### BITB34 — South Brother
**CRD page:** [Three Brothers & Central Chagos Coral Surveys (Mar 17-20, 2015)](https://coralreefdiagnostics.com/three-brotherscentral-chagos-20150317-to-0320)
**Date:** 2015-03-19 (dive window 13:49:31–14:42:14, `DateTimeOriginal` EXIF on all 27 files, in three bursts: 13:49–13:52, 14:04–14:18 and 14:33–14:42. **The third dive of the day on one continuous camera counter** — BITB32 ran 08:47–09:06 and BITB33 10:44–10:52, which is why two-thirds of this folder's filenames collide with theirs. Flag 2.)
**GPS:** 6.16935°S, 71.54637°E
**Frames reviewed:** 27 of 27 (100% of the site folder, `BiTb34-great site-finished`; none evicted). **Only 11 of the 27 carry CV data.** Every photo-method number below rests on those 11, not on the 27.
**Coral cover:** 61.13% (formal transect surveys; algal cover 27.78%) — **the highest coral cover and the lowest algal cover of any site written in this redo**, and 9th of 104 site-rows across the whole Chagos dataset (median 39.4%) · **% bleached:** 5.45% (`% bleaching (2nd training)`, Mayfield & Dempsey 2026, Oceans). **See flag 6 — Anderson recorded no bleaching before May, and the photographic record does not corroborate this number.**
**Reef type:** fringing, protected, lagoonal (inside, emergent) · **Temperature:** 29.8 °C · 6 transects across 3 depth strata (D2–D4), above the corpus median of 5
**Live hard coral cover across methods** (n=11 photos of 27): transect 61.13% · CoralNet points 32.7% · Coralscapes 34.8%. Deviation **−28.4 to −26.3 pp — the largest method deviation in the entire redo, beating BITB30's −24.3.** **Mean of 3: 42.9% ± 15.8.** Both photo methods read roughly **half** the transect. **The gap is a denominator artefact and it closes almost exactly: renormalised over labelled benthos the mask reads 60.3% and the point set 66.7% against a 61.13% transect. Flag 1.**
**Coral genera identified:** the transect recorded **30 scleractinian genera** — 9th-richest of 104 site-rows, on above-median survey effort, so this is a real assemblage and not a richness floor. *Acropora* dominates at 24.81% absolute (40.6% of all coral cover), then *Favia* 8.37%, ***Stylophora* 5.20%**, *Porites* 3.93%, *Gardineroseris* 2.84%, *Pavona* 2.28%, *Montipora* 2.10%, *Platygyra* 1.93%, *Echinopora* 1.65%, *Pocillopora* 1.64%, *Heliopora* 1.40%, *Favites* 1.04%, and a tail of 21 further genera including *Astreopora*, *Cyphastrea*, *Fungia*, *Galaxea*, *Goniastrea*, *Hydnophora*, *Pachyseris*, *Symphyllia*, *Turbinaria* and the hydrocoral *Millepora*. From the photographs, hedged: ***Acropora*** in four forms (**tabular plates well over a metre**, corymbose/digitate, arborescent staghorn, and a **purple-mauve branching form** in P3190464 and P3190469); **enormous whorled foliose stands compatible with *Echinopora lamellosa* or a foliose *Montipora*** (P3190416, P3190417); **massive and knobby-lobed *Porites*** (P3190391, P3190463); meandroid brains compatible with *Platygyra*/*Leptoria* (P3190393, P3190461, P3190463); cerioid honeycomb compatible with *Favites*/*Goniastrea* (P3190412, P3190461); star-corallite domes compatible with *Astreopora* (P3190391, P3190441, P3190464); **blunt purple-brown digitate colonies compatible with *Stylophora*** (P3190393, P3190462); **fine compact verrucose pocilloporids compatible with the *P. damicornis*/*P. acuta* group** (P3190412, P3190414, P3190441, P3190461); and ridged foliose plates compatible with *Pachyseris*/*Echinopora* (P3190393). Plus *Halimeda* (P3190461), green *Didemnum*-type ascidians (P3190414, P3190427) and red encrusting sponges. Growth-form calls from photographs are given as "compatible with"; no species is asserted. Transect genus names are survey identifications.
