climval 0.2.0 adds data alignment and a CI gate for climate models
Northflow released climval 0.2.0 on PyPI on 13 November 2026. The new version aligns a climate model and its reference before it scores them, and adds a check that stops a build when a model misses its limits. climval is Northflow's open Python library for validating climate models against observations and other models.

One call from two datasets to a score
climval.compare takes a reference and a candidate as NetCDF, Zarr or CSV, local or remote. It puts both on one grid, matches calendars and time steps, converts units, masks missing values in pairs and weights every cell by its area. Each step is recorded with the result, so anyone can see how a score came about.
Three metrics are new: centred RMSE, the ratio of standard deviations and Kling-Gupta efficiency.
Built for ERA5 at full size
A year of hourly ERA5 at 0.25 degrees against a daily model at 1 degree now plans in a few seconds and tens of thousands of tasks. Before, the same job needed 0.4 to 12 million tasks or ran out of memory.
A gate in CI
climval check reads limits per metric from a YAML, TOML or JSON file and returns exit code 1 when a limit is broken. A GitHub Action runs the check, writes a summary to the job and keeps the report.
climval 0.2.0 is on PyPI under Apache 2.0, and installs with pip install climval.


