convert xarray Dataset to rioxarray Dataset?

[wendellwt]

How can I convert xarray's xarray.DataSet to rioxarray's xarray.DataSet?

What I have works because I write out the xarray.DataSet (using xarray) and then read it back in (using rioxarray). Both items say thay are an <xarray.Dataset>. Is there a better way than writing to a file & reading back in just to make it the rioxarray version of Dataset?

This works fine:

ds_xa = xarray.open_dataset( infile,
                decode_coords='all', engine='h5netcdf')

# need xarray here because it has .coarsen()
ds6_xa = ds_xa \
    .coarsen(x0=6, y0=6,boundary='trim') \
      .max()

ds6_xa.to_netcdf('ds6.nc')  # q: why is this write needed?

xr = rioxarray.open_rasterio('ds6.nc')  # q: can I do this directly to dx6_xa?

vr = xr.VIL

# need rioxarray here because it has .shapes()
shapes = features.shapes(vr)

According to the docs, .open_rasterio() does some very nice things when reading a file from disk.

Q: how do I get rioxarray to do those things on an existing (i.e. in-memory) xarray DataSet?

[snowman2]

Is this what you are referring to: https://corteva.github.io/rioxarray/stable/getting_started/crs_management.html ?

[wendellwt]

Thank you, but no, it's not the crs part. It's not as bad as I thought, and it may be in the "it's a feature, not a bug" category.

If I write out a Dataset that has this:

Dimensions:        (x: 6688, y: 3175)
Coordinates:
    grid_mapping0  int32 ...
  * x              (x) float64 -134.4 -134.3 -134.3 ... -61.66 -61.65 -61.64
  * y              (y) float64 53.87 53.86 53.85 53.84 ... 19.38 19.37 19.36
Data variables:
    start_time     datetime64[ns] ...
    stop_time      datetime64[ns] ...
    xml_metadata   |S47067 ...
    VIL            (y, x) float32 nan nan nan nan nan ... nan nan nan nan nan
    VIL_FLAGS      (y, x) float32 nan nan nan nan nan ... nan nan nan nan nan
    PRECIP_PHASE   (y, x) float32 nan nan nan nan nan ... nan nan nan nan nan

and read it back in, I get a Dataset that has this:

Dimensions:        (band: 1, x: 6688, y: 3175)
Coordinates:
  * band           (band) int64 1
  * x              (x) float64 -134.4 -134.3 -134.3 ... -61.66 -61.65 -61.64
  * y              (y) float64 53.87 53.86 53.85 53.84 ... 19.38 19.37 19.36
    grid_mapping0  int64 0
Data variables:
    VIL            (band, y, x) int16 ...
    VIL_FLAGS      (band, y, x) int8 ...
    PRECIP_PHASE   (band, y, x) int8 ...

It appears that:

• Band 1 was created
• data elements were changed from float32 to int8 and int16
• start_time, stop_time, and xml_metadata disappeared (I don't need these, btw)
• nan was changed to 0 (not sure about this one)

So, to replicate the write/read operation, it looks like .VIL.fillna(0) on the original Dataset would suffice. Does this seem right?

[snowman2]

data elements were changed from float32 to int8 and int16

This is because mask_and_scale=True is default with open_dataset. You can disable it if you don't want the data masked and scaled.

nan was changed to 0 (not sure about this one)

This is because mask_and_scale=True is default with open_dataset. You can disable it if you don't want the data masked and scaled.

[wendellwt]

Thank you. That's what I was looking for. I just need to do those operations to the in-memory Dataset.