Deep Vision: AI-enabled Mapping of Vulnerable Marine Ecosystems in the Atlantic

Project overview

Protecting deep-sea biodiversity requires robust, spatially explicit evidence on where vulnerable marine ecosystems (VMEs) occur and how they are distributed across the ocean basin. However, much of the Atlantic deep sea remains poorly characterised, despite the existence of extensive underwater imagery and seabed mapping data collected over many years.

This project brings together international partners to revolutionise how deep-sea biodiversity is mapped at scale by combining artificial intelligence, biological observations and high-resolution seabed data. Led in part by Plymouth Marine Laboratory, the project focuses on generating the evidence needed to inform legal protections for vulnerable marine ecosystems across the Atlantic Basin.

Project aims

The project aims to provide decision-ready scientific evidence to support the protection of deep-sea biodiversity by:

• Identifying and mapping vulnerable marine ecosystem indicator taxa across the Atlantic.
• Developing predictive habitat suitability models for VMEs at basin scale.
• Delivering accessible, policy-relevant outputs to inform legal and management frameworks.

How the project works

Pooling and standardising existing data
Project partners combine extensive archives of underwater imagery from across the Atlantic Ocean. These images are annotated with organism identifications to create a shared, quality-controlled dataset of deep-sea biodiversity observations.

AI-enabled identification of VME indicator taxa
Artificial intelligence models are trained and applied to the imagery dataset to automatically detect and classify VME indicator taxa. This approach enables rapid, consistent analysis of large volumes of data that would be impractical to process manually.

Population density mapping
The AI-derived observations are converted into georeferenced population density data for VME indicator taxa at known locations throughout the Atlantic Basin.

Habitat suitability modelling
Biological data are integrated with abiotic environmental information—primarily high-resolution multibeam bathymetry—to build habitat suitability models. These models are then applied across areas of the Atlantic where suitable seabed data exist, allowing prediction of VME distribution over a significant proportion of the basin.

Impact and outcomes

The project delivers new, basin-scale evidence to support the legal protection of vulnerable marine ecosystems in the deep sea. Key outcomes include:

• Maps of predicted VME distributions to inform marine protected area design and spatial management measures.
• Evidence to support international decision-making on deep-sea conservation, including on the high seas.
• Data products and visualisations that can be used by NGOs and other interest-holders advocating for stronger legal protections for deep-sea biodiversity.

By combining AI-enabled analysis with high-resolution seabed data, this project significantly advances our ability to protect deep-sea ecosystems at the scale required to meet global biodiversity and sustainability goals.