Severe · 6 min
How to read ensemble spread when a hurricane is five days out
The mean track is the least useful line on the chart. Here is what the spread is actually telling you, and how to act on it.
Ensemble Weather AI Research ·
When a tropical cyclone is five days from landfall, every news graphic shows a single line with a cone around it. That line is an average of many model runs, and it is the one position almost no individual model predicted.
An ensemble forecast is a set of separate runs, each started from a slightly different initial state or produced by a different model. Where they agree, the atmosphere is behaving predictably. Where they fan out, small differences today become large differences later — and that fanning is itself the forecast.
A tight cluster at day five is a genuinely confident forecast, and you can commit resources to it. A bimodal spread — half the members going left of a landmass, half going right — is not a forecast with a wide error bar. It is two different outcomes, and the mean sits in the gap between them where nothing will actually happen.
This is why we return every member rather than a single number. A utility positioning repair crews does not want the average landfall. It wants to know whether there is a branch of the ensemble that puts the eye over its service territory, and what fraction of members take it.
The practical rule: look at the shape of the distribution before you look at its centre. Count the members, not the line.
Every response carries the spread
Ensemble members, model weights and p10/p50/p90 on every value — including on the free tier.