Weather History
How accurate the data is
Weather History does not show records from a weather station on your street. Historical weather comes from reanalysis: measurements from thousands of stations, satellites and buoys, merged into a model and computed onto a grid of cells. This page says what follows from that — what you can rely on, where to expect differences, and what to do when a number looks wrong.
Where a number comes from
Historical data comes from the ERA5 reanalysis, computed by the European centre ECMWF and served by Open-Meteo. A reanalysis takes every measurement available for a given day — from ground stations, satellites, balloons, ships and buoys — and lets a physical weather model work out the state of the atmosphere over the whole planet.
The result is not a map of stations but a grid of cells. Depending on the period a cell is roughly nine kilometres across (from 2017 on) to twenty-five (earlier years), and holds a single set of values for its whole area. When you ask about a place, a land cell with a similar elevation is picked and the values are then statistically adjusted to the actual elevation of the spot, using a 90-metre elevation model. That is why two neighbouring towns get different numbers — but the difference between them is derived from the terrain, not measured on site.
In other words: the number you see is the best reconstruction of that day from the model and everything that was measured that day. A real thermometer at that spot may have read something slightly different at that hour.
Treat a single hour in 1963 as a good estimate, and the shape of the whole day as reliable.
What it means in practice
- The shape of the day is reliable. When it warmed up, when the rain started, how hard the wind blew — the course of the day matches.
- A single hour is a good estimate. A degree or two of difference from the nearest station is a normal property, not an error.
- Mountains, valleys and coasts differ more. Temperature is adjusted to the elevation of the spot, but rain, wind and cloud cover apply to the whole cell. A valley inversion, the lee of a hill or a sea breeze are too small for the model to see.
- A local shower may be missing. A thunderstorm over one village is smaller than a cell. The model spreads it over the surroundings, or misses it altogether.
- Older data is less certain. Before 1979 there were no satellite observations and fewer stations. Numbers from the 1950s are coarser than yesterday’s.
- A forecast is a forecast. The nine days ahead come from MET Norway and each further day is less certain — which is why the app draws them outlined, not as filled cards.
In the app
Every result screen carries the line “ERA5 reanalysis — measurements merged into a model grid and interpolated to this place” at the bottom, with a link to the accuracy note. The source of a number is never more than one glance away from the number itself.
What the app is not
It is not a station archive or an official record. It is not suitable as evidence for an insurance claim, for court, or for any decision where tenths of a degree matter. National weather services hold the actual station records for that.
Why reanalysis, then? Because it is the only thing that can answer a question about any place on Earth on any day since 1950 — including a village where no station ever stood. That breadth is paid for in precision, and the app does not pretend otherwise.
When a number looks wrong
Compare it with the record of the nearest official station for the same day, not with memory or a garden thermometer. If the difference looks like an error rather than a property described above, write in with the place and the date — it can be checked.
Accuracy is something to keep working on, and concrete examples help with that more than anything else.