Sorry, that I misuse this platform for my silly question:
I have a 
dataset
Dimensions: (time: 1, band: 4, y: 8192, x: 4096)
Coordinates:
* time datetime64[ns] 8B 2023-06-15
* band <U5 80B 'Red' 'Green' 'Blue' 'Alpha'
* y float32 33kB 7.064e+06 7.064e+06
* x float32 16kB 1.194e+06 1.194e+06
Data variables:
band_data (time, band, y, x) uint8 134MB dask.array<chunksize=(1, 4, 1024, 1024), meta=np.ndarray>
spatial_ref int64 8B ...
I added to "band_data" a '_CRS' attribute as dict with 'url', 'wkt' & 'projjson' infos (EPSG 3857), saved the whole thing as 
file and hoped 
can open it in 
However seems to have trouble finding the coordinate reference and assumes the origin is 0,0 (Extent 0.0, -8192.0 : 4096.0, 0.0).
Now, my question: where do I store the "_CRS" dict correctly? Do I also need to add a GeoTransform infos? If yes, where?
@Mixed_Pixels I think the easiest solution may be to use the rioxarray methods:
https://corteva.github.io/rioxarray/html/rioxarray.html#rioxarray.rioxarray.XRasterBase.set_spatial_dims (I think it should take care of the ordering)
https://corteva.github.io/rioxarray/html/rioxarray.html#rioxarray.rioxarray.XRasterBase.set_crs (should work if you pass "EPSG:3857")
https://corteva.github.io/rioxarray/html/rioxarray.html#rioxarray.rioxarray.XRasterBase.write_transform (tell gdal/qgis where the raster should be placed)
They will manage the specifics for you, and you could use the output object as example if you wish to do these things manually
Philipp Gärtner
Ignacio Borlaf-Mena