The coupled ocean-wave component of ECMWF's HRES — the same centre Forecast Euro draws for the atmosphere, pointed at sea state instead. Eleven fields: significant wave height, three different periods (peak, mean and mean zero-crossing), mean wave direction, and the one thing no other source here publishes — the sea state's height split into six period bands, 10-12 s through 25-30 s. 3-hourly to +144h. Open the explorer →

ECMWF's machine-learning forecast, carried the same way Forecast Euro carries it for the atmosphere: same colours, same instrument as ECMWF Wave, so the interesting question — where the two disagree — is something you can actually see — including on the six period bands, which AIFS publishes too. 6-hourly to +240h, the longest range on the site. AIFS publishes no peak-period field at all, so its period page shows the mean period and says so rather than quietly labelling a mean as a peak. Open the explorer →

NOAA/NCEP's global wave model, coupled with GFS and run operationally four times a day — the open-data baseline most U.S. marine forecasts, including the National Weather Service's offshore and high-seas products, are built on. The richest field set here — sixteen: the combined sea, the wind sea, and three separate swell trains, each with its own height, period and direction, plus the 10 m wind driving all of it. WAVEWATCH III is the only source on the site that partitions swell three deep. 3-hourly to +168h. Open the explorer →

Environment Canada's Global Deterministic Wave Prediction System — the fourth global model this site carries, and the one measured on the fewest assumptions: its own field names (HTSGW, PWPER, PWAVEDIR) are unambiguous, unlike some of its neighbours'. Seventeen fields: combined sea, wind sea, two swell partitions, both a peak and a mean direction — where those two disagree, two wave trains are crossing — plus 10 m wind and sea-ice concentration, which is where a wave forecast stops. It is also the only source here that publishes surface Stokes drift: the net drift the waves themselves give the surface water, which is not the ocean current but is what moves floating debris, spilled oil and a person in the water on top of it. 3-hourly to +168h, twice a day (00Z, 12Z). Open the explorer →

The UK Met Office's global wave model — WAVEWATCH III again, but on the Met Office's own 25 km grid and forced by their Global 10 km atmospheric run rather than by GFS. That makes it the useful second opinion on this site: two independent forecasts of the same physics, differing only in the wind that drove them, which is where most of the spread in a wave forecast actually comes from. Nineteen fields, the fullest set here — combined sea with three period statistics (peak, Tm01 mean, Tm02 zero-crossing), the wind sea, three swell trains, and the 10 m wind on the wave grid itself. 3-hourly to +144h, twice a day: 00Z and 12Z are the only runs that go the distance, so the 06Z and 18Z runs, which stop at +71h, are left alone rather than made the timeline shorter every six hours. The one NetCDF source on the site, and the one that packs a whole day of hours into each file. Open the explorer →

Environment Canada's Regional Deterministic Wave Prediction System, run separately for Superior, Huron & Michigan, Erie and Ontario at about 1 km — each lake's own unrotated grid, on its own bathymetry, not a crop of one shared Great-Lakes raster. A freshwater basin builds a steeper, shorter-period sea than the open ocean does, which is exactly what a global model tuned for the ocean gets wrong — and why the Great Lakes carry their own section on this site. The same seventeen fields as GDWPS, Stokes drift included, at forty times the resolution. Hourly to +48h, four times a day. Open the explorer →