Mapping Coral Bleaching With AUVs and AI

Marine biologists are revolutionizing coral reef conservation by combining autonomous underwater vehicles (AUVs), photomosaics, and segmentation artificial intelligence to accurately measure coral bleaching across vast atolls. Traditional diver-based surveys are limited by time, depth, and human endurance, making it nearly impossible to monitor entire reef systems effectively. By deploying AUVs to capture thousands of overlapping seafloor images, scientists can stitch these photos into massive, high-resolution maps known as photomosaics. Advanced segmentation AI then analyzes these maps pixel by pixel to identify coral species and quantify the exact percentage of bleached tissue, providing rapid, large-scale insights into reef health.

Data Collection with Autonomous Underwater Vehicles

Surveying a large atoll requires covering hundreds of square kilometers, a scale unattainable by human scuba divers. Marine biologists program AUVs to navigate predetermined grid patterns just a few meters above the reef. Equipped with downward-facing, high-definition stereo cameras and strobe lighting, the AUV maintains a constant altitude and speed using Doppler velocity logs and acoustic sensors. As it glides over the atoll, the vehicle captures thousands of crisp, overlapping photographs of the benthic environment, documenting deep and remote reef zones without human intervention.

Creating High-Resolution Photomosaics

Once the AUV is retrieved, the thousands of individual images must be combined into a single, continuous map. Biologists use specialized photogrammetry software employing Structure-from-Motion (SfM) algorithms. The software identifies matching visual features across overlapping photographs and calculates the exact position and orientation of the camera for each shot. This process stitches the images together into a geographically accurate, three-dimensional photomosaic. These massive visual datasets allow researchers to view an entire stretch of an atoll as one seamless image, while retaining the ability to zoom in and see millimeter-scale details on individual coral colonies.

Automated Analysis via Segmentation AI

Analyzing a photomosaic manually would take researchers months of painstaking work. To accelerate this process, biologists utilize semantic segmentation, a deep learning technique. Researchers train convolutional neural networks (CNNs) by feeding them thousands of manually labeled images of healthy corals, bleached corals, algae, and sand.

Unlike basic image recognition, which simply draws a box around an object, segmentation AI classifies every single pixel in the photomosaic. The AI traces the exact, complex boundaries of each coral colony. It differentiates between living, pigmented coral tissue and the stark white skeletons of bleached corals, while also filtering out non-coral elements like white sand or rubble that could skew the data.

Quantifying Atoll-Wide Bleaching

After the segmentation AI finishes processing the photomosaic, it generates quantitative metrics on the reef's composition. Biologists receive precise calculations detailing the total surface area of the reef, the percentage of living coral cover, and the exact proportion of coral that is currently bleached. Because the AUV collects GPS-linked data, these metrics are mapped spatially across the atoll.

By repeating this process over time, scientists can track the progression of bleaching events or measure reef recovery at the colony level. This integration of robotics, imaging, and artificial intelligence allows marine biologists to pinpoint thermal refuges, identify resilient coral species, and direct conservation efforts where they are needed most across massive marine ecosystems.