Coastal seaweed plays a critical role in sustaining biodiversity, supporting fisheries and mariculture, providing raw material for food and pharmaceutical industries, and sequestering carbon, yet it also poses ecological and socio-economic challenges due coastal degradation and over exploitation. Accurate mapping of seaweed extent is therefore essential for both ecological monitoring and informed coastal policy. Conventional remote sensing techniques, however, are often limited in coastal zones due to spectral confusion with water, sediments, and other vegetation. This study addresses these limitations by integrating freely available Landsat-8 imagery with advanced deep-learning segmentation methods to provide a scalable approach for seaweed monitoring. Focusing on Palghar district, Maharashtra, India, we analyzed 231 cloud-free images spanning 2020-2024. Key spectral indices Floating Algae Index (FAI), NDVI, and a customized Seaweed Enhancement Index were derived to enhance detectability of seaweed canopies. Two segmentation architectures, U-Net and VGG-UNet, were trained using augmented datasets, with FAI serving as the primary input feature. Model performance was assessed with multiple accuracy metrics, including IoU, Dice coefficient, precision, recall, and F1-score.
Results demonstrated a clear advantage of VGG-UNet over U-Net, achieving precision of 0.90 and F1-score of 0.91 compared to U-Net’s 0.43 precision and 0.54 F1-score. Seaweed extent was reliably quantified, with Dhanu showing 9.38% cover (5.54 km²) and Kelwa 7.26% (4.28 km²). Validation against field-based observations confirmed the robustness of the method, even in spectrally complex coastal environments. Beyond technical advances, the study highlights important policy implications. By offering a cost-effective and scalable tool, deep-learning segmentation can support coastal governance frameworks in balancing conservation with economic utilization. Applications include monitoring of invasive species such as Sargassum, early-warning systems for excessive biomass accumulation, seaweed vegetation and growth monitoring and spatial planning for seaweed mariculture and sustainable natural resource harvesting. Integrating such AI-driven monitoring into marine and coastal management policies could reduce ecological risks, enhance climate resilience, and provide decision support for resource allocation. Overall, this study demonstrates that combining Landsat-8 data with deep learning models not only advances remote sensing science but also offers practical pathways for evidence-based coastal ecosystem management and sustainability.
Keywords: Seaweed, Deep Learning, Landsat-8 Imagery, Remote Sensing, Coastal Governance, Marine policy