Isobaria: publishing how wrong your own product is
isobaria.com- Problem
- Weather apps give you a number and keep quiet about the margin. When the models disagree with each other —which is exactly when the forecast is worth least— you never find out: you see the same figure with the same confidence.
- Solution
- Isobaria cross-references three models (ECMWF IFS, ICON and GFS) into an agreement index that drops when they disagree, and publishes a verification page with each model's measured error. When the first independent validation against AEMET weather stations qualified the earlier method, it was published anyway, with its limitations spelled out.
- Outcome
- The June 2026 verification compared 19,560 forecast-observation pairs across eleven official stations, with the evidence archived immutably and the breakdown by model, horizon and province open to anyone. The product can be audited from the outside, which is the opposite of what the sector does.
- Next.js
- TypeScript
- Open-Meteo
- AEMET