**Representative photos:** P3190470.jpg, P3190389.jpg, P3190464.jpg, P3190419.jpg, P3190416.jpg

**BITB34 is the best reef in the redo, and it is not close.** At 61.13% coral against 27.78% algae it carries nearly half again the coral of BITB30 (42.5%), the previous best, and the imagery is emphatic from the first frame to the last. **P3190470 may be the single finest reef frame in the collection** — a field of enormous emerald tabular *Acropora* plates, six or seven of them a metre and more across, laid edge to edge and overlapping to the limit of visibility with barely a hand's width of pavement between them. **P3190389** is the same reef seen obliquely: tan and gold tables tiered up a slope in four or five ranks, corymbose *Acropora* packed into every gap, dark surgeonfish threading the plates. **P3190386** stacks olive-green plates over a pavement of *Pocillopora*-type clumps and CCA, and **P3190472** — the highest-cover scored frame, 63.3% live by points and 59.5% by mask — closes the dive on the same continuous plate pavement. By eye the wide frames here carry 50–70% living tissue. **The 61.13% transect is entirely believable; the photo methods' 33–35% is not, and explaining that is one of this entry's two jobs.**

**Where the reef is not plated it is foliose, and that is the site's signature.** **P3190416 and P3190417 are the two most striking colonies of the redo so far** — single mounded stands two to three metres across built entirely of tightly whorled purple-brown foliose plates, thousands of them, spiralling out from the centre like stacked crockery and terminating in a clean pale growing margin. Both are compatible with *Echinopora lamellosa* or a foliose *Montipora* (the transect records both, at 1.65% and 2.10%), and **neither frame carries a CV score.** Between the plates and the whorls the reef is a dense mosaic: **P3190463 and P3190464** are the highest-cover wide frames at the site, pavement invisible beneath corymbose and digitate *Acropora*, mauve staghorn, *Platygyra*-type brains, *Astreopora*-type domes and a huge pink-brown knobby-lobed *Porites*; **P3190462** runs a whole foreground of blunt gold-and-mauve digitate colonies compatible with *Stylophora*, the site's third-ranked genus at 5.20%. **P3190391** is the massive-coral frame — lobed *Porites* fingers, faviid domes and purple-crusted submassives shoulder to shoulder, a diver hanging in the blue above.

**The macrofauna is the best of the redo, and it is rays.** **Six of the 27 frames are given to spotted eagle rays** compatible with *Aetobatus ocellatus*. **P3190419 is the best — three rays in silhouette against open blue water, wings spread, whip tails trailing** — while P3190429, P3190430, P3190432, P3190434 and P3190435 follow a single animal (in 430 and 432 with a remora on its back) planing low over a white sand flat, two more on the horizon of P3190434. **P3190418** holds three trevally in silhouette with two further ray wings behind. Beyond the rays: **three trevally compatible with *Caranx melampygus*, flanks stippled electric blue, over the plate reef in P3190422**; **large schools of humpback red snapper compatible with *Lutjanus gibbus* in P3190468 and P3190469**; **a sergeant-major school compatible with *Abudefduf* in P3190466**; a dense silverside or juvenile-fusilier cloud pouring over the reef in **P3190427**; blue-green chromis, parrotfish, dark surgeonfish and butterflyfish throughout. **The large dark oval silhouettes on the sand horizon of P3190429, P3190430 and P3190432 are not callable** — humphead wrasse and a large parrotfish are both compatible, and the BIEG04 lesson applies.

