Seagrass meadows, including eelgrass beds, are expected to contribute to blue carbon, defined as carbon sequestered and stored by the ocean, and to provide co-benefits with ecosystem services such as fisheries. Although the restoration of seagrass meadows has long been undertaken by local non-profit organizations, it is necessary to annually monitor the distribution of seagrass meadows in an inexpensive, simple, and high-resolution manner for the certified assessment of carbon dioxide sequestration.
In this study, we proposed a method for mapping seagrass meadows distributed from subtidal to intertidal zones by employing consumer-grade aerial drone photography and deep learning of Feature Pyramid Network (FPN). Extensive efforts were made to create a four-season dataset for ground truth on the Futtsu Coast in Tokyo Bay located at Futtsu City, Chiba Prefecture. We proposed a universal method for mapping seagrass meadows distributed from subtidal to intertidal zones. Compared to U-Net, FPN was capable of more accurate classification. We proposed optimal pre-processing methods associated with the application of filters. We also discussed the generalization of the model and the applicability and limitation of using each seasonal data.
Chen, J. and Sasaki, J. Mapping of subtidal and intertidal seagrass meadows via application of the Feature Pyramid Network to unmanned aerial vehicle orthophotos. Remote Sensing, 13(23), 4880, 2021. DOI