Weather Forecast Models Guide
minuteTemp tracks 20 weather models from agencies around the world. Here's what they are and how they differ.
What Is a Forecast Model?
Numerical Weather Prediction (NWP) models are complex computer simulations that use atmospheric physics equations across three-dimensional grids to predict future weather. Each model divides the atmosphere into a grid of cells and simulates how temperature, pressure, moisture, and wind evolve over time.
Different models use different grid sizes (resolution), cover different geographic areas, and update at different intervals. Higher resolution models capture local effects better but can only forecast a shorter time horizon. Lower resolution models can see further into the future but miss local details.
US-Focused Models (NOAA/NCEP)
These models are produced by NOAA/NCEP and focus on US weather prediction:
| Model | Resolution | Horizon | Updates |
|---|---|---|---|
| HRRR (High-Resolution Rapid Refresh) | 3km | 48 hours | Every hour |
| NBM (National Blend of Models) | 3km | 10 days | Every hour |
| GFS (Global Forecast System) | 25km | 16 days | Every 6 hours |
| NAM (North American Mesoscale) | 5km | 84 hours | Every 6 hours |
HRRR: Highest resolution US model. Updates hourly with a 48-hour horizon. US only.
NBM: Bias-corrected blend of multiple models. Applies statistical post-processing to raw model output.
GFS: Primary US global model. Covers the full 16-day forecast horizon at 25km resolution.
NAM: Regional model covering North America. 5km resolution captures more local terrain effects than GFS (25km).
Global Models (ECMWF)
ECMWF operates two global forecast systems:
ECMWF IFS25km · 10 days · Every 6 hours
Operated by the European Centre. Frequently cited in verification studies as a top-performing global model.
ECMWF AIFS25km · 10 days · Every 6 hours
AI-enhanced version of IFS. Combines traditional physics with machine learning.
Note: ECMWF data via Open-Meteo may have a slight delay compared to direct ECMWF access.
AI/ML & International Models
minuteTemp also tracks AI models and forecasts from weather agencies around the world, providing independent perspectives:
| Model | Provider | Resolution | Notes |
|---|---|---|---|
| GraphCast | Google DeepMind | 25km | AI/ML model. Different strengths than traditional models, especially for extreme events. |
| GEM | Canada (ECCC) | 10–25km | Useful for cross-border weather patterns affecting northern US cities. |
| ICON | Germany (DWD) | 13km | Strong European model with global coverage. |
| ARPEGE | France (Meteo France) | 25km | French global model with 4-day horizon. |
| JMA GSM | Japan (JMA) | 55km | Coarser resolution but provides an independent global perspective. |
| UKMO | UK (Met Office) | 10km | High-resolution Unified Model. 7-day horizon. |
| GRAPES | China (CMA) | 25km | Chinese global model. 10-day horizon. |
| ACCESS | Australia (BOM) | 40km | Australian global model. Useful for southern hemisphere perspective. |
Models by Forecast Horizon
Different models cover different time ranges. Here's how they compare:
HRRR
Highest resolution (3km), hourly updates capture rapidly evolving conditions.
HRRR + NBM
HRRR provides 3km detail; NBM adds statistical bias correction.
NBM + ECMWF IFS
NBM is bias-corrected for US stations; ECMWF provides an independent global perspective.
GFS + ECMWF IFS
These models extend to 10-16 days. Forecast skill decreases significantly beyond day 7.
Model Consensus
When 10 or more models agree on a temperature within a narrow range, that's a strong signal of confidence. minuteTemp highlights model consensus on the dashboard.
Conversely, when models diverge significantly — especially when HRRR and ECMWF disagree by more than 2°F — uncertainty is high. This doesn't mean the forecast is wrong; it means the atmosphere is in a state where small differences in initial conditions lead to very different outcomes.
See reading the dashboard for how to spot consensus and divergence visually.
Resolution Explained
Model resolution describes the size of each grid cell. Smaller cells = more detail:
- •3km (~2 miles) — Neighborhood-level. Can capture sea breezes, urban heat islands, terrain effects.
- •10-13km (~6-8 miles) — City-level. Good detail for most purposes.
- •25km (~15 miles) — Regional. Captures large-scale weather patterns but misses local effects.
- •55km (~34 miles) — Synoptic-scale. Broad patterns only. Individual city conditions are interpolated.
3km models (HRRR, NBM) capture local effects — sea breezes, urban heat islands, terrain-driven variations — that can shift temperatures by several degrees compared to coarser grids.
Note