comparer where does not filter raw_data

[otzi5300]

cmp_2m = cmp.where(cmp.data['z']==-2).rename({cmp.name: "1m"})

cmp_2m.raw_model_results contains all z levels

modelskill/src/modelskill/comparison/_comparison.py

Lines 912 to 914 in c50518c

	d = self.data.where(cond, other=np.nan)
	d = d.dropna(dim="time", how="all")
	return Comparer.from_matched_data(d, self.raw_mod_data)

This is an issue if we want to plot timeseries for the comparer.

cmp_2m.plot.timeseries()

will plot all model depths

[ecomodeller]

The tricky part here is that where() delegates directly to xarray — the condition is built against cmp.data, which raw_mod_data doesn't share. Replaying an arbitrary xarray condition on the raw data isn't really feasible.

Have you considered whether sel() could be the right place to solve this instead? It already filters raw_mod_data for time and model selection. Supporting generic aux variables there (like z) would give a clean API without exposing internals — e.g. cmp.sel(z=-2) rather than cmp.where(cmp.data['z'] == -2). Cheers, 🤖

[otzi5300]

Good idea!

I have come across another issue and possibly a bit more tricky. If we filter on specific comparer time, the functions will error if the raw data does not contain the exact same timestamp. This should be a relatively common issue since we interpolate the mod-obs pairs to model but keep the timestamps in the raw_data.

A possible fix is to show the nearest raw_model_time.