Rabehi, W.; El Amin Larabi, M.; Benabbou, O.; Kreri, S., and Dellani, H., . Sandy beach mapping using a deep learning approach: Potential method for automated monitoring of Algerian coastal erosion.
Highly exposed to the Algerian stream and the growing sea level rise, the occidental Mediterranean basin is undergoing a repetitive case of erosion, especially on the Algerian coast where the sandy beaches are the most vulnerable due to the lack of their sediment supplies, a phenomenon caused by several factors such oueds/river stations and dams (sediment holders), but also the extinction of sediment sources such coastal dune eroded by extensive urbanization. In this critical situation of clear ecosystem imbalance, spatial mapping of sandy beaches is the first, key step for overall erosion monitoring on the Algerian coast; indeed, even complicated because of the radiometric similarity with several land cover classes such as artificialized and bare areas. The classical classifiers used in remote sensing are often enhanced by manual correction to provide an optimal result, which is time-consuming and cannot be reproducible in multidate mode as an automatic approach for the scientific community. Instead, deep learning methods can provide an efficient and quick tool for the extraction of sandy beach areas using a massif sampling despite the geomorphological variability of beach's sand. The aim of this contribution is to provide a large-scale cartography of sandy beaches over all the Algerian coast using a reproducible approach and open access data (Sentinel-2) to provide a largely missing national data and also to report the efficiency of this technic to the scientific community.