Philippines - Verde Island Passage: Kirby's Rock

Setting

Kirby's Rock is a true pinnacle, and Anderson's frames resolve it cleanly into a shallow top at 2-3 m and steep current-exposed flanks dropping into blue water (e.g., Image 4).

4

General benthic features

Hard coral cover is high on the pinnacle top and upper flanks - tabular and plating Acropora, large massive Porites domes, brain corals, branching Acropora, and Pocillopora thickets, foliose plates, and one big massive colony photographed with polyps fully extended (e.g., Image 1).

1

Lavender, white, and cream Dendronephthya, gray-green colonial ascidians, Tubastraea cups, and abundant crinoids, including a striking white feather star, fill the remaining space (e.g., Image 3).

3

Coral cover

The frames Anderson labeled by zone quantify the gradient: the pinnacle top scores 60.5% by CoralSCOP computer vision from 95 separate colony masks and 38.1% by DINOv3, the 2-3 m shallow top 43.1% and 15.8%. Cover comes from two computer vision methods, neither preferred. DINOv3 point classification gives 23.3% hard coral (95% CI 21.2-25.8) from all 73 frames and 1,460 points. CoralSCOP computer vision, its full run now finished, gives 27.7% across all 73 frames (median=24.4%, bootstrap 95% CI 24.1-31.6), against the 28.4% from the 20-frame subset the map's 25.8% was built on. The published spread is 5.0 points and the completed run narrows it to 4.4, but CoralSCOP still sits above the DINOv3 confidence interval, so the two methods disagree here. The frames show why: CoralSCOP's highest image, PA011176 at 78.2%, is a field of white and lavender soft coral where DINOv3 reads 18.6%, a 60-point gap on a single frame. The two models fail differently. CoralSCOP computer vision reads high on low-cover reefs - a median=1.68x a co-located Reef Check diver survey, 4.6x at worst - and it masks soft coral as readily as stony coral. DINOv3 is strong in-domain (6-class macro-F1 0.62, hard coral recall 0.82) but collapses off-domain: cross-domain point accuracy 0.17-0.25, ceiling near 35-50%; on the two topside frames in this set it still returned 10.2% and 20.1% hard coral for photographs of a boat and the sky. Neither is a diver survey or a transect; both are provisional and a better model will supersede them. Coral cover: 26% — photo-derived, and the two methods broadly agree (23% vs 28%). Two independent computer vision passes over all 73 of Anderson's frames from this dive give a DINOv3 calibrated point classifier reading of 23.3% (95% CI 21–26%, block bootstrap) and a CoralSCOP computer vision of 27.7%, a spread of 4.4 percentage points. The map publishes their mean so sites can be ranked against one another, but the 23–28% range is the honest statement of what is known here: the two products have different denominators and should not be collapsed into one measurement. This number comes from CoralSCOP and DINOv3, which Anderson considers a stopgap; he is evaluating other models and will regenerate the figure once a better one is identified.

Stressors

There is no site-wide bleaching signal; the reef reads healthy and structurally complex, with the usual background of turf, rubble pockets, and old mortality on the deeper flanks.

Fish and other fauna

It carries the strongest fish showing of the Anilao and Verde group: dense clouds of anthias streaming over the wall, and tight formations of blue-striped snapper and yellow fusiliers (e.g., Image 2).

2

A vivid red encrusting sponge hosts a scatter of brilliant yellow sea cucumbers (Colochirus robustus) - identified from Anderson's own filename, and worth flagging because at a glance they read as nudibranchs. He also photographed a frogfish and a large white dorid nudibranch on ascidian stalks.

Comparisons with other data

No independent survey was made at this exact site; the nearest ones are listed in the table below.

Dive record

The first dive of the three-tank Verde Island Passage day on 1 October 2016; all 73 available frames were reviewed, two of them topside boat shots.

Other notes

The frame named "outside Kirby's Rock" is in fact a surface boat shot, so the deeper surrounding slope is documented only within the numbered sequence.