How close is a global tide model to the tide tables?
Tide tables for a port are made from that port's own tide gauge. Years of readings are broken into their repeating components, and the components are run forward to predict the future. It works very well, and it only works where there is a gauge.
Our tides API takes a different route. It uses FES2022, a global model of the ocean tide built by LEGOS, NOVELTIS and CLS for the French space agency CNES.[1] It follows the FES2014 atlas, whose design is described in a 2021 paper.[5] The model covers every coastal and offshore point on Earth, so it can answer for a beach that has never had a gauge. We wanted a number for how close it gets to the tables.
The test
We took fifteen NOAA tide stations and one month, October 2026. For each station we pulled NOAA's predicted high and low waters[2] and computed our own at the station's coordinates. Then we matched each NOAA high or low with ours and measured two things: how far apart they were in time, and how far apart in height. Both sets of heights are relative to mean sea level.[3]
That gave about 120 high and low waters per station.
| Station | Mean range | Time difference | Height difference |
|---|---|---|---|
| La Jolla, CA | 1.13 m | 2.8 min | 5.6 cm |
| Boston, MA | 2.79 m | 3.7 min | 5.8 cm |
| Atlantic City, NJ | 1.17 m | 3.7 min | 9.8 cm |
| Monterey, CA | 1.05 m | 4.1 min | 5.3 cm |
| Duck, NC | 0.95 m | 4.8 min | 16.5 cm |
| Neah Bay, WA | 1.64 m | 6.5 min | 9.0 cm |
| Charleston, SC | 1.53 m | 7.1 min | 13.0 cm |
| San Francisco, CA | 1.21 m | 7.1 min | 4.4 cm |
| San Juan, PR | 0.33 m | 18.5 min | 6.4 cm |
| Anchorage, AK | 7.59 m | 19.5 min | 84.9 cm |
| Honolulu, HI | 0.38 m | 22.7 min | 4.3 cm |
| Key West, FL | 0.38 m | 30.6 min | 12.7 cm |
| The Battery, New York | 1.35 m | 47.2 min | 11.5 cm |
| Galveston Pier 21, TX | 0.34 m | 50.5 min | 15.1 cm |
| Seattle, WA | 2.28 m | 57.5 min | 11.2 cm |
The time column is the mean absolute difference. The height column is the root mean square difference.
Open coasts
Eight of the fifteen stations come in at about seven minutes or better. They face open water or sit in a harbour with a wide mouth, and they have a tidal range of about a metre or more. At those, the heights are within 4 to 17 cm.
For planning a walk across a tidal flat, timing a photograph, or deciding when to launch a kayak, that is close enough. The weather will move the real water level by more than that on most days.
Sounds, inlets and narrow entrances
Seattle and New York are the two stations where the times differ most, and always in the same direction. Our high and low waters come 57 minutes before NOAA's at Seattle and 47 minutes before at the Battery.
The cause is geography. The tide reaches Seattle by travelling the length of Puget Sound, and it reaches the Battery through the Narrows. That journey takes time. A global model has grid cells a few kilometres across, so at these stations it reports the tide near the entrance, before it has travelled up the channel. Anchorage, at the head of Cook Inlet with a range over seven metres, shows the same effect in height: 85 cm.
Global tide models share this. A review of the leading ones found their errors against observations grew from about 0.9 cm in the deep ocean to 5.0 cm on continental shelves and 6.5 cm at the coast.[4] For a point at the head of a sound, the local tide tables are the right source for timing.
The small-range stations are a different story
Honolulu, San Juan, Key West and Galveston all have a mean range under 40 cm, and all four show timing differences of 18 to 50 minutes.
When the range is that small the tide curve is nearly flat at the top, and a few centimetres of disagreement can slide the peak a long way sideways. At Honolulu the heights agree to 4.3 cm, the best in the table, while the times are 23 minutes apart. The time of high water there is not a number anyone can pin down tightly, and it matters less, because the water is barely moving.
What we take from this
On an open coast with a real tidal range, the model is within a few minutes and several centimetres of the tide tables, with no gauge involved.
Look at the map before you pick a point. If it faces open water, use the prediction as it is. Up an estuary, inside a sound, or behind a narrow harbour entrance, the times can differ by up to an hour, and the local tables are the ones to use.
Reproducing this
NOAA's predictions are public. This is the request for Boston:
curl "https://api.tidesandcurrents.noaa.gov/api/prod/datagetter?product=predictions&begin_date=20261001&end_date=20261031&datum=MSL&station=8443970&time_zone=gmt&units=metric&interval=hilo&format=json"
And ours, for the same point:
curl -H "Authorization: Bearer $THIRDTRAIL_KEY" \
"https://thirdtrail.life/v1/tides/extremes/?lat=42.35389&lon=-71.05028&start_date=2026-10-01&days=31"
A month in the past needs a paid plan. On the free tier you can compare the next two days instead.
Sources
- AVISO+, CNES. "FES2022 (Finite Element Solution) Ocean Tide." https://www.aviso.altimetry.fr/en/data/products/auxiliary-products/global-tide-fes/release-fes22.html
- NOAA Center for Operational Oceanographic Products and Services. "CO-OPS API for Data Retrieval." https://api.tidesandcurrents.noaa.gov/api/prod/
- NOAA CO-OPS. "Tidal Datums." https://tidesandcurrents.noaa.gov/datum_options.html
- Stammer, D., Ray, R. D., Andersen, O. B., Arbic, B. K., et al. (2014). "Accuracy assessment of global barotropic ocean tide models." Reviews of Geophysics, 52(3), 243-282. https://doi.org/10.1002/2014RG000450
- Lyard, F. H., Allain, D. J., Cancet, M., Carrère, L., and Picot, N. (2021). "FES2014 global ocean tide atlas: design and performance." Ocean Science, 17(3), 615-649. https://doi.org/10.5194/os-17-615-2021
Method: NOAA high and low water predictions for 1 to 31 October 2026, MSL datum, GMT. thirdtrail predictions from FES2022 using 34 constituents at each station's published coordinates. Each NOAA event was matched to our nearest event of the same type within three hours. This compares two predictions, not a prediction with observed water levels. Tide predictions exclude weather. Informational use only, not for navigation.