Detection of coastlines in SAR images using wavelet methods

To monitor, map, and model sediment transport in big river estuaries, complete maps of bottom topography are needed. This information is obtained by extracting shorelines from synthetic aperture radar (SAR) images of the European remote sensing satellites (ERS-1R), tide gauge information, and heights from a dynamic tide model. The aim of this paper is to show how coastlines can be derived from SAR images by using wavelet and active contour methods. In a first step, an edge detection method suggested by Mallat et al. (1992) is applied to ERS SAR images to detect all edges above a certain threshold. A blocktracing algorithm (BA) then determines the boundary area between land and water. A refinement is achieved by local edge selection in this coastal area and by propagation along the wavelet scales. In a final step, the refined edge segments are joined by an active contour algorithm. A model solution for land water segmentation is constructed from a SAR image by visual inspection. The error is estimated by comparing the results from the algorithms to this model solution. The area under investigation is the Elbe estuary, which is located in the intertidal zone of the German Bight. Due to high rates of sediment transport in this area, local government agencies have to spend large amounts of money to keep shipping lanes open. The demonstrated method is shown to be suitable for operational use by applying it to a data set of 24 SAR images covering a time span of approximately five years.