## Location

University of Utah

## Start Date

5-10-1999 10:45 AM

## Description

Understanding the polar ice regimes is fundamental to the understanding of global climate and other geophysical processes. Sea ice characteristics can be grouped into a number of .general sea ice classes. Multisensor data from NSCAT, ERS-2, and SSM/I is reconstructed into enhanced resolution imagery for use in ice type classification. The resulting 12-dimensional data set is linearly transformed through principal component analysis to reduce data dimensionality and noise levels. An iterative statistical data segmentation algorithm is developed using maximum a pasteriori techniques. The conditional probability distributions of observed vectors given the ice type are assumed to be Gaussian. The cluster centroids, covariance matrices, and a priori ·distributions are estimated from the classification of a previous temporal image set. An initial classification is produced using centroid training data and a weighted nearest neighbor classifier. Though validation is limited, the algorithm results in an ice classi1ication which is subjectively superior to a conventional k-means approach.

Multisensor Microwave Sea-Ice Classification

University of Utah

Understanding the polar ice regimes is fundamental to the understanding of global climate and other geophysical processes. Sea ice characteristics can be grouped into a number of .general sea ice classes. Multisensor data from NSCAT, ERS-2, and SSM/I is reconstructed into enhanced resolution imagery for use in ice type classification. The resulting 12-dimensional data set is linearly transformed through principal component analysis to reduce data dimensionality and noise levels. An iterative statistical data segmentation algorithm is developed using maximum a pasteriori techniques. The conditional probability distributions of observed vectors given the ice type are assumed to be Gaussian. The cluster centroids, covariance matrices, and a priori ·distributions are estimated from the classification of a previous temporal image set. An initial classification is produced using centroid training data and a weighted nearest neighbor classifier. Though validation is limited, the algorithm results in an ice classi1ication which is subjectively superior to a conventional k-means approach.