Mauritius - Anemone Garden

Setting

Not a coral reef but a true anemone garden: large bommies on a sand-and-algae lagoon flat carpeted nearly wholesale by magnificent-type sea anemones (consistent with Heteractis magnifica), attended by clouds of dark Dascyllus-type damselfish (e.g., Image 3, Image 11, Image 12, Image 13, Image 14, Image 16).

31112131416

Coral cover

Manually scored hard coral cover was just 2.7%, limited to scattered Pocillopora and massive Porites-type colonies (e.g., Image 4, Image 7, Image 18).

4718

No error terms since neither CoralScop nor DinoNet could reliably distinguish anemone from scleractinian corals. Coral cover — site cover estimate (manual point count and CoralNet, weighted equally); the other automated methods are rejected here. The published 2.9% is the site cover estimate from Anderson Mayfield's Mauritius manuscript: his manual point count (2.7%) and CoralNet, weighted equally. A DINOv3 cross-check was not run for this site, so there is no second estimate to compare it against and no way to tell from this record alone whether the two models would have agreed. Both models have known weaknesses, and they are different weaknesses. CoralSCOP computer vision reads HIGH on low-cover reefs: across twelve sites with a Reef Check diver survey within 1 km, the median CoralSCOP figure is 1.68x the survey value and the worst case is 4.6x. The DINOv3 point classifier is strong on reefs resembling its training data (6-class macro-F1 0.62, hard coral recall 0.82) but degrades badly on an unseen reef — cross-domain point accuracy measured at 0.17-0.25 against Anderson Mayfield's own annotations, with a documented ceiling near 35-50%. Neither is a diver survey, neither is a benthic transect, and both should be read as provisional indices that a better model would supersede. Anderson keeps an eye out for newer CV models and will refresh this CoralSCOP and DINOv3 figure once a more trustworthy option is available.

Fish and other fauna

Highlights included a trumpetfish, honeycomb grouper, whitemouth moray, longfin bannerfish, spiny cushion stars, and sea cucumbers (e.g., Image 1, Image 2, Image 10, Image 17).

121017

Comparisons with other data

A published figure above about 70% usually reflects the surveyors choosing the most coral-rich part of the reef, so read it as a best-patch number rather than a site mean.