Ever wondered why Windy and Windguru show different numbers for the same beach on the same afternoon? The answer almost always comes down to which model is doing the talking. Understand the three big global models and the mystery disappears, and you become a sharper forecast reader in the process.
What a weather model actually is
A global weather model divides the atmosphere into a 3D grid, feeds in current observations, and runs the physics forward in time. Two numbers decide how useful the output is for us:
- Resolution (grid spacing): how fine the grid is. A coarse grid smooths over the local terrain, sea breezes and gust structure that make or break a kite session. Finer is generally better for spot-level wind.
- Update frequency (run cadence): how often the model is re-run with fresh data. More runs per day means you catch shifts sooner.
One more concept worth knowing: deterministic vs ensemble. The deterministic run is the single “best guess.” The ensemble runs the model many times with slightly nudged starting conditions to show the spread of possible outcomes. When the ensemble fans out widely, the forecast is low-confidence, which matters a lot for trip planning.
The three models at a glance
| Model | Run by | Resolution | Updates / day | Particular strength |
|---|---|---|---|---|
| ECMWF | European Centre for Medium-Range Weather Forecasts | ~9 km HRES / ~18 km ensemble | 2 | Most accurate medium range (~3-10 days) |
| GFS | US NOAA | ~13 km deterministic / ~33 km ensemble | 4 | Fast, free, the default behind many apps |
| ICON | Germany’s DWD | ~13 km global / ~6.5 km ICON-EU | 4 | Central Europe, mountains, wind & convection |
The resolutions and run schedules here are based on published model specs, but agencies upgrade their models every so often, so treat the figures as current-at-writing rather than permanent. One older comparison from Windy.app, for instance, cited ECMWF at ~14 km and GFS at ~27 km, which tells you the numbers tighten up over time as the models improve.
ECMWF: the accuracy benchmark
ECMWF is the one forecasters reach for when they want the best medium-range guidance. Its high-resolution deterministic run sits at roughly 9 km, the ensemble around 18 km, and it updates twice a day. It’s widely considered slightly more accurate than GFS overall, and the edge shows up most in the 3-to-10-day window, which is exactly the window you care about when booking a kite trip or deciding whether next weekend looks worth driving for.
The catch for hobbyists is access. ECMWF data isn’t as freely and openly distributed as GFS, so apps that surface it often gate it behind a paid tier. If your app lets you toggle ECMWF on, it’s usually worth a look, especially for multi-day planning.
GFS: the free workhorse
GFS, run by the US NOAA, is the model most of us see by default without realizing it. It’s global, runs at about 13 km deterministic (with a coarser ~33 km ensemble), and updates four times a day. Its two big advantages are speed and price: it’s fast and it’s free, which is why so many apps and websites build on it.
It’s perfectly good for day-to-day “is there wind tomorrow” checks. Where it loses ground to ECMWF is at longer ranges, where small early errors compound. For a session two or three days out, GFS is usually fine. For a forecast a week out, lean on ECMWF if you can see it.
ICON: the European specialist
ICON (Icosahedral Nonhydrostatic) comes from Germany’s DWD. The global version runs around 13 km and updates four times a day, but the part kiters in Europe should care about is ICON-EU, a higher-resolution regional version at roughly 6.5 km. That finer grid pays off for Central Europe, mountainous terrain, and wind-and-convection-driven situations like thermal sea breezes and frontal passages that coarser global models blur out.
If you ride in Europe and your app offers ICON or ICON-EU, it’s a genuinely useful third opinion, not just filler.
Which model is “most accurate”?
The short version: ECMWF leads in the medium range, but no model is best everywhere or every day. Accuracy depends on the region, the weather pattern, and how far out you’re looking. A high-resolution regional model like ICON-EU can beat a global model for a specific European spot, while ECMWF’s ensemble will give you the most honest read on a week-out trip decision.
This is why experienced kiters don’t marry one model. They look at two or three and read the relationship between them.
The convergence rule: the only forecasting trick you really need
Here’s the practical payoff. Pull up the same spot and time on ECMWF, GFS and ICON:
- If they agree (similar speed, similar direction, similar timing), confidence is high. Trust it, pick your kite, and go.
- If they diverge (one says 18 kt (33 km/h / 21 mph) cross-onshore at noon, another says 12 kt (22 km/h / 14 mph) two hours later), the atmosphere is genuinely uncertain. Expect a volatile, possibly gusty session, keep your kite choices flexible, and recheck after the next model run.
Model divergence isn’t a bug to be annoyed at. It’s information, the forecast telling you the day is borderline.
How this shows up in your apps
You don’t run these models yourself; your apps do. Windy is the standout for this exact job because it lets you switch ECMWF, GFS and ICON layers side by side, which makes the convergence check trivial. Windguru is widely understood to run GFS plus its own higher-resolution and regional models, though we’d point you to Windguru’s own docs for the exact current lineup rather than stating it as fact. For the full rundown of which app does what, see our guide to the best wind and forecast apps for kitesurfing.
Once you know which model you’re looking at, the next skill is reading the columns properly: average vs gust vs direction. That’s covered in how to read a wind forecast for kitesurfing, and if you’re still sizing up whether a forecast is even rideable, start with what wind speed you need to kitesurf.
Bottom line
ECMWF is your best single bet for medium-range accuracy, GFS is the free workhorse running quietly behind most apps, and ICON (especially ICON-EU) is the European specialist worth a look. But the real skill isn’t picking a favorite, it’s reading whether they agree. Convergence means confidence; divergence means caution. Build the habit of glancing at more than one model and you’ll get burned by far fewer “the forecast lied to me” days.