Create a rioxarray-enabled DataArray without reading tiff

[jessjaco]

Is there a way to create a rioxarray-enabled DataArray "from scratch", i.e., without having a file that can be opened via open_rasterio? Should I create a DataArray and then set somehow the crs and transform manually? I am currently saving a tiff file just to reopen it, like this (the height, width, crs, and transform params are known)

with rasterio.open(
    file,
    "w",
    driver="GTiff",
    height=nrow,
    width=ncol,
    count=1,
    dtype="uint8",
    crs=epsg,
    transform=transform,
) as dst:
    dst.write(np.ones((nrow, ncol)), 1)

[snowman2]

    xds = xarray.DataArray(
        np.ones((nrow, ncol), dtype="uint8"),
        dims=("y", "x"),
    )
    xds.rio.write_transform(transform, inplace=True)
    xds.rio.write_crs("EPSG:4326", inplace=True)
    xds.rio.write_coordinate_system(inplace=True)

[jessjaco]

Thanks, that put me on the right track but it doesn't have the coords set. This is what the DataArray looks like when I load it with open_rasterio:

<xarray.DataArray (band: 1, y: 6736, x: 4210)>
[28358560 values with dtype=uint8]
Coordinates:
  * band         (band) int64 1
  * x            (x) float64 2.601e+05 2.601e+05 ... 3.863e+05 3.863e+05
  * y            (y) float64 4.907e+06 4.907e+06 ... 4.705e+06 4.705e+06
    spatial_ref  int64 0
Attributes:
    scale_factor:  1.0
    add_offset:    0.0

And this is what the one you suggested is

<xarray.DataArray (y: 6735, x: 4210)>
array([[1, 1, 1, ..., 1, 1, 1],
       [1, 1, 1, ..., 1, 1, 1],
       [1, 1, 1, ..., 1, 1, 1],
       ...,
       [1, 1, 1, ..., 1, 1, 1],
       [1, 1, 1, ..., 1, 1, 1],
       [1, 1, 1, ..., 1, 1, 1]], dtype=uint8)
Coordinates:
    spatial_ref  int64 0
Dimensions without coordinates: y, x

But I think this will work:

# get the x and y  coords (may be a simpler/better way to do this)
x, _ = rasterio.transform.xy(transform, np.arange(ncol), np.arange(ncol))
_, y = rasterio.transform.xy(transform, np.arange(nrow), np.arange(nrow))

da = xarray.DataArray(
        np.ones((1,nrow, ncol), dtype="uint8"),
         dims=("band", "y", "x"),
         coords=dict(x=x, y=y, band=[1])
)
da.rio.write_transform(transform, inplace=True)
da.rio.write_crs(epsg, inplace=True)

Which produces:

<xarray.DataArray (band: 1, y: 6735, x: 4210)>
array([[[1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        ...,
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1]]], dtype=uint8)
Coordinates:
  * x            (x) float64 2.601e+05 2.601e+05 ... 3.863e+05 3.863e+05
  * y            (y) float64 4.907e+06 4.907e+06 ... 4.705e+06 4.705e+06
  * band         (band) int64 1
    spatial_ref  int64 0

I've just got to figure out if that once cell difference in the y-dimension means anything (I think it's a rasterio quirk which is noted in the docs)

[snowman2]

rioxarray.rioxarray.affine_to_coords should help.

[jessjaco]

Yes that works well thanks