Merge Datatree siblings when they are compatible

[jonas-spaeth]

Hey,

I've come across datatree only recently, but it I already see many use-cases in my work, so thanks for the effort!

A minimal example of what I'd like to have:

• I have data from two models (say 'us' and 'eu'), and they have different x and y coordinates → therefore I store them as siblings in a datatree dt (structure: "forcasts/us", "forcasts/eu") instead of a single Dataset
• After some manipulation, both Datatrees are actually compatible, e.g., because I averaged over x and y
• for further analyses, I'd like to concat the two Datatrees in a single Dataset along the dimension model, something like xr.concat([dt["us"].ds, dt["eu"].ds], dim="model").assign_coords(model=["us", "eu"]) (for example, I could then use dt['forecasts'].ds.t2m.plot(hue='model'))
• it would be nice to allow such an operation, e.g., via dt.concat_leaves(), where the result is a datatree "forecasts"

A very similar use case would be:

• I have forecast runs where the resolution changes after day 15 of the integration
• I could store them as a datatree ("forecast/short-range", "forecast/medium-range")
• after doing some manipulation, e.g., spatial averaging, both forecast ranges could be compatible, and one sibling stores leadtime days 0-14 and one stores days 15-46
• it would be cool to have again something like dt.merge_leaves() to have a new dataset with a continuous leadtime

If something like that is already possible I apologize for my ignorance.

Again, thanks for putting this together.

Cheers, Jonas

[keewis]

see #192 for some discussion on collapsing subtrees in general

[flamingbear]

Closed in favor of #192