Reproducibility of CESM with different start dates

[tcraig]

I notice in compset B, if just change run_startdate, the outputs of the very beginning few days are quite similar but not the same. Small differences are fou nd around South American coast and regions around Indonesia. So the model will read something depends on the run_startdate? If so, which file it will read? If not, why the outputs are different? The model will get some random values depend on this run_startdate?

 

[tcraig]

The run_startdate sets the initial date of the model integration. we are aware of some small differences that are introduced with different start dates. these are related to calculation of orbital parameters and other calendar issues. the differences should initially be on the order of round off different but those differences will grow fairly quickly and quickly lead to a different ensemble member of the same climate. we agree that a nice feature would be if different start years produced the same identical results, but that is not currently a feature we consider high priority.

 

[maythywinter@australia_edu]

Thank tcraig, your post is very helpfull !