ACCESS vs ECMWF vs GFS
Australia's own model, the European heavyweight and the American workhorse compared — resolution, accuracy, cost and when to trust each.
ACCESS, ECMWF and GFS are three of the global weather models that shape Australian forecasts, and they differ in who runs them, how finely they see the atmosphere and how they perform. In short: ECMWF is generally the most accurate global model, GFS is the freely available American workhorse, and ACCESS is Australia's own system — tuned for our region and the basis of official BOM warnings, but not released as open data.
Meet the three models
- ACCESS — the Australian Community Climate and Earth-System Simulator, run by the Bureau of Meteorology. Built on the UK Met Office Unified Model and run at global (ACCESS-G, ~12 km), regional and city resolutions.
- ECMWF (IFS/HRES) — the deterministic high-resolution model of the European Centre for Medium-Range Weather Forecasts, at roughly 9 km. Consistently rated the world's leading global model.
- GFS — NOAA's Global Forecast System from the United States, at about 13 km, run four times a day and free to everyone.
Side by side
| ACCESS-G | ECMWF (HRES) | GFS | |
|---|---|---|---|
| Operator | Bureau of Meteorology (Australia) | European Centre (Europe) | NOAA (United States) |
| Approx. resolution | ~12 km | ~9 km | ~13 km |
| Runs per day | Multiple | Twice (00z, 12z) | Four (00/06/12/18z) |
| Typical range | ~10 days | ~10 days | ~16 days |
| Open gridded data? | No public feed | Yes | Yes, fully |
| Regional strength | Tuned for Australia | Global, very strong overall | Global, noisier at long range |
Resolution: how finely each model sees the atmosphere
Resolution is the grid spacing — how far apart the model's calculation points sit. ECMWF's roughly 9 km global grid is the finest of the three, which helps it resolve fronts and coastal detail a little more sharply. ACCESS-G and GFS sit close together near 12–13 km. Finer is generally better, but only up to a point: the physics, the data assimilation and the sheer quality of the starting snapshot matter just as much as raw grid spacing. ACCESS also runs finer regional and city grids over Australia that outperform any global model for local detail such as sea breezes and mountain effects.
Accuracy: which one wins?
By the standard verification scores that meteorological agencies publish, ECMWF is generally the most accurate global model in the medium range, and has been for years. That is not a marketing line — it reflects long-running, publicly tracked verification. GFS is a capable model and its four-times-daily, fully open output makes it the backbone of countless free apps, but it tends to be noisier and less consistent than ECMWF at longer ranges.
ACCESS is harder to slot into a simple ranking. As a global model it is competitive; its real advantage is the high-resolution, Australia-focused configurations and the fact that Bureau forecasters run and tune it locally. It is the model behind the warnings your state emergency service acts on. "Best global model" and "best model for a Severe Weather Warning over the Illawarra" are different questions with different answers.
Cost and availability
This is where the three diverge most for everyday users. GFS is completely free and open, which is why it appears everywhere. ECMWF has moved much of its core data to open access, so third-party sites can show it. ACCESS, by contrast, is not published as an open gridded feed — the Bureau delivers it through its own MetEye, app and warning products. That is why consumer tools like Windy can show ECMWF and GFS but not ACCESS.
It also explains the design of comparison sites. Because a site like checkweather.io cannot ingest ACCESS, it compares the three open global heavyweights — ECMWF, GFS and ICON — measures how much they agree, and links back to BOM for the official Australian picture. No black boxes: you can see each model's take and how confident the agreement is.
Update frequency and forecast range
The models also differ in rhythm. GFS runs four times a day (00, 06, 12 and 18 UTC) and reaches out to about 16 days, so it refreshes often and looks a long way ahead — though that far horizon is more curiosity than forecast. ECMWF's flagship high-resolution run updates twice daily and extends to around 10 days, prioritising quality over quantity. ACCESS-G runs on its own schedule to support the Bureau's forecast cycle. For a user, more frequent runs mean a forecast can update sooner as new observations arrive; but a fresher run of a noisier model is not automatically better than an older run of a sharper one.
Which handles Australian conditions best?
Australia poses distinctive challenges — weather generated over the data-sparse Southern Ocean, explosive summer convection, tropical cyclones in the north and the peculiar dynamics of East Coast Lows. ECMWF's global skill translates well to the large-scale drivers, but ACCESS's regional and city configurations, tuned to Australian geography, often capture local detail that a single global grid smooths away. GFS provides a valuable independent check. In practice, the strongest Australian forecast is not any one of these in isolation — it is a forecaster reading all of them and reconciling where they differ.
Which should you actually trust?
For a single best guess two to five days out, ECMWF is the smart default. For official Australian warnings, cyclones and fire danger, BOM (running ACCESS plus its own judgement) is the authority — full stop. And for gauging confidence, the honest move is not to crown one model but to compare them: when ACCESS-driven BOM guidance, ECMWF and GFS all point the same way, trust it; when they split, treat the forecast as genuinely uncertain and wait for the next run.
The bottom line
ECMWF edges the accuracy crown among global models, GFS wins on openness and update frequency, and ACCESS wins on local tuning and its role behind Australia's official warnings. The models are not rivals so much as complementary witnesses — and the smartest forecasting reads all of them.