The research group of Dr. Chen Hui from the School of Marine Science and Engineering, Nanjing Normal University, and the Jiangsu Provincial Key Laboratory of Coastal Ocean Environment–Land Evolution and Ecological Construction, has published a research paper titled "Human impacts versus environmental drivers of seagrass loss in the Nansha islands: Evidence from domain-adaptive deep learning" in the International Journal of Applied Earth Observation and Geoinformation (CAS Q1 Top, IF: 8.6). This study focuses on the typical blue carbon ecosystem of seagrass beds in the Nansha Islands. Addressing bottlenecks such as remote location, scarce samples, complex marine environment, and difficulties in long-term monitoring, the research proposes a novel seagrass identification method called SeagrassMNet, driven by the synergy of physical constraints and deep learning, offering a new approach for shallow-sea ecological remote sensing monitoring.
The research group integrated adversarial domain adaptation and multi-temporal feature alignment strategies, effectively overcoming the challenge of poor model generalization in complex marine environments. Results show that the model achieved an overall accuracy of 95.89% and an IoU of 92.56% in seagrass identification tasks, significantly outperforming mainstream models and demonstrating strong cross-region and cross-year transferability. Using this method, the group constructed a multi-temporal distribution map of seagrass beds in the Nansha Islands, systematically revealing a significant seagrass degradation process: the total seagrass area decreased from 6.89 km² to 3.48 km², a overall reduction of 47.7%. Further analysis indicates that while environmental factors such as sea surface temperature anomalies and sea surface height changes show significant correlations with seagrass changes, human activities—particularly artificial island construction and coastal modification—exhibit stronger correlations with seagrass loss, highlighting the important role of anthropogenic impacts in regional seagrass degradation.
This achievement improves the accuracy of shallow-sea ecological identification under scarce sample conditions methodologically, fills the gap in long-term continuous mapping of seagrass in the Nansha Islands in application, and deepens the scientific understanding of the synergistic mechanisms between natural drivers and anthropogenic disturbances in seagrass degradation. It can provide important scientific and technological support for blue carbon resource assessment, marine ecological protection, and fine-scale island management in the South China Sea.