pvtend.tendency.load_climatology

pvtend.tendency.load_climatology(clim_path: Path, engine: str = 'netcdf4', chunks=None) → Dataset[source]

Load climatology, auto-detecting the file layout.

Fallback order:

  1. Single merged file

  2. Per-var-per-month files

  3. Per-variable files

chunks=None (default) keeps the lazy-backend layout. Pass a dask chunk spec (e.g. {"day":1,"hour":1}) to back the variables with dask: each .sel(month,day,hour).values then computes and releases its slice instead of retaining it in the per-variable backend cache. This is what the PPVI workers use — without it a long-lived worker accumulates one ~0.4 GB month-chunk per accessed month/var (~19 GB/worker), throttling the cgroup.