Nothing on this site is our own weather model. We fetch published forecasts from national meteorological agencies, keep them separate so each can be judged on its own, and show you which one said what.
Every card below links to the raw response for a spot you choose — the same URL our worker calls, with the same variables and units. If our number and theirs ever disagree, that link is how you find out.
Raw-data links for
The aggregator
how the data reaches usAgencies publish forecasts as GRIB2 — a compact binary format built for supercomputers, not browsers. Open-Meteo decodes it, interpolates to a coordinate and serves JSON. We use their free non-commercial tier.
We ask for each model in a separate request rather than letting Open-Meteo merge them. It costs four calls instead of one, and it means an outage in a single model cannot quietly contaminate the others or muddy the per-model scoring the archive exists for.
A licensing note, stated plainly. Open-Meteo's free tier is non-commercial use only. Wind Window carries no advertising and no subscription, so it is compliant today. If that ever changes, this pipeline moves to their commercial plan or goes direct to ECMWF's open data — which is free for commercial use, but arrives as GRIB2 and has to be decoded here. That decision gets made before any revenue, not after.
The Bureau of Meteorology
no forecast, but the observations are hereAustralia's own Bureau of Meteorology runs a global model, ACCESS-G, and it used to be the fifth column in the matrix. It is not there any more, and the reason is unglamorous: it stopped returning data.
Open-Meteo still accepts bom_access_global as a model id — the request
succeeds and comes back a well-formed HTTP 200 — but every value in the response is
null, at every location we have tested, in Australia and elsewhere. That is a domain
carrying no data rather than a malformed request, and there is no successor id. So
ACCESS-G is out of the model list, recorded in config/models.json with the
evidence, and worth re-checking periodically. It was the only Australian-agency model
in the set, which is a real loss for a site about Australian wind.
Going direct to the Bureau instead is not a drop-in fix. Their model output is distributed under a data licence agreement rather than an open licence, and it arrives as GRIB2 which would have to be decoded here. That is a project, not a config change.
Their observations are here, and they matter more
The Bureau's automatic weather stations are a different product with different terms, and those we do use — and now show. Ten of them are read every half hour: Fawkner Beacon, St Kilda Harbour, Frankston Beach, Point Wilson and South Channel Island on Port Phillip; Swanbourne, Hillarys, Garden Island, Mandurah and Rottnest in WA. They appear in the Live stations panel on the forecast page, and they are what our forecasts get scored against.
This is also the answer to "where is the Baywinds data". Baywinds is not a data source: it is a long-running display of these same Bureau masts — Fawkner Beacon, St Kilda, Frankston, South Channel, Point Wilson — and says so in its own footer. It has been doing it since the page was hosted by the University of Melbourne's School of Earth Sciences. Reading the Bureau directly gets the same stations, at full history, without scraping anybody's site.
The terms, and why we are inside them. The Bureau's data-feeds page says its anonymous products "are free to access and are not for commercial use". Its copyright page adds that, absent a specific licence, you may use its content "for personal use, or use within your organisation" but "must not supply it to any other person or use it for any commercial purpose".
Wind Window carries no advertising, no subscription, no affiliate links and no revenue of any kind. That puts it in the non-commercial category, which is the same footing community sites like Baywinds have published on for years. Every reading is credited to the Bureau and linked back to them, and the underlying feed is public and free for anyone to read.
Monetisation is intended, and it changes this. The condition above is the non-commercial status, not the attribution — so on the day this site earns anything, republishing these readings stops being permitted. At that point the choice is a Bureau data licence agreement, or turning the live display off and going back to using the observations only to score our own forecasts, which is a single config flag. The same day settles Open-Meteo's free tier, which carries the identical restriction. We would rather write that down here than discover it later.
Why a club station is still worth more
Bureau stations are excellent and they are not sited for us. Fawkner Beacon is a mid-bay platform; Rottnest is 43 m up on an exposed island, which reads systematically high against any mainland beach and is why it is never used to calibrate one. A club anemometer sits on the water people actually ride, and a club can grant redistribution rights outright — which removes the monetisation problem entirely rather than deferring it. That is why we are still asking clubs.
Everything we work out ourselves
derived, not fetched| Row | What it is | How it is worked out |
|---|
Anything in this table is computed by us from the models above. None of it is an independent measurement, and none of it has been verified against observed wind yet — see how it works.
Attribution and licences
required, and correct- ECMWF IFS and AIFS — © ECMWF, licensed CC-BY-4.0. Neither ECMWF nor Open-Meteo endorses this site or its interpretation of their data.
- GFS — NOAA, public domain as a work of the US Government.
- ICON — Deutscher Wetterdienst, CC-BY-4.0.
- Station observations — Bureau of Meteorology, © Commonwealth of Australia. Anonymous feeds, free to access for non-commercial use; republished with attribution while this site carries no revenue. See above.
- Marine wave data — via Open-Meteo. Display only, unverified.
- Coastlines and estuaries — © OpenStreetMap
contributors, licensed ODbL.
natural=coastlineandnatural=water, clipped per region and simplified to about 50 m at build time. Natural Earth 1:10m was tried first and was too coarse — it generalises a bay by a kilometre or more, which put several spots on the wrong side of their own coastline.
Observations obtained from partner stations may be stored and used to compute skill scores. Only the derived scores are ever published — raw third-party observations are never served from a public endpoint, and that boundary is enforced in the code, not just promised here.
The archive as a dataset
every forecast we have ever published, by monthEvery run is archived the moment it is issued, before it can be scored. That archive is exported monthly as CSV: one row per spot, model, issue time and forecast hour, with wind, gust, direction, temperature and cloud. The current month is rewritten nightly.
Licences travel with the rows: ECMWF IFS and AIFS are CC-BY-4.0, NOAA GFS is public domain, DWD ICON asks for attribution, all via Open-Meteo. No measured wind is included; the Bureau's readings are held under non-commercial terms and stay in our database.
Loading the index…