Abstract
Among the basic geophysical ocean parameters which can be derived from multichannel passive microwave data are sea ice concentration and ice (or surface) type. Ice concentration can be derived using the Bootstrap algorithm with estimated precision of about 2% to 10% in generally thick consolidated ice areas where the emissivities are relatively stable. In the formalism, the reference brightness temperature of water is assumed to be constant while that of sea ice is allowed to vary according to the location of the input data relative to the multichannel distribution of the consolidated ice data points. Larger uncertainties are expected in marginal ice zones, leads, polynya regions, and meltponded regions where the emissivities of the ice surfaces are considerably more unpredictable. An unsupervised cluster analysis technique, aided by a neural network, has also been developed to provide complementary information about the ice cover. This technique makes use of the observed clustering of radiometrically distinct data points in n-dimensional brightness temperature space, where n represents the number of channels. Atmospheric and surface effects are minimized with the use of a neural network back propagation technique. The technique is shown to provide consistent identification of the various ice regimes in the marginal ice zones, the seasonal ice region, and the perennial ice regions. Observed temporal changes in the ice concentration and classification appear to be realistic and are consistent with known seasonal growth, decay and advection characteristics of the sea ice cover.
