AI and Advanced Imaging Track Coral Reef Changes
Researchers at the University of Miami Rosenstiel School’s Aircraft Center for Earth Studies have deployed two advanced imaging and artificial intelligence systems to transform coral reef monitoring. Published in Frontiers in Marine Science, the studies demonstrate how automated optical and computational tools can rapidly quantify reef health across vastly different scales, enabling faster conservation responses. The first approach, Airborne Fluid Lensing, overcomes the persistent challenge of ocean wave distortion in aerial photography. By capturing thousands of rapid images and computationally reconstructing the seafloor, the system achieves centimeter-scale resolution in waters up to twenty meters deep. Researchers applied the technology to map over five square kilometers of Tumon Bay, Guam, in 2022 and 2024. The pre- and post-typhoon comparisons revealed stark ecological shifts following Category 4 Typhoon Mawar in May 2023. Massive coral cover declined by fifty-nine percent, fore-reef habitat dropped by thirty-five percent, and competing algae expanded by one hundred and five percent. To process the imagery, a deep-learning model was trained using manually labeled data from volunteers and marine biologists, achieving eighty-eight percent classification accuracy across the entire mapped area without requiring retraining for the follow-up survey. Complementing the aerial surveys, the second study introduces PICOGRAM, an open-source AI system designed to automate the analysis of routine underwater photographs. Traditional manual tracing of coral colonies is time-intensive and delays critical data collection. PICOGRAM identifies individual colonies, traces boundaries, and estimates coral cover directly from standard survey photos. Rather than relying on extensive hand-labeled datasets, the model learns from algorithmically generated masks based on visual features like color and texture. When evaluated against independent expert annotations across varied reef environments, PICOGRAM achieved eighty-seven point five percent overlap with manually drawn boundaries under familiar conditions and eighty-two point five percent at novel sites. Coral cover estimates deviated from expert measurements by an average of just one point two percentage points. The system incorporates a confidence scoring mechanism that allows scientists to correct uncertain outputs with minimal manual intervention, streamlining quality control. NASA has validated the tool at Application Readiness Level 8, and NOAA has since integrated it into operational monitoring programs. Together, these technologies address the growing volume of reef imagery that exceeds manual analytical capacity. Fluid Lensing provides basin-wide ecological baselines and damage assessments, while PICOGRAM delivers colony-level metrics from routine dive surveys. By converting imagery into precise, repeatable measurements, the systems allow resource managers to pinpoint degradation, evaluate restoration efficacy, and allocate conservation funding with greater precision. Researchers note that integrating these tools with diver observations and environmental telemetry will create a comprehensive monitoring framework. Ultimately, rapid, high-fidelity reef assessment supports coastal resilience, marine fisheries, and tourism economies dependent on healthy coral ecosystems.
