Add very minimal xarray plugin for engine="rasterio"

[alexamici]

Use xarray plugin infrastructure to add support for:

>>>importxarrayasxr>>>xr.open_dataset("myfile.jp2", engine="rasterio")
<xarray.Dataset>Dimensions: (x: 19087, y: 3932)
Coordinates:
*band (band) int641*y (y) float64 ...
*x (x) float64 ...
spatial_refint640Datavariables:
band_data (band, y, x) float32 ...

also add automatic engine selection for files with .tif and .geotif extensions.

>>>xr.open_dataset("myfile.tif")
<xarray.Dataset>Dimensions: (x: 19087, y: 3932)
Coordinates:
*band (band) int641*y (y) float64 ...
*x (x) float64 ...
spatial_refint640Datavariables:
band_data (band, y, x) float32 ...

• ClosesENH: Add xarray entrypoint #197
• Tests added
• Fully documented, including docs/history.rst for all changes and docs/rioxarray.rst for new API

[codecov]

Codecov Report

Merging #281 (511d99e) into master (74c2c0f) will decrease coverage by 1.08%.
The diff coverage is 0.00%.

❗ Current head 511d99e differs from pull request most recent head 18d7bd8. Consider uploading reports for the commit 18d7bd8 to get more accurate results

@@ Coverage Diff @@## master #281 +/- ##
==========================================
- Coverage 93.68% 92.59% -1.09% 
==========================================
Files 12 13 +1 Lines 1362 1378 +16 ==========================================
Hits 1276 1276 - Misses 86 102 +16 

Impacted Files	Coverage Δ
rioxarray/xarray_plugin.py	0.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 74c2c0f...18d7bd8. Read the comment docs.

[snowman2]

Use xarray plugin infrastructure to add support for:

>>>importxarrayasxr>>>xr.open_dataset("myfile.jp2", engine="gdal")
<xarray.Dataset>Dimensions: (x: 19087, y: 3932)
Coordinates:
*y (y) float64 ...
*x (x) float64 ...
spatial_refint640Datavariables:
band1 (y, x) float32 ...
Attributes:
scale_factor: 1.0add_offset: 0.0grid_mapping: spatial_ref

also add automatic engine selection for files with .tif and .geotif extensinos.

>>>xr.open_dataset("mygile.tif")
<xarray.Dataset>Dimensions: (x: 19087, y: 3932)
Coordinates:
*y (y) float64 ...
*x (x) float64 ...
spatial_refint640Datavariables:
band1 (y, x) float32 ...
Attributes:
scale_factor: 1.0add_offset: 0.0grid_mapping: spatial_ref

One thing I am noticing in these examples is that the attributes for the band data variables are on the Dataset. If you go this route, it would be important to make sure that those attributes are pushed to the DataArray.

[alexamici]

One thing I am noticing in these examples is that the attributes for the band data variables are on the Dataset. If you go this route, it would be important to make sure that those attributes are pushed to the DataArray.

You are right, I used .to_netcdf and started looking for fix-ups, but I missed the fact that array attributes where promoted to global attributes.

Anyway the whole PR is work in progress. If you like the idea we can discuss the direction to go here, or in the main issue.

[alexamici]

@snowman2 The current implementation is just to see where I'd be aiming following pydata/xarray#2844

>>>importxarrayasxr>>>xr.open_dataset("test/test_data/input/cog.tif", decode_coords="coordinates") # or decode_coords=True<xarray.Dataset>Dimensions: (band: 1, x: 500, y: 500)
Coordinates:
*band (band) int641*y (y) float642.715e+062.714e+06 ... 2.565e+062.565e+06*x (x) float641.635e+061.635e+06 ... 1.784e+061.784e+06Datavariables:
spatial_refint64 ...
band_data (band, y, x) int16 ...
>>>xr.open_dataset("test/test_data/input/cog.tif", decode_coords="all")
<xarray.Dataset>Dimensions: (band: 1, x: 500, y: 500)
Coordinates:
*band (band) int641*y (y) float642.715e+062.714e+06 ... 2.565e+062.565e+06*x (x) float641.635e+061.635e+06 ... 1.784e+061.784e+06spatial_refint64 ...
Datavariables:
band_data (band, y, x) int16 ...

I didn't manage to add the "grid_mapping": "spatial_ref" to band_data.encoding yet, but once we do it the resulting netCDF will be the same in both cases.

The best course of action would be to implement decode_coords="all" in open_rasterio though.

[alexamici]

@snowman2 I'll close the PR for now, waiting for a resolution of #282. Thanks for your attention!

[snowman2]

@alexamici are you aware of the target release date for xarray 0.18?

[snowman2]

@alexamici am I good to squash and merge? (rasterio 1.2.2 was just released and looks like it is breaking things, so I am not worried about test failures related to this PR).

[alexamici]

Release should be "soonish", but we have no target date to my knowledge.

Anyway the changes are totally backward compatible, since che xarray_plugin.py is not imported in older version of xarray and will only be imported by upcoming releases.

[alexamici]

You can merge now for me.

[snowman2]

Thanks @alexamici 👍