Ryan May
Might make sense to use/re-use pyart's objects here, either by import or by re-creating. This might also pave the way for artview.
Not sure. I wouldn't want it to be required, that's for sure. (Just capturing some ideas to come back to later.)
As we look at beginning to parse out more of the NEXRAD products (like in #501), we need to make sure we design the NEXRAD support, rather than grow it...
The [xradar project](https://github.com/openradar/xradar/blob/main/xradar/model.py) has a DataTree-based software model (based on CFRadial 2.1) that we should adapt for `Level2File` (and the polar Level 3 products).
Still need modified k-index.
Implementation challenges, and their performance implications, aside (this would mean new derivative operations that are only used for periodic cases), any idea how to determine this robustly using netCDF-CF metadata?...
This [on-going work in xarray](https://github.com/pydata/xarray/issues/7031) might at least possibly pave the way for simplicity of implementation, and just leave good ol' metadata as the challenge.
The idea in that change is that masked should behave the same as nan. I have personally never had a problem with nan/masked essentially not having units.
So the original intent was to allow users to mask a value without worrying about units (or, more importantly, allowing code that doesn't know about Pint to work without error...
I think I have a reasonably clean fix in #1608.