Forecast models

16 models.More clarity.

Compare global and regional forecasts, explore their coverage, and see where they agree at your spot.

Global models
4Every spot on Earth
Regional models
12Detail down to 1.1 km
Live networks
9Stations and buoys

Why multi-model

Weather models describe the same atmosphere using different grids, physical assumptions, observations and update cycles. On a settled afternoon they may agree. When a front approaches or coastal showers develop, their forecasts can diverge.

WAVOA keeps each model separate. Every NOW card and meteogram follows one named model end-to-end. Switch the tab to see another model’s forecast instead of hiding the differences in a single average. NBM is the labelled exception: NOAA produces that blend upstream, and WAVOA displays it as its own model.

Four global models cover every spot, including ICON-Global at 13 km. Twelve regional models add detail down to 1.1 km where available. The global wave layer has a 9.2 km grid, with groundswell separated from windsea.

The app has 16 model tabs fed by 17 upstream feeds, with NBM’s mainland and Hawaii domains combined in one tab. Marine layers and nine observation networks sit alongside them: measurements from stations and buoys are clearly distinguished from forecasts.

How a forecast reaches your screen

Direct from the agency. Model feeds come from the meteorological services that run them: ECMWF, NOAA, DWD, Météo-France, MeteoSwiss, ItaliaMeteo, DMI, ECCC and SMN Argentina. Marine fallbacks are described separately below.

Native time steps. Models keep their own grids and time steps. Resampling a coarse forecast onto a finer grid would not create real weather detail, so WAVOA preserves the differences you came to compare.

Seven days of model history. At each refresh, the analysis and earliest hours of the latest run are joined to the stored history. Each model builds a rolling seven-day hindcast on the same timeline as its forecast. Scroll back to compare that model’s past values with measured conditions at your spot.

Estimated gusts are marked. AIFS and SMN-WRF do not publish a gust field. WAVOA estimates those gusts and marks them with ~. For the other models, the gust field comes from the provider.

When a cycle fails. If a provider has not published a new run or a transfer fails, the previous cycle stays available. Its freshness indicator continues to age, so older data remains identifiable.

Explore the marine sources →

Coverage map

Explore regional coverage and the finest model grid reaching each point. Select a model to see its domain, or click the map to inspect the models covering that location. The app shows a model when it has actually produced data for the selected spot.

Global — every spot on Earth4
Regional models12

Click anywhere on the map to list every model available at that exact point — including where you would drop a custom spot.

Shaded by the finest model reaching each point. Pick a model to see its full domain, including where a finer one covers it — or click anywhere on the map to list every model available there.

Footprints are indicative. HRRR and NBM are Lambert conformal, DINI is a Lambert fan and HRDPS is rotated-pole, so their real domains are curved rather than rectangular.

Availability is per spot, not per rectangle: the app shows a model tab only once that model has actually produced data for your location.

Global — every spot on Earth

All four global models cover the world, including custom spots you add yourself. ICON-Global provides a 13 km grid everywhere, alongside the approximately 28 km grids of ECMWF, GFS and AIFS.

AIFS has its own tab because its machine-learning approach differs from the physical equations used by IFS, even though they start from the same ECMWF analysis. Keeping their forecasts separate makes disagreements visible.

Regional — finer detail where available

The twelve regional models are listed from finest to coarsest. Their coverage often reaches beyond the region in the model name. Check the additional coverage column and the map for the places you ride.

That brings 2.2 km coverage to the Croatian and Tunisian coasts, 2.5 km to parts of East Africa and the Caribbean’s South American coast, and 4 km to Chile and Uruguay. The Egyptian Red Sea gains roughly 11 km detail, Morocco’s Atlantic coast 6.5–11 km, and the Aegean 6.5 km. Availability depends on the model producing data at the individual spot.

Waves and sea temperature

Waves, tide and sea temperature use their own layers. Changing the selected wind model does not replace the sea-state layer. Wave coverage is global, including regions without a high-resolution regional wind model.

  • Copernicus Marine waves — 9.2 km worldwide. Wave height, peak period and direction, with a separate swell partition to distinguish groundswell from local windsea. The layer also includes a seven-day history.
  • Wave fallbacks. NOAA GFS-Wave at 0.25° fills locations masked as land by the Copernicus grid. The Open-Meteo marine service is the final fallback.
  • Sea-surface temperature. NOAA GeoPolar at 5 km and MUR at 1 km.
  • Tide and currents. Open-Meteo marine data, refreshed daily to match the upstream cadence.

Live measured data

A forecast describes expected weather; an observation measures what happened. WAVOA retrieves nine observation networks and shows station and buoy measurements beside forecasts, making it easier to judge the model against actual conditions.

How observations reach the app

  • Direct from the operator. Measurements are retrieved through the public APIs provided by the agencies operating each network.
  • Matched by distance. Land stations are matched to spots within 15 km; offshore buoys within 30 km. Outside those distances, WAVOA does not substitute a remote reading.
  • Visible freshness. Each observation displays its age, including the operator’s publishing delay and WAVOA’s collection cycle. The network intervals below describe the upstream observations.
  • Ten days retained, seven displayed. Recent observations share a timeline with model history so you can compare forecasts with measurements at your spot.

The nine observation networks

* Wind measurements are used from offshore buoys only. Sheltered harbour stations may understate the wind experienced on open water.

Coverage limits and model choices

The table describes the regional models currently integrated into WAVOA. Where a regional model is unavailable in the app, the four global models remain available. North and East African regional coverage is listed above. Each provider’s data access and licence conditions can be found in the sources section.

Regional — finer detail where available →

Model agreement

The spread / confidence indicator compares the models available at a spot. It summarises how far their forecasts differ; it is not a guarantee that any particular forecast will be correct.

  • Wind speed. The standard deviation of 10 m wind speed for the same forecast hour is expressed as a percentage of the mean. Below 10% is a tight spread, 10–25% is split, and above 25% is a wide disagreement.
  • Wind direction. The angular standard deviation of 10 m wind direction is assessed; differences above 30° are flagged.
  • Waves. The comparison also considers significant wave height and peak period.

The models are not fully independent: ECMWF and AIFS share an analysis, while HRRR and NBM share some upstream inputs. A broader catalog helps reveal disagreements between different systems, for example AROME and ICON-CH1 in the Alps, or HRDPS and HRRR around the Great Lakes.

Providers and licences

The agencies below provide the model and observation data described in this guide. Follow their own documentation and terms for the conditions that apply to a particular dataset. Full attribution for WAVOA is collected on our licences page.

Read the full attributions and licences →

WAVOA processes these sources into its forecast experience. Rights to reuse, adapt or redistribute underlying data depend on each provider and dataset. Consult the linked terms for any separate use of the data.