Research prototype · public data · not validated · not for navigation.

Methods and limitations

Prototype 2, methodology p2-0.1, built 2026-10-11.

This page explains how each number in Prototype 2 is counted, what has been tested, and what the data cannot tell. It is written from the build's own reports, so it changes when they do.

What it covers

21 study regions, from the Gulf of Maine to American Samoa, plus part of the Mid-Atlantic Ridge: 149,754 hexagons in all.

• Hexagons. H3 cells at resolution 6 (about 36 km²). Finer cells at resolution 8 (about 0.7 km²) show depth class and mapping detail.
• Coastal hexagons count only their seafloor: GEBCO grid cells marked as land are left out.

Data

The build reads each dataset's terms from Aethon's licence registry before it uses any file. In the prototype, Sources (at the end of each section) opens the record, file and licence behind every value.

How each number is counted

• Measured, interpolated, predicted. GEBCO's Type Identifier grid says where each depth in its 15-arc-second grid came from. A hexagon's shares are the fractions of its seafloor grid cells with each kind of code: measured (10, 11, 12, 13, 14, 15, 16, 17), interpolated (41, 42, 43, 44) and predicted from satellite gravity (40, 45). The average hexagon has 77% of its seafloor measured (76% if every grid cell is counted once instead, coastal land included). A type code says where a depth came from, not how accurate it is.
• Survey records. These are distinct surveys whose footprint or track reaches the hexagon. A survey listed in more than one layer is counted once. Footprint coverage is the area of the union of footprints, so overlaps are not counted twice. Dates are as each catalogue gives them. The oldest, NOAA charting surveys dated 1835, have a year only.
• Seen. A dive track from the included records (NOAA Okeanos Explorer ROV dives and USGS field activity 2017-001-FA) crosses the hexagon. Tracks cross 321 of the 149,754 hexagons. A camera sees a strip a few metres wide, so "seen" means seen in part. A hexagon with no track has no camera observation in these records, but other operators' dives (E/V Nautilus, MBARI, Schmidt Ocean Institute, university ships) are not read yet.
• Near the seabed. This is the World Ocean Atlas 2023 long-term mean at the deepest standard level in its square: temperature on a 0.25° grid, oxygen on a 1° grid. It is flagged where that level lies above the hexagon's seafloor.
• Gaps. There are eight fixed sentences, each shown only when its rule is met:
• Predicted only: at least 90% of the seafloor is predicted.
• No survey record: no record in the survey catalogues read.
• Old record: the latest survey is from before 2000.
• Survey not in the grid: footprints cover at least 25% of the hexagon while GEBCO counts under 10% as measured.
• Coarse resolution: shown everywhere, with the finest mapping there is.
• Few observations: 2 or fewer near-seabed temperature observations.
• Sources disagree: GEBCO and GMRT differ by more than 100 m, or GEBCO calls the depth predicted where GMRT holds ship data.
• Not integrated: near-bottom (AUV) mapping, sample, biology or geochemistry.

Each threshold was moved 20% either way to see how much the gap counts change. - Possible next observations. These name what could address a gap. They are not ranked and not validated.

Tests

• Trackline test, written down before it ran. How much does adding NCEI's older trackline record change what Aethon says is unsurveyed? The rule: a change of more than 5 percentage points in the share of hexagons with no survey record. Mid-Atlantic Ridge: 34.1% → 20.5%. New Jersey: no change. It was decided on the Mid-Atlantic Ridge and the New Jersey shelf; the 19 regions added later are reported beside them. Catalogue coverage shapes what Aethon calls unsurveyed: every "no survey record" sentence now says so. Read the test.
• Depth between surveys (Prototype 1, Ridge only). Depth estimated between ship surveys was off by a median of 147 m, more farther from a measurement. It does not apply to the shelf. Read the test.
• Anomaly scan, exploratory (not written down in advance). In 303 hexagons GEBCO and GMRT differ by at least 50 m and 10% of the depth. The data explain 272 of them (GEBCO is not measured there, or the ground is steep); 31 are not explained by anything read. And 12 rises show only in the satellite prediction. These are leads to check, not findings. Read the scan.

What it cannot tell you

• Records are counted, not how densely the seafloor was measured. A track is one line of depths.
• GEBCO's type codes describe its own compilation, not every survey ever run.
• Seen covers two dive collections.
• Only interpolated depth on the Ridge has been tested against measurements.
• Methane seeps (New Jersey) are cleared for use but not yet shown.
• NCEI trackline and multibeam use rests on NCEI's stated terms. Written confirmation has been requested.
• It is not for navigation or for anything involving safety at sea.

Reproducing it

Code, configuration, the licence registry, a checksummed download manifest and every report are kept together. One command fetches the data and one builds all 21 regions (about five minutes). The repository is private for now; anyone who wants to rerun it can ask for access.

Questions and corrections: hello@withaethon.com.