**Both target pocilloporid groups are present here, which is unusual for Chagos.** The CRD site page records ***P. acuta*/*P. damicornis* (ChPd/a22–25) and *Seriatopora* spp. (ChSe14–15)** at this leeward fringing reef, with **ChPa22 and ChPd24 genotyped and ChPd23 and ChPd25 sampled but not genotyped**. **Note that the transect scored *Seriatopora* at zero** — the genus was collected and observed but never fell under a transect point, which is exactly the kind of presence a cover survey cannot record. The imagery is consistent and unusually well supplied with candidates: **four of the 27 frames are close-up flash macros composed on a single pocilloporid colony**, which is how a sampled colony gets documented. **P3190441** is a compact cream-white verrucose colony among tan neighbours; **P3190461** a pale pink-and-cream colony between a *Platygyra*-type brain and a cerioid *Favites*-type sheet, with *Halimeda* around it; **P3190412** a cream-yellow colony beneath a honeycomb massive; **P3190414** a larger tan colony with pale extended polyps, ringed by green *Didemnum*-type ascidians. All four are compatible with the compact *damicornis*/*acuta* habit rather than the blunt *verrucosa* form seen at BITB27. **No tag, label or marker is visible on any of them, so which — if any — is ChPa22 or ChPd24 needs Anderson's judgement.** Nothing in the folder reads as *Seriatopora*.

**Health is very good and there is no survey apparatus anywhere.** Structural loss is old: Coralscapes puts dead coral at 4.43% of image area and rubble at 1.25%, and what dead framework there is has already been resurfaced in pink and mauve CCA — the pavement in P3190461, P3190463 and P3190412 is almost entirely CCA-plated old framework. **No transect tape, quadrat, slate or collecting gear appears in any of the 27 frames** — the second survey-only folder of the redo after BITB30 — although three frames show a diver (P3190391 in the water column, **P3190432 and P3190434 photographing the eagle ray with a compact camera**), and four are close-up macros composed on a colony rather than on the reef (flag 9). **No COTS, no disease lesions and no predation scars anywhere in the folder.**

<b>Assemblage composition.</b> **This paragraph rests on 11 of the site's 27 frames.** Across those 11 the Coralscapes masks put live hard coral at 34.8% of image area, of which massive and meandroid is the largest component at 35.5%; the rest breaks down as other hard coral 18.4%, tabular <i>Acropora</i> 17.6%, <b><i>Stylophora</i> 9.3%</b>, <i>Pocillopora</i> 6.9% and <i>Acropora</i> (non-tabular) 6.9% of live cover. **That spectrum is independently corroborated by the transect**, which puts massive and meandroid genera at 34.3% of coral cover (vs the mask's 35.5%) and <i>Stylophora</i> at 8.5% (vs 9.3%) — **the only site in the redo where a Coralscapes <i>Stylophora</i> class can be checked against a transect that actually recorded the genus.** The independent 30-point annotation of the same 11 frames (330 points) gives live hard coral as 32.7% of all points and apportions it <i>Acropora</i> 45.4%, other hard coral 25.0%, <i>Porites</i> 24.1%, <i>Pocillopora</i> 5.6% of live coral, against a transect <i>Acropora</i> share of 40.6%. Non-coral cover is dominated by turf 8.5% and crustose coralline algae 5.2% of all points. Structural loss registers as dead coral 4.43% of image area and rubble 1.25%. Bleaching stands at 2.12% of annotated points — **7 points in total**, split <i>other hard coral</i> 1.52% and <i>Acropora</i> 0.61% — while Coralscapes reads bleached at 0.05% of image area, effectively zero. Frame-to-frame live cover runs 0–63.3% by point count and 0–59.5% by mask, and **the two methods agree with each other well here: r = 0.724, against 0.716 corpus-wide and 0.278 at BITB30.** Conspicuous fauna and sessile invertebrates appear as sponges in 5 frames and giant clams in 1 frame.

