Coral Reef Diagnostics

Indonesia - Halmahera: Kusu Drift (round 2)

⬇ Download this page as a report card
05
Photo: Anderson Mayfield/Coral Reef Diagnostics
Country
Survey date(s)
2026-08-13
Coral cover
43%
Claude computer vision (105 photos) · other methods below · median 47%
Observations
Heat stress
Most recent heat load record
26 Oct 2026 · Bleaching Watch
5-year thermal stress load
5-year mean max. degree heating weeks (DHW)
2021–2025
NOAA Coral Reef Watch — North Maluku, Indonesia virtual station
i 5-year mean max. DHW is the mean of the annual maximum DHWs over the five complete years named above, not conditions today.
More infoNorth Maluku, Indonesia (NOAA CRW virtual station)
Photographs
106
50 shown on this page
Site summary
Read this summary in:

Kusu Drift's second dive opens in the same hazy, low-visibility water that plagued the earlier flooded-camera dive, but develops into a colorful and densely built reef stretch — vivid salmon and magenta staghorn coral fields and stacked tabulate tables. One pale cauliflower-form colony is flagged as possibly bleached or a natural pale morph, not part of a wider pattern. A blind Claude computer-vision point count puts coral cover around 43%.

Max. depth dived
25.1 m
dive-log maximum
Water temp. during survey
26–27°C
range logged on the dive
Bottom time
43 min
in at 07:36

These are unedited originals; edited versions will replace them.

This page shows 50 of the 106 photographs from this site, including every one the text cites.

18 of the 50 photographs on this page are shown beside the text that discusses them under "Site description" below.

147910121415161718212223262728293031333536373839404144454648

Photo credit & attribution

Site description

Setting

Kusu Drift, round 2 (Halmahera dive 16), opens in hazy, low-visibility water on a coralline-pink and sponge-encrusted wall with divers and a rope/line visible, and cover in this deep stretch is patchy and low (e.g., Image 2, Image 11, Image 43, Image 47, Image 53).

Anderson's field note: “absolutely ripping currents for about 20 minutes straight, ended up in beautiful shallows.”

General benthic features

After the opening stretch the reef develops steadily into a colorful and densely built reef stretch: extensive fields of fine-branching staghorn corals in vivid salmon-pink, magenta, and pale lavender run near-continuous across dozens of frames (e.g., Image 3, Image 13, Image 19, Image 24, Image 34, Image 42, Image 51, Image 52).

Large overlapping table corals recur through the second half of the dive, often stacked several deep, alongside massive tan-brown domes and boulders threaded through the branching fields (e.g., Image 6, Image 3, Image 20, Image 32, Image 5, Image 54).

Gorgonian fans, a soft coral/anemone-like cream rosette, and scattered fish (chromis, damsels, an eel-shaped fish crossing sand) round out the fauna.

Coral cover

Stressors

One pale-white cauliflower-form colony sits among normally pigmented tan-green neighbors in one frame — isolated and hedged as possible partial bleaching or a natural pale morph, not part of a wider pattern. No disease, crown-of-thorns, or predation was seen in any of the 105 frames reviewed.

Comparisons with other data

We found no other surveys at this site as of 2026. If you would like to share yours, please email us.

CV model comparison

Mean 43%, median 47% across 105 photos.
More infoA few low-cover photos (sand, rubble, or open bottom) pull the mean down; the median is the typical photo. No computer-vision method yet estimates coral cover accurately on new images, so treat this as indicative rather than a measurement; it is revised as the methods improve.
Headline: Claude computer vision — a blind point count of this site's own photographs.
More infoDINOv3 and CoralSCOP were run on the same photos and are listed for comparison. No diver transect exists for this trip, so agreement between the rows is not accuracy, and no computer-vision method yet estimates coral cover accurately on new images; treat every row as indicative.
MethodHard coral cover (%)Photos (n) ± SE
In-house DINOv331.8n = 105 · SE 1.1
CoralSCOP unedited37.9n = 105 · SE 1.6
Claude43.0n = 105 · SE 2.4

The two computer-vision methods agree within 6.1 pp (31.8–37.9%) and read 5.1–11.2 pp below Claude's blind point count (43.0%); the headline is Claude's blind point count because no transect or manual count exists here.

Data & links

All images from Coral Reef Diagnostics. Data are from Coral Reef Diagnostics.

© Coral Reef Diagnostics.