Urban water body detection from the combination of high-resolution optical and SAR images

This paper proposes an automated water body detection method to delineate detailed water bodies from high-resolution satellite images. It consists of three steps: a) coarse water mask detection from optical imagery using unsupervised classification; b) water mask refinement using backscatter value from synthetic aperture radar (SAR) images; and c) advanced morphological filtering to produce a final water mask. The experiments over Calgary Alberta demonstrate the importance of each step and show the advantages of this method relative to traditional methods, namely, its high degree of accuracy, ability to be full-automated, stability and potential for transferability. It is designed for water mask detection at sub-meter accuracy for industrial and governmental users undertaking hydraulic modeling in an urban environment.