Forecast Model run timing.
- 2 hours ago
- 3 min read
FORECAST TIMING 101 (BoM • ACCESS • ECMWF • GFS) — why different apps show different answers
If you’re seeing different outcomes between BoM, Windy, TV, and other apps, it’s usually NOT the “model being wrong” — it’s the run timing, processing delay, and apps being one run behind.
The main models (what they are) BoM ACCESS (Australia) • BoM’s modelling suite focused on Australian performance. • Best use: short-range detail, coastal effects, storms/rain bands (especially higher-res versions).
ECMWF (Europe) • Global model that’s often strong on the big-picture pattern. • Best use: synoptics and mid-range trends (the “shape” of the week).
GFS (USA/NOAA) • Global model that updates frequently and is widely distributed. • Best use: quick situational awareness and trend checking vs ECMWF/ACCESS.
ACCESS models you’ll see in Australia (the differences) ACCESS-G (Global) • Global coverage, good for broad pattern + longer-range context.
ACCESS-R (Regional) • Regional Australia domain, usually sharper than global for Aussie detail in the first few days.
ACCESS-C (City / Convective scale) • Very high-resolution, short-range, location-focused detail (great for local timing/structure, but only short range).
ACCESS-CE (City Ensemble) • A “bundle of runs” around ACCESS-C to show uncertainty (spread = confidence check).
ACCESS-TC (Tropical Cyclone) • A relocatable higher-resolution domain used for tropical cyclone guidance.
Run times (the “Z time”) and why it matters Models run at set “Z” times: • 00Z, 06Z, 12Z, 18Z In AEST that’s roughly: • 00Z = 10am AEST • 06Z = 4pm AEST • 12Z = 10pm AEST • 18Z = 4am AEST
But here’s the key: the model run time is NOT when you’ll see it. After a run finishes, it still needs processing, packaging, and distribution.
Why Windy and many apps are often “one run behind” Most apps ingest model data on a schedule. So even after a new run starts, you can still be viewing the previous completed ingest. Windy (and many others) can lag because: • the run has to finish • data must be processed into their format • servers distribute it worldwide Result: you can easily be looking at an older run while a newer run exists.
Rule of thumb: If an app/site doesn’t clearly show the model run timestamp (the Z time), you can’t verify freshness. If they won’t show the run time — dump the app.
Why TV and radio can be behind too Broadcast workflows are built around deadlines: • they pull model data • build graphics • time it for a segment By the time it airs, a newer run may already be out. And yes — sometimes they use older runs because it’s easier (or laziness).
Paid data vs public feeds Some services pay for faster access, higher detail, or more consistent delivery (ECMWF and some BoM data services). That can mean “newer” data shows up sooner on paid platforms than free ones.
What we do on Wally’s Weather — Daily Edition Our Daily Edition stays aligned with the latest BoM ACCESS updates as they become available, so you’re not stuck looking at yesterday’s run dressed up as “today”.
Quick check you can do right now:
Find the model label (ACCESS / ECMWF / GFS)
Find the run time (00Z/06Z/12Z/18Z)
If the run time isn’t visible, you have no way to confirm freshness
Known apps and runs
Old runs = about 4 to 6 hours older than the current run
Most recent = current run but still could be 4 to 6 hours old, check update time
Meteologiz - old runs
Windy.com - old runs
TropicalTidbits - most recent
NCEP Model Guidance - most recent
BOM Meteye - most recent (Wally purchases this)
Earth null school - most recent
Earth zoom - unknown but only has GFS or Icon and looks old
Ventusky - unknown
FlowX - unknown
Clime - unknown
References
Comments