Automatic Shoreline Detection from Video Images by Combining Information from Different Methods

• Published in: Remote sensing (Basel, Switzerland) Vol. 12; no. 22; p. 3717
• Main Authors: Ribas, Francesca, Simarro, Gonzalo, Arriaga, Jaime, Luque, Pau
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 12.11.2020
• Subjects: Accuracy; Algorithms; Automation; Beaches; bias; Cities; Coastal management; coastal regions; Coastal waters; Coastal zone management; Coasts; Computation; detection; Error analysis; evolution; Grain size; humans; information; Mediterranean region; Methods; monitoring; Morphology; Remote sensing; shoreline detection; Shorelines; signal filtering; Spacetime; video monitoring; weighted combination; Mediterranean Sea; Mediterranean region
• ISSN: 2072-4292; 2072-4292
• DOI: 10.3390/rs12223717

Properly registering the time evolution of the shoreline—the coastal land-water interface—is a crucial issue in coastal management, among other disciplines. Video stations have shown to be powerful low-cost tools for continuous monitoring of the coast in the last 30 years. Despite the efforts of the scientific community to get algorithms able to properly track the shoreline position from video images without human supervision, there is not yet an algorithm that can be recognized as fully satisfactory. The present work introduces a methodology to combine the results from different shoreline detection algorithms so as to obtain a smooth and very much improved result when compared to the actual shoreline. The output of the introduced methodology, which is fully automatic, includes not only the shorelines at all available times but also a measure of the quality of the obtained shoreline at each point (called self-computed error). The results from the studied beaches—located in the region of Barcelona city (Spanish Mediterranean coast)—show that such self-computed errors are in general good proxies of the actual errors. Using a certain threshold for the self-computed errors, the final computed shorelines have RMSE (Root Mean Squared Errors) that are in general smaller than 2.5 m in the great majority of analysed images, when compared to the manually digitized shorelines by three expert users. The global RMSE for all dates and beaches is of 1.8 m, with a mean bias <1 m and percentage of retrieval success >95% of the points.