**Flags:** (1) **THE −28 pp GAP IS A DENOMINATOR ARTEFACT AND THE CORRECTION LANDS ON THE TRANSECT. THIS IS THE HEADLINE FINDING.** Both photo methods read roughly half the transect — 32.7% and 34.8% against 61.13% — because **50.9% of the point budget goes to no benthic class at all (water 21.5%, fish 1.2%, Unknown 28.2%: 71, 4 and 93 of 330 points) and Coralscapes calls 40.7% of image area `background_dark` plus 1.5% fish.** A transect measures cover **of the substrate**; both photo methods report cover **of the image**. Renormalise over labelled benthos and the gap all but vanishes: **Coralscapes live over (100 − background − fish) = 60.25% against a 61.13% transect, a residual of −0.9 pp — the closest any renormalised photo estimate has come to a transect anywhere in the redo — and live over labelled benthic points = 66.67% (70.99% including bleached), i.e. +5.5 pp.** This is the fifth low-reading site at which dropping the unlabelled pool lands the estimate on the transect (after BIEA26, BITB27, BITB28, BITB29) and by far the most extreme. Renormalising over labelled benthos before any transect comparison should now be corpus-wide standard. **Do not cite BITB34 as evidence that photo methods under-read coral cover.** (2) **16 OF 27 FRAMES WERE NEVER ANNOTATED, AND THE ANNOTATIONS DO NOT EXIST TO RECOVER.** All 27 are on disk, none evicted, but only 11 carry CV data — and the split is perfectly clean: **the 11 scored frames are exactly the 11 filenames unique to this folder, and all 16 unscored filenames are shared — 10 with `BiTb32-finished` and 6 with `BiTb33-finished`**, the two earlier dives of 2015-03-19 on the same camera counter. BITB32 and BITB33 won every collision, so **those 16 rows in the comparison CSVs read the other folders' pixels, not BITB34's** (the collided pairs are byte-different, and BiTb32's P3190416 is timestamped 08:47:49 against BITB34's 14:07:00). CoralNet only ever held one image per filename, so **a re-key recovers nothing**; the visual review of all 27 is the stronger evidence. **Had those 16 rows been naively pooled into BITB34, the site would read 21.0% by points and 20.0% by mask — 40 pp below transect, twice as wrong.** (3) **VERDICT ON THE SUBSET QUESTION: THE 16 UNSCORED FRAMES ARE NOT HIGHER-COVER — THEY ARE BIMODAL, AND ON AVERAGE SLIGHTLY POORER. THE −28 pp GAP IS REAL, NOT A SAMPLING ARTEFACT.** This is the BITB28 answer, not the BITB27 answer, and it needed testing because at a 61% site the alternative would have mattered enormously. **Nine of the 16 unscored frames are the highest-cover imagery at the site** — P3190416, P3190417, P3190427, P3190463, P3190464, P3190466, P3190468, P3190469 and **P3190470**, all 50–70% living tissue by eye, none carrying a CV number. **But six are the eagle-ray sequence over bare white sand — P3190418, P3190429, P3190430, P3190432, P3190434, P3190435 — carrying essentially no coral at all**, with the flash macro P3190441 (~45%) making up the rest. Coarse by-eye means: **unscored 16 ≈ 36%, scored 11 ≈ 45%.** The scored set is, if anything, the higher-cover. **The 11 scored frames are a fair — even generous — sample of this reef, and the shortfall must be explained by flag 1, not by which filenames happened to be unique.** (4) **BLUE-WATER DILUTION IS MINOR — REJECTED AS THE EXPLANATION.** Only 2 of 11 scored frames are ≥40% water (**P3190419 at 86.7%**, the three-eagle-ray blue-water frame, and P3190422 at 46.7%). Dropping both lifts CoralNet 32.7→37.4 and Coralscapes 34.8→40.8 — **still 23.7 and 20.3 pp below transect, under a sixth of the gap closed.** No BIEA20-style sign flip. (5) **A PATCHY REEF IS REJECTED.** From 13:49 to 14:42 every wide benthic frame carries dense tabular *Acropora*, whorled foliose stands or packed corymbose thickets; if anything the imagery suggests the 61.13% transect is **conservative**. (6) **BLEACHING — THE PUBLISHED 5.45% IS NOT CORROBORATED BY THE PHOTOGRAPHS, AND THE DIVER RECORDED NONE.** This is a **March** site. Anderson recalls **no bleaching on the March or April dives — only May**, and the published data agree: half-month means run 6.1 / 8.2 / 8.4 / 10.0% through March–April, then **31.8% in early May**, 3–4× every earlier period. **The 5.45% is reported here as published, but should be read as baseline noise in an annotation set that is ~76% robot prediction, not as observed bleaching.