Update dimension ordering to match NetCDF-CF conventions

Open jdbcode opened this issue 2 months ago • 1 comments

Discussed in https://github.com/google/Xee/discussions/196

^{Originally posted by tylere December 13, 2024} Most Python libraries that plot raster images use a (y, x) or (lat, lon) ordering of dimensions. This is the convention used by Matplotlib's pyplot with images, and many libraries that utilize it (Xarray, rasterio).

(lat, lon) ordering is also used for higher dimensional data that follows the netCDF-CF conventions, which are a standard for climate forecast data. Specifically:

https://ferret.pmel.noaa.gov/Ferret/documentation/coards-netcdf-conventions Order of dimensions: If any or all of the dimensions of a variable have the interpretations of "date or time" (a.k.a. "T"), "height or depth"(a.k.a. "Z"), "latitude" (a.k.a. "Y"), or "longitude" (a.k.a. "X") then those dimensions should appear in the relative order T, then Z, then Y, then X in the CDL definition corresponding to the file.

By contrast Xee currently creates an Xarray dataset with dimensions in the order (time, lon, lat) in which the lon and lat coordinates are swapped. This requires the user to process (transpose) the data before using Matplotlib or similar plotting libraries.

This convention difference may have contributed to several the the Xee issues that have been raised (#45, #171, #174).

I think it may be worth changing Xee's coordinate ordering to (time, lat, lon) in order to match the convention used by netCDF-CF and Matplotlib. I realize this will be a breaking change for some users, but it will make the library easier to use for future users.

Thoughts?