AI Used to Understand Scallop Ecology.

A recent peer-reviewed study led by Coonamessett Farm Foundation (CFF) demonstrates how Artificial Intelligence (AI) can be used to analyze massive underwater imagery datasets, revealing new insights into the ecology and behavior of Atlantic sea scallops (Placopecten magellanicus).

This new study shows how AI models can dramatically expand the ability to fully utilize these

large optical datasets. Researchers from CFF partnered with Kitware, a technology company with expertise in technical computing and AI, to develop and apply AI tools capable of rapidly

analyzing large optical datasets collected during Habitat Mapping Camera (HabCam) surveys.

Kitware has been developing the open-source computer vision platform Video and Image Analytics for Marine Environments (VIAME) since 2017. Through funding from the Sea Scallop Research Set-Aside grant program, CFF and Kitware improved AI models within VIAME to detect sea scallops, flounder species, and benthic habitat types.

Using these AI models, CFF scientists analyzed over 1.3 million images collected during HabCam V3 surveys in 2019 and 2021, in a fraction of the time needed for manual annotations.

The large-scale analysis enabled researchers to investigate rare behaviors like scallop swimming, across different age classes and habitat types. The analysis revealed that scallop

densities change as scallops mature in relation to different benthic habitat types including

gravel, shell hash, bryozoans, sea star beds, sand dollar beds, and sand waves. Models of

swimming behavior confirmed that juvenile scallops swim more frequently than adults, and this behavior is influenced by benthic habitat type.

Understanding scallop recruitment, survival, and habitat preferences has become increasingly

important as Atlantic sea scallop landings have declined and assessment surveys have

documented decreasing biomass across fishing grounds. This study highlights the potential for

AI to fully leverage existing optical datasets, offering a more efficient and comprehensive

approach to ecological monitoring moving forward.

The new paper, entitled “Benthic habitat influences sea scallop distributions and swimming

behavior based on underwater imagery and machine learning”, is available by following this link.