** The per-frame pattern is the familiar one in its most extreme form yet: **all 7 bleached points sit on just 2 of the 11 scored frames, and 6 of the 7 (86%) on P3190414 alone — a close-up flash macro of a healthy tan pocilloporid with pale extended polyps, ringed by white CCA, pale sponge and green ascidians, scored 20.0% bleached.** The seventh is on P3190412, a second flash macro. **Nine of eleven frames carry no bleached point at all**, and Coralscapes reads bleached at 0.054% of image area site-wide (max 0.30%, on P3190414). **Remove P3190414 and the point-based signal falls from 6.1% to 0.9% of coral points.** Report P3190414 as annotation provenance, not observation. (7) **THE PUBLISHED DATASET ALREADY DOCUMENTS THIS GAP.** The supplement's own image-based estimate, `coral cover (%; 2nd training)`, reads **41.40% against its own 61.13% transect**, recorded as a 38.49% difference. **The −28 pp is therefore not an artefact of the CV rerun** — it is the same denominator problem, already visible in the published numbers, and the corpus-wide −15 pp offset in `CHAGOS_MULTIMETHOD_README.md` is very likely the same thing. (8) **BOTH TARGET POCILLOPORID GROUPS PRESENT — *P. acuta*/*P. damicornis* (ChPd/a22–25) AND *Seriatopora* (ChSe14–15) — but the transect scored *Seriatopora* at 0.00%.** Genus present, collected and genotyped, yet invisible to the cover survey. **The four flash macros P3190412, P3190414, P3190441 and P3190461 are the candidate sampled-colony frames; none carries a visible tag or marker, and identifying ChPa22 / ChPd23 / ChPd24 / ChPd25 among them needs Anderson's judgement.** (9) **NO SURVEY APPARATUS AT ALL — 0 of 27 frames carry a tape, quadrat, slate or collecting gear.** Three show a diver (P3190391, P3190432, P3190434); four are close-up flash macros (P3190412, P3190414, P3190441, P3190461). **Unlike BITB30, the macros here read HIGHER than the wide frames** — among the scored 11, three macros average 32.2% by points and 48.0% by mask against 32.9% and 29.9% for the eight wide frames, because flash on a full-frame benthic subject removes the water column flag 1 is about. (10) **THE MASK GETS THE COMPOSITION RIGHT WHILE GETTING THE DENOMINATOR WRONG** — growth-form spectrum within 1.2 pp of the transect on both massive/meandroid and *Stylophora*, absolute cover 43% low. **The strongest evidence in the redo that the photo methods fail at scaling, not recognition**, and the argument for flag 1. (11) **P3190419 (three spotted eagle rays in silhouette) and P3190470 (a field of enormous emerald tabular *Acropora*) are flagged for illustrative use**, with P3190416, P3190389 and P3190464. **P3190419 doubles as a control on the BITB30/P3180419 pathology**: 86.7% water, 0.0% live coral by both methods — exactly the frame type that carried 29% of BITB30's bleaching signal — **and here it carries no bleached point at all.** (12) **P3190416 and P3190417's whorled foliose stands are the largest single colonies in the redo**, compatible with *Echinopora lamellosa* or a foliose *Montipora*; both unscored, neither callable to genus. (13) **Qualify "algal cover 27.78%" wherever it is quoted**: 12.29 pp is crustose coralline algae, 8.92 pp turf, 3.93 pp turf-sediment, against **1.30% fleshy macroalgae** (corpus median 2.82%) and 0.17% cyanobacteria. Not an algae-dominated reef in any competitive sense. (14) **Manifest cross-check: `coralnet_annotated: 11` is CORRECT here** — a useful counter-example to BITB30 flag 16. No per-frame depth data; the site record gives 3 depth strata and a nominal depth of 27. (15) **No COTS, no disease lesions, no predation scars and no bleached colonies visible in any of the 27 frames.**

**Benthic data:** [Download BITB34 benthic survey data (CSV)](../benthic_data/BITB34_benthic.csv) — scleractinian cover by genus with full names, plus substrate and algal categories (bare substrate, CCA, turf, macroalgae, invertebrates), from the LOF BIOT-1 survey (Mayfield & Dempsey 2026, *Oceans*, supplemental file S1).