Scientists Are Planning to Chart Every Living Thing in the Atlantic’s Deep Sea — All of It

Last June, over forty marine scientists from eleven different countries gathered in a crowded conference room at a hotel in Nice to discuss a problem that has been building for decades: we are actively working to protect areas of the Atlantic Ocean that we have never really seen. Governments are required by law to safeguard vulnerable marine ecosystems in areas where they are known or likely to occur. Professor Kerry Howell of Plymouth Marine Laboratory made it clear that day that the key word is “known.” Conservation efforts currently end at the boundaries of what has already been documented by researchers. By default, everything else is outside the protection boundary, including the unexplored sponge beds, the unmapped canyons, and the cold-water coral gardens that no ROV has ever been to.

The Atlantic deep-sea biodiversity mapping project aims to bridge that gap. Under the larger Challenger 150 program, the project is led by Plymouth Marine Laboratory and the University of Aveiro in Portugal. Its goal is to create thorough, high-resolution models of species distribution throughout the Atlantic floor, from the relatively well-studied regions of the North Atlantic to the Central and South Atlantic, where there is insufficient data to be truly concerning. The ultimate goal is what scientists are referring to as a “spatial digital twin” of the ocean floor: a comprehensive, dynamic digital model that can guide conservation choices, spatial management, and the practical task of determining the areas where human activities like bottom trawling and deep-sea mining are most likely to result in permanent harm.

It’s worth taking a moment to consider the scope of the goal. The second-largest ocean in the world is the Atlantic. Its deep-sea ecosystems include sponge communities that have been constructing their structures over geological time, black corals that may be over 4,000 years old, and deep-sea reefs that serve as shark and marine life nurseries. Statutory fisheries programs, which are intended to monitor fish stocks rather than the overall health of ecosystems, are the main source of current data collection. As a result, a library of species and habitat types that are present, functional, ecologically significant, and completely undetectable to the legal frameworks designed to protect them remains virtually unrecorded.

The Bezos Earth Fund has provided Howell’s team with funding to use artificial intelligence for deep-sea biodiversity monitoring, which is what makes the project’s scale possible. Two related projects are already in progress: Coral Cartography, which was introduced in April 2026 at the AAORIA Forum and specifically focuses on mapping Atlantic cold-water corals to support area-based management decisions, and Deep Vision, which uses AI to analyze thousands of hours of previously unanalyzed ROV video and photographic data from archival sources. Decades of underwater surveys have produced a significant amount of imagery, the majority of which is stored in institutional archives without being examined. The process of identifying and cataloging species from that footage would simply take longer than the conservation timeline permits without machine learning tools that can process it at scale.

The practical difficulties are substantial. The Nice meeting spent a lot of time resolving the technical issue of getting imagery collected by various research institutions to speak a common language. This is because different research institutions have used different equipment, different standards, and different species classification systems. A standardized operational taxonomic unit system—a common framework for identifying and labeling what researchers find—is essential, according to Dr. Lara Atkinson of the South African Environmental Observation Network. It is necessary for the meaningful combination of data from Norway and Brazil. The topic of acceptable error rates in automated species identification was brought up by Nils Piechaud during his presentation on AI applications. This may seem like an academic issue until you consider the possibility that a policy document’s incorrect identification of a coral species could leave an ancient reef garden unprotected.

This mapping project feels more urgent than it might otherwise because of the connection to other struggles in ocean policy. In a way that scientific publications seldom manage, the David Attenborough documentary Ocean has raised public awareness of bottom trawling. In addition to harming the ecology, trawl gear that is dragged over unexplored deep-sea reef systems destroys equipment and costs the fishing industry money. Both issues are simultaneously resolved by knowing the exact location of the reefs. Professor Howell has been making this point clearly, and it’s the kind of argument that usually makes an impact.

There’s a sense that, after years of being discussed in isolation, the components of this coalition—researchers from eleven nations, AI tools processing archival footage, and coral mapping projects launching at international forums—are finally coming together. Howell framed it simply: either they align now and complete it correctly, or they can keep doing this as a jigsaw, picking up individual pieces in individual places. For a very long time, the deep sea has been patient. Whether science can advance more quickly than the threats is still up for debate. At least the map is being created, though.