A short explanation of how the prediction is made, what feeds it, and how much to trust it.
This page predicts sea level at Avonmouth, on the Severn Estuary near Bristol, for any date from 1970 to 2037. The prediction has two layers: a purely astronomical tide calculation that needs no weather information and works for any date in that range, and an optional weather correction that adds the effect of air pressure and wind when real or forecast weather is available.
The 32 harmonic constants driving the astronomical prediction were derived from real sea-level measurements at Avonmouth, taken every 15 minutes from 1 January 2002 to 10 April 2012.
Sea level rises and falls mainly because of the changing gravitational pull of the moon and sun as the Earth rotates and the moon orbits. That combined effect can be broken down into a set of steady, individually-predictable oscillations called harmonic constituents — the best-known is M2, the twice-daily lunar tide.
The 10+ years of Avonmouth gauge data were run through a least-squares harmonic analysis (the same class of method national tide-table authorities use) to work out the size and timing of 145 of these constituents at this specific location, then the 32 largest were kept. Because the underlying astronomy is predictable indefinitely, this part of the prediction needs no live data at all — it's exact arithmetic, not a forecast.
The Bristol Channel funnels the tide into an unusually large range: the M2 constituent alone contributes over 4 metres of amplitude here, among the largest tidal ranges anywhere in the world.
Actual sea level is the astronomical tide plus a "surge" — extra height (or a deficit) driven by weather, mainly low air pressure and onshore wind piling water up the estuary. This is most visible during storms, where surge can add well over a metre to the predicted tide.
A machine-learning model (gradient-boosted trees) was trained to predict this surge from air pressure and wind, using two weather sources:
The model was trained on 2002–2010 data and tested on 2011 to April 2012 — data it never saw while training — to check it actually generalises rather than memorising. It measurably improves accuracy, most noticeably during storms.
On that held-out test period, the astronomical-only prediction reconstructs the real record to within about 0.31 m RMS error. Adding the weather correction brings that down to about 0.24 m; roughly half of that improvement is genuine weather skill (isolated by testing the model on weather alone, with the tide's own state withheld), the rest reflects the harmonic model's own small residual imprecision.
A few things worth knowing before reading too much into any single number this site shows you:
This is an independent, experimental project, not an official tide table. For navigation, safety planning, or flood-risk decisions, use the UK Hydrographic Office's Admiralty EasyTide or the Environment Agency's flood warning service instead.