An iterative approach to multisensor sea ice classification

Abstract

Abstract-Characterizing the variability in sea ice in the polar regions is fundamental to an understanding of global climate and the geophysical processes governing climate changes. Sea ice can be grouped into a number of general classes with different characteristics. Multisensor data from NSCAT, ERS-2, and SSM/I are reconstructed into enhanced resolution imagery for use in ice-type classification. The resulting twelve-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 likelihood (ML) and maximum a posteriori (MAP) techniques. For a given ice type, the conditional probability distributions of observed vectors 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 nearestneighbor classifier. Though validation is limited, the algorithm results in an ice classification that is judged to be superior to a conventional -means approach.

Polar Applications of Spaceborne Scatterometers

“…Those using pencil-beam scatterometers have focused primarily on MY and FY ice classification [48], [62], [63], while others working with fan-beam systems discriminate additional ice types [4], [39], [43], [61], [64], [65]. Other researchers have employed additional sensors to aid in classification [66]- [69].…”

C-band Scatterometers and Their Applications

“…Several studies have investigated the applicability of scatterometer data in various cryosphere research areas for instance; mapping snowmelt extent (Wismann et al, 1997;Wismann, 2000), snow accumulation in Greenland , snow cover over the Northern Hemisphere (Nghiem et al, 2001), frozen terrain in Alaska (Kimball et al, 2001). Other studies have used scatterometer data for determination of freeze/thaw cycles in Northern Latitudes (Bartsch et al, 2007), spatial and temporal variability of sea ice , classification of sea ice in Polar Regions (Remund et al, 2000), deriving the surface wind-induced patterns over Antarctica by measuring the azimuthal modulation of backscatter . In winter when soil surface freezes, dielectric properties of the soil changes significantly which results in low backscatter values.…”

Sea ice mapping method for SeaWinds

“…Bayes risk is related to the probability that a hypothesis (the pixel belongs to class ) is correct given the observation . The decision rule selects the hypothesis ( for sea ice and for ocean) that results in a minimum expected value of Bayes risk (2) where is an arbitrarily assigned loss for selecting , given that (the pixel belongs to class ) is correct. For MAP criterion, is zero for and unity for .…”

“…The Bayes detection formulation in (5) allows for a simple pixel-based test of ensemble sea ice and ocean histograms [fourdimensional (4-D)], where the loss terms essentially shift the threshold of the decision. The choice to assume spatial uniformity for the probability, but not for the loss terms in (2), is somewhat cavalier but will be justified in Sections III-A to III-D.…”