Model Agreement
When the world's two best forecasts disagree, you should know
Serious sailors already do this by hand: open two apps, flip between GFS and ECMWF, and try to eyeball where the stories differ. OpenSail does that comparison properly, along your actual route, with the same interpolator on both models, and renders the verdict where you plan.
Every forecast is a guess. The disagreement is the error bar.
A single forecast presented confidently is the most dangerous artifact in passage planning. The model is not lying to you; it is giving you one plausible future with the uncertainty stripped off. Most weather apps stop there, because uncertainty is hard to render and harder to sell.
OpenSail's position is that the uncertainty is the product. When GFS and ECMWF agree through +30 hours and split after, that single fact restructures a passage: you trust the first night, you build margin into the second, and you know exactly which hours to re-check when the next runs publish.
The comparison is honest by construction. Both models are sampled with an identical interpolator along the same track, run times are paired within a disclosed tolerance, and when only one model carries a field, the app says no comparison is possible instead of faking one.
On your route, not a global map
A global disagreement heat map is trivia. What changes a decision is knowing the models split near the Gulf Stream axis at the hour you cross it. Every computed route is re-sampled against the second model, and each leg gets an agreement verdict, so divergence is weighted by where your boat will actually be and when.
Pair it with the departure planner: a window where the models agree beats a marginally faster window where they do not.
Route: Ft. Lauderdale → Freeport Models: GFS 06z · ECMWF IFS 00z (±2h pairing) +00h → +30h agree (divergence < 0.15) +30h → +42h diverging (0.15 – 0.35) +42h → +48h ⚠ split (peaks 0.52 near the Stream axis) ⚠ No disagreement possible for 'currents': only one selected model forecasts it.
What it does not do
Honest limitations
- Two deterministic models are not an ensemble. Agreement between GFS and ECMWF raises confidence; it does not guarantee the atmosphere read the same forecast.
- Divergence tells you where uncertainty lives, not which model is right. Treat split zones as planning margin, not as a tiebreak to gamble on.
- Comparisons are only offered on fields both models natively forecast, at run times that pair within tolerance. When that fails, you get an explicit refusal, not a quiet blend.
Frequently asked questions
Which weather model is best for sailing, GFS or ECMWF?
Neither, universally, and anyone who gives you a one-word answer is selling something. ECMWF's IFS usually leads global verification scores; GFS runs four times a day, publishes faster, and carries fields ECMWF's open data does not. The honest answer changes with region, season, and weather regime, which is exactly why OpenSail shows you where they agree and where they split instead of quietly picking one.
What is model disagreement?
Two models given the same physics and slightly different starting conditions drift apart as the forecast runs forward. Where they still agree at +48 hours, the atmosphere is behaving predictably and the forecast deserves more trust. Where they diverge, the situation is genuinely uncertain, and your plan should carry margin there.
Is this an ensemble forecast?
No, and we say so on the page rather than in a footnote. This is a deterministic pair: the operational GFS and ECMWF IFS runs compared directly. A full ensemble spread (ECMWF EPS, GEFS) is a bigger data problem and is on the roadmap. Two independent flagship models disagreeing is still the single most useful uncertainty signal a sailor can get today.
What does the divergence number mean?
It is a normalized difference of the models' wind fields sampled along your route with an identical interpolator, so neither model gets an accidental advantage from resampling. 0 means the models say the same thing at that point of your passage; higher values mean materially different wind speed or direction at the hour you would actually be there.
Can I see disagreement without planning a route?
Yes. The map has a disagreement view for browsing, but the route-sampled version is the one that changes decisions, because it weights the disagreement by when and where you will actually be.
What happens when a field cannot be compared?
The app tells you. If only one selected model natively forecasts a field, OpenSail reports that no honest comparison is possible for it rather than synthesizing one. Model runs are also time-paired within a disclosed tolerance, so you are never comparing a fresh run against a 12-hour-older one silently.
Keep exploring
Stop flipping between two apps to build your own error bar
The live map runs real GFS and ECMWF data right now, free, no account needed.
Compare the models