Abstract: Seagrass possesses extremely high ecological value. To reveal the dynamic distribution patterns of seagrass beds in typical tropical marine areas, this study took the coastal waters of Manado, Indonesia as the research area and constructed a comprehensive dataset integrating temporal features based on Sentinel-2 Level-2A multispectral satellite imagery data. Through preprocessing steps including spatiotemporal filtering, cloud masking and spectral index calculation, combined with the spectral response characteristics of seagrass pixels, classification technology was adopted to achieve high-precision extraction and dynamic monitoring of seagrass bed distribution. In the classification process, the Recursive Feature Elimination (RFE) method was first used to optimize and screen feature indices, and then the Light Gradient Boosting Machine (LightGBM) was applied to realize the accurate classification of seagrass beds. The experimental results showed that the method proposed in this study achieved high precision in extracting the distribution information of seagrass beds in the research area; the area of seagrass beds in the study area fluctuated to a certain extent from 2020 to 2025 (variation < 13%), indicating the overall stability of the seagrass ecosystem. This study verified the feasibility and effectiveness of using optical satellite remote sensing image classification technology to monitor seagrass beds in complex coastal environments, and provided scientific support for the protection of seagrass ecosystems and the management of blue carbon resources in tropical marine areas.

Key words: seagrass, Sentinel-2 satellite remote sensing monitoring, feature fusion, classification technique, temporal and spatial distribution change, Manado region, Indonesia