As part of Blue Carbon ecosystems (BCEs), detached macrophytes can be transported to the coast due to current and wave actions, and then deposited on the shore as beach wrack. To date, the role of beach wrack in the material cycle in BCEs is still unclear. In order to track the fate of beach wrack, this study conducted a monitoring survey on a semi-sheltered beach in Odense Fjord (Denmark) using camera trap data. Deep learning with a VGG network architecture was used to classify the image dataset acquired by the camera trap. The VGG network demonstrated the capability to identify beach wrack from different beach scenes, and the method can provide results on large datasets within a short time (187 images analyzed within 5 min) compared to manual identification of images. By combining the VGG detection with color-based segmentation, beach wrack coverage was determined. To evaluate the impact of ambient conditions on wrack deposition on the shore and relocation back to the sea, wind (including speed and direction), water temperature, and tidal amplitude were analyzed as environmental variables. Partial least squares regression (PLSR) analysis revealed that micro-tidal action with an average amplitude of 0.41 m accelerated the movement of floating macrophytes between the shore and the sea. Despite being exposed to the prevailing southwesterly winds (average speed of 11 m/s), the beach was sheltered due to the location in the inner part of Odense Fjord, limiting the transport of drifting macrophytes from sea to the shore. By using the camera trap to conduct continuous monitoring, this study presents a labor-saving and practical approach to track the dynamics of detached macrophytes deposited on the shore. Furthermore, the application of deep learning in image identification provides a study case for using a large image dataset to assist in ecological studies of dynamic environments.
Read full abstract