Aerial maps record a 59% drop in massive coral across Guam's Tumon Bay after Typhoon Mawar

Tumon Bay was imaged in 2022 and again in 2024. Typhoon Mawar, a Category 4 storm with 140 mph winds, hit Guam in May 2023 ^[6]. The two surveys are about two years apart ^[1], so anything else that stressed the reef in that window also ends up in the change figures. Soufyane Bouchelaghem, a postdoctoral associate at ACES and the study's first author, treats the result as storm damage. "Repeating the survey after Typhoon Mawar turned that map into a precise measurement of damage across Tumon Bay," Bouchelaghem said ^[15]. I'd expect the storm explains most of a loss that size. Two snapshots cannot say how much of it, though. Answering that would take a comparison reef the storm missed.

The scale comes from the classifier, which sorted roughly 99% of the 2022 imagery by itself ^[2]. At 88% accuracy ^[10], about 12% of its calls were wrong ^[3]. For a habitat map, that is usable. A change measurement needs more care, because a decline is one classified map subtracted from another. Errors that stay the same in both years tend to cancel. Errors that shift between years do not, and the model was run on the 2024 images without retraining ^[11], after the mix of coral and algae it had to sort had changed sharply ^[7][8]. Running unchanged on new imagery is the right test of whether a reef classifier works outside a paper. It is also the step where the 88% is least certain to carry over.

The aerial method is a neat piece of physics. Waves bend light coming up from the seafloor, so conventional aerial photos show reefs blurred or shifted out of place ^[4]. Fluid Lensing, which Ved Chirayath developed while at NASA ^[18], shoots thousands of frames in quick succession and uses the magnifying effect of the passing waves to reconstruct a sharp image computationally ^[4]. "We mapped an entire reef rather than sampling a small portion of it, and we did it at a resolution capable of distinguishing individual coral colonies," Bouchelaghem said ^[14]. Conventional assessments rest on a limited number of dive sites. These maps record change location by location across the bay, including the sections that survived ^[13].

Chirayath, who directs ACES ^[1], sees both studies as answers to a throughput problem. "Coral reefs are changing at a pace that requires us to collect and analyze information much more efficiently," Chirayath said ^[16]. The published account says the tools could help managers assess damage after hurricanes and bleaching and send limited conservation resources where they are most needed ^[20]. Steering money depends on cost as well as speed. The account does not give a cost per survey for either method, break the 88% down by habitat class, or report how closely the model matched the freediver photo transects at the 10 field sites ^[12].