Hybrid GFS Ensemble
| Provider: | National Centers for Environmental Prediction, NOAA (USA) |
| Model scope: | Global |
| Update frequency: | every 6 hours |
| Resolution: | 0.25°, 15.0nm, 27.8km |
| Model duration: | 41 forecasts starting at 0 hr, ending at 10 days |
| Parameters: | pressure, wind, temperature, vertical velocity, 250 mb, 500 mb, 850 mb, ensemble standard deviation |
| GRIB model date: | Tue Feb 10 18:00:00 2026 UTC |
| Download date: | Wed Feb 11 00:44:24 2026 UTC |
| Download delay: | 6hr 44min |
Note: the Download delay is the amount of time required for the GRIB model to compute its forecast and then for the LuckGrib cluster to download the data and make it available. The LuckGrib delay is generally less than 10 minutes, the remainder of the delay is the model compute time.
Introduction to ensemble models
At the risk of oversimplifying, weather models come in two main flavors: deterministic and ensemble.
A deterministic model does its best to capture the current atmospheric state and simulate its evolution forward in time. Inevitably, inaccuracies creep in: initial conditions can never be measured perfectly, and the physics (or ML) simulations are approximations.
Ensemble models address this uncertainty by running many slightly perturbed versions of the same forecast simultaneously. Small variations are introduced in the initial conditions and/or model physics, producing a spread of possible outcomes. The idea is that the true future state is likely somewhere within that spread. Areas where members agree indicate higher confidence; wide disagreement signals greater uncertainty.
NOAA is now running (at least) three different global ensemble models. The first of these has been around for a while, the following two have just been introduced.
- GEFS — Long-standing, 31 members based on the GFS model.
- AIGEFS — New ML-based, 31 members derived from AIGFS.
- Hybrid GEFS (HGEFS) — Combines all 62 members (31 GEFS + 31 AIGEFS) for the best of both worlds.
The Hybrid GEFS (HGEFS)
HGEFS merges the physics-based GEFS ensemble with the newer ML-based AIGEFS ensemble, creating a 62-member hybrid system.
The Hybrid GEFS offers data describing the ensemble average, along with the standard deviation. These two parameters offer a simple way to study the likely forecast outcome, along with a measure of its confidence.
For technical background, see the NOAA repository document here. The abstract highlights a key advantage of the ML component:
… Recently, data-driven machine learning weather prediction (MLWP) models not only produce superior forecasts compared to traditional numerical weather prediction (NWP) models but also run with magnitudes smaller computing resources and are significantly faster than traditional NWP.
For more on ML in weather forecasting, see the short discussion here.
HGEFS produces forecasts four times daily (00Z, 06Z, 12Z, 18Z), with output every 6 hours out to 10 days.
Available parameters
Currently, HGEFS offers a limited but useful set of ensemble products:
- Means (ensemble average) for key fields such as sea-level pressure, 10 m wind and temperature.
- Spreads (standard deviations) for the same fields—indicating forecast uncertainty (low spread = high agreement/confidence; high spread = greater uncertainty).
Many diagnostic parameters available in GEFS (e.g., full individual member fields, wave/swell, vorticity) are not yet provided in HGEFS. As with other early ML-based systems, the parameter set is expected to expand over time.