EGU21-15204, updated on 03 May 2023
https://doi.org/10.5194/egusphere-egu21-15204
EGU General Assembly 2021
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Vessel-based optical data and artificial intelligence for sampling mega-plastic concentrations on the high seas

Robin de Vries, Matthias Egger, Thomas Mani, and Laurent Lebreton
Robin de Vries et al.
  • The Ocean Cleanup, Research Team, Netherlands (robin.devries@theoceancleanup.com)

Remote sensing of marine debris has seen recent successes in coastal regions. However, these approaches focus on the detection of large accumulations of marine debris, often mixed with organic waste and related to events. Individual large plastic items (macroplastics, > 50cm) in remote marine environments are a substantial part of the marine debris surface mass budget, yet  remain poorly quantified.

Current knowledge on the accumulation of macroplastic debris at the ocean surface is mostly limited due to methodological constraints. Macroplastics are typically too large for collection by neuston trawls. Furthermore, the relatively small sea surface area typically investigated during offshore research expeditions often is too small to account for the low areal concentrations of macroplastics. Given the importance of macroplastic in the global ocean plastic mass balance, quantitative information on the spatiotemporal distribution of macroplastics afloat in the surface ocean are urgently needed.

By now, our location-enabled action camera's on-board vessels of opportunity have recorded a vast amount of optical data from the North Pacific and North Atlantic Ocean (approximately 1 million images). By selection and labelling of occurrences of debris in images, we have trained an object detection and localization algorithm. We use the camera’s intrinsic parameters to estimate relevant sampling parameters, such as size and distance of each object detected. An overview of numerical concentrations is generated by combining the object detection solution with bulk processing of the optical data. The first results are promising and well-comparable to sampling methods applicable to smaller debris size classes, such as surface neuston nets.

How to cite: de Vries, R., Egger, M., Mani, T., and Lebreton, L.: Vessel-based optical data and artificial intelligence for sampling mega-plastic concentrations on the high seas, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15204, https://doi.org/10.5194/egusphere-egu21-15204, 2021.

Corresponding displays formerly uploaded have been withdrawn.