# Pre-registration: seals as oceanographers under Antarctic winter sea ice

Written 2026-11-08 ~22:30 UTC, before any profile was counted. WOD files were still downloading; MEOP list_profiles.csv was opened only to read its header.

## Question

Of all temperature profiles ever taken under Antarctic winter sea ice, what share came from instruments carried by seals (and other marine mammals), compared with ships and with profiling floats (Argo)?

## Data

- **World Ocean Database (WOD, NOAA NCEI)**, geographic files for the WMO 10° squares south of 60°S (36xx, 37xx, 38xx, 56xx, 57xx, 58xx), downloaded 2026-11-08. Instrument classes:
  - **seals** = APB (animal-borne profiles; WOD ingests MEOP);
  - **ships** = CTD + OSD (bottles) + XBT + MBT (ship-lowered or ship-dropped);
  - **floats** = PFL (profiling floats, almost all Argo);
  - **other** = GLD (gliders), DRB (drifting buoys, ice-tethered), MRB (moorings), UOR.
- **NSIDC Sea Ice Index (G02135)** monthly sea-ice concentration GeoTIFFs, 25 km polar stereographic south grid.
- **MEOP-CTD 2024-03-08 release** `list_profiles.csv`, as a cross-check on the seal count (MEOP has profiles WOD may not yet hold, and vice versa).

## Definitions (fixed now)

- **South of 60°S:** latitude < −60.
- **Winter:** June, July, August, September.
- **Years:** 1979–2025 (the satellite sea-ice record; 2026 is incomplete in WOD).
- **Under ice:** the profile's position falls in a 25 km cell whose NSIDC monthly concentration for that year-month is ≥ 15 %. Cells flagged land, coast, missing or pole-hole are excluded from the count, not counted as ice-free.
- **Profile:** one WOD cast with a valid position and date; any variable counts. Duplicate flagging is WOD's own.

## Primary test

Seal share = seals ÷ (seals + ships + floats + other), over all under-ice winter profiles, 1979–2025.

- **PASS** if the seal share is **≥ 60 %** (a clear majority, not a bare one).
- **FAIL** otherwise.

## Secondary measures (reported whichever way they come out; not part of the verdict)

1. **Coverage, not count.** Seals sample several times a day, so profile counts favour them. Grid the under-ice region into cells of 1° latitude × 3° longitude; count year-months × cells touched by each class. Report the seal share of all touched cell-months, and the share touched *only* by seals.
2. **Era split:** the same share for 2004–2025 (the seal-tag era) and 1979–2003.
3. **MEOP cross-check:** the same filter applied to MEOP `list_profiles.csv`; report MEOP count vs WOD APB count.
4. **Depth:** share of under-ice winter profiles reaching ≥ 500 m by class (seals often stop shallower than ships and floats).

## If the primary test fails

Test 3–5 pivots on the loaded data before any REJECT: share by depth band, by month, by sector (Weddell, Ross, East Antarctica, Amundsen–Bellingshausen), regions reached only by seals, and a mixed-layer result from seal data alone. Pivots are labelled as such on the page and in VERDICT.md.

## Known caveats written down before counting

- Argo floats under ice keep profiling but have no GPS fix; their positions are often interpolated afterwards. WOD keeps them at the interpolated position. This could move some float profiles in or out of ice cells.
- Monthly concentration blurs the ice edge; a profile at the edge in a month the edge moved may be misclassified.
- Prior art: published MEOP papers state seals supplied a large share of Southern Ocean profiles. The page will cite that and present this count as an independent recount, not a discovery of the fact itself.

## Amendment 1 (2026-11-08 ~22:50 UTC, still before any count)

- MBT and UOR files are fetched for the same squares; they were named above but missing from the first download.
- **Coast sensitivity (secondary, not the verdict):** profiles whose cell is coast, land or missing are excluded by the primary rule. Seals work on the shelf, so the exclusion may fall unevenly. Report excluded counts by class, and a sensitivity share in which each excluded profile takes the value of the nearest valid ocean cell (within 3 cells, 75 km).
- **Duplicates:** count seal (APB) casts sharing date, hour and position rounded to 3 decimals; report the share computed with those duplicates removed beside the primary. If WOD's under-ice seal count is well above MEOP's, treat it as a duplicate warning.
- **Float positions:** report float (PFL) under-ice counts by year, to show whether WOD keeps Argo profiles taken under ice without a GPS fix.