REFS pressure-level ensemble forecasts: CONUS at 3 km

GribStream Code: refsprslev

NOAA RRFS ensemble forecasts on 14 pressure levels from 1000 to 250 hPa: five members at 3 km over CONUS, hourly through 60 hours. Temperature, moisture and circulation fields from the pre-implementation feed.

Forecast at a glance

Compare 850 hPa temperature and moisture, mid-level winds and geopotential height, and upper-level vorticity. Combine these profiles with the surface-member dataset to examine each forecast scenario from the ground upward.

How RRFS generates these forecasts

NOAA’s Rapid Refresh Forecast System version 1 uses the limited-area Finite-Volume Cubed-Sphere (FV3) dynamical core within the Unified Forecast System. Its North American domain has 3 km horizontal grid spacing; these datasets expose the CONUS output on a 1,799 × 1,059 Lambert conformal grid. The system targets short-range severe convection, heavy precipitation, winter weather and other hazards for which the timing and structure of weather systems matter. RRFSv1 complements HRRR, which remains in operation. NOAA GSL: RRFSv1; NCEP: RRFS products.

Why the five members differ

The five current-cycle members vary initial conditions, lateral boundary conditions and model physics relative to the deterministic RRFS forecast. Initial conditions represent uncertainty in the starting atmosphere; boundary conditions affect weather entering the regional domain; physics choices change the treatment of processes such as clouds and turbulence. Following the same member across fields preserves its evolving weather scenario. NWS SCN 26-48.

What this ensemble includes

GribStream’s REFS member datasets contain the five individual RRFS ensemble forecasts from one initialization. NOAA’s RRFS Ensemble Forecast System (REFS) statistics combine current and six-hour-old RRFS deterministic and ensemble runs, with current and older HRRR runs over CONUS and Alaska. An average or threshold count across these five members therefore has a different membership from NOAA’s published REFS mean or probability products. NWS SCN 26-48.

Interpreting resolution and uncertainty

A 3 km grid can represent mesoscale structure, but grid spacing is not a guarantee of storm-location accuracy or a promise to resolve individual tornadoes. Clouds, turbulence and other unresolved processes still require parameterization; RRFSv1 also uses convective parameterization. Do not equate “convection-allowing” with “all convection explicitly resolved.” The physics and evaluation context are documented by NOAA: RRFSv1 physics development and NCEP/EMC evaluation.

With five members, an unweighted threshold count changes in steps of 20 percentage points. That arithmetic is a sample frequency, not a calibrated event probability. Shared model errors can remain even when all five agree. For site decisions, examine nearby cells, adjacent forecast hours and successive runs, and verify against observations. This archive spans a changing pre-implementation system, so it is not a stationary climatology or a skill ranking against HRRR, HREF or a global ensemble.

What pressure levels add

The pressure product samples each member on 14 fixed pressure surfaces from 1000 to 250 hPa; these are output levels, not the model’s native vertical layers. Temperature and moisture help describe air masses and instability, while wind and geopotential height describe the flow steering weather systems. Use the same member and valid time when combining levels or joining surface fields. Sparse pressure sampling limits detailed sounding diagnostics, and pressure surfaces near the ground can intersect terrain. The table below gives all eight parameters and their levels.

History and source revisions

The new CONUS archive begins with the 9 September 2026 12 UTC run. Historical surface files also include FRZR freezing-rain accumulation, giving 42 parameter names across the archive versus 41 in the current product. NOAA's October 15 revision removed that accumulation and changed instantaneous precipitation-type processing. Compare precipitation-type verification across that change with care. The pressure product contains eight parameters throughout the reviewed archive.

How this fits the RRFS family

Members 1–5 are individual forecasts. The deterministic RRFS datasets remain separate. NOAA's ensemble means, spread and probability products are a later addition, as are Alaska, Hawaii and Puerto Rico. The refsprslev name now refers to this CONUS product; the retired prototype's North America grid and history are not carried forward.

Companion member dataset: refs2dfld · Deterministic RRFS 2D · Deterministic RRFS 3D

NOAA source and operational schedule

These are NOAA's pre-implementation member forecasts. NWS schedules operational implementation for 3 November 2026, subject to weather-related delay. NOAA provides the data under CC0 and requests attribution. GribStream preserves the published values and member identities.

NOAA / AWS · NWS SCN 26-48 · CC0-1.0

Parameters