Help / NOAA models
NOAA models: choosing one for your task
GFS, GEFS, AIGFS, AIGEFS, HGEFS and NAEFS: what they are for, how their forecasts differ and how to read ensemble products on the portal.
Start with your task and available data. A model name, finer grid or AI label alone does not guarantee a better forecast.
Models covered here
This guide covers six atmospheric families connected to the portal's data system. GFS is labelled NOAA in some source names. NAEFS is a joint US–Canadian product also distributed by NOAA; its global fields are not limited to North America.
Inclusion in this guide does not imply that every field is available in every tool. Choose from published layers and the model list of the particular tool; time, level and parameter must all be available.
Quick choice by task
| Task | Starting point | What to check |
|---|---|---|
| Point weather, travel, outdoor work | GFS; AIGFS to compare the large-scale scenario | Near-surface level, valid time, run freshness and current observations. |
| Wind and temperature aloft, route profile | GFS with an available vertical profile | One model run, vertical reference and arrival time at each point. |
| Planning several days ahead, weather-window consistency | GEFS; AIGEFS as an additional ensemble | Mean together with spread; individual members and probabilities where available. |
| Combining physical and AI perspectives | HGEFS | Mean and spread; it is not independent of GEFS and AIGEFS. |
| Comparing ensembles from different weather centres | NAEFS | Product type, bias correction and available parameters. |
| Paraglider, drone or balloon launch; local gusts and storms | Model for the overall situation, then local observations and profiles | Terrain, wind at operating height and equipment limits; a global grid does not resolve every local detail. |
GFS — the main physical forecast
The Global Forecast System calculates atmospheric evolution using physical equations. It provides one deterministic scenario: a useful starting point for weather maps and vertical profiles.
On this portal it supports temperature, wind, pressure, humidity and available cloud-field analysis. For a route, use levels actually present; read diagnostic icing or turbulence fields together with their stated method.
GEFS — a physical ensemble
The Global Ensemble Forecast System has 31 atmospheric members. Variations in initial conditions and model treatment produce different weather scenarios.
Use GEFS to examine the consistency of a trend: whether cooling, a wind shift or a weather window persists across scenarios. Read mean geavg together with spread gespr; an individual member is just one possibility.
AIGFS — one AI-based forecast
The Artificial Intelligence Global Forecast System is a deterministic weather model based on GraphCast and adapted by NOAA. It produces numerical atmospheric fields; it is separate from the portal's text AI chat.
Use it to compare the large-scale scenario with GFS. NOAA reports improvements in several metrics but identifies tropical cyclone intensity limitations in version 1.0. AI therefore does not imply better performance for every phenomenon.
AIGEFS — an AI-based ensemble
The Artificial Intelligence Global Ensemble Forecast System has 31 members. The portal's data system supports its mean avg and spread spr fields.
Use it as an additional perspective on forecast uncertainty and compare with GEFS at the same valid time. Agreement between means does not rule out large differences among individual members.
HGEFS — physical and AI ensembles together
The Hybrid Global Ensemble Forecast System combines 31 GEFS and 31 AIGEFS members: 62 in total. It is a hybrid ensemble product.
Use its mean and spread to consider both approaches together. GEFS, AIGEFS and HGEFS are not three independent votes: HGEFS already includes the first two.
NAEFS — a US–Canadian ensemble
The North American Ensemble Forecast System combines the US GEFS and the Canadian global ensemble. The connected product applies bias correction and provides mean geavg and spread gespr.
Use it to compare scenarios from different weather centres and assess the consistency of large-scale conditions. It shares a component with GEFS, so their agreement is not fully independent evidence either.
NOAA forecast range and product grids
These are NOAA product ranges, not a promise of data availability on the portal. Output grid spacing differs from internal model resolution and from actual forecast accuracy.
| Model | NOAA forecast range | Product grid and time step |
|---|---|---|
| GFS | Up to 16 days | The portal system supports 0.25°, 0.5° and 1° grids. NOAA describes hourly output through +120 h, then every 3 h. |
| GEFS | Up to 16 days; a separate extension of the 00 UTC cycle reaches 35 days | 0.25° and 0.5° products are used. The time step depends on lead time and product. |
| AIGFS / AIGEFS | Up to 16 days (+384 h) | 0.25°; every 6 h. |
| HGEFS | Up to 10 days (+240 h) | 0.25°; every 6 h. |
| NAEFS | Up to 16 days | The connected bias-corrected product uses 0.5°; the time step depends on lead time. |
Reading mean, spread and members
Example: a mean temperature of +15 °C with a spread of 1 °C expresses different member agreement from the same mean with a spread of 5 °C. Writing +15 ± 5 °C does not automatically provide a guaranteed interval or frost probability.
Spread of U and V represents uncertainty in wind components. It is not a physical wind vector for balloon trajectories or drift calculations. A trajectory needs a scenario's wind; mean wind is usable only as an explicitly labelled average scenario.
| Label | Meaning | What it does not mean |
|---|---|---|
| mean / avg / geavg | The average across ensemble members. | Not necessarily a realizable or most likely scenario; extremes may be smoothed. |
| spread / spr / gespr | A measure of member dispersion, usually standard deviation, in the field's units. | Not a probability percentage or guaranteed forecast error. |
| control / member | The control or an individual perturbed member. | Not the whole ensemble or a guaranteed outcome. |
| probability | Likelihood of a specified event or threshold exceedance, if such a product is published. | Cannot be read directly from mean or spread without additional data and assumptions. |
Comparing models fairly
- Choose the same point or route and forecast valid time.
- Check the run time: matching valid times can come from different model runs.
- Match parameter, units and level: 10 m AGL, 850 hPa and SFC are different surfaces.
- For precipitation, match accumulation intervals; for ensembles, match the statistic.
- Check missing values and freshness. A missing field is not zero or evidence that a hazard is absent.
- If models disagree, retain alternative scenarios and check observations instead of automatically selecting the most favourable forecast.
When additional data is needed
For decisions immediately before a flight or outdoor work, supplement model fields with current observations and relevant official reports. METAR/SPECI are aerodrome observations and TAF is an aerodrome forecast; these are different products, not global model names.
HRRR, RAP, NAM, NBM and other NOAA products are outside this list of connected atmospheric families. Roshydromet is also represented in the system but is not part of NOAA. Space weather is separate from atmospheric forecasting.
Check your understanding
Three common model-selection questions.
HGEFS and GEFS means agree. Is that independent confirmation?
Not fully: HGEFS includes GEFS members. Account for shared ensemble components.
Does a temperature spread of 4 °C mean a 4% chance of error?
No. It is dispersion in degrees, not a percentage probability.
NOAA produces 16 days, but the map shows only three. Where is the rest?
The portal shows the range published in its data system. The source model's capabilities and the loaded dataset can differ.
Official sources
Specifications were checked against NOAA/NCEP/NWS. Task suggestions are portal guidance, not an official NOAA ranking of models.
- NOAA/NCEP: GFS — physical model and forecast output
- NOAA/NCEP: GEFS — GEFS ensemble and forecast range
- NWS: Service Change Notice 25-89 — AIGFS, AIGEFS and HGEFS composition, grid and lead times
- NCEP: AIGEFS/HGEFS verification — ensemble verification and specifications
- NOAA: AI-driven global weather models — AI model purpose and version 1.0 limitations
- NCEP: NAEFS — international ensemble and bias correction
- NCEP: Model Analysis and Guidance products — NAEFS composition and global coverage
- NCEP: Ensemble Background — interpreting ensemble statistics