Researchers from the University of Hawaii at Manoa have developed a cutting-edge AI-based surveying technology that keeps track of the health of coral reefs from the air due to the increasing deterioration of coral reefs throughout the world.
The researchers have created a novel technique for identifying and monitoring coral reef halos, which are rings of bare sand around reefs, using deep learning models and high-resolution satellite images powered by NVIDIA GPUs.
Future of Coral Reef Conservation
“Coral reef halos are a potential proxy for ecosystem health,” said Amelia Meier, a postdoctoral fellow at the University of Hawaii . These halo patterns, which are visible from orbit, provide scientists and environmentalists with a rare chance to monitor large and remote locations. We can routinely evaluate halo existence and size in almost real-time to gauge ecosystem health thanks to AI.
Understanding Reef Health Through Sea-ing Clearly
Recent findings show that reef halos, formerly only associated with fish grazing, might also be a sign of a healthy predator-prey ecology. While some herbivorous fish feed on algae or seagrass close to the boundary of the protective reef, hunters scour the bottom for burrowing crustaceans, exposing the nearby sand.
These dynamics suggest that the region provides a wide variety of ocean dwellers with a plentiful supply of nutritious food. If the geometry of the halo changes, there may be an imbalance in the marine food chain and/or an unfavorable reef environment.
Warm Water
Coral reefs support about 1 million aquatic species and provide habitat, food, and nidification grounds while making up less than 1% of the ocean’s surface. The economic value of commercial and subsistence fishing, tourism, coastal storm protection, and the production of antiviral chemicals for drug discovery research totals roughly $375 billion yearly.
However, nutrient pollution, ocean acidification, and overfishing pose a danger to the health of the reefs. Increases in coral bleaching and infectious diseases are also caused by escalating climate change and the thermal stress that results from a rising ocean.
Scientists estimate that by 2050, all coral reefs will face risks, with many being in grave danger. Currently, more than half of the world’s coral reefs have already been destroyed or severely damaged.
AI: Opening Up New Horizons
Global conservation efforts depend on being able to detect changes in reef haloes. However, keeping track of these changes requires a lot of work and time, which restricts the number of surveys that researchers can conduct annually. Remote reef access presents additional difficulties.
The researchers developed an AI program that uses data from international satellites to recognize and quantify reef haloes, providing conservationists with a chance to preventatively address reef damage.
They created a dual-model framework using two different kinds of convolutional neural networks (CNNs) using Planet SkySat imagery. They built a Mask R-CNN model that can recognize the boundaries of the reef and halo, pixel by pixel, by relying on computer vision techniques for picture segmentation. The regions of both are then classified and predicted using a U-Net model that has been trained to distinguish between the coral reef and halo.
The researchers trained and tested hundreds of annotations on the coral reef models using the TensorFlow, Keras, and PyTorch libraries.
The CNNs run on an NVIDIA RTX A6000 GPU, assisted by a cuDNN-accelerated PyTorch framework, to manage the task’s high computing demands. The NVIDIA Academic Hardware Grant Program participants that got the A6000 GPU were researchers.
In less than two minutes, the AI tool rapidly locates and counts 300 halos across a 100 square kilometer area. A human annotation takes around ten hours to do the same activity. The model can also traverse varied and complex halo patterns with an accuracy of roughly 90% depending on the area.
Processing hundreds of photos might take a long time, however making use of the NVIDIA GPU drastically sped up the process.
One difficulty is that the model’s accuracy may be constrained by the picture resolution. Low-resolution course-scale photography makes it harder to see reef and halo borders and leads to less precise projections.
Environmental Monitoring Must Increase
The researchers are investigating the connection between species composition, reef health, and halo presence and size using this novel technique. They’re investigating the connection between sharks and halos right now. The team hopes to estimate shark population from orbit if their predicted predator-prey-halo interplay is confirmed.
