Climate models, put to the test.
climval is an open source Python library that checks climate model output against ERA5, CMIP6 and CORDEX. It is the quality check on the Earth observation data that enters our pipelines.
A model is only as good as its comparison.
Climate projections drive decisions on flood defences, farming and energy. Comparing them with observations is routine work that is often done by hand, a little differently each time.
climval makes that comparison the same every time, so results from different teams can be set side by side.
What it reads.
- Reference data
- ERA5 reanalysis, CMIP6 model ensembles and CORDEX regional downscaling.
- Station data
- Weather stations through Open-Meteo and Meteostat, in the groundtruth notebook.
- Your own model
- Any simulation output that fits the same structure.
How it works.
- Load
- Model output and reference data are read into one typed structure.
- Compare
- RMSE, mean absolute error, mean bias, Pearson correlation and the Taylor skill score, with Taylor diagrams.
- Report
- Scorecards export as HTML, JSON and Markdown, the same way for every team that runs them.
groundtruth: ERA5 against the stations.
Daily air temperature at two metres, 2015 to 2020, at five stations in five climate zones.
- Stations
- Oslo, Madrid, Nairobi, Toronto and Sydney.
- Output
- A scorecard per station, time series, a world map and a report you can export.
- Setup
- None. It runs in Binder in the browser.
In figures.
- PyPI
- Install with pip install climval
- Apache 2.0
- Open source licence
Version 0.1.0, installable today.
climval 0.1.0 was published on PyPI in February 2026 under Apache 2.0. In March 2026 we published groundtruth, a notebook that checks ERA5 against weather stations, and it runs in Binder with no credentials.
Who it serves.
- Climate scientists comparing models.
- Teams that depend on climate data and want to know how far to trust it.
