Draft

When does a new run arrive, and why the delay?

A run is named after its start time, but the files appear hours later. Real delays for the models in the catalog.

What “00Z, 06Z, 12Z, 18Z” mean

Most global models are started four times a day, at 00, 06, 12 and 18 UTC (“Z” stands for Zulu time, which is UTC). A “12Z run” is the forecast whose starting time, the time of the analysis, is 12:00 UTC. It describes the atmosphere from 12:00 UTC onward, but the forecast is not available at 12:00: it has to be computed first. Some regional models start more often: every three hours or even every hour.

Why there is a delay

  • Waiting for data. Observations arrive late, so the analysis waits for a cut-off time.
  • Building the analysis. Blending observations with the last forecast takes time, even on a supercomputer.
  • Running the forecast. The model works through its time steps. The early forecast hours are finished first, which is why they appear before the later ones.
  • Post-processing and upload. Output is converted to files, such as GRIB, and copied to public servers.

Real delays from this catalog

+0 h+2 h+4 h+6 h+8 h+10 h+12 hrun start time (for example 12:00 UTC) →HRRR0 h 50 m → 1 h 18 mICON-D20 h 44 m → 1 h 21 mAROME2 h 05 m → 2 h 45 m ≈ICON2 h 40 m → 3 h 28 mGFS3 h 32 m → 5 h 14 mGEFS3 h 46 m → 6 h 28 mECMWF IFS7 h 34 mECMWF ENS7 h 40 mJMA GSM9 h 31 m ≈first filelast file
Time from the start of a run to the first and the last forecast file appearing, for selected models, from the schedule data in the catalog.

For example GFS first files appear about 3 h 32 m after the run time, ECMWF’s IFS about 7 h 34 m, and the fine-scale ICON-D2 only 0 h 44 m. Fine-scale models that cover small areas often come out sooner than global ones. Figures marked ≈ are estimates.

TermMeaning
First fileThe first forecast step is available. The next few hours of the forecast can already be used
Last fileThe whole run is available, to the end of the forecast
Run time (cycle)The start time of the run, in UTC
Valid timeThe time a forecast value is for. Valid time = run time + forecast step

The Dashboard and Timeline pages show the next expected time for each model, and the History page shows how reliable each model’s publication has been.