Tidal Channel Identification from Remotely Sensed Data Using Spectral and Shape Characteristics

Mapping of tidal channels is significant and fundamental in intertidal hydrodynamic studies. This paper is an attempt to develop a new approach for tidal channel identification from remotely sensed data, by accounting for the spectral and shape properties of channels. The method begins with segmentation of the image ensuring similarity of neighbouring pixels in each segment. A few training pixels selected in the channels locate the training channel segments. The image is classified in non-channel and channel segments based on the statistical similarity, which is established here by T statistic, of unknown segments to known channel segments. It is obvious at this stage, owing to the spectral complexity of tidal regions, that several non-channel areas are also classified as channels. The curvilinear shape of channels, which is established using shape indices, is used to eliminate the falsely classified non-channel areas. The paper presents the methodology and some initial results obtained. It also outlines the scope for further research. * Corresponding author