A fuzzy-possibilistic scheme of study for objects with indeterminate boundaries: application to French Polynesian reefscapes

Abstract

This communication describes the study of an ecological system using remote-sensing data and image-analysis tools derived from possibility theory. Possibility theory enables the construction of membership functions using a multisource fusion algorithm. The sources of information are the sampled training stations. We test:to see if the possibilistic algorithm is able to provide results with an accuracy at least equal to that provided by traditional probabilistic-classification algorithms. Then, for each pixel,we analyze the hierarchy of membership degrees output by the fusion to study the spatial structure of an ecosystem composed of objects: that lack precise boundaries. We characterize patches or gradients,:boundary rates, and transition states. As an example, a scheme of analysis for underwater reefscapes at Moorea Island, French Polynesia, is proposed, The nonparametric multisource fusion method has an accuracy of 82% (overall normalized-percentage agreement), while a probabilistic maximum-likelihood classifier has an accuracy of 73%. The analysis of the hierarchy of membership degrees indicates that almost 25% of Moorea Island lagoon is heterogeneous, composed of real boundaries, transition states, and fragmented zones.