Coral Cover: St. Eustatius 2026

Shaded value = in the CV mean (shade deepens with cover) Gray italic = shown for comparison but left out of the CV mean: the gold standard (the reference value) and the manually annotated in CoralNet CoralNet benchmark annotations (reference values, not models), Coralscapes (not trusted), and any value that reuses the reference's own data n = frames behind the value — = not run or not applicable (see each region's notes) Site names and method names link out (new tab)

Coral Cover: St. Eustatius 2026

4 sites · 299 photos

No gold standard.

Site	Date	Headline (site page)	Claude	DINOv3	Coralscapes v1	Coralscapes v2	CoralSCOP	Mean ± SD of CV	Range
The Ledge to Hangover	2026-02-24	2.9%Claude	2.9^a,fn=88	17^a,bn=88	17^a,cn=87	21^a,cn=88	26^a,d,en=88	15 ± 11.6	2.9–26
STENAPA Reef	2026-02-24	4.6%Claude	4.6^a,fn=103	18^a,bn=103	7.2^a,cn=102	7.5^a,cn=102	20^a,d,en=103	14 ± 8.5	4.6–20
Double Wreck	2026-02-25	5.6%Claude	5.6^a,fn=46	16^a,bn=46	13^a,cn=46	14^a,cn=46	26^a,d,en=46	16 ± 10.3	5.6–26
Chien Tong wreck	2026-02-25	3.8%Claude	3.8^a,fn=62	17^a,bn=62	6.0^a,cn=62	5.6^a,cn=62	13^a,d,en=62	11 ± 6.5	3.8–17
All sites (unweighted mean)		—	4.2	17	11	12	21	14 ± 8.8	4.2–21

Notes for St. Eustatius 2026 (10)

[a] Value from the map's computer-vision method tables (denominator and photo set as on the site's page).
[b] DINOv3 (in-house, frozen Alor/Indonesia model) = mean hard-coral probability over 20 synthesized points per photo, NO water/fish class, so not a benthic denominator (from the map's site cover table).
[c] Coralscapes (v1 EPFL-ECEO/coralscapes-vit-b-dpt; v2 coralscapesv2-dinov3-vitb-lora-dpt) is not considered reliable for these photos: shown, never averaged. Area fraction, not a point count.
[d] CoralSCOP = per-frame sum of colony-mask area fractions (zero-filled; overlapping masks double-count), averaged over frames with an ok row. Area index, not a point count: denominator differs from any point-count reference. It is also unstable frame to frame (the same subject has scored 94.1% in one photo and 23.3% in another), and the mask sum overstates cover most on the healthiest reefs.
[e] CoralSCOP here = the site page's frame-matched row: run on exactly the photos behind the Claude cell (per-frame values joined by file; capped at 100% per photo; zero-filled). Comparison only.
[f] Claude = blind 7-call point count on the UNEDITED photos, 20 points per photo, (HC+BL)/(20 − WATER − FISH − OBST), unknown points kept in the denominator; mean across photos.
Mean/Range: Mean ± SD and Range are over the CV cells that enter the mean (CoralNet where independent, DINOv3, CoralSCOP unedited/edited, Claude). Gold standard, frames manually annotated in CoralNet, Coralscapes and same-measurement cells are shown in gray italic but never averaged.
Headline (site page): the coral-cover figure the site's page, report card and map pop-up print, with its method; it is never averaged into the CV mean.
Methods: Claude (Anthropic); DINOv3 (Meta; paper; used here as the frozen backbone of the in-house classifier); Coralscapes v1 and v2 (EPFL-ECEO; dataset paper, code); CoralSCOP (Zheng et al., CVPR 2024); CoralNet.
Sites: names drop the region prefix (hover for the full name) and link to the site's CRD page, or to the regional summary page where the site has none. Date = survey date from the map pins (first–last visit); sites are listed earliest first, undated sites last.

All regions