---
title: "The Bristol Channel and a very confused datum"
author: "Ivo Sandoval (@nullisland)"
date: 2026-07-11T08:33:38.256Z
updated: 2026-07-11T08:33:38.256Z
canonical: "https://jot.place/@nullisland/the-bristol-channel-and-a-very-confused-datum"
description: "Chart Datum at Avonmouth sits 6.50 m below Ordnance Datum Newlyn. Here is the arithmetic that goes wrong when a height file does not name its zero."
tags:
  - "cartography"
  - "evidence"
  - "field-notes"
  - "measurement"
  - "metrology"
---

# The Bristol Channel and a very confused datum

A contractor sent me a bathymetric surface for a reach of the Severn Estuary in March, and the only thing wrong with it was a number that was not in the file. Everything else was excellent: dense soundings, sensible gridding, a proper uncertainty layer. Nobody had written down which zero the heights were measured from. In that estuary, that omission is worth six and a half metres.

## Fourteen metres of anything is a lot

Avonmouth is a Standard Port, so its tidal levels are published.

| Level | Height above Chart Datum |
| --- | --- |
| Mean high water springs | 13.2 m |
| Mean high water neaps | 9.8 m |
| Mean low water neaps | 3.8 m |
| Mean low water springs | 1.0 m |

Mean spring range 12.2 m, mean neap range 6.0 m. You will also see 14 m and 15 m quoted for this coast, and those figures disagree because they are different quantities, not because anyone measured badly.[^1] Mean spring range at a named port is one number. The extreme range, highest astronomical tide down to lowest, is always larger. And figures published for "the Severn Estuary" tend to be taken further upstream where the funnel has had more distance to work in; about 15 m is the usual quote for the estuary as a whole. The Bay of Fundy averages around 16 m and holds the record without much argument. Whether the Severn comes second or third depends on whether you count Ungava Bay in northern Quebec as a separate contender, and I have never found two sources that agree.

## A depth on a chart is not a depth

In UK waters, Chart Datum is at or very near Lowest Astronomical Tide: the lowest level that can be predicted to occur under average meteorological conditions and under any combination of astronomical conditions. Both halves matter. Astronomical, so no storm surge. Any combination, so it is a level the sea reaches essentially never, and in many years not at all.

A charted sounding of 2.4 m therefore does not tell you there is 2.4 m of water. It tells you the seabed sits 2.4 m below a surface the sea almost never touches. To get an actual depth you add the predicted height of tide, also published above Chart Datum, and the two numbers add only because they share a zero.

Good design, incidentally. Every charted depth is a worst case, so a mariner who ignores the tide entirely is being conservative rather than being dead. It fails only when a second dataset arrives carrying a different zero.

## Two zeros, six and a half metres apart

Ordnance Datum Newlyn is the datum for land levelling in England, Scotland and Wales, fixed at mean sea level as recorded by the Newlyn Tidal Observatory between 1915 and 1921. The site was chosen for sitting in granite bedrock, being clear of major rivers, and better representing open ocean levels. It replaced Ordnance Datum Liverpool, and the difference between the two was determined to be 0.13 ft, which tells you something pleasant about the Second Geodetic Levelling.

Chart Datum is not a national datum at all. It follows LAT, and LAT follows the local tidal range, so the offset between the two zeros is a local quantity.

| Port | Chart Datum relative to ODN |
| --- | --- |
| Avonmouth | 6.50 m below |
| Liverpool | 4.93 m below |
| Newlyn | 3.05 m below |
| Portsmouth | 2.73 m below |

Look at Newlyn. Even at the tide gauge that *defines* Ordnance Datum Newlyn, Chart Datum is 3.05 m below it, because ODN is a mean sea level and Chart Datum is a low water level. There is nowhere the two coincide, and the spread across those four ports alone is 3.77 m.

## Where it goes wrong, with real numbers

Take a berth at Avonmouth. The chart gives a sounding of 2.4 m. The Admiralty prediction for the high water in question is 11.4 m above Chart Datum. Four ways to work out the depth of water at high water, two of them right.

| Method | Bed level | Water level | Depth | Error |
| --- | --- | --- | --- | --- |
| Everything in Chart Datum | 2.4 m below CD | 11.4 m above CD | 13.8 m | none |
| Everything in ODN | 8.90 m below ODN | 4.90 m above ODN | 13.8 m | none |
| Chart sounding read as ODN, gauge feed in ODN | 2.4 m below ODN | 4.90 m above ODN | 7.3 m | 6.5 m too shallow |
| Survey bed in ODN, Admiralty tide in CD | 8.90 m below ODN | 11.4 m above CD | 20.3 m | 6.5 m too deep |

The two correct routes agree exactly, as they must, because a depth is a difference between two surfaces and it does not care which zero you subtract from both. The wrong routes are each out by precisely the datum offset, in opposite directions, and here is what should make you uneasy: neither wrong answer looks wrong. Both 7.3 m and 20.3 m are plausible depths at that berth at that state of tide. No range check, plausibility filter, unit check or schema validation catches either. The units are right. The arithmetic is right. Only the reference surface is wrong, and the reference surface is the one piece of metadata that almost never travels with the numbers.

Row three cancels a movement that would have been fine. Row four sends the ship.

:::warning
The failure is regional, which is why it survives so long. The same mistake at Portsmouth costs 2.73 m, quite likely to be swallowed by whatever under-keel clearance you were already carrying. So a pipeline written and tested on the south coast passes, ships, and works for years. The Bristol Channel is where a latent datum bug finally gets a big enough lever.
:::

## What I ask before I open somebody else's soundings

- [ ] Which vertical datum, named explicitly, not implied by the folder the file lives in
- [ ] If Chart Datum, at which port, because the offset is local and there is no national value
- [ ] Sign convention: are depths positive downward or negative upward
- [ ] Which horizontal frame, and whether the vertical component was converted with it or quietly left alone
- [ ] Epoch of the survey, because a seabed under a 12 m range does not sit still
- [ ] Whether a datum shift has already been applied, so I do not apply it twice

The last one catches me most. A missing conversion is often obvious, because the result is absurd. A conversion applied twice gives you a number that is still plausible and is now wrong by exactly double, and it will sit in a dataset for years being confidently cited.

There is proper machinery for this, and the Vertical Offshore Reference Frames work out of UCL exists precisely to relate Chart Datum to the ellipsoidal and land datums offshore, where there is no gauge to interpolate between. But machinery is not the fix. The fix is refusing a height file that does not name its zero, which sounds officious right up until the first time you nearly ground something.

The same arithmetic bites from the other direction in fieldwork, where being in the right place on the right day at a time two hours out means the animals you came for are under a metre of water. Wren's @lowtide/rockpool-survey-prussia-cove is this problem where the penalty is a wasted morning.

A depth is a difference between two surfaces. Name both, or you have not recorded a depth. You have recorded a number.

[^1]: This is the honest reason published tidal trivia contradicts itself, and it generalises far beyond tides. When two reputable sources give you 14 m and 15 m for the same estuary, the useful question is almost never which is right. It is which quantity each of them thinks it is reporting, and that part is usually undocumented.