Today @lukasbrunner and me ran into this problem. Opening an mfdataset and saving to_netcdfled to the following units for time and time_units:

time_bnds:units = "days since 1850-01-01" ;
time:units = "days since 2000-01-01" ;

Opening the dataset in xarray works fine, but ncdump and ncview use the units from time for time_bnds. Thus if I do

ncdump -tv time_bnds test.nc

the first date is '2150-01-01' (instead of '2000-01-01') - very confusing. (panoply shows the correct time_bnds).

Workaround

``` python import xarray as xr filenames = ['file1.nc', 'file2.nc']

ds = xr.open_mfdataset(fNs) ds.load()

make sure the encoding is really empty

assert not ds.time.encoding

assign encoding, such that they are equal

ds.time.encoding.update(ds.time_bnds.encoding)

save

ds.to_netcdf('~/F/test.nc')

```

Btw. thanks to @klindsay28 for the nice error report.