rio.reproject fails for timedelta64['ns'] with NaT

[anonymousForPeer]

Hi,

I try to reproject a few datasets and some of them have dtype timedelta64['ns'] with nan values 'NaT'.

import xarray as xr
import rioxarray
from rasterio.enums import Resampling

repro = dataarray.rio.write_crs('EPSG:4326')
resample = repro.rio.reproject('EPSG:3035', resolution=(1000,-1000), resampling=Resampling.bilinear)


ValueError: Could not convert object to NumPy timedelta

If I try to set the nodata value it dos not work either

repro.rio.write_nodata('NaT', inplace=True)


ValueError: could not convert string to float: 'NaT'

According to the rioxarray description of reproject it uses the internal nodata value - which works fine for datat arrays that have no timedelta64['ns'] dtype.

Does anyone know how to overcome this?

Additionally the rio.reproject() seems to change my dataarray.time dimension dtype so that ncview can't read it anymore.

[snowman2]

timedelta is not supported. I recommend converting to float64 before reprojecting.

[anonymousForPeer]

OK, thank you for the clarification.