Site-Dependent classes for the classification of intertidal sediments

The biophysical properties of intertidal sediments highly affect the stability of an intertidal area. Therefore,the characterization of intertidal sediments according to major biophysical properties is essential. Remote sensing technology has been offering great alternatives to classical data collection methods, where imagery is frequent and gives full spatial coverage of the study area. Various methods have been used to characterize sediment properties using remotely sensed imagery, including the highly popular pixel-based supervised classification. The typically considered biophysical properties are grain-size distribution, organic matter content, moisture content, and chlorophyll a content. To carry out a supervised classification, classes have to be specified in advance. In literature, these classes have been selected using various scientific or case-dependent justifications due to the challenging fuzzy nature of sediment classes. For example, the limits between wet and dry or sandy and muddy sediments are not easily defined as hard boundaries. Due to this case-dependent nature of the classification, comparing classified imagery of different study sites or different images of the same site has not been practical. This paper addresses the possibility of finding site-dependent classes, instead of case-dependent classes for the classification of the IJzermonding, an intertidal flat in Belgium. This is carried out using field spectra of various years. These spectra indicate the possibility of obtaining thresholds for the different properties using unsupervised classification. On this basis, thresholds for sediment classes are chosen. In a last step, the image is classified in a supervised manner by means of the Bayesian Pairwise Classifier approach using the thresholds set in the previous step. Finally, the classification accuracy is compared with respect to other used thresholds from literature.