Multi Sensor Sea Ice Classification
Dinessen, Frode; Killie, Mari Anne; Breivik, Lars-Anders
Norwegian Meteorological Institute, NORWAY
The Norwegian Ice Service provides ice charts covering the Arctic waters from the east cost of Greenland to Novaya Zemlya, each weekday. The charts are mainly based on a manual interpretation of SAR (Synthetic Aperture Radar) data from Radarsat-2 recorded in ScanSAR Wide dual polarization (HH/HV) mode and acquired from the morning overpass. Increased activity in the Arctic has led to a demand for more frequent updates of the ice conditions close to the marginal ice zone. An automatic algorithm to supplement the current ice charts are developed. The algorithm is based on a multi sensor approach using a Bayesian technique to combine classification results from an AVHRR ice edge classification and a Radarsat-2 SAR classification. The AVHRR product is running operational at the Norwegian meteorological institute as part of the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF) project. The product discriminates between ice/water/cloud. The Bayesian SAR classification algorithm utilize dynamic updated ice statistics from the manual derived ice charts. Statistics for mean and standard deviation are calculated from the very close drift ice, close drift ice and open water classes in the latest ice chart. For the purpose of open water/ice discrimination we have used data from both HH and HV channels. The values are normalized to a fixed incidence angel. The use of cross-polarized data from the HV channel has improved the ability to automatic separate between water/ice but the low signal to noise value in this channel gives rise to some image artifacts due to ScanSAR image processing. In the pre-proceessing this effects are reduced by subtracting estimated noise values. Results from the automatic classification will be presented.