Why different software and commands are producing different results in calculating daily mean from ERA5 hourly data?

Question

Basically, this is a repost or follow-up from How to calculate daily average from ERA5 hourly netCDF data? I have four atmospheric variables, viz. dew point temperature, longwave radiation, shortwave radiation, and wind speed (calculated from u10 and v10 wind components). All are on an hourly scale and I need to convert them into the daily scale (note that here I've given the example of dew point temperature only). In order to do this first I applied the following command

cdo daymean -shifttime,-1hour d2m.nc d2m_wb.nc and got the following result.

cdo sinfo d2m_wb.nc
   File format : NetCDF2
    -1 : Institut Source   T Steptype Levels Num    Points Num Dtype : Parameter ID
     1 : unknown  unknown  v instant       1   1       475   1  F64  : -1
   Grid coordinates :
     1 : lonlat                   : points=475 (19x25)
                              lon : 85.5 to 90 by 0.25 degrees_east
                              lat : 21.5 to 27.5 by 0.25 degrees_north
   Vertical coordinates :
     1 : surface                  : levels=1
   Time coordinate :  25904 steps
     RefTime =  1900-01-01 00:00:00  Units = hours  Calendar = gregorian  Bounds = true
  YYYY-MM-DD hh:mm:ss  YYYY-MM-DD hh:mm:ss  YYYY-MM-DD hh:mm:ss  YYYY-MM-DD hh:mm:ss
  1949-12-31 23:00:00  1950-01-01 11:00:00  1950-01-02 11:00:00  1950-01-03 11:00:00
  1950-01-04 11:00:00  1950-01-05 11:00:00  1950-01-06 11:00:00  1950-01-07 11:00:00
  1950-01-08 11:00:00  1950-01-09 11:00:00  1950-01-10 11:00:00  1950-01-11 11:00:00
  1950-01-12 11:00:00  1950-01-13 11:00:00  1950-01-14 11:00:00  1950-01-15 11:00:00
  1950-01-16 11:00:00  1950-01-17 11:00:00  1950-01-18 11:00:00  1950-01-19 11:00:00
  1950-01-20 11:00:00  1950-01-21 11:00:00  1950-01-22 11:00:00  1950-01-23 11:00:00
  1950-01-24 11:00:00  1950-01-25 11:00:00  1950-01-26 11:00:00  1950-01-27 11:00:00
  1950-01-28 11:00:00  1950-01-29 11:00:00  1950-01-30 11:00:00  1950-01-31 11:00:00
  ..................................................................................
   ................................................................................
   ................................................................................
   .................
  2020-10-31 11:00:00  2020-11-01 11:00:00  2020-11-02 11:00:00  2020-11-03 11:00:00
  2020-11-04 11:00:00  2020-11-05 11:00:00  2020-11-06 11:00:00  2020-11-07 11:00:00
  2020-11-08 11:00:00  2020-11-09 11:00:00  2020-11-10 11:00:00  2020-11-11 11:00:00
  2020-11-12 11:00:00  2020-11-13 11:00:00  2020-11-14 11:00:00  2020-11-15 11:00:00
  2020-11-16 11:00:00  2020-11-17 11:00:00  2020-11-18 11:00:00  2020-11-19 11:00:00
  2020-11-20 11:00:00  2020-11-21 11:00:00  2020-11-22 11:00:00  2020-11-23 11:00:00
  2020-11-24 11:00:00  2020-11-25 11:00:00  2020-11-26 11:00:00  2020-11-27 11:00:00
  2020-11-28 11:00:00  2020-11-29 11:00:00  2020-11-30 11:00:00  2020-12-31 23:00:00
cdo    sinfo: Processed 1 variable over 25904 timesteps [6.03s 37MB

The result (daily mean) seemed very dubious to me (since in timestamp it has 11:00:00). Therefore, I asked the same in the CDO forum and they suggested shifting back 1 min instead of 1 hour.

cdo daymean -shifttime,-1min d2m.nc d2m_wb.nc

Here, the timestamp (of d2m_wb.nc) shows 12:00:00 instead of 11:00:00. I was not sure whether the result is correct or not. So, I tried to cross-check this using the following script in python (3.8.5).

import xarray as xr
hourly_d2m = xr.open_dataset(r'E:\Data\Rda\d2m\d2m.nc')
d2m = hourly_d2m['d2m']
d2m= hourly_d2m.shift(time=-1).dropna(dim='time',how='all') 
sds = d2m.resample(time='D').mean(dim='time')
sds.to_netcdf('E:\Data\Rda\d2m\daily\d2m_wb.nc')

There I got the 00:00:00 timestamp after calculating the daily mean of d2m_wb.nc. Now, the question is different software and commands are producing different results of a common objective (i.e. daily mean). I'm completely confused and looking for a suitable solution or clarification. Thank you.

Answer 1

Consider re-reading the answer I gave to your previous question. As I said then, it is somewhat arbitary which timestep software uses in the aggregate output. This is why CDO gives users the option to change it. For example, you can set it to the first or final timestep being averaged over if you want. xarray uses a slightly different assumption about which timestep to choose. Neither is right or wrong in terms of the time step to choose. They both achieve the same objective, which is to give you a time average with the correct day.