Vol. 8, No. 2, 158-179, 2009

Abstract
Intertidal sediment stability is related to physical, biological, and chemical properties of sediments. To characterize these properties, extensive field work is required. Since field sampling on intertidal flats can be inefficient, unsupervised analysis of remotely sensed data offers an alternative. In this study, three unsupervised classification techniques were explored for the extraction of sediment characteristics from airborne hyperspectral data: k-means, the Gustafson-Kessel algorithm, and the mixture of Gaussians model. Simulated datasets based on real data were built and utilised to examine the suitability of the techniques for sediment characterization. The issues of intra-class variability, spectral dimensionality, and the choice of the number of clusters were investigated. The study showed that unsupervised classification methodologies can be used for sediment characterization, and that their performance depends on intra-class variability and feature selection. The mixture of Gaussians model was revealed to be the most suitable of the three techniques. Finally, a hyperspectral image of an intertidal study area was successfully classified in an unsupervised manner using the mixture of Gaussians technique.

ISSN 1729-3782