GribStream Blog

NOAA schedules RRFS and REFS operations for August 31, 2026

May 13, 2026 |

rrfs2dfld rrfsprslev refsprslev namconusnest hrrr rap

NOAA/NWS has set August 31, 2026 at 12 UTC for operational RRFS and REFS, replacing NAM, SREF, HREF, most HiresW domains, and NAM MOS.

NOAA/NWS has now put a date on one of the biggest regional-model transitions in years: RRFS and REFS are scheduled to become operational on August 31, 2026, beginning with the 12 UTC cycle.

That same implementation retires a large part of the legacy regional suite: NAM, SREF, HREF, HiresW for CONUS, Alaska, Hawaii, and Puerto Rico, and NAM MOS. The Guam HiresW domain is the major exception called out in the notice.

For GribStream users, this is the point where RRFS stops being only an experimental dataset to watch and becomes the system NOAA is using to consolidate storm-scale regional guidance. If your product, dashboard, forecast comparison, or model workflow still depends on NAM, HREF, SREF, or HiresW, the migration window is no longer hypothetical.

The important dates

There are two dates to track:

June 9, 2026, on or about: NOAA expects a real-time RRFS and REFS feed to become available on NOMADS in parallel paths.
August 31, 2026 at 12 UTC: RRFS and REFS are scheduled for production implementation, and the legacy systems are scheduled for retirement on the same day.

NOAA includes the usual operational caveat: if the date falls under a Critical Weather Day, Enhanced Caution Event, or other significant weather constraint, implementation moves to the next eligible weekday at 12 UTC.

What RRFS changes

RRFS is NOAA's next-generation rapidly updated, convection-allowing regional forecast system. The operational configuration described by NWS is a 3 km North America system with deterministic forecasts and companion ensemble guidance.

The deterministic RRFS schedule is broader than the legacy systems it replaces:

00/06/12/18 UTC cycles: forecast out to 84 hours
other hourly cycles: forecast out to 18 hours
full North America domain: 3 km
CONUS and Alaska subsets: 3 km
Hawaii and Puerto Rico subsets: 2.5 km
fire-weather run: separate relocatable 1.5 km domain

That matters because RRFS is not just a replacement file stream. It changes the baseline model family for short-range, storm-scale workflows: severe convection, aviation weather, precipitation timing, winter hazards, wind guidance, fire-weather diagnostics, and high-resolution regional ML inputs.

What REFS changes

REFS is the ensemble product generation system built around RRFS output. It replaces HREF as the main high-resolution ensemble product family.

The NWS notice says RRFS itself produces five ensemble forecast members through 60 hours for the 00/06/12/18 UTC cycles. REFS then combines output from current and 6-hour-old RRFS deterministic and ensemble cycles. For CONUS and Alaska, REFS also includes two HRRR members from current and 6-hour-old cycles.

The products are the kinds of ensemble fields HREF users already care about: means, spread, probability-matched means, localized probability-matched means, probabilities, ensemble agreement scale products, and CONUS flash-flood recurrence interval exceedance probabilities.

The big customer-facing difference is scope and cadence. REFS runs to 60 hours, compared with HREF's 48 hours, and NWS says REFS will generate products for CONUS, Alaska, Hawaii, and Puerto Rico for every 00/06/12/18 UTC cycle.

What this means on GribStream

GribStream already serves the current experimental RRFS family:

RRFS 2D Fields: rrfs2dfld
RRFS Pressure Levels: rrfsprslev
RRFS Ensemble: refsprslev

Use those datasets today for side-by-side testing, selector discovery, and migration planning. The live GribStream RRFS catalog is the safest place to check exact variables because field availability, levels, and product split matter more than the model name alone.

The August 31 12 UTC cycle should be treated as a model-system boundary in your own validation. If you compare RRFS against NAM, NAM Nest, HREF, SREF, or HiresW, keep the cutover date explicit in backtests, alert thresholds, and model-comparison dashboards.

As NOAA moves the real-time feed into NOMADS parallel and then production directories, GribStream will track the operational feed shape and update support around the stable production products. We will keep existing GribStream dataset codes stable where possible, but the operational distribution may expose domains, product groups, and ensemble outputs differently from the experimental AWS feed.

Migration checklist

If you use legacy regional guidance, start the comparison now:

NAM or NAM Nest users: compare pressure-level fields, 2 m temperature, 10 m wind, precipitation, cloud, visibility, ceiling, and storm diagnostics against rrfsprslev and rrfs2dfld.
HREF users: compare ensemble mean, spread, probability, and probability-matched products against REFS as product coverage becomes available.
SREF users: treat this as a higher-resolution ensemble migration, not a one-for-one product rename.
HiresW users: check both domain coverage and variable coverage. The Guam HiresW domain is expected to remain, while CONUS, Alaska, Hawaii, and Puerto Rico are scheduled for retirement.
Operational customers: preserve a side-by-side period before August 31 so thresholds, model blends, and dependent models can be recalibrated.

The practical risk is not only that an old file path disappears. RRFS and REFS are different modeling systems with different physics, domains, cadence, lead-time structure, and ensemble membership. Applications that depend on model-specific behavior should validate forecast behavior, not just request syntax.

Why this is more than a data-feed change

The public weather community is already treating this as a real forecasting transition. NOAA's own testbed writeup frames RRFSv1 and REFS around direct comparisons with NAM, NAM Nest, and HREF, while noting that future MPAS-based RRFS development continues beyond v1. Local forecast teams are also starting to explain the change to viewers; WRAL's coverage, for example, points to small-scale features, precipitation timing, and the loss of familiar NAM behavior in cold-air damming setups.

That is the right way to think about the change. RRFS is a new operational baseline for a class of decisions that used to be split across several regional models. For GribStream customers, the best move is to test early, keep the August 31 boundary visible, and migrate with the exact fields and domains your workflow actually uses.