Why do weather forecasts disagree?
Two apps, two different answers for the same afternoon. The reason is not carelessness — it is that they are running different models, and disagreement is useful information.
Weather forecasts disagree because different apps run different models, each starting from a slightly different snapshot of the atmosphere and solving the physics with different code. Add the atmosphere's inherent chaos and, over New Zealand, sparse observations across the Southern Ocean, and small differences quickly grow into different forecasts. Disagreement is not a mistake — it is a signal about how confident anyone can be.
They are not all the same forecast
The most common misunderstanding is that there is one official forecast that apps copy. There is not. Behind the scenes, apps draw on different global models — most often ECMWF (European), GFS (American) or ICON (German) — and each is an independent simulation of the whole planet's atmosphere. One app showing ECMWF and another showing GFS will naturally differ, in the same way two well-informed experts can reach different conclusions from the same evidence.
Four reasons forecasts diverge
1. Different starting data
Every forecast begins with a snapshot of the current atmosphere, assembled from satellites, weather balloons, aircraft, ships and surface stations. Each centre blends these observations differently, so no two models start from exactly the same picture. Because the atmosphere is chaotic, even tiny differences at the start grow into large differences days later — the famous “butterfly effect” described by meteorologist Edward Lorenz.
2. Different physics and resolution
Models represent clouds, rainfall, turbulence and terrain with different equations and at different grid spacings — roughly 9 km for ECMWF, 13 km for GFS and ICON. A finer grid resolves mountains and coastlines better. These design choices mean the models genuinely “see” the world slightly differently.
3. The atmosphere is chaotic
This is the deep reason. The atmosphere is a chaotic system, so predictability has a hard ceiling of around two weeks. The further ahead you look, the more the models diverge — not because they are getting worse, but because the future genuinely becomes less certain.
4. Global versus local models
A global model and a high-resolution local model — such as those MetService runs over New Zealand — will differ, especially for mountain rain and channelled wind. The local model usually captures terrain effects that the global one smooths away.
Why New Zealand sees more disagreement
Some of this is universal, but New Zealand tends to see wider spread than the well-observed Northern Hemisphere. Weather arrives from the west across the enormous, data-sparse Southern Ocean, where there are very few surface observations to pin the models down. They rely heavily on satellite data, which leaves more room for the models to differ about exactly what is coming and when. Fast-moving systems in the westerlies and sharp terrain contrasts then magnify small timing errors into visible disagreements between apps. An ex-tropical cyclone tracking towards the upper North Island is the classic case: models can differ by hundreds of kilometres on where it goes, and that single uncertainty ripples through every app's forecast for the region.
Disagreement is information, not noise
Here is the shift in thinking that makes you a better forecast reader: when models disagree, that tells you something valuable — the atmosphere has not settled on one outcome. When they agree, confidence is high. Professional forecasters have used this idea for decades through “ensembles”, running a model many times with slightly nudged starting conditions to see how much the forecasts spread. If nearly every ensemble member shows rain, the forecast is close to a sure thing; if half show rain and half show sun, the honest message is that it genuinely could go either way.
Seen this way, the disagreement between two apps is a low-tech version of the same signal. It is the atmosphere refusing to commit — and that is worth knowing, because it tells you whether to book the outdoor wedding with confidence or to have a wet-weather plan ready.
| What the models are doing | What it means for you |
|---|---|
| ECMWF, GFS and ICON broadly agree | High confidence — plan on it |
| Models agree on pattern, differ on timing | Trust the trend, hold the detail loosely |
| Models clearly split | Genuine uncertainty — keep plans flexible |
How checkweather.io uses the disagreement
Rather than hiding the differences behind one confident-looking line, checkweather.io does the opposite. We line ECMWF, GFS and ICON up side by side and turn how closely they agree into a plain-English confidence score, so you can see at a glance whether a forecast is rock-solid or a coin toss. We also keep an accuracy scoreboard that grades each model against the weather that actually happened. No black boxes — just an honest view of where the models agree and where they part ways.
What to do when your apps disagree
When two apps give different answers, do not simply pick the one you prefer. Instead, ask how far ahead you are looking (disagreement grows with distance), check whether it is timing or the whole pattern that differs, and lean on the model with the better local track record — often ECMWF for the medium range. And whenever a warning is involved, go straight to MetService, New Zealand's official forecaster and the only source of Severe Weather Watches and Orange and Red Warnings. Two forecasts disagreeing is not a problem to be annoyed by — it is the atmosphere telling you how sure to be, and once you learn to read it that way, a divided forecast becomes one of the most useful things you can see.