Future reconsideration of features on MetPy's xarray accessor and its place in the ecosystem

[jthielen]

Following up on #2361, given that all of MetPy's current xarray accessor functionality is rather general-purpose (grid projections, CF coordinate awareness, unit handling, etc.) and that efforts like cf-xarray and pint-xarray have implemented many of those features in independent packages (as well as expanding beyond what MetPy does), it would be worth reconsidering on the time horizon of MetPy v2.0 what, if anything, MetPy should have in its own accessor.

As they exist now, the combination of cf-xarray and pint-xarray handle much of the features we need, just excluding those related to grid projections and related CRS-type operations. So, for those existing packages, the main question seems to be longevity and continued maintenance. With the yet-missing-elsewhere features, given the progress in xarray on explicit indexes, I have a feeling that a suitably lightweight (i.e., not rioxarray/rasterio) "geospatial coordinate utility" package will arise to support geospatial indexing concerns (probably aligned with xoak). So, I wouldn't be surprised if, by the time for v2.0 rolls around, there will be sufficient overlap of other packages to obviate the need for MetPy's own accessor utilities, and we can instead simply add dependencies on those packages.

That all being said, perhaps new features will be added to MetPy's accessors (such as those related to #3) between now and v2.0, which would mean keeping the accessors around for that reduced set of features rather than getting rid of them entirely.