Data

What is in the archive, how far back it goes and what we do to it before you see it.

Variables

Code	Variable	Unit	Resolution	From
`t2m`	Air temperature at 2 m	°C	hourly	1940
`td2m`	Dew point at 2 m	°C	hourly	1951
`rh`	Relative humidity	%	hourly	1951
`pmsl`	Pressure at mean sea level	hPa	hourly	1940
`ws10`	Wind speed at 10 m	m s⁻¹	hourly	1949
`wd10`	Wind direction at 10 m	degrees	hourly	1949
`prcp`	Precipitation total	mm	hourly, daily	1940
`snwd`	Snow depth	cm	daily	1955
`sun`	Sunshine duration	min	hourly	1961
`gsr`	Global solar radiation	W m⁻²	hourly	1981

Six further derived variables (heating degree days, growing degree days, apparent temperature, vapour pressure deficit, wind chill, humidex) are computed on request rather than stored.

Quality control

Five stages, applied in order. A value that fails any stage is flagged, never dropped — you can always request the raw series with qc=none.

Range check against physically plausible bounds per variable and latitude band.
Step check: rejects impossible hour-to-hour jumps, tuned per variable.
Persistence check: flags a sensor reporting a constant value for longer than expected.
Spatial consistency against the nearest five stations within 60 km.
Internal consistency, for example dew point above air temperature.

Bulk downloads

Whole-network snapshots are published as Parquet on the first of each month. Compressed sizes below; each file has a companion SHA-256 and a Croissant metadata sidecar.

Bundle	Period	Size	Format
Hourly, all variables	1940–2026	214 GB	Parquet, partitioned by year
Daily aggregates	1940–2026	11 GB	Parquet
Station metadata	current	18 MB	CSV, GeoJSON
Normals 1991–2020	reference	340 MB	Parquet, NetCDF