The _FillValue encoding seems to be preserved with your desired value in my example: ``` $ ncdump -h result.nc netcdf result { dimensions: time = 366 ; y = 200 ; x = 200 ; variables: int64 time(time) ; time:units = "days since 2000-01-01 00:00:00" ; time:calendar = "proleptic_gregorian" ; short Rg(time, y, x) ; Rg:_FillValue = -1s ; Rg:long_name = "HWSD sub sum content" ; Rg:units = "percent wt" ; Rg:valid_range = 97., 103. ; double latitude(y, x) ; latitude:_FillValue = -99999. ; double longitude(y, x) ; longitude:_FillValue = -99999. ;

// global attributes: :Conventions = "CF-1.0" ; :content = "HARMONIZED WORLD SOIL DATABASE; first it was aggregated to one global file; then the missing areas were filled with interpolated data; then separate tiles were extracted" ; :scaling_factor = "20" ; ``` Is the time encoding important for your land surface model? That does change in my example (see the units attribute); you might need some special logic to handle that.

Thanks @tommylees112 -- note that the reindex and ffill methods do not operate in-place, so you'll need to assign the results of them to new variables. You can also do them all in one step: ``` In [1]: import xarray as xr; import pandas as pd; import numpy as np

In [2]: ds = xr.open_dataset('Rg_dummy.nc')

In [3]: times = pd.date_range("2000-01-01", "2000-12-31", name="time")

In [4]: ds['time'] = np.array([times[0]])

In [5]: ds2 = ds.reindex(time=times, method='ffill')

In [6]: ds2.to_netcdf('result.nc') ``` Regarding the issue saving to files -- I can reproduce that issue with older xarray versions. It is related to

2512 and was fixed in #2513 (i.e. it works with the master version of xarray). The good news is this bug only applies to saving ds in my example, not ds2, so you should be able to do this with your current setup just fine.

@tommylees112 I had a look at your dataset and noticed that the time coordinate was not encoded in a way that allows xarray to automatically decode the time values into datetimes. Specifically, the units attribute is not in the format of some unit of time (e.g. 'seconds') since a given reference date. ``` $ ncdump -h Rg_dummy.nc netcdf Rg_dummy { dimensions: time = 1 ; y = 200 ; x = 200 ; variables: double time(time) ; time:_FillValue = NaN ; time:standard_name = "time" ; time:units = "day as %Y%m%d.%f" ; time:calendar = "proleptic_gregorian" ; short Rg(time, y, x) ; Rg:_FillValue = -1s ; Rg:long_name = "HWSD sub sum content" ; Rg:units = "percent wt" ; Rg:valid_range = 97., 103. ; double latitude(y, x) ; latitude:_FillValue = -99999. ; double longitude(y, x) ; longitude:_FillValue = -99999. ;

// global attributes: :Conventions = "CF-1.0" ; :content = "HARMONIZED WORLD SOIL DATABASE; first it was aggregated to one global file; then the missing areas were filled with interpolated data; then separate tiles were extracted" ; :scaling_factor = "20" ; } `` This is whyds.time` is an array of floats when you load the file in. We could discuss how to address that, but depending on your broader needs here that might not be a major concern.

Regarding assigning a new value to the time coordinate, the following should work: python ds['time'] = np.array([times[0]])