The ceramic gadget resembles a multi-armed starfish and is roughly the size of a dinner plate. Before it is deployed, it contains juvenile corals that have been grown in aquaculture tanks, sometimes for months before anyone trusts them to the ocean. It is composed of white alumina ceramic and has a rough surface to give young corals something to hold. Work is wasted if one of these is dropped in the wrong place. The coral won’t survive if you drop it on bare rock with inadequate light, poor water flow, excessive silt, or an area where crown-of-thorns starfish are active. Australian researchers have been using artificial intelligence to address the difficulty of knowing where to drop it—at scale, over a reef system spanning more than 2,300 kilometers.
That issue is now being addressed through the Deployment Guidance System, which was created under the Reef Restoration and Adaptation Program. On a surface vessel, the system installs cameras and real-time AI processing. As the boat passes over a reef site, the system uses high-resolution imagery to scan the seafloor below, finds microhabitats that match what young corals need to establish themselves, and, with an accuracy of about a meter, releases coral seeding devices at the precise moment the vessel passes over an ideal location. The decision-making process is based on years of ecological and oceanographic research that AIMS scientists have compiled into the circumstances that truly predict coral survival. The AI is not speculating. No human team working by eye from a boat deck could match the speed and consistency at which it applies accumulated professional knowledge.
AIMS in Townsville, Queensland University of Technology, James Cook University, CSIRO, the University of Queensland, and Southern Cross University are all part of the RRAP consortium, which is jointly financed by the Great Barrier Reef Foundation and the Australian Government’s Reef Trust. In order to determine whether coral seeding can be supplied at the scale the reef actually requires—not hundreds of corals across a study site, but potentially millions across large sections of damaged reef—the program’s Pilot Deployments Program started conducting practical trials on the southern reef around the Keppel Islands in 2025. Automation is the only way to reach that scale. The ground cannot be covered by human divers and snorkelers planting coral by hand. The only mechanism that makes the math even remotely feasible is the AI system.
In tandem with the restoration endeavor is the monitoring component of the project. Through its Long-Term Monitoring Program, AIMS uses manta tow surveys to monitor changes in coral cover throughout the northern, middle, and southern reef regions on 124 reefs each year. All three locations had high to extreme bleaching prevalence, according to data from 2024–2025, which was gathered following the fifth mass bleaching episode since 2016 and was described as the greatest in spatial footprint ever recorded on the reef. AIMS also created the AI monitoring platform Reef-Cloud, which automates the annotation of coral survey photos. Previously, this procedure required skilled taxonomists to manually go through the video. The period between data collection and useful ecological insight is greatly reduced when that stage is automated.
In a different but related endeavor, scientists at the University of South Australia are developing a multimodal monitoring platform that tracks coral bleaching, juvenile coral density, water quality, and overall ecosystem health in something closer to real time by combining satellite imagery, underwater video and photos, remote sensing data, and machine learning. The objective is a worldwide monitoring system that would provide reef managers with useful information without having to wait for the completion and processing of the next seasonal survey. It’s still unclear if that system can function consistently in all of the reef’s environmental variations, including varying depths, water clarity, coral assemblages, and lighting. Regarding the difficulty, the study team is realistic.
