New imaging and AI tools sharpen coral reef monitoring in Guam
Two University of Miami-led studies describe complementary ways to measure coral reef change: airborne Fluid Lensing mapped Tumon Bay, Guam, at centimeter scale, while the PICOGRAM system analyzed standard underwater photographs. The Tumon Bay maps compared 2022 and 2024 surveys around Typhoon Mawar, and PICOGRAM’s estimates closely matched expert assessments in tests. The tools could help managers target reef monitoring, damage assessment and restoration decisions with more detailed evidence.
The story
Researchers at the University of Miami Rosenstiel School’s Aircraft Center for Earth Studies reported two imaging approaches intended to improve coral-reef observation at different scales. In Tumon Bay, Guam, they used NASA Airborne Fluid Lensing and artificial intelligence to map more than 5 square kilometers of reef habitat. The technique combines many rapidly captured aerial images to counter wave-driven distortion and produced imagery with features as fine as half a centimeter in water up to about 20 meters deep. The team surveyed the bay in 2022 and again in 2024, enabling a comparison before and after Category 4 Typhoon Mawar struck Guam in May 2023. Their maps recorded a 59% decline in massive coral cover, a 35% decline in coral fore-reef habitat and a 105% increase in algae. Volunteers and marine biologists labeled about 1% of the 2022 imagery through NASA’s NeMO-Net; the resulting deep-learning model classified the remaining images with 88% accuracy and was applied to 2024 images without retraining. Separately, ACES researchers and collaborators at NOAA and the University of Hawai‘i at Manoa developed PICOGRAM, an open-source AI system that identifies coral colonies, traces their boundaries and estimates coral cover from underwater photographs. In tests against 137 images independently analyzed by marine ecologists, it achieved 87.5% overlap with expert outlines under familiar conditions and 82.5% at previously unseen reef sites.
Why it matters
Reef managers must often decide where to survey, protect or restore corals after storms, bleaching events or disease outbreaks, but manual image analysis can take months. A whole-reef map can show where losses and surviving sections are located rather than extrapolating from a limited number of dive sites. At the colony scale, a system that flags its uncertainty lets specialists focus their time on questionable results instead of tracing every visible coral by hand. These uses remain conditional on local validation and human quality control, but they could make repeated monitoring more practical where image collections exceed available analysis capacity.
Evidence and context
Coral monitoring produces information at very different scales, from aerial surveys covering kilometers of reef to diver photographs showing individual colonies. Conventional aerial imagery is complicated by waves that bend light between the seafloor and a camera, while underwater photo surveys can generate more images than researchers can feasibly outline manually. The two studies address those separate bottlenecks: Fluid Lensing computationally reconstructs a clear aerial view, and PICOGRAM creates training masks from image characteristics including color, texture, edges and spatial context. The article notes that coral reefs support fisheries and tourism, provide marine habitat and reduce wave energy reaching coastlines, making reliable evidence on reef condition relevant to both ecosystem management and coastal protection.
Limits and unknowns
The results should not be read as a universal measure of reef loss or as proof that either tool will perform identically everywhere. The Fluid Lensing comparison concerns one full reef system, Tumon Bay, measured in 2022 and 2024; its AI classification was trained on labels covering about 1% of the earlier imagery and reported 88% accuracy. PICOGRAM performed less strongly at reef sites it had not previously encountered, with 82.5% overlap with expert outlines versus 87.5% in familiar conditions. It also requires scientist review of low-confidence outputs, and tracking individual colonies’ growth, recruitment and mortality is described as requiring further development.
Follow the story
Share it, watch the explainer, or continue with the reporting trail.
Story timeline
Initial Tumon Bay survey
Researchers surveyed Tumon Bay, Guam, using airborne Fluid Lensing imagery.
Typhoon Mawar
Category 4 Typhoon Mawar struck Guam between the two Tumon Bay surveys.
Follow-up Tumon Bay survey
Researchers resurveyed the bay, allowing a location-by-location comparison with the 2022 map.
New imaging and AI tools sharpen coral reef monitoring in Guam
A short visual explainer for readers who want the story in a different rhythm.
Highlights
- Two ways to monitor reefs
- Tumon Bay mapped twice
- Maps reveal change
- More targeted reef decisions
Transcript
Two Miami-led studies describe sharper airborne reef mapping and AI analysis of ordinary underwater photographs.
Researchers mapped over five square kilometers of Guam's Tumon Bay before and after Typhoon Mawar.
Those maps showed surviving coral alongside a 59 percent decline in massive coral cover.
Reef managers could use those maps and PICOGRAM's colony outlines to target surveys and restoration.
At unseen sites, PICOGRAM achieved 82.5 percent expert overlap; uncertain results still need scientist review.