computing climatologies

[taylor13]

I’m in the process of documenting how the date range is recorded in filenames for CMIP7 and want to make sure I understand what is requested for climatological variables (with time coordinates identified by the labels time2, time3, and time4).

As I understand it:

• time2 represents a climatological annual cycle recording 12 climatological monthly mean values based on multiple simulated years; thus, the length of the time dimension for the data reported is 12.
• time3 represents the monthly climatological diurnal cycle (sampled hourly), recording 24 values for each month. thus, the length of the time dimension for the data reported is 24*number of months of requested data.
• Time4 represents the monthly mean of daily maximum values; thus, the length of the time dimension for the data reported equals the number of months of requested data.

I hope you can check that I’ve understood this correctly so I can then document it.

Also, I'm curious for time3, does the data request ask that these be reported for the entire length of the run 24*N values (N=number of years), or is it for a limited portion of the run (say, 1 year)?

thanks.

[JamesAnstey]

I hope you can check that I’ve understood this correctly so I can then document it.

@taylor13 AFAIK that's all correct.

Also, I'm curious for time3, does the data request ask that these be reported for the entire length of the run 12*N values (N=number of years), or is it for a limited portion of the run (say, 1 year)?

I assumed it was the entire length of the run, but since you asked I took a closer look. There are 5 time3 variables, all requested only by the CFMIP Opportunity "Clouds, circulation and climate sensitivity: extension for process-level studies" for the AFT amip and amip-p4k experiments (plus 3 non-FT experiments). It specifies no time subsets, and I gather it will be compared to satellites, so the length of the amip run (which starts in 1979) sounds reasonable. (As fairly bespoke variables, I wonder if CFMIP coordinates internally who will provide these?)

Searching metagrid just now I found only one CMIP6 institute providing E1hrClimMon variables, and it seems that's what they did:

netcdf rlutcs_E1hrClimMon_GFDL-CM4_amip_r1i1p1f1_gr1_197901010000-198501010000-clim {
dimensions:
        bnds = 2 ;
        lat = 180 ;
        lon = 288 ;
        time = UNLIMITED ; // (72 currently)
        time_of_day_24 = 24 ;
        time_of_day_edges_24 = 25 ;
variables:
        float rlutcs(time, time_of_day_24, lat, lon) ;

(together files in the dataset cover the 1979-2015 period)

[taylor13]

Thanks @JamesAnstey for all this good information. I also have now had a further look at the CF conventions. It turns out if you wanted a mean diurnal cycle for each month averaged over several years, the cell_methods would have to include "time: mean over years" (see example 7.13).
There is only one correct interpretation of the cell_methods attached to variables with time dimension "time3", which is to report the diurnal cycle for individual months. So, 24*N samples.

Somewhat independent of the above, I think the correct way to report the data is to define a single time coordinate, not two time coordinates as in the GFDL output above. I'll ask the CF community to confirm (on their discussion board).