RRFS
The Rapid Refresh Forecast System.
RRFS is NOAA's new convection-allowing model. It is what takes the app's timeline past 48 hours and out to 84.
It does not replace HRRR. That is a common assumption and it is wrong. HRRR stays in operations, and RRFS retires a different set of models entirely.
RRFS is in preview. It becomes operational at 1200 UTC on 14 October 2026, a date that has already slipped twice: it was first set for 31 August, then 6 October. Until then its output is experimental, published on a preliminary basis, and can change or stop without notice. Weigh it accordingly, and weigh it less the further out you look.
At a glance
| Run by | NOAA / NCEP |
|---|---|
| Grid spacing | 3 km |
| Domain | North America |
| Dynamical core | FV3 |
| Initialized | Every hour |
| Forecast length | 18 hours from every hourly run; 84 hours from the 00, 06, 12 and 18 Z runs |
| Status | Preview; operational from 14 October 2026 |
| In the app | Timeline hours 49–84, and any earlier hour HRRR has not posted yet |
What it actually replaces.
NOAA runs a crowd of overlapping short-range models, each with its own grid, schedule and quirks. RRFS and its ensemble sibling REFS consolidate that crowd. Under NWS Service Change Notices 26-47 and 26-48, these are retired at 1200 UTC on 14 October 2026:
| NAM | North American Mesoscale — fully retired |
|---|---|
| SREF | Short-Range Ensemble Forecast — retired |
| HREF | High-Resolution Ensemble Forecast — retired, replaced by REFS |
| HiresW | High-Resolution Window — retired, except the Guam domain |
| NAM MOS | NAM Model Output Statistics — retired |
HRRR and RAP are not on that list. Both remain operational. HRRR is not merely surviving, either: over the CONUS and Alaska domains it supplies two of REFS's members, from its current and 6-hour-old runs. The two models are meant to run alongside each other, which is exactly how this app uses them.
RRFS is built on the FV3 dynamical core, the same one NOAA uses for its Global Forecast System. HRRR is built on WRF. That is why RRFS is a new system rather than another version of HRRR.
Why it earns the last 36 hours.
HRRR's long runs stop at 48 hours. RRFS reaches 84 from the same 00, 06, 12 and 18 Z cycles. That is the difference between planning tomorrow and planning the day after.
Each hour of the timeline has an owner. HRRR owns hours 1–48, and RRFS owns hours 49–84. If the owner has not posted a frame for an hour yet, the other model fills in, so a late HRRR run does not leave a hole in the near hours. Since HRRR is staying, this split is the permanent arrangement rather than a stopgap.
The app draws both models onto the same map frame with the same color table, so the handover at hour 48 does not show up as a step in the shading.
When frames appear.
RRFS posts later than HRRR, which is what you would expect from a run nearly twice as long over a larger domain. Measured on NOAA's parallel feed in September, an RRFS run started posting 80 to 115 minutes after its cycle time. A long run reached hour 84 about 2 hours 50 minutes after it.
For every hour, the app uses the newest run that has posted it. Until a new run reaches an hour, the previous run's frame stands in, and the color bar under the slider shows which run each hour came from. An hour stays empty only when no recent run covers it. Either way, the slider keeps its full length.
How far to trust hour 80.
Not very far, and that is not a criticism of the model. A convection-allowing forecast three days out will not tell you where a storm will be. What it will tell you is whether a system is expected to be over your route at all, and roughly when. That is enough to decide whether to look harder at that day.
Use the far end of the timeline to choose which day to fly. Use the near end to choose the route.
Compared with what actually happened
The site publishes forecast frames against the observed radar for the same valid time, so you can judge how the models did rather than take anyone's word for it.
Sources
- NWS SCN 26-48, implementation of RRFS and REFS, updated 9 September 2026.
- NWS SCN 26-47, retirement of NAM, SREF, HREF, HiresW and NAM MOS, updated 9 September 2026.
- NOAA Global Systems Laboratory, RRFS version 1.
- NOAA Environmental Modeling Center, HRRR.
The only thing you know for certain about the forecast is that it is wrong.
A forecast is wrong in the details by definition, and a preview model doubly so. RRFS output in this app is experimental. Radar Forecast is a planning aid, not a certified EFB. Always get an official preflight briefing.