Shoreline extraction on Java’s North Coast based on the satellite imagery: utilizing U-Net segmentation across diverse coastal characteristics

The North Coast of Java, a densely populated region with major urban centers, faces significant anthropogenic pressure and is highly vulnerable to recurrent flooding and land subsidence. These factors accelerate shoreline changes, necessitating effective shoreline monitoring for coastal zone management, risk mitigation, and sustainable development. Various methods, such as field survey based and remote sensing, have been utilized to extract shoreline position. This study used the U-Net Segmentation Technique to accurately delineate shorelines from Sentinel-2 satellite imagery across this dynamic region. The U-Net model, a sophisticated deep learning technique widely recognized for its efficacy in semantic segmentation of remote sensing imagery, was implemented to distinct water and land areas. The available image dataset was rigorously partitioned into training (70%) and independent validation (30%) subsets to ensure robust model evaluation, achieving Intersection over Union (IoU) of 92%. Subsequently, precise shoreline positions were extracted by applying the Laplacian of Gaussian Edge Detection algorithm. Notably, the model’s performance exhibited variability contingent on the specific coastal morphology and land cover characteristics: achieving high accuracy in relatively homogeneous sandy coastal areas and demonstrating moderate to good performance in regions characterized by gravelly substrates and infrastructural development. However, the model encountered inherent challenges in accurately segmenting complex muddy coastal zones, particularly those extensively modified by the presence of fishponds, which often exhibit spectral similarity to shallow water. This research underscores the considerable potential of the U-Net deep learning architecture for enabling consistent and cost-effective yearly shoreline observation at a regional scale, providing essential and timely spatial data for governmental and authorized parties to inform evidence-based decision-making and effectively address pressing environmental issues, including coastal erosion and inundation.