Multisensor microwave sea-ice classification

A knowledge of polar sea-ice extent, type, and location is a valuable tool in understanding many geophysical processes. This paper presents an iterative statistical technique for ice classification of multisensor remote sensing imagery. NSCAT, ERS-2, and SSM/I data are used to construct enhanced resolution images. Preprocessing consists of data standardization and principal component analysis dimensionality reduction. The iterative algorithm is developed based on maximum a posteriori (MAP) method with the assumption of a Gaussian conditional probability density of each ice type. The technique is applied to data collected during a six day imaging interval in September of 1996. For this application, the algorithm performs better than nearest neighbor, iterative maximum likelihood, or K-means approaches.