High Latitudes
Bulgin, Claire1; Eastwood, Steinar2; Merchant, Christopher3
1University of Edinburgh, UNITED KINGDOM; 2Met No, NORWAY; 3University of Edinburgh/University of Reading, UNITED KINGDOM

The European Space Agency created the Climate Change Initiative (CCI) to maximize the usefulness of Earth Observations to climate science. Sea surface temperature (SST) is an essential climate variable to which satellite observations make a crucial contribution, and is one of the projects within the CCI programme. SST retrieval is dependent on successful cloud clearing and identification of clear-sky pixels over ocean. At high latitudes image classification is more difficult due to the presence of sea-ice. Newly formed ice has a temperature close to the freezing point of water and a dark surface making it difficult to distinguish from open ocean using data at visible and infrared wavelengths. Similarly, melt ponds on the sea-ice surface make image classification more difficult. We present here a three-way Bayesian classifier for the AATSR instrument classifying pixels as 'clear-sky over ocean', 'clear-sky over ice' or 'cloud' using the 0.6, 1.6, 11 and 12 micron channels. We demonstrate the ability of the classifier to successfully identify sea-ice and consider the potential for generating an ice surface temperature record from AATSR which could be extended using data from SLSTR.