ERA5: what is a climate reanalysis?

A reanalysis is not a station record, and it is not a simulation either. It is the two combined: take a modern weather forecasting model, replay the entire past through it, and at every time step feed in every observation available on that date — surface stations, buoys, ships, radiosondes, aircraft, and satellites from the 1970s onwards.

The model supplies physical consistency and fills the gaps; the observations keep pulling it back towards what was actually measured. The result is a complete grid with no holes, where station networks leave entire continents nearly empty.

The word "analysis" comes from meteorology: it is the name of the reconstructed initial state before each forecast. "Re-analysis" means all those past analyses have been redone, in one pass, with a single version of the model — and that last point is what makes the exercise valuable.

Operational weather models are improved continuously. Stacking the analyses produced across decades would make every improvement a step change in the series — and a step change is indistinguishable from a change in climate.

By replaying the whole past with a frozen version, a reanalysis removes that source of spurious trends. It is what allows 1950 to be compared with 2020 without the comparison being mostly about progress in modelling.

Consistency is not perfect for all that: the quantity and quality of observations have themselves changed enormously. The satellite era is far better constrained than the 1940s, and polar or oceanic regions early in the record lean more on the model than on measurement.

Two figures circulate, and both are correct. The NATIVE resolution of the model producing ERA5 is about 31 kilometres; the archive itself is distributed on a regular 0.25° grid, roughly 28 kilometres, and that is the grid this site queries. The order of magnitude is what to remember: a few tens of kilometres.

Each cell carries a single value representing the average over that area — not the value at its centre, and not that of any particular point.

The consequences are concrete. Two cities twenty kilometres apart often fall in the same cell and therefore share exactly the same data. Sharp relief is smoothed: a valley and the ridge above it are averaged together although their climates differ. An urban heat island, which plays out over a few kilometres, does not appear at all.

This is why this site reports no difference between two cities below half a degree: under that threshold you are reading the resolution of the grid, not a difference in climate.

ERA5 is excellent for averages, totals, long trends and comparison between regions of the world. For those uses, its homogeneous coverage and temporal depth have no public equivalent.

It is the wrong tool for a microclimate, for a very local extreme, or for any use that requires data from a named station — a contract that specifies "the station at such-and-such airport" needs that station, not a grid-cell average. The records shown on this site are ERA5 cell records: mutually consistent and consistent worldwide, but they will not match exactly the official records of a national weather service, which rest on thermometers.

ERA5 is produced by ECMWF under the Copernicus Climate Change Service (C3S), and redistributed under the CC-BY 4.0 licence. That is what lets this site recompute, reformat and republish the data freely, provided the source is credited — which every page does.

What Climate Memory contributes is therefore not the raw data, which is public: it is the computation of normals, trends and indicators across tens of thousands of localities, and their presentation in a readable form.